A Sensitivity Analysis of Different Regression Models for Competing Risks Data Through a Simulation Study

• Activity Number: 553
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Yuliang Liu* and Charity J. Morgan and Gary R. Cutter
• Companies: The University of Alabama at Birmingham and The University of Alabama at Birmingham and The University of Alabama at Birmingham
• Keywords: competing risks ; sensitivity analysis ; proportional hazards ; model misspecification ; model comparison

Abstract

In order to compare the performance of three regression models: Cox's Proportional Hazard (PH) model , Exponential Accelerated Failure Time (AFT) model and Fine-Gray Proportional Subdistribution Hazard (PSH) model for the estimation of hazard ratio of event of interest for the competing risks data by different approaches, we conducted a sensitivity analysis of these models through a simulation study. The competing risks datasets were generated by simulations following latent failure or bivariate random variables methods. We found that both the Cox PH model and then exponential AFT model have better performance for the estimate of the hazard ratio for an event of interest for data generated either by the simulation using latent failure or bivariate random variables methods as compared to the Fine-Gray PSH model. While the Fine-Gray PSH model yields biased estimates of the hazard ratio, this model gives unbiased estimates of the time-averaged subdistribtuion hazard ratio. Thus, we confirmed that the Fine-Gray PSH model still provides meaningful results even if the model has been misspecified by incorrectly assuming proportional subdistribution hazards.