Bayesian Extensions to Generalized Linear Mixed Effects Models for Household Tuberculosis Transmission

Avery McIntosh, from Boston University School of Public Health will present the Department of Statistical Science seminar with a talk entitled, "Bayesian Extensions to Generalized Linear Mixed Effects Models for Household Tuberculosis Transmission".

Avery McIntosh is a PhD candidate in biostatistics at Boston University School of Public Health. His principal research is on Bayesian models for tuberculosis transmission and spatio-temporal mapping of multidrug-resistant tuberculosis. His interests include the pedagogy of statistics and resampling techniques for nonparametric inference.

Abstract: Household contact studies are a mainstay of tuberculosis transmission research in particular, and of infectious disease research in general. The estimation of risk factors for latent tuberculosis transmission in household contacts of a primary case assumes that newly infected persons contracted the disease from a household contact. However, genotyping of secondary cases with active disease has shown that a substantial portion of contacts can become infected from sources other than within the household. Simulation of household contact data shows that not accounting for community-acquired infection in household contact studies can lead to substantial bias in risk factor estimates. We present a novel approach to generating unbiased estimates of household risk factors in the presence of community infection. This method has the added benefit of estimating the underlying risk of community infection for all persons in the study area. Finally, we employ the method in a household contact study of tuberculosis transmission in Vitoria, Brazil to estimate the effect of several common risk factors on disease transmission, and to estimate the overall risk of infection from the community. Preliminary results show that the method gives estimates similar to two other standard methods for two risk factors in the Brazil study, and gives estimates notably different from the competing models for two other predictors of disease transmission. These results, along with results from simulated data, indicate that standard models for tuberculosis transmission likely underestimate risk for two commonly measured predictors of disease transmission. We intend to employ this model for the estimation of community risk of infection in several distinct geographic locations, and to create a statistical software package in the R environment for easy implementation of the model.

Fri, 29 Apr 2016 - 13:00

Lecture Theatre 3, PD Hahn, Upper Campus, UCT