Bridging Network Science and Causal Inference to Improve Public Health Response to Outbreaks and Endemic Diseases

Harvard University D/B/A Harvard School Of Public Health, Boston MA

Investigators

Linked publications, trials & patents

Abstract

Understanding the dependencies among individuals with regard to disease status is critical to investigating spread of communicable disease as well as effectiveness of mitigation strategies. Rather than assuming that the members of a population of interest are fully mixed, as is commonly done with mass-action models in infectious disease epidemiology, the network approach enables individual-level specification of contact patterns by considering the structure of connections among the members of the population. By representing individuals as nodes and contacts between them as edges, network representation of populations facilitates the identification of individuals who drive epidemics and thereby provides insight into promising interventions to control disease spread. Network science in its early days focused exclusively on static networks; hence, use of network science methods to study disease spread was often limited to the study of dynamical processes on static networks. Recent advances in epidemic control underscore the need to move research focus to dynamic networks, and study of temporal networks has emerged as its own subfield. The growing literature on dynamical processes on dynamic networks underscores a variety of challenges–for example, nonpharmacological interventions, such as mask wearing and social distancing, promoted during the peak of the COVID-19 pandemic. Factors that currently limit the public health benefit of network analyses include lack of available network data (especially on dynamic networks) and lack of suitable quantitative methods for uncovering network mechanisms, accommodating partially observed networks, and making causal inferences from incomplete observational network data. The goal of the proposed project is to develop such methods in order to increase the pace and accuracy of research in infectious diseases and endow the public health community with these important tools. Our specific aims are the following. Aim 1: To identify individual-level mechanisms for the formation and evolution of dynamic, fully observed person-to-person contact networks. Aim 2: To develop Bayesian statistical inference methods for settings in which networks are only partially observed. Aim 3: To develop methods that permit causal inferences on networks in settings where the outcome of one participant may depend on the exposures, covariates, or outcomes of others. We subdivide the final aim into two distinct components: Aim 3a: To develop a Bayesian approach for causal inference of network effects; Aim 3b: To extend the approach of Aim 3a to allow for partially observed networks or networks whose properties are known with error. We also propose to implement these methods in open-source software, using a combination of R and Python, to facilitate their dissemination and adoption. We believe that the research proposed here can help harness the type of network data we may realistically have available to address the types of causal questions that are at the core of public health.