Nonlinear Mixture Models for Identifying Early Markers of Neurological Diseases

• Activity Number: 304 - Risk Applications for Disease, Toxicology, and Biomarker Modeling
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Risk Analysis
• Author(s): Qinxia Wang* and Ming Sun and Yuanjia Wang
• Companies: Columbia University, Department of Biostatistics and J.P. Morgan Chase, Compliance Analytics and Columbia University
• Keywords: Mixture model; Generalized linear mixed effects model; EM algorithm; Item response theory; Disease progression; Parkinson's disease

Abstract

Current diagnosis of neurological disorders often relies on late-stage clinical symptoms. Recent research suggests that changes in clinical markers may occur in a temporal ordering and can assist early diagnosis. We propose a nonlinear mixture model to identify informative disease markers, their order of occurrence, and the relationship between the occurrence and subject characteristics. Our method fits ordinal biomarkers jointly to estimate their age-dependent trajectories in longitudinal data. To capture between-subject heterogeneity, a latent binary variable in the mixture model indicates disease susceptibility. Susceptible patients are assumed to have ordinal outcomes in an adjacent-category logit model. It allows for subject-and-marker-specific inflection points to indicate a critical time when the fastest degeneration occurs. In addition, it uses subject-specific vulnerability scores shared among markers to improve efficiency. The model can be estimated with the EM algorithm. Simulation studies are conducted under various scenarios. We apply our method to data from Parkinson's Progression Markers Initiative, and show utility to aid early personalized diagnostic decisions.