Phylogenetics, causal inference, and model diagnostics for longitudinal studies of infectious disease transmission and control

Ohio State University, Columbus OH

Investigators

Abstract

Project Summary/Abstract Pairwise survival analysis is a set of statistical methods for studying the transmission of disease in house- holds and other groups of close contacts. It correctly accounts for the transmission of disease from person to person, giving it greater accuracy and statistical power than statistical methods borrowed from chronic dis- ease epidemiology such as logistic and Poisson regression. Pairwise regression models can simultaneously estimate the effects of covariates (e.g., vaccination status, age, antiviral prophylaxis, or wearing a mask) on susceptibility to infection and on infectiousness, producing detailed epidemiological insights that can inform public health interventions. This project aims to extend these methods in three ways. First, we will develop efficient numerical methods to incorporate pathogen genetic sequences into parameter estimation. This information can reduce bias and variance in parameter estimates by providing partial information about who- infected-whom, but statistical inference with these data can be computationally expensive. We propose efficient Bayesian Markov chain Monte Carlo (MCMC) algorithms that use a pruning algorithm to generate proposals for transmission trees. Second, we will develop methods for assessing goodness of fit for pairwise regression models and for calculating secondary attack risks and infectiousness profiles, which are important determinants of the severity of an epidemic and the most effective methods of control. The SAR is the probability of transmission from an infectious person to a given susceptible contact, and it is an important determinant of the potential severity of an epidemic. The infectiousness profile tells us how infectiousness changes over the course of the infectious period, so it determines how long an infected person needs to take precautions around other people and how effective nonpharmaceutical interventions like quarantine and isolation can be. The epidemiological insights pairwise survival models generate are reliable only if they fit the data well, so epidemiologists and statisticians need to be able to evaluate and compare these models in a manner similar to standard regression models in biostatistics. Finally, this project will develop graphical and statistical methods to assess confounding and selection bias in infectious disease transmission data. To inform public health interventions, we must understand the requirements for reliable causal inference.