Flexible Cure Rate Modeling Under Latent Activation Schemes

• Activity Number: 66
• Type: Contributed
• Date/Time: Sunday, August 7, 2005 : 4:00 PM to 5:50 PM
• Sponsor: ENAR
• Author(s): Freda Cooner*+ and Sudipto Banerjee and Bradley P. Carlin and Debajyoti Sinha
• Companies: University of Minnesota and University of Minnesota and University of Minnesota and Medical University of South Carolina
• Keywords: Cure rate models ; Cure fractions ; Latent activation schemes ; Moment generating functions ; Bayesian hierarchical models ; Markov Chain Monte Carlo algorithms

Abstract

With rapid improvements in medical treatment and health care, many survival datasets reveal a substantial portion of patients cured. Extended survival models called cure-rate models account for the probability of a subject being cured. Popular cure models can be classified broadly into the classical mixture models of Berkson and Gage (1952) or the hierarchical classes of Chen, Ibrahim, and Sinha (1999). Recent developments in formulating Bayesian hierarchical cure models have evoked significant interest regarding relationships and preferences between these two classes of models. Our present work proposes a unifying class of cure-rate models that facilitates flexible hierarchical model-building while including both existing cure model classes as special cases. This unifying class also elucidates the relationship between classical and hierarchical cure models. Issues such as regressing on the cure fraction and propriety of the associated posterior distributions also are discussed. Finally, we offer a simulation study and illustrate with two datasets that reveal our model's ability to distinguish among underlying mechanisms that lead to relapse and cure.