Flexible Multivariate Joint Model of Longitudinal Intensity and Binary Process for Medical Monitoring of Frequently Collected Data

• Activity Number: 254 - Contributed Poster Presentations: Section on Bayesian Statistical Science
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Resmi Gupta*
• Companies: Cincinnati Children's Hospital Medical Center
• Keywords: Multivariate joint model; shared random effects model; unbalanced longitudinal data

Abstract

A frequent problem in longitudinal studies (LS) is that subjects may be assessed at self-selected, irregularly spaced points in time.This induces bias especially if the availability of data is directly related to the outcome.We developed a multivariate joint model (MVJM) in a mixed outcomes framework to minimize irregular sampling bias in LS.We demonstrate our approach on data involving blood glucose (BG) monitoring throughout pregnancy and preterm birth (PB) risk among women with type 1 diabetes.A multivariate linear mixed sub-model (LMSM) for the longitudinal outcome (BG),a Poisson model for the sampling intensity (SI), and a logistic regression model for binary process (PB) is specified. The association between models is captured through shared random effects.MCMC methods were used to fit the model. MVJM provided better prediction, compared to joint model with LMSM(ignoring (SI)) and a two stage (TS) model (AUC(MVJM) = 0.63, AUC(JM) = 0.57, AUC(TS)=0.59). Higher BG level led to 36% lower odds of PB (OR(MVJM(BG))=0.64(0.39, 0.70).Higher SI led to higher chances of PB (OR(MVJM(SI))=2.31(1.53, 4.9).A simulation study is presented to illustrate the effectiveness of the MVJM approach