Transcript Document

Punkaj Gupta, MBBS
Division of Pediatric Cardiology
Arkansas Children’s Hospital
March 26, 2015
• None for all authors
• VPS data was provided by the VPS, LLC.
No endorsement or editorial restriction of
the interpretation of these data or opinions
of the authors has been implied or stated
Building Trust in the Power of
“Big Data” For Outcomes
Research to Serve the Public
Good
• Study of the end results of particular health
care practices and interventions
• Uses retrospective, non-interventional
data from existing multi-center databases
• The nation is spending over $800 billion
dollars on health care, yet very little is known
about what that $800 billion is buying
• Outcomes research helps us understand the
most effective and efficient ways to provide
high quality health care
• Existing data may be used to conduct studies
that are not amenable to a randomized trial
format
• Existing data often describe “real-world” care
and may be used to define practice variation
• Nickname in computer science, business,
and public policy for the application of
sophisticated analytic techniques to large and
rapidly growing databases
• In medicine applicable to electronic health
records, clinical registries, and administrative
databases
• Virtual PICU Performance System (VPS,
LLC): ~ 1 million ICU patients from 130
Pediatric ICUs
• The Pediatric Health Information System
(PHIS): ~ 3 million patients from 43 freestanding children’s hospitals in United
States
• Big data provides great potential for
extracting useful knowledge to achieve the
‘triple aim’ in health care
– better care for individuals,
– better care for all, and
– greater value for dollars spent.
Okun S, McGraw D, Stang P, et al. Discussion Paper: Making the Case for
Continuous Learning From Routinely Collected Data. Institute of Medicine
• Health care lags behind other industries in
leveraging advances in information
technology and analytical techniques.
• If “Big Data” using databases like VPS,
LLC applied to health care, it would
potentially improve quality and efficiency
of the system.
• Affordable Care Act: Incentives are
increasing for stakeholders (including
clinicians, insurers, purchasers, and
patients) to collect, analyze, and exchange
health care information
• Study two examples from VPS, LLC
database
• Demonstrate strength and weakness of
“Big Data” through these examples
Punkaj Gupta, MBBS; Xinyu Tang, PhD; Casey Lauer, BA;
Robert M. Kacmarek, PhD, RRT; Tom B. Rice, MD;
Barry P. Markovitz, MD, MPH; Randall C. Wetzel, MBBS
• Little is known about the effects of clinical
education, and hospital structure on
medical outcomes in children with critical
illness
• Increasing concerns regarding trainee
inexperience as a contributing factor to
outcomes in children with critical illness
• Similar concerns for non-university, and
non-free standing children’s hospitals
providing a lower level of care for critically
ill children
• To evaluate outcomes associated with
training programs (such as residency or
fellowship training), and hospital structure
(such as free-standing children’s or
university hospital) using the Virtual PICU
Systems (VPS, LLC) Database
•
•
•
•
Odds of ICU mortality
Time to ICU discharge
Odds of mechanical ventilation
Time to liberation from mechanical ventilation
• The Virtual PICU Systems (VPS, LLC) is an
online pediatric critical care network
• Prospective observational cohort from ~130
PICUs with interrater reliability (IRR) testing >
95%
• Formed by NACHRI (now part of CHA),
Children’s Hospital of Los Angeles, and
Children’s Hospital of Wisconsin
• Patients <18 years of age admitted to one
of the participating PICUs in the VPS
database were included
• Patients with both cardiac (cardiacmedical and cardiac-surgical), and noncardiac diagnoses were included
!
336,323 patients
108 hospitals
Residency
260,191patients
77 hospitals
!
Residency + Fellowship
166,362 patients
38 hospitals
!
Residency + Fellowship +
University
148,635 patients
34 hospitals
!
Residency +
Fellowship +
University +
Freestanding
Children’s
119,833 patients
22 hospitals
Non-Residency
Non-Fellowship
Non-University
Non-Free Standing
Children’s
47,891patients
25 hospitals
• Patient characteristics and outcomes
summarized between the study hospitals
and control hospitals
• Multivariable logistic regression models
and Cox proportional hazards models
were fitted to evaluate association of
training programs and hospital structure
with study outcomes
• A total of 308,082 patients from 102
centers were included
• Patients in the study hospitals had greater
severity of illness (PIM-2 and PRISM-3
scores), and had higher incidence of
cardiopulmonary resuscitation
• Compared to the control groups, resource
utilization was also greater among the four
hospital categories, e.g.,
– the use of mechanical ventilation and
– high frequency ventilation, and
– use of arterial and invasive central lines
• Compared to patients in control hospitals,
patients in the four hospital categories
were older, and had significant
comorbidities, such as
– developmental disorder
– genetic syndrome
– low birth weight
– prematurity
• ICU mortality was significantly lower
among the study hospitals- as compared
to the control hospitals
• Despite caring for more complex and
sicker patients, time to ICU discharge was
shorter among the study hospitals- as
compared to the control hospitals
• Could not account for the potential impact
of variables such as– hospital structure and process measures,
– training or availability of ICU personnel, or
– nursing factors on study outcomes
• Our study did not address the financial
burden of training program or hospital
structure as an outcome measure.
