Statistical Inference for the Treatment Effect in Functional AFT Model with Longitudinal Trajectories of Time-Varying Covariate

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Jincheng Shen*+ and Lu Wang and Jeremy Michael George Taylor
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: AFT model ; functional analysis ; treatment by indication

Abstract

Typical survival models that include longitudinal covariates as time-varying variables only use the information at event times and assume the hazard rate at a given time is determined by current covariate levels. We consider a situation where the hazard rate may depend on the entire history as captured by the covariate trajectory. Motivated by a prostate cancer recurrence study, we investigate estimating the coefficient for the treatment effect on cancer recurrence while also including PSA in the model. Specifically, we propose a partial functional accelerated failure time model, where the treatment variable is included in the model parametrically, and the whole history of individual longitudinal PSA trajectory is incorporated as a functional part. The asymptotic properties of the estimated treatment effects are derived using empirical process theory. We demonstrate the finite sample performance by simulation studies, and illustrate the method using the prostate cancer data.