Artificial Intelligence in Health Care

Stanford Medicine researchers have developed an artificial‑intelligence model, SleepFM, that uses a single night of gold‑standard polysomnography data to estimate a patient’s long‑term risk for more than 100 diseases, including dementia, heart disease, stroke, kidney disease, several cancers and even overall mortality. Trained on nearly 600,000 hours of sleep recordings from over 60,000 sleep‑clinic patients and linked to up to 25 years of their electronic health records, the model scanned around 1,000 disease categories and could predict 130 with what the team calls 'reasonable accuracy.' Co‑senior author James Zou says the work shows "sleep contains far more information about future health than we currently use," but stresses the system does not output human‑readable explanations for its risk scores. Outside expert Dr. Harvey Castro, who was not involved, cautions that SleepFM is a research breakthrough but "not yet a bedside tool" and that ranking risk is not the same as predicting outcomes or guaranteeing a disease will occur. The peer‑reviewed study, partly funded by the National Institutes of Health and published in Nature Medicine, underscores both the promise and the current limits of AI‑driven early‑warning tools that mine routine clinical data for hidden health signals.