Proportional Hazards Regression for the Analysis of Failure Time Data with Outcome-Dependent Sampling and Dependent Censoring

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Hui Zhang*+ and John David Kalbfleisch and Douglas E. Schaubel
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: Cox model ; Dependent censoring ; Outcome-dependent sampling ; Survival analysis

Abstract

Outcome-dependent sampling (ODS) is an efficient and cost-saving sampling scheme, wherein subjects selected into the study based on the outcomes of interest (i.e., death, survival). Most methods for analyzing ODS-based data have an underlying assumption that subjects are censored in a manner independent of the failure rate; an assumption which is often violated. For example, renal failure patients may withdraw from dialysis due to failing health, an issue which could produce substantial bias if ignored. In this paper, we develop a novel double-inverse-weighting scheme which combines weights corresponding to the probability of remaining uncensored and the probability of being sampled. The proposed estimators of the regression parameter are shown to be consistent and asymptotically normal. The proposed methods are applied to the Dialysis Outcomes and Practice Patterns Study (DOPPS) data.