New EHR‑Based AI Model Predicts 10‑Year Type 2 Diabetes Risk in Millions

Researchers presented a novel electronic health record (EHR)‑based machine‑learning model at the 2026 American Diabetes Association Scientific Sessions in New Orleans. The retrospective cohort study analyzed 3,365,464 adults aged 18‑70 receiving care at Kaiser Permanente Northern California between 2012 and 2024. Using routine clinical, demographic and publicly available socioeconomic data, the hazard‑based super‑learning model generated individualized risk estimates for 1, 3 and 10 years. The model achieved an area under the curve of 0.886 in training and 0.883 in validation, with a sensitivity of 74 % and specificity of 82 % at a high‑risk threshold of >1.2 % risk. investigators plan prospective testing to see if deployment boosts enrollment in diabetes‑prevention programs and reduces incidence.