Parametric Regression Models for Cumulative Incidence Functions Based on Generalized Odds-Rate Models and Modified Logistic Functions

• Activity Number: 343
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Haiwen Shi*+ and Yu Cheng
• Companies: University of Pittsburgh and University of Pittsburgh
• Keywords: competing risk ; cause-specific hazard function ; cumulative incidence function ; long-term incidence ; modified three-parameter logistic mode ; parametric modeling

Abstract

We propose a new regression model for cumulative incidence functions (CIFs) in competing risk data. Fine and Gray (1999) developed a semiparametric regression model for CIFs and Jeong and Fine (2007) extended a generalized odds-rate model to the competing risk setting with Gompertz models as the baseline CIFs. We propose to use a modified logistic function (Cheng, 2009) for the baseline CIFs in the parametric modeling. The inference procedure relies on standard maximum likelihood methods. Extensive simulations are run to evaluate the performance of the coefficient estimators from our proposed model, the Gompertz model and the semiparametric model. We also apply the three models to Cache County Study data examining the effects of covariates on the CIF of dementia. Our numerical results suggest that our parametric model be more flexible than the Gompertz model and have comparative performance with the semiparametric model for the primary cause. In addition, our parametric model considers the fact that the cumulative probabilities from different causes should eventually add up to one. The other methods overlooked the fact and may lead to misleading results for competing causes.