Exploration of Excess Mortality by ICD9 DX Dx with more

Download Report

Transcript Exploration of Excess Mortality by ICD9 DX Dx with more

Exploring QUEST Mortality
Understanding the Baseline Data and Using Clinical Advisor and
Quality Manger to Create Actionable Hypotheses for Intervention
Eugene Kroch, Ph.D., Vice President and Chief Scientist
Richard A Bankowitz, MD MBA FACP, Vice President and Medical Director
1
Topics
▪ Baseline reports
▫ Model comparison
▫ Variation across hospitals
▫ Size effects
▪ Trending
▫ Two-year time frame
▫ General trends
▫ Trend ranges and volatility
▪ Palliative care patterns
▪ Exploring Potential Drivers of Mortality using
Clinical Advisor or Quality Manager
2
QUEST Mortality Measure
Observed
Actual
=
Index
O/E Ratio
Expected
Predicted
Index > 1 :
Actual mortality is greater than predicted
(opportunity)
Index < 1 :
Actual mortality is less than predicted
3
Measuring Risk (alternatives)
APR-DRG
Severity Classification
Base APR-DRG
Age
Gender
Discharge status
Diagnoses
Procedures
Birth weight
4 Levels of:
•Severity (resource demand)
•Risk of mortality
CareScience
Risk Prediction
Clinical
Principal Diagnosis (terminal digit)
Severity Weighted Comorbidities
Procedures
Urgency of Admission
Neonatal Birth Weight
Demographic
Age, Gender
Household Income
Facility Type
Race
Discharge Disposition
Referral and Selection
Admission Source (e.g Transfer in)
Payor Class
Travel Distance
Facility Type
4
Summary of Model Differences
Aspect
CareScience
APR-DRG
Risk Scaling
Continuous
4 Buckets in APDRGs
Specification (structure)
Stratified Regressions
Decision-Tree Logic
Variables
Clinical/Demog/Selection
Subset of CSI factors
Secondary Diagnoses
CACR (complication adj)
Selected SDx
Population Stratification
Diagnosis
DRG
Calibration Data
All Payor State & Client
Perspective (Client)
Statistical Inference
Regression-based errors
Cell means
5
Illustration of Precision
Under APR-DRGs patient 2 is lumped together with Patient 1, even
though under continuous severity scaling patient 2 is more like patient 3.
APR-DRG severity buckets
2
1
Patient 1
Patient 2
3
4
Patient 3
CareScience continuum
Continuous Severity Scale
Patient 1
Patient 2
Patient 3
6
Baseline O/E Variation across Hospitals
▪ Baseline: 161 hospitals – 2006q3 to 2007q2
▪ CareScience and APR-DRGs are very close (next slide)
CareSci APR-DRG
Mean
0.99
0.96
Median
0.95
0.90
Top quartile
0.82
0.77
▪ Cross hospital range = 0.50 to 2.00
▫ All 12 hospitals with O/E ratios > 1.35 are relatively small
(smallest third in size)
▫ Not so for 16 hospitals with O/E ratios < 0.65
7
Baseline Comparison of O/E Ratios
APR-DRG vs. CareScience
2.4
2.2
Correlation = 94%
2.0
APR-DRG
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.4
0.6
0.8
1.0
1.2
1.4
CareScience
1.6
1.8
2.0
2.2
2.4
8
Baseline Distribution of O/E Ratios
Distribution of O/E Ratios
35
30
Frequency
25
20
15
10
Smaller hospitals
5
0
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
O/E ratio
1.3
1.4
1.5
1.6
1.7
1.8
1.9
9
O/E Trends
▪ 8 quarters: 2005q3 to 2007q2
▪ Overall pattern
▫ O/E ratio falls by about 12% over the 8 quarters
▪ Trend range
▫ For 4-quarter moving averages
▫ 40% decline to 20% increase
▪ Volatility
▫ Time volatility is inversely related to size (correlation is
about -50%)
▫ Quarter-on-quarter O/E changes greater than 0.4 are
concentrated in smaller hospitals (<1000 disch. per qtr.).
