Bayesian Variable Selection for the Analysis of High-dimensional Censored Data

• Activity Number: 271
• Type: Contributed
• Date/Time: Tuesday, August 9, 2005 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Naijun Sha*+ and Mahlet G. Tadesse and Marina Vannucci
• Companies: The University of Texas at El Paso and University of Pennsylvania and Texas A&M University
• Keywords: Bayesian variable selection ; MCMC ; Normal approximation ; Accelerated failure time model

Abstract

We are now in an era of massive automatic data collection where thousands of variables are measured systematically on a few samples. The major task is to identify variables associated with the outcome of interest. A number of variable selection methods have been developed in the context of linear regression and classification; however, few contributions have been made for the analysis of high-dimensional data $(p \gg n)$ with censored survival outcomes. In this paper, we propose a Bayesian variable selection approach for accelerated failure time (AFT) models. We consider several survival distributions. We adopt a data augmentation approach to impute failure times for censored observations and specify mixture priors on the regression coefficients to identify promising subsets of variables. We make use of normal approximations and conjugate priors, which allow us to work with a marginalized likelihood and build a faster MCMC procedure. The proposed methods provide selection on the variables as well as prediction of the survivor function. We describe strategies for posterior inference and explore the performance of the methodology on simulated and real datasets.