Data-Driven Multimodal Subtyping Reveals Differential Cognitive Risk and Treatment Effects in the All of Us Cohort

INTRODUCTION: Cognitively unimpaired (CU) adults show substantial variation in their risk of developing mild cognitive impairment (MCI), yet most subtyping approaches emphasize neurobiological or cognitive features rather than modifiable risk factors. METHODS: We applied a novel integrative subtyping approach to 121,322 CU adults aged [≥]50 years in the All of Us Research Program. Baseline comorbidities, lifestyle behaviors, and social determinants of health were jointly modeled to identify subtypes. Subtype specific MCI risk was evaluated using Cox model, and medication effects was assessed using double/debiased machine learning model. RESULTS: Four subtypes with distinct medical, behavioral, and social vulnerability profiles were identified. A cardiometabolic depressive subtype showed the highest MCI risk and the greatest benefit from cardiometabolic medications, whereas subtypes dominated by behavioral or social vulnerability showed limited benefit. DISCUSSION: Integrative subtyping based on upstream risk factors reveals differential cognitive risk and treatment response, supporting more targeted and equitable dementia prevention strategies. Key Words: multimodal integrative subtyping; differential cognitive risk; treatment effect heterogeneity; All of Us Research Program