Stanford AI uses a night's sleep to predict 130 diseases

By EngineAI Team | Published on January 14, 2026
Stanford AI uses a night's sleep to predict 130 diseases
SleepFM, a new AI foundation model developed by Stanford researchers, can predict more than 130 medical disorders, including Parkinson's, dementia, and heart attacks, from a single overnight sleep recording.

The specifics:

The model was trained using 600K hours of sleep data from 65K subjects, examining heart rate, respiration, muscle signals, and brain waves.

The model identified an imbalance in bodily signals, such as a beating heart and a brain in deep slumber, as a symptom of impending illness.

The scientists tested predictions across more than 1,000 illness categories by connecting sleep data to 25 years' worth of Stanford Sleep Clinic health records.

Parkinson's disease (89% accuracy), dementia (85% accuracy), heart attacks (81% accuracy), and overall mortality risk (84% accuracy) were all predicted by SleepFM.

Even though we sleep for a large portion of our lives, there is still a lot to discover about what data from that period may show. Overnight recordings may serve as an early warning system, according to SleepFM, and as wearable technology advances, predictive health monitoring may eventually shift from sleep labs to your wrist.