Overview
- Stanford Medicine’s study, published in Nature Medicine, introduces SleepFM, an AI model that estimates long‑term disease risk from a single polysomnography recording.
- The model was trained on nearly 600,000 hours of multimodal sleep data from more than 60,000 sleep‑clinic participants, with signals split into five‑second segments for analysis.
- Pairing sleep studies with up to 25 years of electronic health records, the system identified about 130 diseases it could predict with reasonable accuracy.
- Performance was strongest for cancers, pregnancy complications, circulatory conditions and mental disorders, with reported C-indexes such as 0.89 for Parkinson’s, 0.85 for dementia, 0.84 for hypertensive heart disease, 0.81 for heart attack, 0.89 for prostate cancer, 0.87 for breast cancer and 0.84 for mortality.
- Researchers and outside experts caution that results vary by condition, the cohort reflects clinic populations, interpretability remains limited and future work includes external validation and testing with wearable data; the study reports partial NIH funding.