B1157
Title: Scalable Bayesian joint models for proportion outcomes and informative observation times
Authors: Ya Su - Virginia Commonwealth University (United States) [presenting]
Sanvesh Srivastava - The University of Iowa (United States)
Dipankar Bandyopadhyay - Virginia Commonwealth University (United States)
Abstract: Electronic health records (EHR) data when the visiting process is informative has gained much attention recently. A Bayesian joint modeling approach is considered for proportion outcomes via a mixed effect model and informative observation times via a counting process with an intensity function with frailty. The EHR data could include a large number of patients, and together with the intrinsic high dimension of the parameter space, it poses a challenging task for any MCMC sampler to function well. A divide-and-conquer approach is adopted with a simple adjustment on the likelihood in each subset followed by an easy combination step to approximate the posterior samples based on the original posterior. Simulation and real data analysis reveal the efficiency the algorithm achieves while maintaining accuracy.
