Regression Modeling of the Cumulative Incidence Function with Missing Causes of Failure Using Pseudo Values

• Activity Number: 130
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Margarita Moreno-Betancur*+ and Aurelien Latouche
• Companies: Inserm Centre for Research in Epidemiology and Population Health and CNAM/Inserm Centre for Research in Epidemiology and Population Health
• Keywords: Competing risks ; Cumulative incidence function ; Inverse probability weighting ; Missing cause of failure ; Multiple imputation ; Pseudo-values

Abstract

Competing risks arise in survival analysis when patients may fail from several causes. Regression strategies to determine the effects of prognostic factors on crude survival quantities often assume that the cause of failure is known for all patients, but this is seldom the case. Excluding patients with a missing cause may lead to biased estimates. Several authors have addressed the problem of modeling the cause-specific hazards in this setting, but little attention has been given to direct modeling of the cumulative incidence function, defined as the probability of the event by a given time. We derived a class of regression models for this function in the missing cause setting, encompassing key models such as the Fine and Gray and additive models, by proposing two extensions of Andersen-Klein pseudo-value approach. The first extension is based on inverse probability weighting and the second on multiple imputation. Variance estimation and asymptotic properties were considered and small-sample performance was evaluated through an extensive simulation study. We analyzed data from an ECOG breast cancer treatment clinical trial to illustrate the practical value of the proposed methods.