Transcript Does The Chronic Care Model Work? - UW Family Medicine & Community

Developing Population Health Clinical
Informatics System Requirements to Support
Primary Care Delivery and Quality Improvement
Brian Arndt MD
Lawrence Hanrahan PhD MS
Jonathan Temte MD PhD
Marc Hansen MD
George Mejicano MD MPH
John Frey MD
David Simmons MPH
November 7, 2008
Presentation Objectives
1.
2.
3.
4.
Review the basics of medical informatics and its
domains.
Review progress to date on early collaborations in
clinical informatics between the UW DFM and the
WI Department of Public Health.
Review population health informatics including best
practices in data analysis.
Explore ways in which health information technology
can build a critical bridge between primary care and
the public health care system.
2
Medical Informatics
• Definition: the systematic application of computer
science and technology to medical practice, research, and
medical education
• Scope includes the conceptualization, design,
development, deployment, refinement, maintenance, and
evaluation of systems relevant to medicine
Medical Informatics Domains
Bioinformatics
Molecular
Cellular
Genetic
Imaging
Informatics
Clinical
Informatics
Public Health
Informatics
Tissues
Organs
Individual
Patients
Population
Health
Adapted from Shortliffe
Medical Informatics Hierarchy
Public Health
Informatics
Clinical
Informatics
Imaging
Informatics
Bioinformatics
Adapted from Shortliffe
Illustration: Public Health Alerts & Reporting
• Introductory statement printed each week in Public
Health Reports, 1913-1951:
“No health department, state or local, can effectively
prevent or control disease without knowledge of when,
where, and under what conditions cases are occurring.”
• Despite being mandated by law, communicable disease
reporting is poor – incomplete, inaccurate, and delayed
6
The current state of the art …
DIARRHEA ALERT VIA EMAIL CHAIN:
Amanda Kita (Public Health)
→ Mike Holman (UWMF Employee Health)
→ Sue Kaletka (DFM Administration)
→ Mark Shapleigh (Clinic Manager)
→ Brian Arndt (Clinician)
→ Patient w/ diarrhea
New Clinical Information Systems Needed!
We’ve officially arrived at the point of no return!
Clinical and Population Health Informatics
Diffusion Model
Data
Interpretation
Data
Analysis
Clinical
Systems
Data
Collection
Data
Interpretation
Information
Dissemination
Population
Health
Data
Collection
Clinical Informatics
Two Way Information Flow
Data
Analysis
Public Health Information Flow
Wisconsin
Department
of Health Services
EMR Data
Central Server
WI Dept of Public Health
!
EMR Alerts
practice alert in
EMR if patient
presents with
symptoms
matching
condition
10
Public Health Alerts through the EMR
(Legionella Outbreak)
Chronic Care Model
Community
Resources
& Policies
Health System
Health Care Organization
SelfManagement
Support
Informed,
Activated
Patient
Delivery
System
Design
Decision
Support
Productive
Interactions
Clinical
Information
Systems
Prepared,
Proactive
Practice Team
Improved Outcomes
12
Clinical Information Systems
• ID chronic conditions that require both proactive and reactive care
• Diabetes, CHF, asthma, metabolic syndrome, etc
• The conditions to follow often are dictated by larger systems (ie, local health
plans with Pay for Performance programs)
• Also consider conditions that may progress further
–
–
–
–
Impaired fasting glucose or gestational diabetes → Type 2 diabetes
Hyperlipidemia → Coronary artery disease
Elevated BP w/o HTN → HTN
Overweight → Obesity
• Develop algorithms to appropriately identify patients
– Billing data is usually not enough – consider addition of lab data, prescription
medication data, EMR problem list abstraction, etc
13
Develop Registries
• Organize clinic subpopulation data to plan quality improvement
efforts and to facilitate new care processes
– What about determining comorbidity score (ie, Charlson) from
administrative data to target patients at highest risk of mortality?
• Many EMRs are adequate for managing individuals, but cannot
manage populations well
– Practices should think about this functionality when purchasing an EMR
• Registries can be created in the absence of a fully functional EMR
with other commonly available software
– Microsoft Excel, Microsoft Access, etc
– Physicians Plus Insurance Corp. currently uses DocSite
– What are the algorithms insurers use to identify our patients with a
particular condition?
