Social Network Structure Influences Disease Transmission

Transcription

1 Social Network Structure Influences Disease Transmission presented by Bryan Lewis MPH Network Dynamics and Simulation Science Laboratory, Virginia Bioinformatics Institute, Virginia Tech Kofi Adasi, Chris Barrett, Keith Bisset, Dick Beckman, Deepti Chafekar, Jiangzhou Chen, Stephen Eubank, Bryan Lewis, Achla Marathe, Madhav Marathe, Paula Stretz, and Anil Vullikanti

2 Overview Construction of synthetic population with mobility Introduce Epi.Fast tools Comparison of different networks Epi curves / Attack Rates (similar) Vulnerability description Difference in Vulnerability distribution Sub-graphs of Vulnerability distributions Future work: Dense subgraph finding

3 Simulating realistic interactions Assignment of synthetic population to homes consistent with census Assignment of activities to individuals consistent with demographic features Gravity model used to determine locations of each individual s locations (constrained by trip distances)

4 Social Networks EpiSims uses this data to calculate each individual s event-to-event risk of infection (completely dynamic) Efficiencies can be gained by precomputing network and implementing graph theory algorithms Structure highly dependent on individual s behavior, subject to change as epidemic progresses VA_newriver_contacts.smallslice graph of VA_newriver_contacts.smallslice.uel

5 Epi.Fast methods Set of algorithms to calculate the disease percolation through contact structure Can be configured for different disease models, initial seedings, and output formats Distributed architecture allows calculations on very large populations Yields 1-fold speed improvements

6 Studying the Influence of Network Structure Consider several different synthetic populations generated with identical methods from different underlying demographics Simulate epidemics using identical disease modeling and seeding methods Compare and contrast results Epi curves / Attack rates Structure of actual epidemic networks

7 Epidemics in Structured populations 4 2 Infections 6 8 New River Valley Population Epicurves Days at various levels of transmissibility

8 Epidemics in Random Populations (Chicago-like) 1 2 Infections Random Population (based on Chicago) Epicurves Days at various levels of transmissibility 8 1

9 Epidemics in Random Populations (Virginia-like) 2 4 Infections Random Population (based on VA NRV) Epicurves Days at various levels of transmissibility 8 1

10 Attack Rates Proportions Infected 1% NRV-Full Outbreaks Random-NRV-Full Outbreaks Random-Chicago-Full Outbreaks NRV-All iterations 75% Random-NRV-All iterations Cumulative Infection Random-Chicago-All iterations 5% 25% % Transmissibility

11 Analytic approach Negligible compute times allow greater flexibility in approaches to study the network s effect on the epidemic s progress Through multiple iterations can assess vulnerability of each node in the contact network Compare the connectedness of the node measured through degree and total contact-time with it vulnerability

12 Vulnerability by connectedness bins

13 Goal Identify characteristics of nodes with vulnerability that is unusually high/low compared to others in the 25% 4% Dz flow Dz flow Workplace 1 Workplace 2 Structural Demographic Find common characteristics of high flow and low flow subgraphs Classroom Day care Dz flow 45% Highly Vulnerable Node Dz flow 6%

14 Vulnerability analysis Vulnerability scores can be adjusted across connectedness bins Vulnerability scores can be mapped onto contact network Dominant disease conduits can be identified VA_newriver_contacts_node-415_depth-2.subgraph graph of VA_newriver_contacts_node-415_depth-2.subgraph.uel 7431

15 Vulnerabilities Realistic structure VA Newriver (protopop) - Vulnerability distribution by txmibility 3 25 Frequency (~15K total) 2 Trans = 1 Freq Trans = 2 Freq Trans = 3 Freq 15 Trans = 4 Freq Trans = 5 Freq Trans = 6 Freq Trans = 7 Freq Vulnerability

16 Vulnerabilities Random structure Random Graph based on VA_nrv - Vulnerability distribution by txmibility 35 3 Frequency (1K total) 25 Trans = 1 Freq 2 Trans = 2 Freq Trans = 3 Freq Trans = 4 Freq Trans = 5 Freq 15 Trans = 6 Freq Trans = 7 Freq Vulnerability

17 Work in Progress Apply dense subgraph vulnerability analysis of contact networks Classify by shape and common characteristics unusually vulnerable (and unvulnerable) graphs Improve Epi.Fast methods to handle dynamically changing networks