Transcript Julie Swann - Institute for People and Technology

Analytics for Public Health to Design
Efficient, Effective, Equitable Systems
Julie Swann, PhD
Industrial and Systems Engineering
Georgia Institute of Technology
Fall 2013
2
Health Analytics
• What is it?
▫ A set of methods from the mathematical, engineering,
physical, and social sciences to understand, predict,
optimize, and/or evaluate a system using real or
generated data. (Can include Biosurveillance)
• What types of tools?
▫
▫
▫
▫
Statistics and probability
Optimization with mathematical modeling
Simulation of systems with uncertainty
Economics and financial analysis
• What goals?
▫ Efficiency (cost)
▫ Effectiveness (outcomes)
▫ Equity (disparities)
3
Health Analytics
Understand
Predict
Optimize
Evaluate
•What is the status quo of the system in question?
•What is happening over time, geographical space, or subpopulations?
•How can we quantify Cost, Outcomes, and Disparities?
•What factors are associated with quantifications of (Cost, Outcomes, or Disparities)?
•If you make a change to the system, what will the impact be?
•Can you predict where the greatest needs are or how they will change over time?
•What policy change would be most effective at decreasing cost, improving outcomes, or
reducing disparities?
•What are the most appropriate interventions for a given situation (behaviors, network
infrastructure, screening policies, etc.)?
•What interventions improve outcomes or disparities the most with the lowest cost?
•Can the impact of an intervention be projected over time, space, or subpopulations?
•How can predictions of need or cost be validated?
•For implemented interventions, how well do they perform on outcomes?
4
Some Data Sets
Medical Claims Data
Medicaid (children & pregnant
women, GA + 13 other states, 20052009)
Access, Disparities,
Baseline,
Interventions
Electronic Health Record
Queries on specific projects
(Children’s Healthcare of Atlanta
and VA)
Costs, Outcomes,
Trends
Electronic Monitoring
Monitoring in NICU and PICU at
Children’s Healthcare of Atlanta
Associations, Who
and How Long
Disease Registries
Cystic Fibrosis
Access, Outcomes,
Trends
Disease Progression
“Natural History” Models;
Agent-based simulations
Screening Policies,
Interventions
National Surveys or
Examinations
NHANES, BRFSS, HCUP KIDS
Predictions
geographically
State Databases
GA’s Oasis, HCUP SEDD and SIDD
Small-Area
Variations in Cost
General
Census, National Provider Index
Supply and
Demand
5
Research Examples
• Population screening policies for diseases
▫ Hepatitis C, HIV, Cancers, Newborn Screening
▫ Who? What ages? How often?
• Interventions for diseases spreading across a network
▫ H1N1a pandemic using agent-based modeling
▫ Post-campaign evaluation of distribution system
• Quantifying and explaining access to care and disparities
▫ Pediatric patients across a network
• Predicting disease prevalence in small areas
▫ Childhood obesity
• Decision-support tools
▫ Catch-up scheduling for vaccinations
• Work covers multiple disciplines at GT including collaborations with
CDC and other health entities, representing many researchers not
present today
6
Population Screening
• Screening to diagnose disease can
▫ Increase treatment success or slow disease
▫ Reduce secondary infections
• Trade-offs
▫ May be costly to test many people
▫ Test or overtreatment may involve some risk
• Research Question:
▫ Who should be screened and how often,
given their risk characteristics and the specifics of the
disease?
• Simulation can be used to examine screening policies
across a population for the current and future cost
and outcomes
▫ “Natural History” of a disease populated with clinical
findings on progression, rates, etc.
• Approach has been used for cancers, HIV, Hepatitis C,
newborn screening, etc.
7
Disease Epidemics
• Influenza and other diseases
can spread quickly over time or space
• Use modeling to aid decision making
▫ Policies (e.g., school closure, quarantining)
▫ Planning (e.g., vaccines, treatments)
▫ Response (e.g., supply prepositioning,
points of distribution, etc.
Day 90, Forecast
Community 1
Household 1 Household 2
• Simulation: SEIR-type models can represent each
Work place
individual’s status (Susceptible to Recovered), and
agent-based simulation allows interactions over network
▫ Studied mutations, seasonality, traveling, and policies
• Post-campaign analysis
School
Household 3
Household 4
Community 2
▫ State distribution systems and associated leadtimes
affected vaccination rates at the state level (statistical
analysis)
▫ Access to vaccines in shortage period showed some
inequities
(optimization and statistics)
• Future: Cholera and other studies
Estimated Scarcity
8
Healthcare Access
1.
2.
3.
4.
2. Infer disparities
Main determinants of
Access to Pediatricians are:
Income, Education,
Population density and
Segregation
Measure access.
Infer disparities.
Design interventions.
Link to outcomes.
