An Integrative Approach for Subtyping Mental Disorders Using Multimodal Data

Integrating multimodal biological, cognitive, and clinical data is crucial for uncovering distinct psychiatric disease subgroups, enabling precision diagnosis, personalized treatment, and more targeted drug development. However, a significant gap remains between traditional clustering approaches and the growing need for advanced methods that can integrate and jointly analyze multimodal biological and clinical data to achieve more biologically meaningful subtyping. This study introduces the Mixed INtegrative Data Subtyping (MINDS) method, a Bayesian hierarchical joint model designed to identify subtypes of Attention-Deficit/Hyperactivity Disorder (ADHD) and Obsessive-Compulsive Disorder (OCD) in adolescents using multimodal data from the Adolescent Brain Cognitive Development (ABCD) Study. MINDS integrates clinical assessments, neuro-cognitive measures, and neuroimaging biomarkers while simultaneously performing clustering and dimension reduction. By leveraging Polya-Gamma augmentation, we propose an efficient Gibbs sampler to improve computational efficiency and provide subtype identification. Simulation studies demonstrate the superior robustness of MINDS compared to traditional clustering techniques. Application to the ABCD study reveals more reliable and clinically meaningful subtypes of ADHD and OCD with distinct cognitive and behavioral profiles. These findings show the potential of multimodal model-based clustering for advancing precision psychiatry in mental health.