Joint Models for Longitudinal Quantitative and Binary Data: An Application in Psychiatry

• Activity Number: 265
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Hanga Galfalvy* and Maria A. Oquendo and J. John Mann
• Companies: Columbia University and Columbia University and Columbia University
• Keywords: psychiatry ; joint model ; longitudinal data ; mixed effect model

Abstract

In longitudinal studies of psychiatric illness, there are often multiple outcomes of interest that are cross-correlated with each other and subject to measurement error. Separate analyses of the outcomes can lead to biased and inefficient estimates and increased prediction error. Joint models have been recently developed for longitudinally measured quantitative and binary outcomes. In this study, we apply these models to a longitudinal study of mood disorder patients, where subjects display different patterns of recurring depression, life stress and suicidal behavior, during a 2-year follow-up, and each of these is known to influence the others, besides depending on trait characteristics of the individual patient. We compare the results to those from single-outcome models, using proc glimmix in SAS. We find increased efficiency in estimation in the joint model compared to the corresponding single-outcome models. However, translating some of the most interesting single-outcome models (notably, those that test associations between time lagged terms) into joint models is difficult, therefore more research is needed in this area.