10
Overall Trend over 8 Quarters
O/E Ratio Trend
1.2
1.1
O/E Ratio
1.0
Moving Avg
0.9
Mean O/E ratio has
fallen about 12%
0.8
0.7
0.6
2005 Q3
2005 Q4
2006 Q1
2006 Q2
2006 Q3
Year and Quarter
2006 Q4
2007 Q1
2007 Q2
11
Strong Mortality Declines
Aurora Medical Center - Kenosha
Strong Mortality Declines
2.0
NW Alabama Health Care Authority (Helen Keller)
(larger hospitals)
Baptist Memorial Hospital-North Mississippi
St. Mary's Medical Center
1.8
Baltimore Washington Medical Center
North Mississippi Medical Center
Baptist Memorial Hospital-Memphis
O/E Mortality Ratio
1.6
1.4
1.2
1.0
0.8
Note Bapt Mem
0.6
0.4
05q3-06q1
05q3-06q2
05q4-06q3
06q1-06q4
06q2-07q1
Overlapping quarters
06q3-07q2
06q4-07q2
12
Trend Break Example
Trend Break Example
Baptist Memorial - Memphis
1.6
Baptist Mem Hosp
4-quarter MA
Mortality O/E Ratio
1.4
1.2
1.0
0.8
0.6
0.4
2005 Q3
2005 Q4
2006 Q1
2006 Q2
2006 Q3
Year and Quarter
2006 Q4
2007 Q1
2007 Q2
13
Distribution of Palliative Care Coding
Hospital Palliative Care Rate
Coding Variation
90
80
Half of hospitals have
less than 2 per thousand
70
Frequency
60
50
40
30
20
10
0
2
4
6
8
10
12
14
16
18
20
Palliative Care Rate per Thousand
22
24
26
28
14
Palliative Care Mortality Distribution
Hospital Mortality Rates
for Patients under Palliative Care
30
Mean = 53%
25
Frequency
20
15
10
5
0
10%
20%
30%
40%
50%
60%
Mortality Rate
70%
80%
90%
100%
15
QUEST Mortality Drill Down Report to be Released
End of April
16
Exploring Drivers of Mortality
▪ Goal
▫ Explore in-patient mortality by finding ACTIONABLE clusters – IE
patient cohorts in which mortality rates might be improved with an
intervention (Part of a PDCA cycle)
» Common cause – systemic problems
» Special cause – isolated but important causes
▪ Definition
▫ Excess Deaths = Total deaths in excess of predicted by the risk
adjustment model = (obs % - exp %) * N patients
▫ Excess Deaths can be “negative” in this definition
▫ Therefore sum of all non-negative Excess Deaths over all patient
subsets will be greater than hospital-wide results (hospital-wide obs
– hospital-wide exp) * Total Discharges
▫ In other words, there are always pockets of opportunity
▪ Approach
▫ Use CA or QM to determine excess death by categories
» Admission Source, Age, Principal Dx, APR-DRG or DRG, severity, other
17
A Tale of Two Hospitals
▪ Two Sample Hospitals
▫ Hospital 1: > 375 beds, non-teaching, urban, o/e < 1.00, 2nd Qrtle
▫ Hospital 2: < 375 beds, non-teaching, urban, o/e > 1.00, 3rd Qrtle
▪ Questions
▫ What conditions are associated with excess mortality across the
entire hospital population? Conditions can be primary or secondary
conditions (e.g., sepsis is not always coded as primary diagnosis)
▫ Is there evidence for special cause or common cause variation by
common groupings?
» Admission source, care progression, age, principal dx, etc.
▪ Goal
▫ Determine top three or four focus areas in which to implement
PDCA cycles to improve in-patient mortality
18
Hospital 1: Excess Death by Admit Source –
Aggregate
Notice the hospital-wide o/e is < 1.00 and very close to TPT
NO Excess Deaths by any given admission source
No evidence of special cause variation at hospital-wide level
Source: Clinical Advisory Quality Reports with Excess Deaths added - see Appendix
19
Hospital 1: Excess Deaths by Age Group –
Aggregate Level
Possible special cause variation in patients over 84 years old
Source: Clinical Advisory Quality Reports with Excess Deaths added - see Appendix
20
Hospital 1: Excess Mortality by Primary Dx
Hospital-Wide Excess Deaths (partial) sorted by excess deaths
Remember this hospital has
an O/E = 0.88. However,
there are still many pockets of
opportunity.
