Stanford scientists have developed a revolutionary AI model called SleepFM, which can predict the risk of developing over 100 health-related conditions years in advance based on data from just a single night’s sleep.
According to the study’s authors, the body emits a massive amount of signals during sleep. Until now, polysomnography (a comprehensive sleep study) was primarily used only to diagnose sleep disorders. However, SleepFM views this data through a much broader lens.
The model was trained on nearly 600,000 hours of sleep data collected from 65,000 individuals. It operates on a principle similar to AI models like ChatGPT. While ChatGPT learns human language based on text, SleepFM learns the “language of sleep” by analyzing physiological data, including brain activity, heart rate, respiration, and eye and leg movements.
Researchers compared the model’s predictions with 25-year medical histories. The results were impressive: SleepFM predicted several diseases with high accuracy, including:
-
Parkinson’s disease (89% accuracy);
-
Dementia (85%);
-
Heart attack and hypertension;
-
Cancers (including prostate and breast cancer);
-
Mental health disorders.
Interestingly, the model provides the most accurate predictions when signals from different organs are inconsistent with each other—for example, when the brain is “asleep” but the heart is functioning as if the person were awake.
In modern medicine, models with 70% accuracy are considered useful; however, SleepFM exceeded 80% in many categories. In the future, this technology will allow doctors to detect diseases long before they manifest and take preventive measures.
