Passive early screening for Alzheimer's disease and related dementias using EHR comorbidity patterns.

Early identification of Alzheimer's disease and related dementias (ADRD) remains limited by specialized tests and late-stage diagnosis. The Zero-burden Risk Assessment (ZeBRA) is an AI-driven score that predicts incident ADRD up to a decade before diagnosis using only routine electronic health record (EHR) data, without laboratory tests, imaging, or questionnaires. Trained on 487,989 cases and 12,483,718 controls from nationwide U.S. insurance claims and validated on held-out National samples and two independent cohorts, ZeBRA achieved AUC = 0.93 and 0.83 in the 50+ cohort for 1-year and 10-year horizons, respectively, with positive likelihood ratios exceeding 10 in the National 50+ held-out cohort at 95% specificity and stable discrimination over time. Performance was consistent across age, sex, race, and ethnicity subgroups. In a prospective feasibility pilot, higher ZeBRA scores showed concordance with lower Montreal Cognitive Assessment (MoCA) scores, indicating greater cognitive impairment (R = -0.78, 95% CI: -0.94 to -0.37). Compared with prior EHR-based models, ZeBRA provides superior accuracy, cross-site generalizability, and noise-corrected interpretability via our novel Λ-OR attribution metric. Scalability and low burden suggest application in population-level early detection and presymptomatic trial enrichment.