Semiparametric Bayesian Analysis of Matched Case-Control Studies When Exposure Variable are Measured with Error

• Activity Number: 53
• Type: Contributed
• Date/Time: Sunday, August 3, 2003 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Stat. Sciences
• Author(s): Samiran Sinha*+ and Malay Ghosh and Bhramar Mukherjee
• Companies: University of Florida and University of Florida and University of Florida
• Keywords: conditional inference ; Dirichlet Process ; Gibbs sampler ; logistic model ; prospective logistic model ; Metropolis-Hastings

Abstract

The paper considers semiparametric Bayesian analysis of matched case-control studies where the exposure variables are subject to measurement error. The case-control data consist of a binary disease variable, a completely observed vector of covariates and a vector of exposure variables measured with error. The prospective model of disease status is usual logistic model. The standard approach is to assume that there is no stratum effect on the distribution of exposure variable. In contrast, we allow stratum effects for the distribution of exposure variable across the different strata. As a result,even under the prospective logistic model, the joint conditional likelihood of disease status and exposure variable involves nuisance parameters which grow in direct proportion to the number of strata. To address this, we assume a Dirichlet Process prior with a mixing normal distribution for the stratum effects and estimate all the parameters within a fully Bayesian paradigm. The Bayesian procedure is implemented using the MCMC technique. Examples are given to illustrate the method.