Print

Modeling and Simulation of Complex Social Dynamics for Catastrophic Event Preparedness and Response 

1/1/2012

epstein_joshua_1

Joshua Epstein’s agent-based computational modeling project will powerfully advance the DHS mission of mitigating and preparing for catastrophic events and health crises.

Abstract

Objective

 

The Modeling and Simulation of Complex Social Dynamics for Catastrophic Event Preparedness and Response Project advanced the DHS Mission of mitigating and preparing for catastrophic events and health crises, such as Pandemic Flu (H1N1) and bioterror releases of airbornechemical or biological agents. The project began with four major lines of work containing the development and application of activities that wereunderway in PACER I.

The original four lines were:

1. Deepening the representation of Behavioral Factors in Crisis Preparedness and Response modeling, incorporating current research incognitive science in collaboration with Dr. John Cacioppo, the Tiffany and Margaret Blake Distinguished Professor and Director of Centerfor Cognitive and Social Neuroscience at the University of Chicago, and recognized expert on mass behavior. Dr. Epstein and Dr.Cacioppo co-hosted a workshop in July, 2011 to further the field of Computational Social Neuroscience. Prior to this event, they collaborated through a series of meetings to discuss and identify key factors of cognition and decision making to be represented withinEpstein's models. Specifically, they addressed emergent social structures, the neural, hormonal, cellular, and genetic mechanisms thatsupport them, and the consequent behaviors (Cacioppo & Berntson, 1992, 2004; Cacioppo et al., 2000, 2002; Cacioppo et al., 2007;Decety & Cacioppo, 2010). These extend Epstein's published work on the effect of fear contagion on epidemic dynamics (Epstein, et al2008).

2. Continuing development of the award-winning PACER Large-Scale Agent Model applied to the Science of Surge. This has a US National version (300 million software individuals) and a new Global Scale version (6.5 billion individuals) with unprecedented performance.PACER is truly the world leader in this area.

3. Continued development of Hybrid Plume-Agent Modeling for Large-Scale Urban Evacuation Design. This models the interaction ofairborne toxic plumes and human evacuation dynamics (e.g., traffic congestion, compliance with shelter-in- place directives), visualizingthem both in 3-D. This work was published in PLoS ONE, Volume 6, Issue 5.

4. Continued collaboration with Dr. Ron Brookmeyer, Professor of Biostatistics, and Nick Reich at The Johns Hopkins Bloomberg School ofPublic Health, on the development of real-time statistical tools for estimating disease parameters for novel pathogens, essential for theoptimal timing of interventions. Transitions between institutions delayed the pace of this particular collaboration (Reich from the JohnsHopkins University to the University of Massachusetts Amherst School of Public Health and Health Services; Epstein from the BrookingsInstitution to the Johns Hopkins University).

During Year 2, the fourth line of work was discontinued and the research scope was extended to include the objectives of a) incorporating abroader range of behavior (fear, distrust, biased risk appraisal, emotional contagion) in the modeling, and b) making the interactive large scalemodel ILSAM Graphical User Interface (GUI) more transparent and concretely useful to DHS. During Year 3, the research scope was furtherincreased to include the development of a method to test model parameters, the establishment of formal collaboration with the Institute forComputational Medicine (ICM) in Computational Social Neuroscience, the creation of a new Twitter-based tool for public health surveillancemonitoring, and the refinement of the GUI and technical requirements for the ILSAM. The Twitter-based tool for public health surveillancemonitoring was introduced to address our discontinued goal of real-time statistical tools for estimating disease parameters for novel pathogens

Completed Project Overview

 

YEAR ONE: We were extremely successful in Year One, achieving all of the stated goals and adding new areas of research. We completed and published twoof our most complex models, significantly expanded our understanding and incorporation of social neuroscience, continued the development ofreal-time statistical tools for estimating disease parameters for novel pathogens, and developed the ILSAM and IRED (published in the Journal of Homeland Security).


YEAR TWO: Following our successful first year, we increased our Year Two scope to involve new areas of research. We continued the three major lines ofwork to further the development and application of activities that were underway in PACER I and developed our work in the following ways:

− Expert Review Recommendations: Incorporated the suggestions and revisions identified during the subject matter expert review of themodeling and simulation project.

