Bayesian Median Regression for Censored Survival Data via Transform-Both-Sides Model

• Activity Number: 390
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Debajyoti Sinha*+ and Jian-Chang Lin and Stuart R. Lipsitz
• Companies: Florida State University and Florida State University and Brigham and Women's Hospital
• Keywords: Transform-both-sides model ; Dirichlet process ; survival analysis

Abstract

For survival data, median regression offers an useful alternative to the methods based on popular proportional hazards and accelerated failure models. Existing frquentist median regression methods for survival data often require restrictive assumptions and hard to implement complex algorithms involving nonlinear and discontinuous equations. We propose new parametric and semiparametric Bayes methods for estimating the parameters of median regression using a model with same optimal and unknown transformation to both log-survival and median. Via theoretical and simulation studies, we show better performance and properties of our estimators compared to existing estimators in various situations including where our modeling assumption is not correct. We illustrate the implementation of our estimation estimation methods and model diagnostics via reanalysis of a small-cell lung cancer study.