Primary Faculty Mentor Name

Laurent Hébert-Dufresne

Project Collaborators

Sarah Shugars, Laurent Hébert-Dufresne, Kirk Dombrowski

Status

Graduate

Student College

College of Engineering and Mathematical Sciences

Program/Major

Computer Science

Primary Research Category

Engineering & Physical Sciences

Secondary Research Category

Social Sciences

Tertiary Research Category

Biological Sciences

Presentation Title

Viable Targeted Immunization Strategies in Complex Networks: Addressing the Boundary Specification Problem

Time

12:10 PM

Location

Chittenden Bank Room

Abstract

In today’s highly connected world, contagion processes on networks are of interest and concern across the sciences, with applications ranging from biological disease, to computer viruses, to fake news. A large body of research suggests that network-based intervention strategies can be effective and cost-efficient approaches to curtailing harmful contagion in these myriad settings. Approaches based around breaking or weakening potential infection pathways of structurally important individuals are generally referred to as targeted immunization strategies, but can take many forms including vaccination, treatment, increasing access to risk prevention apparatus, and more. Theoretical and simulation-based research into these strategies often involves a number of assumptions about the quality of available data which are usually unrealistic when working with social networks. With the results of such research reflecting the efficacy of immunization strategies under these assumptions, practitioners aiming to incorporate targeted immunization in network interventions are left with little guidance on how best to apply these ideal approaches in their real-world settings. One such assumption involves the testing of these strategies on networks considered to be closed systems, whereas social network data is usually collected within some location or organizational context, excluding but acknowledging the connections outgoing from this context. In this project, we examine the effect of the boundary specification problem on the efficacy of targeted immunization strategies. Using both simulated networks and a real-world case study, we expand on a recent method which explores the full-network consequences of immunization strategies developed based on incomplete network samples—adapting the method to networks sampled in ways mimicking how boundaries are traditionally drawn in network research. The simulated network is generated via a metapopulation matrix, connecting multiple realistic local networks. The case study uses data from a mixed-methods research project focused on social networks and risk behaviors among people who inject drugs (PWID) in rural Puerto Rico. Here, location has a strong influence on network composition, but many PWID also have ties with people in nearby localities, and sometimes, as far away as the continental U.S. The efficacy of immunization strategies was benchmarked at multiple levels of boundary specification in terms of simulated final epidemic size, as both a proportion of the entire global network as well as a proportion of the subnetwork which was sampled and partially immunized. Our findings indicate that as boundary specifications shrink the sampled subnetwork relative to the size of the global network, immunization strategies implemented within these specifications fall substantially in efficacy. In terms of protecting the entire network, targeted immunization on subnetworks with the smallest boundary conditions performed only marginally better than random immunization. However, while worse in performance than when used with complete networks, targeted immunization strategies still remained somewhat effective when implemented with boundary specifications at most scales. Thus, continued research and more complete sampling methods are encouraged.

This document is currently not available here.

Share

COinS
 

Viable Targeted Immunization Strategies in Complex Networks: Addressing the Boundary Specification Problem

In today’s highly connected world, contagion processes on networks are of interest and concern across the sciences, with applications ranging from biological disease, to computer viruses, to fake news. A large body of research suggests that network-based intervention strategies can be effective and cost-efficient approaches to curtailing harmful contagion in these myriad settings. Approaches based around breaking or weakening potential infection pathways of structurally important individuals are generally referred to as targeted immunization strategies, but can take many forms including vaccination, treatment, increasing access to risk prevention apparatus, and more. Theoretical and simulation-based research into these strategies often involves a number of assumptions about the quality of available data which are usually unrealistic when working with social networks. With the results of such research reflecting the efficacy of immunization strategies under these assumptions, practitioners aiming to incorporate targeted immunization in network interventions are left with little guidance on how best to apply these ideal approaches in their real-world settings. One such assumption involves the testing of these strategies on networks considered to be closed systems, whereas social network data is usually collected within some location or organizational context, excluding but acknowledging the connections outgoing from this context. In this project, we examine the effect of the boundary specification problem on the efficacy of targeted immunization strategies. Using both simulated networks and a real-world case study, we expand on a recent method which explores the full-network consequences of immunization strategies developed based on incomplete network samples—adapting the method to networks sampled in ways mimicking how boundaries are traditionally drawn in network research. The simulated network is generated via a metapopulation matrix, connecting multiple realistic local networks. The case study uses data from a mixed-methods research project focused on social networks and risk behaviors among people who inject drugs (PWID) in rural Puerto Rico. Here, location has a strong influence on network composition, but many PWID also have ties with people in nearby localities, and sometimes, as far away as the continental U.S. The efficacy of immunization strategies was benchmarked at multiple levels of boundary specification in terms of simulated final epidemic size, as both a proportion of the entire global network as well as a proportion of the subnetwork which was sampled and partially immunized. Our findings indicate that as boundary specifications shrink the sampled subnetwork relative to the size of the global network, immunization strategies implemented within these specifications fall substantially in efficacy. In terms of protecting the entire network, targeted immunization on subnetworks with the smallest boundary conditions performed only marginally better than random immunization. However, while worse in performance than when used with complete networks, targeted immunization strategies still remained somewhat effective when implemented with boundary specifications at most scales. Thus, continued research and more complete sampling methods are encouraged.