• Develop approaches for
projecting system changes
with application to common
conditions (e.g., asthma) and
rare ones (e.g., Cystic
Fibrosis) across different
network types
• Utilizes optimization to
“match” supply and demand
and statistics for analysis
Approved IRBs for Medicaid claims data
across 14 states and 5 years, CF
registry data, and AHRQ hospital
utilization data
1. Access to Pediatricians in GA
1. Estimated distance to Cystic Fibrosis
Centers and uncovered areas
3. Design interventions
Increasing participation of
MD taking Medicaid would
improve Access for
Medicaid patients without
compromising access for
overall population
4. Link to outcomes
Ex, Access affects outcomes.
Use to design more effective
interventions
9
Disease Prevalence
• What is the expected level of disease in each area, so
interventions may be targeted effectively?
▫ Prevalence of pediatric obesity differs by population characteristics
and/or geographically; widespread survey may be costly
• Statistical modeling can be used to project prevalence in
“small” geographical areas”
▫ Build regression to predict probability child is overweight or obese
 NHANES (with examinations) or NSCH (self-reported)
▫ Combine with a simulation of virtual individuals from Census data
• Validated with external data and compared to CDC approach
Results used to target interventions by a large healthcare provider in GA
Prevalence (zip codes)
Number of Children (2-17)
Priority Areas: ~ 80% of the
overweight children in GA
Decision Support: Vaccination Tool
• Many children have late, early or missed vaccinations
• Aim: a freely available and easy to use automated tool for catch-up
scheduling using vaccination history and feasibility rules
• Approach: dynamic programming in Excel or other platforms
• Online Childhood Scheduler (Keskinocak, Pickering, and collaborators)
https://www.vacscheduler.org/
http://www.cdc.gov/vaccines/schedules/easy-to-read/child.html
400,000+ downloads or online visits across all scheduler tools (adult, adolescents,
children)
11
Collaborations with the CDC
• Influenza and Emergency Response
▫ On loan during H1N1 pandemic
▫ Supply chain factors associated with
variation in vaccination uptake by state
▫ Access to vaccine during shortage period
• Obesity and Nutrition
▫ Prediction of pediatric obesity in
geographical small-areas
IPA;
ORISE Fellowship
PERRC Seed Grant
GT-CDC Seed Grant
• Global Health
▫ Tool to evaluate health kits pushed to
points of distribution
• Chronic or Infectious Diseases
▫ Role of Bathhouses and Sex Clubs in HIV
transmission
▫ Screening policies for diseases
• Immunization
▫ Catch-up Vaccination Scheduler
RFP for Small Business w/
subcontract to GT
Blood, Sweat & Tears
12
How Projects Began
• HIV policy analysis (2003)
▫ CDC scientist approached GT faculty they knew socially for assistance
with building a computational model
• Immunization (2007):
▫ GT faculty met with individual CDC scientists to solicit “challenging
problems” that needed solving
• H1N1 Pandemic Projects (2009+)
▫ GT asked to participate in pandemic response based on relationships and
two years of previous research in area
 While there, GT researchers identified other questions of interest, not
being asked by the CDC
• Global Health (2012)
▫ CDC scientists solicited ideas to spend year-end money
• Obesity (2012)
▫ GT faculty approached CDC groups with idea for seed grant until found
right people
13
Conclusions
• Appropriate public health problems
▫ Almost any, and methods may need to be integrated
▫ Much data exists already that can be utilized
• How such research can be performed
▫ Identification of existing or new problem
▫ Availability of data
▫ Sufficient resources
• Results
▫ Understanding of complex problems
▫ Publications of research results and/or tool
development
▫ Efficient, Effective, Equitable outcomes
14
Analytics in Public Health for
“Efficient, Effective, and Equitable” Outcomes
• Julie Swann, PhD
[email protected]
http://humanitarian.gatech.edu
• GT faculty or scientists in area
▫
▫
▫
▫
▫
▫
▫
Turgay Ayer
Rahul Basole
Dave Goldsman
Pinar Keskinocak
Eva Lee
Nicoleta Serban
(+ students and others)
• Other Collaborators
▫
▫
▫
▫
▫
▫
▫
▫
▫
▫
▫
▫
Jim Bost (CHOA)
Michael deGuzman (Children’s NY)
Ozlem Ergun (GT)
Paul Griffin (Penn State)
Anne Fitzpatrick (Emory)
Bruce Lee (Johns Hopkins)
Stephen Onufrak (CDC)
Michael Schechter (VCU)
Michael Washington (CDC)
Pascale Wortley (CDC)
Xingyou Zhang (CDC)
(+ advisees and others)
Education and Collaborations
• Professional education
▫ Health and Humanitarian logistics short courses
▫ Sept 2013 and May 2014
▫ http://humanitarian.gatech.edu
• Student project teams in classes
▫ http://www.isye.gatech.edu/seniordesign
▫ http://humanitarian.gatech.edu
• Graduate student and faculty research
▫ ORISE Fellows
▫ Seed Grants
▫ Grants or Contracts