Bayesian Regression Models for Group Testing Data

• Activity Number: 63
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Christopher S. McMahan*+ and Joshua Tebbs and Timothy Hanson and Christopher R. Bilder
• Companies: Clemson University and University of South Carolina and University of South Carolina and University of Nebraska-Lincoln
• Keywords: Bayesian binary regression ; Group testing ; Pooled Testing ; Latent variables

Abstract

Since the seminal works of Dorfman (1943), group testing, also more commonly referred to as pooled testing, has been widely implemented to more efficiently screen individuals for infectious diseases. Depending on the goals of the study, many pooling schemes have been proposed which address issues to include preserving anonymity in estimation studies, reducing classification cost, and for quality control purposes. In general these goals are accomplished by requiring that a significant number of the individuals be assigned to multiple pools and/or be retested individually. Regretfully, the vast majority of existing group testing regression methods proceed under the assumption that each individual is assigned to exactly one pool, and are therefore inapplicable for analyzing data collected from these more complex studies. To account for more complicated group testing data structures we have developed new Bayesian regression methods that flexibly incorporate virtually all pool response data regardless of the pooling scheme. We illustrate our new methods using chlamydia and gonorrhea data collected in Nebraska.