Estimation of the transition probabilities condition on repeated measures in multi-state models

Abstract

The topic of joint modeling of longitudinal and survival data has received remarkable attention in recent years. In cancer studies for example, these models can be used to assess the impact that a longitudinal marker has on the time to death or relapse. Analyzes of such studies, in which individuals may experience several events, can be successfully performed by multi-state models. The goal of this work is to introduce feasible estimation methods for the transition probabilities conditionally on covariates observed with repeated measures through the use of the landmark methodology and the adaptation of existing methods for joint modeling of longitudinal and survival data. Results of the simulation studies con rm the superiority of the proposed estimator when compared to methods that do not take in consideration the effect of the covariate on the estimated transition probabilities or do not assume all the existence of repeated measures (Breslow estimator).