Online Program Home
My Program

Abstract Details

Activity Number: 172
Type: Contributed
Date/Time: Monday, August 1, 2016 : 10:30 AM to 12:20 PM
Sponsor: Biometrics Section
Abstract #321279
Title: Bayesian Variable Selection for Multivariate Count Data with Excess Zeros: Application to the Pediatric HIV/AIDS Cohort Study
Author(s): Kyu Ha Lee* and Brent Coull and Jacqueline R. Starr
Companies: The Forsyth Institute and Harvard T.H. Chan School of Public Health and The Forsyth Institute
Keywords: Bayesian variable selection ; Markov chain Monte Carlo method ; multivariate regression analysis ; spike-and-slab prior ; zero-inflated Poisson regression

In this paper, we examine the associations between the microbiota and oral health and HIV status from the pediatric HIV/AIDS cohort study (PHACS). The primary outcome is the bacterial count determined by the HOMIM assay. A special feature of the data is the excess number of zeros in bacterial counts from multiple organisms. One approach to analyze the data is to fit a univariate zero-inflated regression model for each of outcomes to account for the non-normal distributions of bacterial taxa that are infrequently observed. However, bacteria exist in communities and certain taxa tend to be found in association with each other and the univariate approach does not take into account underlying dependence among the outcomes. In this paper, we propose a Bayesian variable selection method for multivariate zero-inflated Poisson models that flexibly accommodates the underlying correlation between outcomes into variable selection process. We show via simulation that our proposed method has the satisfactory variable selection accuracy. We successfully apply the approach to PHACS data and identify a subset of species whose prevalence is associated with HIV status and oral health.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2016 program

Copyright © American Statistical Association