JL-Wang

In this talk, we explore issues to analyze longitudinal data which are not observable after an event, such as death, occurs. This triggers informative dropout because the longitudinal data are related to the event-time. Consequently, marginal approaches to model the longitudinal processes will induce bias, and an effective way to remove the bias is to model both the event and longitudinal processes simultaneously. Such an approach is termed joint modeling of longitudinal and survival data in the literature.

We will discuss several intriguing and challenging issues in joint modeling. Typically, a parametric longitudinal model is assumed to facilitate the likelihood approach. However, the choice of a proper parametric model turns out more illusive than in standard longitudinal modeling where no survival end-point is considered. Furthermore, the computational burden and stability are important concerns in the joint modeling setting. To deal with these challenges, we propose several semiparametric random effects model for the longitudinal data and the method of sieves for the survival model.