Stanford's SleepFM Predicts Over 100 Disease Risks from a Single Night's Sleep Data

Stanford Medicine researchers have developed SleepFM, an artificial intelligence model that can predict over 100 health conditions from physiological recordings captured during a single night's sleep. The model was trained on nearly 600,000 hours of polysomnography data from 65,000 participants, leveraging 25 years of follow-up health records from Stanford's Sleep Medicine Center.

SleepFM analyzes multiple physiological signals simultaneously — brain activity, heart rhythms, respiratory patterns, and muscle movements — to identify patterns that correlate with future disease onset. Co-senior author James Zou described the approach: 'SleepFM is essentially learning the language of sleep.'

The model demonstrated particularly strong predictive performance for cancers, pregnancy complications, circulatory conditions, and mental disorders, achieving concordance index scores above 0.8. It excelled specifically at predicting Parkinson's disease, dementia, heart attacks, and several cancer types — conditions where early detection can meaningfully alter outcomes.

The research, published in Nature Medicine, suggests that routine sleep studies could serve double duty: diagnosing sleep disorders while simultaneously screening for a wide range of other health risks. Given that sleep studies are already performed millions of times annually, the approach could provide population-scale disease screening without requiring any additional data collection.