Computational Epidemiology

Size: px

Start display at page:

Download "Computational Epidemiology"

Stephanie Blankenship
5 years ago
Views:

1 Computational Epidemiology as a tool for Understanding the Complex Interactions that Confront Public Health Decision Making Bryan Lewis MPH Network Dynamics and Simulation Science Laboratory

2 Preview Pubic Health: complicated and complex Epidemiology: past and future Decision Making: cost and benefit Investigation of Public Health Decision Making Importance of Representation Design of experiments to inform decisions Analysis of system to inform decisions

3 Acknowledgements

4 Public Health To promote health and quality of life by preventing and controlling disease, injury, and disability. -- CDC mission statement Methods (How): Direct delivery of care, Research, Policy WHO Agencies (Who) : Government, Nonprofits, Academic Institutions

components Yields variety of behaviors Non-linear

5 Public Health Complex Adaptive System Many different components Many interconnections between these components Yields variety of behaviors Non-linear effects Irreducible complexity Illustration by June Mullins

6 Epidemiology Study of epidemics - science of Public Health John Snow first action from epidemiological data Identified source of infection by correlating observed death-rates with the use of the Broad St. pump

7 Computational Epidemiology Study epidemics through computer simulations Design of models Analysis Interpretation High performance computing enable simulations of increasing sophistication

Public Health Decision Making Weighing cost vs.

costs of other decisions Projecting cost and benefit

8 Public Health Decision Making Weighing cost vs. benefit Infringement of rights vs. Public Health Side effects vs. lives saved Financial costs vs. costs of other decisions Projecting cost and benefit difficult People behave differently Diseases behave differently in silico Public Health Laboratory a valuable asset

9 Synthetic Public

Synthetic Disease Finite state machine defining individual s progress through natural history of disease 0.3 symptom1 circulating 1d sympom1 2-

10 Synthetic Disease Finite state machine defining individual s progress through natural history of disease 0.3 symptom1 circulating 1d sympom1 2-5d Detailed representation of disease defining infectiousness, symptoms, recovery, etc. uninfected saved by RX latent long 2d latent short 1d.625 incubating 1d asympt 3-6d 0.5 asymptreat 2-5d 0.19 symptom2 circulating 1d symptom3 circulating 1d symptom1 treated 3-6d symptom2 2-5d 1.0 symptom3 2-5d removed untreated antiviral vaccine symptom2 treated 3-6d symptom3 treated 3-6d

11 Modeling Frameworks EpiSimdemics Discrete-time agent-based simulation platform Agents can change behaviors based on disease state Transparent representations allow direct implementation of interventions Epi.Fast Efficient approximation based on graph theoretic techniques Allows rapid sampling of contact networks to evaluate important interactions and/or individuals Some simplifying assumptions allow 100-fold speed up

12 Activity based modeling Activities leading to the first infection in a household 5a. No intervention [AR = 36.6%] 5c. Close schools at 0.1% (day 30) [AR=2.0%] school work college shop other home Simdemics simulations performed for NIH to test non-pharmaceutical interventions (NPIs) recommended in Federal Pandemic Influenza Plan How the flu moves from home to home Where to best interrupt transmission through community

13 Variability in System 100 iterations from the same set [50] Number of Infections Epi.Fast simulations performed to evaluate general behaviors of this system s init1_trans70_set50_iter1 All initial conditions identical, stochastic effects only

14 Vulnerability 25% 40% Vulnerability of individuals in contact graph Dz flow Workplace 1 Dz flow Workplace 2 Importance of topology care Classroom Dz flow 45% Highly Vulnerable Node Dz flow 60%

15 Who is vulnerable? Vulnerability Measure of discretized contact time and degree

16 Impact of Vulnerability Probability of an Epidemic by Selection of Initially Infected 1 Number of Epidemics (per set) Random High Transmissibility Random Vulnerability High Vulnerability Who is initially infected makes a big difference

17 Computational Epidemiology as a tool for Understanding the Complex Interactions that Confront Public Health Decision Making tmpy Analyzing Complexity tmpx Important information can be buried in vast array of data available

18 Modes of Dynamical System Cluster 1 Cluster 2 Cluster Cluster 4 Cluster 5 Cluster Cluster 7 Cluster Separation of common modes through statistical methods

19 Analysis of modes Cluster curves Cluster curves Cluster curves Vulnerabilities of initially infected # of epidemics in cluster curves Cluster curves Cluster curves Cluster curves

20 Review Problems facing Public Health are complex and require techniques from a wide range of disciplines Structurally valid representations provide greater flexibility in the design and analysis of complex systems Simple examples of how to design experiments and analyze results Computational Epidemiology is an important tool in the Public Health arsenal

Agent-Based Models. Maksudul Alam, Wei Wang

Agent-Based Models. Maksudul Alam, Wei Wang Agent-Based Models Maksudul Alam, Wei Wang Outline Literature Review about Agent-Based model Modeling disease outbreaks in realistic urban social Networks EpiSimdemics: an Efficient Algorithm for Simulating