This research project will develop flexible statistical models for joint analysis of recurrent events and survival time. In long-term, follow-up studies, subjects may experience a recurrent event at multiple times, where the observation of subject-specific recurrent events is terminated by end of the study, or a terminal event such as death. Scientifically relevant problems involving such data structures are prevalent in the biomedical sciences, as well as in econometrics and engineering. A critical issue in analyzing such data is that the history of the recurrent events and the risk of terminal event are interrelated. It is thus important to jointly model the underlying stochastic mechanisms, typically, in the presence of predictor variables that are expected to affect the occurrence of recurrent events and the survival time. A key objective of this project is to expand the inferential and predictive scope of existing techniques for joint analysis of recurrent events and survival time by developing novel statistical models that relax restrictive assumptions of state-of-the-art methods. To facilitate use of the methods by researchers and practitioners, publicly available software will be developed for implementing several of the statistical models. The project will create educational and research training opportunities for graduate students and seek to foster the participation of women and underrepresented groups in the research.
This research project will develop general Bayesian modeling approaches for joint analysis of recurrent events and survival time. The modeling framework builds from Bayesian nonparametric mixtures of Erlang distributions for the survival responses, with covariate effects accommodated more flexibly than proportional hazard regression models. Different classes of joint models will be formulated by combining the nonparametric survival regression modeling methods with parametric models for the covariate-dependent recurrent event point process intensities. The joint models will capture general dependence between the recurrent event and survival processes, while allowing for heterogeneity between subjects. The primary objective is to develop a comprehensive joint modeling framework that significantly improves on model fit and predictive performance relative to the state-of-the-art shared frailty modeling methods. In the context of regression modeling for survival responses, the project will also expand the methodology in the burgeoning field of Bayesian nonparametrics. The research project has a substantial analytic component with regards to study of theoretical properties for the various models, as well as a significant computational component with regards to achieving computationally tractable model fitting. The practical utility of the new methods will be investigated with simulation studies and through applications involving analysis of data from cancer patients.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
