Semiparametric Estimation in the Proportional Subdistribution Hazards Model with Missing Cause of Failure

• Activity Number: 573
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jonathan Yabes*+ and Chung-Chou H Chang
• Companies: University of Pittsburgh and University of Pittsburgh
• Keywords: Competing risks ; doubly-robust ; inverse probability weighting ; missing cause of failure ; proportional subdistribution hazards

Abstract

In analyses involving competing risks, the proportional subdistribution hazards regression model is commonly used to estimate covariate effects of specific risk factors for disease. In some situations however, the actual cause of failure may be unknown or missing. To avoid bias, we develop two semiparametric estimators of covariate effects: the inverse probability weighted complete-case estimator and the augmented inverse probability weighted estimator. We study the properties of these estimators analytically and use simulations to compare their finite sample size performance to that of estimators obtained via a multiple imputation method, a naive complete-case analysis, and a method in which missing cases are treated as an extra failure type. We employ the proposed methods to estimate the effect of the type of induction therapy on the risk of malignancy or malignancy-related death among kidney transplant recipients.