Slide 1

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 2

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 3

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 4

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 5

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 6

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 7

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 8

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 9

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 10

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 11

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 12

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 13

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 14

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 15

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 16

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 17

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 18

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 19

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 20

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 21

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 22

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 23

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 24

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 25

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 26

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 27

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 28

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 29

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 30

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 31

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 32

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 33

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 34

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 35

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 36

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 37

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 38

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 39

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 40

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 41

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 42

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 43

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 44

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 45

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.


Slide 46

“Glucometrics:”
Assessing Quality in
Inpatient Glycemic Management

“Glucometrics”:
Assessing Quality in Inpatient Glycemic
Management
Outline
1. What is ‘quality’ inpatient glucose
management?
2. How should we measure it?
3. Future directions

The 4 Spheres of a Quality Inpatient
Glucose Management Program

Recognition as
a Hospital
Priority

Administrative
support

Clinician
champion

Hyperglycemia
Committee

A Hospital Priority
Why?
• Enhance quality & patient safety
• Competitive advantage
• Cost savings
• The Joint Commission

Institution-Wide Educational Efforts

Physicians

Nursing staff

Pharmacists

Dietitians

Medical
Assistants

Patient Care
• Identification (& coding) of patients

•

Policies & procedures
– Point-of-Care BG Testing
– Institutional BG Targets (ICU, Wards)

– Hypoglycemia protocol
– ICU IV insulin protocols
– Standardized SQ insulin order sets

– Patient education tools
• “Inpatient diabetes management team”
•

Transition to outpatient care (access)

Example of a Glucometrics Report

Graphic Display of Glucometrics Data
MICU/CCU Avg BG Level By Year
160
150
140

Diabetic
All Patients
Non-Diabetic

159.9

158

152.3

150.3

144.9

141.6

141.8

149.6
141.2

133.6
130

133.9

130.8

120
2004

2005
Fiscal Year

www.aace.com/resources/igcrc

2006

2007-Q1

Graphic Display of Glucometrics Data
Diabetic Patients

Non-Diabetic Patients

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Percent

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

All Patients
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%

Percent

300+
200-300
151-200
111-150
81-110
51-80
0-50
% 111-150
% 81-110

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

www.aace.com/resources/igcrc

Patient 2
Patient 1
Patient 3
Patient 4

Blood Glucose (mg/dl)

500

Patient 5
400
300

200
100
50

0

SUN

MON

TUE

WED

THU

FRI

SAT

Measuring Inpatient Glycemic Control:
Special Issues
• Sample site (fingersticks, lab plasma glucose)
• Multiple measures during hypoglycemic or
hyperglycemic “events”
• Varying time intervals of measurement

• Timing in relationship to meals
• Effects of IV fluids (dextrose)
• How to collect glucose measurements?

• How to analyze them?
• How to present data to clinicians/adminstrators?

The 4 Spheres of a Quality Inpatient
Glucose Management Program
Metrics
• Systematic review of hospital BG data

• Analytical models
– What’s the “HbA1c” for glucose control
during a hospitalization?

Metrics Traditionally Used in
the Inpatient Glucose Literature
• Raw blood glucose (BG) average

• % of BGs within a pre-specified range
(80-110, 100-150, <180, <200 mg/dl)
• % of patients with a certain % of BGs within a pre-specified range
•

Hypoglycemia rates (<40, <50, <60, <70 mg/dl)
– % of BGs
– % of patients

•

Hyperglycemic excursions (>180, >200, >300 mg/dl)

– % of BGs
– % of patients
Malmberg 1997; Queale 1999; Capes 2000, 2001; Bhattacharrya 2002; van den Berghe 2001, 2006;
Funary 2003; Krinsley 2003; Goldberg 2004, 2005; Baldwin 2005; McCallister 2005; Kosiborod 2006

Glucometrics Project: Objectives
• Define inpatient glucose quality metrics
• Define ‘units of analysis’
• Compare metric results for the different
analytical units

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Sample
• Yale-New Haven Hospital BG data

• BG data downloaded into relational database for
analysis
– BG values
– Date / Time
– Patient ID
– Hospital ward

• Sample: One general medical ward’s March 2004
BG results (n=1,552)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Metrics Tested
• Mean / Median BG
• % BG in “favorable” range (80 - 139 mg/dl)
• % Hyperglycemia (>300 mg/dl)
• % Hypoglycemia (<60 mg/dl)