− Identify Containment Strategies: Used the LSAM to identify containment strategies that will be robust--that is, effective under a widerange of plausible behaviors, such as fear-induced vaccine refusal, or absenteeism among emergency health care workers.

− Identify Evacuation/Behavioral Strategies: Analogously, levels of shelter-in-place, car-pooling, and traffic-aware routing sufficient topermit effective evacuation of Los Angeles are computable in the Hybrid Model, and will apply to other large urban settings.

− Used cognitive science insights to frame risk communications to minimize morbidity, mortality, economic loss, and social disruptionunder novel stress, uncertainty, limited information, and resource constraints.By the end of Year Two, the suite of models - LSAM, GSAM, ILSAM, IRED, and the Hybrid Plume-Agent Model - had been refined to addressthe provision for the public health and medical needs of the American people in multiple catastrophic event scenarios.



YEAR THREE: Year Three research was centered on the continuous development of the Interactive Large Scale Agent Model (ILSAM), a powerful interactivedesktop simulator. Senior Software Engineer, Jon Parker, modified the original ILSAM to address the main concerns identified in the first ExpertGUI Review. Parker explored and revised the amount of memory needed to successfully run the interactive model, as well as the effects of scalingdown the population and/or number of zip codes. Mac hardware was purchased to allow Parker and his team to remedy compatibility andoperating system complications. A second Expert Review was held June, 2013. Thirteen computer scientists, software developers,epidemiologists, and public/disaster health researchers attended a two-day workshop where they ran, assessed, and amended code for the ILSAMand the Inter-Region Epidemic Dynamics (IRED) Model. These thirteen reviews and the initial review findings are also discussed in Section IV.Representations of Behavioral Factors in Crisis Preparedness and Response modeling continued to include the most recent and relevant research incognitive science. These research areas include modeling of human disease processes, algorithms-based disease risk prediction, and biomedicalimage understanding. Epstein collaborated with ICM faculty, as well as Decision Research, Inc., to further identify, develop, and refine key factorsof cognition and decision making. Year 3 research has primed Epstein's PACER models to incorporate and explore these cognitive science basedbehaviors in a range of catastrophic public health or bioterror crises scenarios. Simulations of vaccine refusal, flight, contagious fear, and theresulting consequences to disease progression can include diverse and changing biological, spatial, and psychological-behavioral information sets.This research is providing insight on boundedly rational behavior under extreme stress and its consequences for crisis management.We developed novel data collection tools to inform our PACER models.

Parker pioneered a Twitter-based framework for tracking public healthcondition trends, improving public health monitoring and surveillance approaches. Unlike existing approaches, the framework provides a broad apriori identification of emerging public health conditions instead of a specific illness (e.g., influenza). This work has been published in theProceedings of the 5th Annual International Conference on Advances in Social Networks Analysis and Mining (ASONAM).Model Tests were conducted using Latin Hypercube Sampling (LHS), a modified monte-carlo method, to sample from parameter space (McKay,Beckman et al. 1979; Blower and Dowlatabadi 1994). As all of the parameters are varied simultaneously, we used partial rank correlationcoefficients (PRCCs) to evaluate the statistical relationships between the parameters and output variables.All of the models developed, and the methods developed to analyze model output, can be used to identify strong nonlinearities, tipping points, andinnovative containment strategies. These models do not eliminate uncertainty or the need for judgment. Even so, they do ensure that judgmentsPACER FINAL REPORT 79are considered against the full range of alternatives in a rigorous fashion to help the decision maker determine which uncertainties are mostimportant and help to bound them.

Under the direction of Dr. Joshua M. Epstein, this project has built models to study and explore Pandemic Flu (H1N1) and the dynamics of diseasespread, as well as developed innovative methods to address bioterror releases of airborne chemical or biological agents. Throughout Years 1, 2,and 3, the suite of models - LSAM, GSAM, ILSAM, IRED, and the Hybrid Plume-Agent Model - were continuously refined to address theprovision for the public health and medical needs of the American people in multiple catastrophic event scenarios. By the end of the project,Epstein and his team achieved the major Research Objectives and expanded the scope of the project by developing more advanced methods of analysis and incorporating the results of two Expert Reviews.