Joint Analysis of Multivariate Longitudinal Ordinal Measurements and Survival Data: An Application to Parkinson's Disease

• Activity Number: 326
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Sheng Luo*+ and Xiao Su
• Companies: University of Texas Health Science Center at Houston and University of Texas at Houston
• Keywords: beta-binomial model ; patient-reported outcomes ; no-U-turn sampler ; Hamiltonian Monte Carlo ; Cox model ; clinical trial

Abstract

Patient-reported outcomes (PRO) have been widely used as primary endpoints in many medical research areas. These outcomes are usually measured longitudinally on ordinal scales. However, a terminal event such as death or dropout can stop the follow-up process in many clinical studies. Moreover, the time to the terminal event may be dependent on the multivariate longitudinal measurements. In this article, we consider a joint model for the correlated outcomes. A beta-binomial regression is used for the multivariate longitudinal ordinal outcomes and a cox proportional hazard model is used for the failure time. These two models are linked via shared random effects. The Bayesian inference is based on samples from the posterior distributions using the novel no-U-turn sampler (NUTS), a variant of Hamiltonian Monte Carlo. Our proposed method is evaluated by a simulation study and is applied to the Long-term Study-1 (LS-1), the largest cohort of patients with early treated Parkinson's disease (PD) with 1,741 participants to determine if creatine (10 g daily) slows the progression of PD.