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Methods: Units of Analysis
“Ward”

n = 1,552

“Patient Stay”

n = 118

[13.2 BGs / stay]

“Patient Day”

n = 467
[3.3 BGs / day]

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400
300

“ward” model

200

X
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

“patient stay model”

300

200
100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Blood Glucose (mg/dl)

500

Patient 2
Patient 1
Patient 3
Patient 4
Patient 5

400

300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Patient 2
Patient 1

Blood Glucose (mg/dl)

500

Patient 3
Patient 4
Patient 5

400

“patient day model”
300
200

100
50
0

SUN

MON

TUE

WED

THU

FRI

SAT

Mean / Median BG
200
186

182

179

180
160

162

160

MEAN
MEDIAN

158

140
120
100
80

Mean/median
of all BGs on
that ward

Mean/median
Mean/median
of each pt’s mean of each pt’s mean
BG during hosp’n BG during each day

60
40
20
0
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

% BGs Within Range (80-139 mg/dl)
50%
45%
40%
35%

33.9%

35.8%
32.2%

30%
25%
20%
15%
10%

% of the mean BGs % of the mean BGs
% of all BGs on
for each patient’s for each patient’s
ward that are
hosp. day that are
hosp’n that are
within target range
within target range within target range

5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hyperglycemic Events (>300 mg/dl)
50%
45%
39.0%

40%
35%
30%
25%
20%
15%

% of all BGs on
ward that are
severely hyper

% of the patient
hospitalizations
where any BG was
severely hyper

C

% of the patient
hosp. days where
any BG was
severely hyper
21.8%

12.8%

10%
5%
0%
WARD

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT STAY

PT DAY

% Hypoglycemic Events (<60 mg/dl)
25%
20%
15%
10%

% of all BGs on
ward that are
hypo

% of the patient
hospitalizations
where any BG was
hypo
7.6%

% of the patient
hosp. days where
any BG was
hypo
4.5%

5%
1.5%

0%
WARD

PT STAY

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

PT DAY

Summary of Results
• Metrics for mean BG, median BG & the % “in target
range” are similar for all three analytical models.
• There were substantial differences between the
models for % hyperglycemia and % hypoglycemia.

– ‘WARD’ model has lowest %
– ‘PATIENT STAY’ model has highest %
– ‘PATIENT DAY’ model is intermediate

Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Other Findings
• Addition of venous plasma lab glucose
measurements to fingerstick data
– Slight reduction in mean glucose values, but not
clinically meaningful
•

Deletion of 1st hospital day of blood glucose
– Slight reduction in mean glucose values, but not
clinically meaningful

• Applying glucometrics to the ICU (‘gold standard’ with
IV insulin infusion) - the realistic maximum % of
patient days within target range is probably ~ 80%

Conclusions
• Glucometrics are useful intermediate outcomes
measures of inpatient hyperglycemia management.
• Perception of performance & quality may depend
upon the unit of analysis
• All 3 Units of Analysis provide useful information

– ‘WARD’ model is the simplest; may be most
useful in operational analyses.
– ‘PATIENT STAY’ model perhaps most useful to
consumers (& risk management).
– ‘PATIENT DAY’ model may be the most
actionable by providers.
Goldberg PA et al. Diabetes Technology & Therapeutics 2006

Inpatient Diabetes Management Team:
Impact on Glucometrics (Before vs. After)
Pt-Day

Metric

Before
Consult

After
Consult

Absolute ∆

Relative ∆

Mean BG

225.1 mg/dl

182.7 mg/dl

-42.4 mg/dl

-18.8%

% in 70-149
Target Range

16.7%

35.3%

+18.6%

+111.4%

% Hyper (>299)

46.7%

22.8%

-23.9%

-51.2%

% Hypo (<70)

12.9%

13.0%

+0.1%

+0.8%

Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Inpatient Diabetes Management Team:
Impact on Glucometrics (vs. Matched Controls)
IDMT

Non-IDMT

-49.5 mg/dl

-16.4 mg/dl

P < .01

P = NS

P = .03*

P = NS

Reduction in Hyperglycemia (>299)

P < .01

P = NS

Increase in Hypoglycemia (<70)

P = NS

P = NS

Mean BG Reduction

Increased cases in

70-149 target range

* McNemar’s Test
Bozzo J et al. 67th Scientific Sessions, American Diabetes Association, Chicago, IL, 2007 (576-P)

Other Proposed Metrics

“Time Average Glucose (TAG)”
300

Glucose (mg/dL) .

