ABSTRACT
It is becoming more and more common in longitudinal studies to collect both repeated measurements of a longitudinal marker and information on the time to an event of interest. Most often, these two processes will be somehow linked and a separate analysis of each is obviously not optimal. An important aspect of joint modeling is the ability to make prediction, and actually to make dynamic prediction, taking into account the evaluation of the longitudinal outcome to update the prediction on the risk of event. With a growing interest on personalized medicine, being able to use all information collected on the potential prognostic marker, both at baseline but also over time, in order make accurate prediction is obviously highly valuable. The question of studying the impact of such dynamic factors on a time-to-event outcome is obviously linked to the question of the inclusion of time-dependent covariates in a survival model.
