Transcript Using computational models to advance the theory and practice of

Public Health Dynamics Laboratory
Using computational models to advance the
theory and practice of Public Health
Mark S. Roberts, MD, MPP
Professor and Chair, Health Policy and Management
Director, PHDL
Briefing for Michelle Dunn, PhD
Senior Advisor, Data Science Training, Diversity and Outreach
Office of the Associate Director for Data Science
National Institutes of Health
PHDL Overview
1
Mental & Computational Models in Public Health
•
"All decisions are made on the basis of models. Most models are in our heads. Mental models are
not true and accurate images of our surroundings, but are only sets of assumptions and
observations gained from experiences ... Computer simulation models can compensate for
weaknesses in mental models" (Forrester, 1994).
Prior individual knowledge
Mental Model
Expert advice
Public
Health
Decision
Mechanistic
Models & Simulations
Data
PHDL Overview
2
Mission of the Public Health Dynamics Lab
• Develop interdisciplinary approaches using
computational models to advance the theory and
practice of Public Health
• Contribute to "Systems Thinking" in the training of
the next generation of health professionals
• Develop and distribute computational tools to be
used to improve real-world population health
decision- making
PHDL Overview
3
A Systems Approach to Public Health
Scale
“Systems Public Health”
108 m Earth
106 m Country
104 m City
102 m Village
100 m Human
10-2m Organ
10-4 m Lymph Follicle
10-6 m Cell
“Systems Biology”
10-8 m DNA
10-10 m Nucleotide
10-12 X Ray
10-14 Atomic Nucleus
PHDL Overview
4
Sponsors and Partners
Models of Infectious Disease Agent Study (MIDAS)
National Center of Excellence
PI: Burke
Sponsor: NIGMS/NIH
(Harvard, Washington, 12 others)
Vaccine Modeling Initiative
PI Burke
Sponsor: Bill & Melinda Gates Foundation
(Imperial and Princeton)
Public Health Adaptive Systems Studies
PI: Potter
Sponsor: CDC
Public Health International Modeling Fellows Program
PI: Grefenstette/Burke
Sponsor: Benter Foundation
www.midas.pitt.edu
www.vaccinemodeling.org
www.phasys.pitt.edu
Benter
Foundation
Partners:
PHDL Overview
5
FRED (Framework for Reconstructing Epidemiological Dynamics)
PHDL Overview
6
Census-matched Synthetic Population
Person = Agent
US Census Data
Each agent is
assigned to
household, school
and workplaces
with other agents
U.S. Population
(112,595,578
households with
289,390,247 people)
LandScan Satellite
Data
DoE School Data
Extract Any
Location
BLS Business Data
PHDL Overview
7
Matched actual demographics
• Iterative proportional
fitting assures that
synthetic attributes are
distributed as real ones
are
• Individual-based models focus on how interactions among individual
simulated agents can result in complex patterns at a population
level. Here is a movie of our team’s first large scale model (of the
possible introduction of avian influenza into the USA).
Ferguson NM, Cummings DA, Fraser C, Cajka JC, Cooley PC, Burke DS.
Strategies for mitigating an influenza epidemic
Nature July 27, 2006; 442: 448-52
PHDL Overview
9
Prediction – Pushing the Boundaries
PHDL Overview
10
Modeling to Inform Policy
• We have seen several example of PHDLderived tools used in policy
– FRED used to estimate mitigation strategies in
2009 Flu epidemic
– Tycho and FRED Measles to inform consequences
of vaccination rates
• Our expertise has provided a platform (PI
Everette James) to conduct research for PA:
“University will assist DPW in monitoring and evaluating changes in health insurance coverage; access to
care; health care and Medicaid financing; state and federal policies, regulations and legislation affecting
health care; and the quality of health care delivery in the Commonwealth’s Medicaid program.”
PHDL Overview
11
FRED Measles
http://fred.publichealth.pitt.edu/measles/
PHDL Overview
12
Fred Ages and Stages
PHDL Overview
13
Goal: High Resolution Detail in FRED
Diseases:
Diabetes
Heart Disease
Hypertension
Asthma
COPD



other chronic dz
Environmental
Characteristics
•
•
•
Pollution
Walkability
Food deserts
PHDL Overview
14
Aspirational future
Social Network/
Behavioral Research
Cardiovascular Disease
Infection/Transmission
Research
Influenza
Unmeasured parameters
Viral
Replication
HIV
.. ...
Clinically complex
model of Hepatitis C
HIV
Mutations
............
.. . .. .
. .. .
.
. .........
.
. . ............. .
. ..
.
.
Geographic/Social Specificity
Resistance to
ART
Legal and
Regulatory
Effectiveness
Of ART
Population
Behavior
CD4 Cell
Count
Viral
Load
Risk of Death
From HIV/AIDS
Health Care
Resources
Disease
in silico Pennsylvania
Simulation
Engine
Health Care
Costs
Interventions
Agent-Based Simulation Model (FRED)
PHDL Overview
15