| Topic： | Bayesian Joint Modeling of Longitudinal Trajectories and Health Outcomes: Latent-Class vs Shared Random Effects Models |
| Speaker： | Prof. Naisyin Wang (University of Michigan) |
| Time： | 2014/12/25 (Thu.) AM10:40 - 11:30 (Tea Time: AM10:20 - 10:40, Room 821) |
| Place： | GENⅢ 837 |
| Abstract： | Joint modeling methods have become popular tools to link important features extracted from longitudinal data to a primary event. In this talk, we consider the use of mixture modeling to accommodate heterogeneity in longitudinal trajectories as well as subject-specific residual variances. Specifically, we assume that each subject's trajectory curve, characterized by random coefficients in mixed effects models, belongs to an underlying latent class with certain class features. We consider the use of either shared random effect or latent class approaches to link specific class specific features to the primary outcomes. We then compare and contrast these two modeling strategies; in particular, we study in detail the effects under mild model-misspecification. One interesting finding is that when the information from the longitudinal data is weak, the latent class approach is more sensitive to model mis-specification. We also find in certain mis-specified settings that the latent class model would result over-optimistic within-sample prediction, which is better than what can be achieved by the use of the true model. However, the same cannot be accomplished in the out-of-sample prediction. This is a unique artifact of latent class approach that is new as far as we know to the existing literature. Finally, we use data collected from women in menopausal transition, trajectory mixture models built on the Bayesian penalized splines framework and multiple examples to illustrate the strengths of the proposed approaches in practice. This is joint work with Bei Jiang, Michael and Mary Sammel. |