Bayesian Variable Selection in Cox Proportional Hazard Models for Survival Data

• Activity Number: 586
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Kyu Ha Lee*+ and Sounak Chakraborty and Jianguo Sun
• Companies: University of Missouri-Columbia and University of Missouri-Columbia and University of Missouri
• Keywords: Cox model ; Elastic-net ; Gamma process ; Microarray ; Variable selection ; Bayesian Analysis

Abstract

In this paper, we propose Bayesian survival models with the help of special shrinkage priors for cases when the explanatory variables are of very high-dimension. The shrinkage priors are obtained through a scale mixture representation of Normal and Gamma distributions. Our proposed variable selection priors correspond to well known frequentist penalties like lasso, elastic-net, grouped lasso, and fused lasso. We assume a discrete gamma process for cumulative baseline hazard in the Cox proportion hazard model. The estimates of the coefficients for the variable selection are computed from the MCMC outputs. The special shrinkage structure in the prior will ensure that the model will be able to group and select similarly behaving covariates together. We hve successfully applied this method on simulated and real right censored survival data sets with microarray covariates.