• Use of ICU Mortality, time to ICU
discharge, and time to liberation from
mechanical ventilation as outcome
measures
Punkaj Gupta, MBBS; Xinyu Tang, PhD; Casey Lauer, BA;
Tom B. Rice, MD; Randall C. Wetzel, MBBS
• Clinical practice variations are common in
children undergoing congenital heart
surgery
• None of the existing literature to-date has
truly compared the volume-outcome
relationship with mechanical ventilation
after pediatric cardiac surgery as an
outcome
• To evaluate the
– odds of mechanical ventilation, and
– duration of mechanical ventilation after
pediatric cardiac surgery
• across centers of varying center volume
using the Virtual PICU Systems (VPS,
LLC) Database
• The Virtual PICU Systems (VPS, LLC) is an
online pediatric critical care network
• Prospective observational cohort from ~130
PICUs with interrater reliability (IRR) testing >
95%
• Formed by NACHRI (now part of CHA),
Children’s Hospital of Los Angeles, and
Children’s Hospital of Wisconsin
• Patients <18 years of age undergoing
operations (with or without CPB) for heart
disease at one of the participating ICUs in
the VPS database were included
• Patients receiving high frequency
oscillatory ventilation (HFOV), or jet
ventilation were also excluded
• Centers with >10% missing data were
excluded
• Average number of cardiac surgery cases
per year for each center
• Study centers were categorized using the
center volume tertiles:
– Low-volume: <175 cases/year
– Medium volume: ≥175 to <275 cases/year
– High-volume: ≥275 cases/year
• Patient characteristics, procedural data,
post-operative outcomes
• Outcomes
– Odds of mechanical ventilation
– Duration of mechanical ventilation after
pediatric cardiac surgery
• Multivariable logistic regression models
and Cox proportional hazards models
used to evaluate the relationship between:
– Center volume and odds of MV
– Center volume and duration of MV
• Models adjusted for patient factors and
center effects
• 10,378 patients from 43 centers were
included
Number of
Centers
Number of
Patients
Low
Medium
High
36
4
3
3,657 (35%)
3,176 (31%)
3,545 (34%)
Low
Medium
High
Mortality
3% (127)
3% (102)
2% (72)
Use of MV
73% (2675)
81% (2576)
68% (2397)
Duration of
MV
24 (8, 96)
27 (8, 99)
45 (19, 119)
Unadjusted
Adjusted
OR (95% CI)
P
OR (95% CI)
P
Low
1.26
(1.14, 1.39)
<0.001
2.68
(2.15, 3.35)
<0.001
Medium
1.78
(1.60, 1.98)
<0.001
1.31
(1.12, 1.52)
<0.001
High
Reference
Reference
• Higher volume centers were associated
with lower odds of mechanical ventilation
in the lower risk patients (STS-EACTS
categories 1-3)
• No significant relationship between center
volume and odds of mechanical ventilation
in the higher risk patients (STS-EACTS
categories 4-5)
Unadjusted
Low
Medium
High
Adjusted
HR (95% CI)
P
HR (95% CI)
P
1.16
(1.10, 1.23)
1.14
(1.08, 1.21)
Reference
<0.001
1.26
(1.16, 1.37)
1.19
(1.11, 1.28)
Reference
<0.001
<0.001
<0.001
• Higher volume centers were associated
with longer duration of mechanical
ventilation in both high risk (STS-EACTS
categories 4-5) and low risk patients (STSEACTS categories 1-3)
• Large clinical practice variations were
demonstrated for MV following pediatric
cardiac surgery among ICUs of varied
center volumes
• Both odds of mechanical ventilation and
duration of mechanical ventilation
following cardiac surgery vary substantially
across hospitals
• Multi-institutional databases can be
powerful tool for doing outcomes research
• If used methodically, database research
can have significant impact on clinical
practice and health care outcomes