A New Class of Joint Models for Longitudinal and Survival Data Accommodating Zero and Nonzero Cure Fractions

• Activity Number: 312
• Type: Contributed
• Date/Time: Wednesday, August 11, 2004 : 8:30 AM to 10:20 AM
• Sponsor: General Methodology
• Author(s): Yueh-Yun Chi*+ and Joseph G. Ibrahim
• Companies: University of North Carolina, Chapel Hill and University of North Carolina, Chapel Hill
• Keywords: joint models ; cure rate models ; random effects ; breast cancer trial

Abstract

Joint models for a longitudinal response process and time-to-event have recently received a great deal of attention in cancer and AIDS clinical trials. Models for repeated measures and event time distribution are assumed to be independent given a common set of latent random effects. We extend the latent process model to multivariate repeated measures, and propose a novel approach to construct the conditional likelihood of the time-to-event, which accommodates both zero and nonzero cure rate structures. The occurrence of a cancer is assumed to be initiated by the mutation of some potential metastasis-competent tumor cells. The rate of generation of metastasis-competent tumor cells over time is modeled as a function of the true longitudinal trajectory as well as baseline covariates, and the promotion time for a potential cell to become a detectable tumor is independently parametrically modeled. A Bayesian paradigm is adopted to facilitate the estimation process and ease the computational complexity. The methodology is applied to a real dataset from an International Breast Cancer Study Group trial.