Stacked Survival Models for Censored Quantile Regression

• Activity Number: 697
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Kyle Rudser* and Andrew Wey and John Connett
• Companies: University of Minnesota and University of Hawaii and University of Minnesota
• Keywords: Survival prediction ; Ensemble ; Survival analysis

Abstract

Inference on quantiles of survival is an attractive alternative to using the hazard ratio for comparing groups that has a meaningful interpretation based on units of time in the censored data setting. Censored quantile regression allows contrasts across groups while adjusting for other factors. Current censored quantile regression methods often rely upon the fairly strong assumptions of unconditionally independent censoring or linearity in all quantiles. We examine the use of stacked survival models in a distribution-free framework for adjusted contrasts of quantiles of survival. By minimizing prediction error, stacking estimates optimally weighted combinations of survival models that can span parametric, semi-parametric, and non-parametric models. As such, the low variance of approximately correct parametric models can be exploited while maintaining the robustness of nonparametric models. We analyze the performance on estimation and inference of constrats of quantiles of survival via simulations and found that using stacked survival models can have potential to provide robust estimation at little loss of precision. We also illustrate the approaches using lung transplantation data.