Bayesian Joint Modeling of Longitudinal Trajectories and Health Outcome: A Broad Evaluation of Mean and Variation Features in Risk Profiles and Model Assessments.

[摘要] This dissertation consists of methodology developments and applications for joint modeling of repeated measurements of health risk factors (i.e., longitudinal trajectories) and health outcome data. In the first chapter, we consider joint models that incorporate information from both long-term mean trends and short-term variability in the longitudinal submodel. We then utilize multiple both shared random-effects (MSRE) and latent class (LC) approaches to predict a binary disease outcome in the primary model. We develop simulation studies to compare and contrast these two modeling strategies; in particular, we study in detail the effects of the primary model misspecification. In the second chapter, we develop a joint modeling method that uses the individual-level longitudinal measurements to predict the occurrence of severe hot flashes in a manner that distinguishes long-term trends of the mean trajectory, cumulative change captured by the derivative of mean trajectory, and short-term residual variability. Our method allows the potential effects of longitudinal trajectories on the health risks to vary and accumulate over time. We further utilize the proposed methods to narrow down the critical time windows of increased health risks. The third chapter is a detailed study of model assessment. We evaluate six Bayesian model assessment criteria in the context of a model that simultaneously considers a set of longitudinal predictors and a primary outcome, connected through either LC or MSRE predictors. We focus on two evaluation aspects: goodness-of-fit adjusted for the complexity of the models, and prediction evaluation based on both training and test samples as well as their contrasts. An interesting result is that when the data are generated under an MSRE mechanism but fit assuming an LC mechanism, a very highly predictive ``artifact;;;; can be generated under certain scenarios. The consequence of this phenomenon is that an over-optimistic classification estimate can be built on such an artifact. The methods developed in all three papers are applied to data from the Penn Ovarian Aging Study, a 13-year longitudinal study. The goal is to explore the associations between reproductive hormone levels (follicle stimulating hormone or FSH) and symptoms in the transition to menopause (severe hot flashes).

[发布机构] University of Michigan

[效力级别] Long-term Trends and Short-term Variability

[关键词] Bayesian Joint Modeling of Longitudinal and Health Outcome Data;Long-term Trends and Short-term Variability;Latent Class and Shared Random Effect;Functional Predictor;Outcome-informed Artifact;Independent Sample Validation;Statistics and Numeric Data;Science;Biostatistics