Joint mixed membership modeling of multivariate longitudinal and survival data for learning the individualized disease progression

Abstract

Patients with Alzheimer’s disease (AD) often exhibit substantial heterogeneity in disease progression due to multiple genetic causes for such a complex disease. Investigating diverse subtypes of neurodegeneration and individualized disease progression is essential for early diagnosis and precision medicine. In this article we present a novel joint mixed membership model for multivariate longitudinal AD-related biomarkers and time of AD diagnosis. Unlike conventional finite mixture models that assign each subject a single subgroup membership, the proposed model assigns partial membership across subgroups, allowing subjects to lie between two or more subgroups. This flexible structure enables individualized disease progression and facilitates the identification of clinically meaningful neurological statuses often elusive in current mixed effects models. We employ a spline-based trajectory model to characterize complex and possibly nonlinear patterns of multiple longitudinal clinical markers. A Cox model is then used to examine the effects of time-variant risk factors on the hazard of developing AD. We develop a Bayesian method coupled with efficient Markov chain Monte Carlo sampling schemes to perform statistical inference. The proposed approach is assessed through extensive simulation studies and an application to the Alzheimer’s Disease Neuroimaging Initiative study, showing a better performance in AD diagnosis than existing joint models.