14
DFM Diabetes Registry
15
Develop Registries
• Assess performance of individual patients and clinicians, clinical
teams, clinics, health systems, and ultimately communities
– Provide regular (and accurate!) feedback for continuous quality improvement
– Reports can be generated to document trends (both improvements and
setbacks)
– Target appropriate clinical interventions based on analysis outcomes using
Plan-Do-Study-Act (PDSA) cycles
16
Population Health Informatics
Data Analysis Best Practices
(Example: Diabetes patients with A1c >7)
Analysis Type
Example
1 – Case series
60% in clinic have A1c >7
2 – Simple comparison
Clinic rate of 60% is higher than
statewide rate of 50%
3 – Comparison + Test
Clinic rate of 60% is significantly
higher than statewide rate of 50%
Utility
Lowest
Low
Medium
4 – Adjusted comparison Age adjusted clinic rate of 60% is
+ Test (ie, adjust for
significantly higher than statewide
principal determinant)
age adjusted rate of 50%
Higher
5 – Multivariate model +
Test (ie, adjust for all
important risk factors /
determinants)
Highest
Clinic rate of 60% adjusted for
age, gender, race, and insurance
status is significantly higher than
comparably adjusted statewide
rate of 50%
Population Health Informatics
Data Analysis Best Practices
• There is limited understanding of disease burden or its risk
factors without formal testing, data tables, charts, graphs, and
maps to display variation
• To create meaningful disease burden displays, each assessment
must be compared to an appropriate reference
• The comparison must be tested (p value, Confidence Interval,
Relative Risk)
• Ideally, other known predictors of risk must be controlled when
comparisons are made
Clinic A1c > 7 Rate or Relative Risk –
Age, Gender, & Race Adjusted
(95% Confidence Intervals)
Clinic A
R
a
t
e
Clinic B
Clinic C
Clinic D
R
R
Clinic E
Risk Compared to State, US, and 2020 Target
Interpretation of Color Gradients
County
Color
Result
Interpretation
Clinic A
Rank highest – Significant
from referent
Disease disparity
Clinic B
Rank high – Not significant
from referent
Possible disease disparity cautious monitoring
Clinic C
Rank same as referent –
Not significant
Possible disease disparity
with room to improve cautious monitoring
Clinic D
Rank low – Not significant
Health advantage - hopeful
monitoring possible
Clinic E
Rank lowest – Significant
Health advantage
Population Health Informatics
Data Analysis Best Practices: A Proposed v1.0
• When the goal is identifying disparity and prediction of risk, it is
appropriate to use automated computer selection algorithms (ie,
backward elimination) built into computer packages1
• Multiple factors are examined and their simultaneous,
independent contribution to health is determined
• The Wisconsin Public Health Information Network (PHIN)
Analysis, Visualization, and Reporting (AVR) system makes this
possible
1Source:
Kleinbaum, Logistic Regression (1994)
The Wisconsin PHIN AVR Portal
The UW DFM Pilot
• De-identified visit records were provided from the Epic EMR over
a 1 year period (N = 309,000)
• Secure, role-based access control
Data Cube
(Structured data for efficient exploration)
Population Health Charting Example
Acute respiratory infections, 480-487
Pneumonia and influenza AND Temperature ≥100 AND
Service Year 2007
Geographic Information System (GIS):
Diabetes Visit Count by Zip Code
Stored Analytic Process:
Logistic Regression Modeling Diabetes Risk
Predicted by Age and Body Mass Index
Diabetes Use Case Proposal:
Population Health Multivariate Model to Support
Primary Care & QI
Outcomes =
Patient
Factors +
Clinician
Factors +
Clinic
Factors +
Obesity
Hypertension
Depression
Diabetes
CVD
Smoking
Alcohol
A1c level
LDL
HDL
BP
Diet
Phys Activity
Process factors
(ie, time to
repeat follow-up)
Age
Gender
Race/ethnicity
Co-morbidities
Medications
Literacy
Culture
Psychodemographics
Insurance
Census block /
tract / zipcode
Age
Location
Gender
Capabilities
Certifications Processes
Specialty
Graduation
date
Years of
practice
Community
Factors
Census block / tract /
zipcode:
Poverty
Education level
Psycho-demographics
(ie, purchasing habits)
Built environment:
Traffic
Recreation / parks
Sidewalks
Restaurant mix
Safety / crime
Fast food sales
Fresh fruit & vegetable
sales / consumption
UW Clinical Informatics Evolution
• Our next steps to develop population health informatics
requirements:
– Core work group will continue literature review & refine proposal
– Focus groups convening 1/9/2009 & 1/16/2009
• Develop paper prototype
– RSS / wider distribution & feedback / CME?
– Pilot testing (starting 7/1/2009)
• Please consider joining us!