Recursive Partitioning and Locally Weighted Conditional Distributions for Censored Quantile Regression

• Activity Number: 569
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Kyle Rudser*+ and Andrew Wey
• Companies: University of Minnesota and University of Minnesota
• Keywords: Quantile regression ; Recursive partitioning ; Trees ; Conditional distribution ; Locally weighted ; Censored

Abstract

Cox proportional hazards regression is the most commonly used survival analysis method for adjusted analysis in practice today. However, the difficult interpretation of the hazard ratio, due in part to its relative measure of association, motivates the use of a direct summary measure with interpretation based on units of time. Censored quantile regression is emerging as a viable complement/alternative. Current censored quantile methods rely upon the fairly strong assumptions of unconditionally independent censoring or linearity in all quantiles. We expand on a recently proposed locally weighted censored quantile regression by Wang and Wang (2009) with a similar approach that uses recursive partitioning to define locality. We also examine performance of an approach based on recursive partitioning that separates borrowing information and forming contrasts. We analyze the performance on estimation and inference via simulations and found that tree based approaches have potential to provide robust estimation at little loss of precision. We also illustrate the approaches using data from a clinical trial on primary biliary cirrhosis.