Abstract

Multivariate longitudinal datasets occur frequently in many applications including medical studies. Modeling such datasets is challenging since the within and between outcomes dependence need to be handled simultaneously. We develop a hierarchical Bayes model where such dependence can be modeled by suitable prior distributions, and variable selection can be performed simultaneously. Our model is flexible since we incorporate different variants of Brownian motions (e.g. fractional Brownian motion, IOU process etc.) to model the subject-specific random effects, and also consider time-varying effects of the covariates. We use Markov Chain Monte Carlo (MCMC) for estimating the model parameters, and use a super-learner by combining different models altogether for better prediction. Our proposed model is used for analyzing a dataset from a clinical trial conducted by Tata Medical center, Kolkata, where lymphocyte counts, neutrophil counts and platelet counts are measured from the patients longitudinally. Extensive simulation studies are also performed to assess the effectiveness of the proposed model over the other traditional models.