Nine Excess Deaths with Sepsis as Primary Dx
Source: Clinical Advisory Quality Reports with Excess Deaths added - see Appendix
21
Hosp 1: Excess Mortality by ICD9 Secondary Dx
Hotpital- Wide Excess Deaths (partial) – sorted by Excess Deaths
Expected
Rate
Notice:
1) Observed and expected
mortality for Palliative Care
Notice:
2) Many other pockets of
opportunity – (note these are
not mutually exclusive
patients)
Source: Clinical Advisory Quality Reports with Excess Deaths added - see Appendix
22
Hosp 1: Excess Mortality by ICD9 Secondary Dx
Hospital- Wide Excess Deaths (partial) – sorted by Clinical Categories
Notice:
Grouping Excess Deaths into
meaningful categories may
help opportunities stand out
Source: Clinical Advisory Quality Reports with Excess Deaths and Clinical Categories added, Clinical Categories are user defined- see Appendix
23
Hosp 2: Excess Mortality
Pareto Analysis by Admit Source (all admits)
Evidence of special cause variation in patients by admit
source. Almost all Excess Deaths are from two sources
Source: Clinical Advisory Quality Reports with Excess Deaths and Clinical Categories added, Clinical Categories are user defined- see Appendix
24
Hospital 2: Excess Mortality- ED Admissions
Pareto Analysis (partial) by Excess Deaths
Clinical Category
Sources of ED mortality: Respiratory, Stroke, Renal, Sepsis,
and “Low Mortality Populations”
Source: Clinical Advisory Quality Reports with Excess Deaths and Clinical Categories added, Clinical Categories are user defined The category
“Low mortality population is based upon the APRDRG expected mortality. “Low Mortality” and “End of Life Care” are arbitrarily defined, not
clinically determined, and are intended to aid analysis only- see Appendix
25
Hospital 2: Excess Mortality – Transfer from hosp
Pareto Analysis (partial) by Excess Deaths
Clinical Category
Sources of Transfer Patient mortality: ? End of life issues
Source: Clinical Advisory Quality Reports with Excess Deaths and Clinical Categories added, Clinical Categories are user defined The category
“Low mortality population is based upon the APRDRG expected mortality. “Low Mortality” and “End of Life Care” are arbitrarily defined, not
clinically determined, and are intended to aid analysis only- see Appendix
26
Hosp 2: Excess Mortality by ICD9 DX – ALL Dx
Dx with more than 5 Excess Deaths – grouped by category (Xcess > 5 deaths)
Clinical Category
Source: Clinical Advisory Quality Reports with Excess Deaths and Clinical Categories added, Clinical Categories are user defined The category “Low mortality population is based upon
the APRDRG expected mortality. “Low Mortality” and “End of Life Care” are arbitrarily defined, not clinically determined, and are intended to aid analysis only- see Appendix
27
Approaching Drivers of Mortality *
Illustrative Examples of Potential Secondary Drivers
GOAL
Potential PRIMARY
DRIVERS
Potential SECONDARY DRIVERS
Early appropriate level of care (ICU)
Early recognition and intervention
Sepsis
Timely transfer to ICU
Elderly and other high risk groups
Early recognition of resp compromise
Hospital – Level
Risk Adjusted
Mortality
(O/E Ratio)
Respiratory
Conditions
Avoidance of VAP
Post operative resp care protocols
Cardiac
Related and
Shock
Rapid response team
Adherence to ACC Protocols
Early transfer to ICU if needed
Improved use of cardiac monitors
*Data mining to examine
top drivers of mortality is
currently in progress
End of Life
Care
Early identification of patients
Proper use of V667 palliative code
Appropriate setting: hospice v acute
28
QUESTIONS?
Eugene A. Kroch
Richard A. Bankowitz
29
Appendix
30
APR-DRG Process Flow
Assign
APR-DRG
Ignore Secondary
Diagnoses Related
to Other Higher Risk
Secondary
Diagnoses
Set Base Score
to Highest
Secondary
Diagnosis Level
Above Score in
Step 2
Primary &
Secondary
Diagnoses
Ignore
Secondary
Diagnoses
Associated with
Primary
Diagnosis
Identify Risk
Codes for
Remaining
Secondary
Diagnoses
Set Minimum
Risk Level Based
on Age or NonOR Procedure
Adjust Risk Code Based Upon
Secondary Diagnoses Severity
Codes
NB: Risk code is mapped into mortality risk based
on the mortality rates from calibration data base.
Step 1
Step 2
Compare Risk Code to
Primary Diagnosis Severity
Code
Step 3
Final Risk
Code
31
CareScience Regression Model
Principal Dx – Pneumonia – one of 142 disease strata
Outcome = age + sex + distance + proc + 
1.0 0.9 -
*
*
*
*
0.8 0.7 -
*
0.6 -
*
0.5 -
*
0.4 0.3 -*
*
*
*
*
*
*
*
*
 = 0.074
*
*
From CS client base sample
*
*
0.2 - *
0.1 |
10
Y
=
dependent variable
|
20
|
30
|
40
|
50
|
60
|
70
|
80
|
90
age
0 + 1X1 + 2X2 + … + nXn
independent variables / explanatory variables
32
Trend Distribution across Hospitals
30
25
Distribution of Hospital O/E Trends
(over 8 quarters)
Mean = -12%
Frequency
20
15
10
5
0
-40% -35% -30% -25% -20% -15% -10% -5%
Trend (%change)
0%
5%
10%
15%
20%
33
Trend Volatility
Maximum Quarterly O/E Change
50
across hospitals
45
40
Frequency
35
30
25
20
15
10
Smaller hospitals
(avg 25% of mean size)
5
0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
O/E Ratio Quarterly Change
0.9
1.0
1.1
1.2
34
Lives Saved by Disease
Reference Period: 2005q3-2006q2
ICD9_Diag
038
518
428
480 - 486
584
433 - 434
162
430 - 432
197
410
571
427
852
153
764 - 765
Sub_Total
Grand Total
Description
Septicemia
Other Lung Dis.
