Estimation of the transition probabilities conditional on covariates with repeated measures: A joint modeling approach

Abstract

In recent years, there has been a significant urge of interest in longitudinal and survival data modeling. This approach holds particular significance in cancer research, where it enables the evaluation of how longitudinal markers influence the event of interest. This paper aims to introduce practical estimation techniques for transition probabilities, conditional on observed covariates with repeated measurements. This innovation allows us to incorporate the trajectory of longitudinal outcomes into regression models by accommodating time-varying covariates for each individual. The results presented in this study confirm the superior efficiency of the proposed methods, which merge existing approaches for joint modeling of longitudinal and survival data with the landmark approach for estimating transition probabilities. These methods outperform approaches that do not fully account the information provided by longitudinal covariate measurements.