All Times EDT
Keywords: Spike-and-Slab LASSO, Longitudinal Data, Multivariate Mixed Endpoints, Variable Selection, Joint Model Estimation, Alzheimer's Disease
Multiple mixed endpoints of different kinds (discrete, continuous, count, ordinal) are ubiquitous in biomedical and biopharmaceutical studies. In part, this may be because (1) a single endpoint, often, is not adequate to describe the disease complexities, (2) investigators are interested in studying the associations among these multiple endpoints, (3) the focus of the study lies in characterizing the treatment effectiveness, or the interest is in investigating the impact of large policy initiatives. There are at least two levels of correlations inherent in this type of especially when dealing with a clustered or longitudinal design - measuring related quantities in the same individual and the relationship among the endpoints. Typically, the standard approach of inference and analysis is to model each outcome separately, ignoring the above correlations. In this presentation, we introduce spike-and-slab LASSO with mixed endpoints (SSLME) to guide estimation and efficient extraction of active predictors in a longitudinal mixed endpoint characterized by high-dimensional covariates. We illustrate our methodology through extensive simulations and real data analysis.