250
200
150
100
50
0
0

24

48
Time (hrs)

72

Persistent Hyperglycemia: An Independent
Predictor of AMI Outcomes
35
30

MACE (%)

25
Quartile 1
Quartile 2
Quartile 3
Quartile 4

20
15
10
5
0
Admission glucose

Persistent hyperglycemia

van der Horst et al. Cardiovascular Diabetology 2007.
30-day major adverse cardiac events (MACE) according to quartiles of admission glucose and persistent hyperglycemia
in MI patients. P value for trend in admission glucose is 0.023 and for persistent hyperglycemia is <0.0001.

‘Hyperglycemic Index’ As a Tool to Assess
Glucose Control: A Retrospective Study

• 10 yr analysis in a 12-bed surgical ICU
• 1779 patients (LOS >4 days)
• 65,528 glucose values

Mathijs Vogelzang, Iwan CC van der Host, Maarten WN Nijsten. Critical Care 2004; 8(3):R122-127

Glucose (mmol/L)

The “Hyperglycemic Index” (HGI)

HGI: 0.73

Time (hours)

•

Calculation of the hyperglycemic index (HGI). All measured glucose values (black dots) and their
corresponding sampling times are taken into account. The average over time is calculated for the
area (shaded) under the glucose curve for hyperglycemic values only. The normal glucose range
is indicated by the hatched area, with 6.0 mmol/L (dotted line) as the cutoff. HGI is the shaded
area divided by the total length of stay. In this case, HGI is 0.73 mmol/L, as indicated by the
dashed line. Note that normal or hypoglycemic measurements do not affect HGI, and thus, they
do not falsely lower this index.

Vogelzang et al. Critical Care 2004

Receiver Operator Characteristic (ROC) Curves
for Different Glucose Measures
ROC
HGI

0.64

Mean BG

0.62

Mean AM BG

0.61

Vogelzang et al. Critical Care 2004

http://glucometrics.med.yale.edu

Glucometrics
•

Institution: Second Hospital

Ward: Medical ICU Type: Adult Internal Medicine ICU 1/3/2005 – 2/1/2005

•

A patient’s glycemic control can be analyzed at three time resolutions. Individual glucoses measure control at the
shortest interval, the time between samples. Mean glucose measures control for the longest interval, the entire
hospital stay. A day’s mean glucose measures control for an intermediate interval, one day. Only this method
has a fixed interval which allows better comparison of one patient to another. The figures below show frequency
distributions; the dark bar on the x-axis between 70 and 149 shows a target or goal glucose range. Percentiles of
the data are shown by the lines and dot over the histogram: 5–––25 •50 75–––95
All Glucose Samples

Patient-stay Means

Glucose (mg/dL)

Patient-day Means

Glucose (mg/dL)

Glucose (mg/dL)

Patient – samples

Patient – stays

Patient – days

Number

1405

64

374

Median*

146

147

146

Mean*

160

155

159

93–275

111–229

104–258

182

118

154

731 52.0

35 54.7

199 53.2

n percent with a glucose < 70

19 1.4

7 10.9

14 3.7

n percent with a glucose >299

47 3.3

9 14.1

26 7.0

5th – 95th percentile*
Spread
n percent with 70  glucose  149 mg/dL*

* For patient-stays and patient-days, summary statistics are computed on the mean glucose of these intervals. For
example, in calculating the medians, the median {of individual patient-samples}, the median {of patient-stay means},
and the median {of patient-day means} are taken. Mean, median, percentile, and spread are stated as mg/dL.
Courtesy of Dr. Prem Thomas, Yale Center for Medical Informatics. http://glucometrics.med.yale.edu

“Glucometrics”: Assessing Quality in
Inpatient Glycemic Management
SUMMARY
1. “Quality” in inpatient glucose management needs to be
better defined.
2. Achieving it requires efforts in 4 spheres: prioritization;
education; patient care; and metrics.
3. Measures of inpatient glucose management are
dependent on the analytical methods employed.
4. It is important for the diabetes community, hospitals,
clinical investigators & the QI experts work together to
better define & validate standardized “glucometrics”
which are meaningful, fair, and actionable.