A new AI model developed by Stanford researchers can predict the risk of over 130 diseases using only sleep patterns, offering a powerful tool for early health monitoring. The model, called SleepFM, was created by a team led by senior co-authors James Zou and Emmanuel Mignot and demonstrates that sleep data holds deep insights into future health conditions, from dementia and stroke to diabetes and heart disease. SleepFM is trained on more than 585,000 hours of sleep data from 65,000 individuals across multiple sleep clinics. The data comes from polysomnography (PSG) recordings—comprehensive sleep studies that capture brain activity, heart rate, muscle movements, and breathing patterns. By analyzing these diverse physiological signals together, the AI learns to detect subtle patterns linked to disease risk. Rahul Thapa, a third-year computer science Ph.D. student and lead author, highlighted the complexity of working with such rich, multimodal data. With each patient’s recording lasting up to eight hours and including dozens of signals, the team faced major technical challenges. They discovered that training the AI across multiple body signals—rather than using traditional supervised learning—yielded better results. To handle missing or inconsistent data, the team developed a novel “leave-one-out” training method, allowing the model to maintain accuracy even when some data was incomplete. “We’re essentially teaching AI to understand the language of sleep,” said Zou. The model identifies population-level trends, showing that certain sleep patterns correlate strongly with future health issues. However, the researchers stress that the predictions are not diagnostic. They reflect relative risk, not certainty, and the model has not been approved by the FDA or validated in clinical settings. Looking ahead, the team envisions integrating SleepFM with wearable devices like smartwatches. While current consumer wearables capture fewer signals than PSG machines, they are rapidly advancing—Apple Watches now offer sleep apnea screening and ECG monitoring. Experts like Chibuike Ukwakwe, a medical and doctoral student researching wearable bioelectronics, believe AI-powered analysis of wearable sleep data could one day support clinical decisions. “This is just the beginning,” said Thapa. “Sleep contains a wealth of physiological information we’re only starting to understand. With AI, we can turn that information into early warnings for disease—potentially transforming how we think about prevention and health monitoring.”