Multistate Models for Recurrent Events and Terminal Events


During the process of colorectal cancer progression, patients may experience new lesions and are censored by deaths. In this talk, we introduce a class of multistate models with error-prone dynamic covariate and semiparametric coefficients to analyse such a disease progression process. Association between recurrent events and terminal events are naturally addressed under the multistate modelling framework. Past event feedbacks are introduced through dynamic covariates, which are observed as longitudinal data with possible measurement errors. Addressing the dynamic features helps to gain better insights into the mechanism governing the occurrence of multistate event. Both time-varying and time-fixed effects from the prognostic factors are considered. To improve efficiency, we adopt the one-step backfitting algorithm. Asymptotic results of the proposed estimators are provided. Simulation study shows that our proposed model and estimation procedure perform well. We apply the model to a phase III clinical trial of metastatic colorectal cancer conducted by the French Federation of Digestive Oncology.

(Joint work with Miss Chuoxin Ma)