Predicting Clinical Events Using Bayesian Multivariate Linear Mixed Models with Application to Scleroderma

• Activity Number: 372 - SPEED: SPAAC SESSION IV
• Type: Topic-Contributed
• Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
• Sponsor: Section on Risk Analysis
• Author(s): Jisoo Kim* and Ami Shah and Laura K. Hummers and Scott Zeger
• Companies: Johns Hopkins University and Johns Hopkins University and Johns Hopkins School of Medicine and Johns Hopkins University
• Keywords: sequentially-updated prediction; Bayesian hierarchical models; longitudinal profiles; multivariate mixed models; scleroderma

Abstract

Scleroderma is a chronic autoimmune disease in which a patient’s disease state is manifest in several irregularly spaced longitudinal measures of lung, heart, skin and other organ systems. The key events of interest represent threshold crossings of two of those longitudinal measures indicating potentially life-threatening loss of sufficient heart and/or lung function. In this paper, we take a Bayesian approach to characterize each individual's future trajectory in multivariate space using mixed models of the multivariate longitudinal outcomes and their dependence on clinically-relevant predictor variables. We calculate the joint posterior distribution of the critical lung and heart events as a functional of the mixed model fixed and random effects and develop a cross-validated, sequential prediction algorithm (CVSP) for multivariate longitudinal data. As additional data are observed during a patient's visit, the algorithm sequentially produces updated prediction distributions for the future longitudinal trajectories and for the two critical events. The updated prediction distributions are obtained without refitting the model by using a K-fold cross-validation method.