Heart Failure
Pneumonia
Renal Failure
Ischemic Stroke
Lung Cancer
Hemorrhagic Stroke
Metastatic Cancer
AMI
CABG
Chronic Liver Dis.
Cardiac Dysrhythmias
Head Trauma
Colon Cancer
Premies
Comparison Period: 2006q3-2007q2
Mortality Lives Lives/ Percent
Cases
Rate
Saved 1000 of lives
95,153
18.6%
1266 13.3 11.7%
65,418
17.1%
1107 16.9 10.3%
168,421
3.2%
694
4.1
6.4%
155,669
3.5%
624
4.0
5.8%
63,125
5.1%
511
8.1
4.7%
96,826
4.0%
325
3.4
3.0%
26,484
11.3%
302 11.4
2.8%
22,166
24.5%
294 13.3
2.7%
21,893
12.5%
260 11.9
2.4%
109,696
6.3%
216
2.0
2.0%
59,513
2.9%
202
3.4
1.9%
16,517
7.7%
176 10.7
1.6%
121,310
2.1%
103
0.9
1.0%
12,772
11.1%
70
5.5
0.7%
18,054
4.1%
68
3.8
0.6%
74,192
3.4%
-38
-0.5
-0.4%
1,091,058
6.6%
5,804
5.3 53.8%
6,132,358
2.0% 10,793
1.8
100%
35
Lives Saved Rate vs. Mortality Rate
18
Other Lung
16
Metastatic
Cancer
Lives Saved per 1000 Discharges
14
Lung Cancer
12
Septicemia
Hemorrhagic
Stroke
Liver
10
8
Renal
6
Pneumonia
4
Head Trauma
Ischemic Stroke
AMI
2
Dysrhythmias
0
0%
5%
10%
15%
Mortality Rate
20%
25%
30%
36
Appendix: How were the Excess Death Tables
Made?
▪
Hospital 1: Excess Death by Admit Source
▫
CA Quality Reports > Mortality Reports > Select QUEST time period > Select Patient Type Inpatient > Drill by Admit Source > Export to
Excel
▫
Add column Excess Death (Mortality – Expected Mortality)* Cases
▫
Sort by Excess Death
▪
Hospital 1: Excess Death by Age Group
▫
▫
▫
▪
Hospital 1: Excess Death by Primary Dx
▫
▫
▫
▪
CA Quality Reports > Mortality Reports > Select QUEST time period > Select Patient Type Inpatient Drill by Principal Dx > ICD9 >
Export to Excel
Add column Excess Death (Mortality – Expected Mortality)* Cases
Sort by Excess Death
Hospital 1 Excess Death by Secondary Dx – Sort by Excess Death
▫
▫
▫
▪
CA Quality Reports > Mortality Reports > Select QUEST time period > Select Patient Type = Inpatient > Drill by Detailed Age Categories
> Export to Excel
Add column Excess Death (Mortality – Expected Mortality)* Cases
Sort by Excess Death
CA Quality Reports > Mortality Reports > Select QUEST time period > Select Patient Type Inpatient Drill by Secondary Dx > ICD9 >
Export to Excel
Add column Excess Death (Mortality – Expected Mortality)* Cases
Sort by Excess Death
Hospital 1 Excess Death by Secondary Dx – Sort by Clinical Grouping
▫
▫
▫
▫
▫
▫
CA Quality Reports > Mortality Reports > Select QUEST time period > Select Patient Type Inpatient Drill by Secondary Dx > ICD9 >
Export to Excel
Add column Excess Death (Mortality – Expected Mortality)* Cases
Sort by Excess Death
Assign categories to the top source of Excess Death – any grouping that is clinical useful will do
Resort by the categories
You may color if you like to enhance visual communication
Note: All Clinical Categories are user defined and are arbitrary, The category “Low mortality population is based upon the APRDRG expected
mortality. “Low Mortality” and “End of Life Care” are arbitrarily defined, not clinically determined, and are intended to aid analysis only. They
are not intended as a substitute for clinical judgment.
37