Baysian Semiparametric Regression Under Competing Risks

• Activity Number: 509
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Xiaolin Fan*+ and Purushottam W. Laud
• Companies: Novartis Pharmaceuticals Corporation and Medical College of Wisconsin
• Keywords: Cumulative incidence function ; Hierarchical models ; Mixture of Polya trees ; Sub-distribution hazard ; Time-dependent covariates

Abstract

Regression on cause-specific hazards is a standard analysis of competing risks data in cases to assess the effects of covariates. As an alternative, Fine and Gray (1999) developed a proportional hazards model directly on cumulative incidence function of the cause of interest. In this presentation we use a mixture of Polya trees process as a prior on the baseline (normalized) cumulative incidence function in Fine and Gray's model. Following the general computation method of Hanson(2006), Metropolis-Hastings chains are employed using the full likelihood function. Simulation results show our estimators are quite comparable with those from the classical method and as robust as the classical method to the choice of the model for the secondary cause of death. Our approach leads to straightforward extensions to a hierarchical structure, time-dependent covariates and time varying coefficients.