Inference of Infectious Causes Using Bayesian Mixture Model with Application to Childhood Pneumonia Etiology Studies

• Activity Number: 644
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Zhenke Wu*+ and Scott Zeger
• Companies: Johns Hopkins University and Johns Hopkins University
• Keywords: latent variable model ; Bayesian clustering ; imperfect measurements ; heterogeneous laboratory measurements ; case-control design ; pneumonia etiology

Abstract

The Pneumonia Etiology Research for Child Health (PERCH) study examines the current distribution of pneumonia-causing bacterial or viral pathogens in developing countries. The measurements are comprised of multiple multivariate binary observations on a list of J pathogens, with varying degrees of precision and missingness. We assume each observation is a draw from mixture model with each mixture component being explained by one type of infectious pathogen causes. We develop a flexible clustering approach to group these heterogeneous measurements into a priori unknown number of mixture components. Besides, the method assumes sparse disease etiology. Assessments of clustering strengths and model fit can be carried out using posterior samples drawn by Gibbs Sampler. This framework allows important covariates and prior knowledge about laboratory procedures to be explicitly incorporated. We demonstrate the method's promising operating characteristics with simulation studies tailored to the motivating scientific problem. Finally, we present the results of estimating pediatric infectious pneumonia etiology from the PERCH study.