A newly developed artificial intelligence model can identify the risk of more than 100 diseases using data from just one night of sleep, according to researchers at Stanford University.
The model, called SleepFM, analyses detailed sleep signals to detect early signs of serious health conditions. Researchers report that the system could transform how physicians screen for disease and assess long-term health risks.
SleepFM is a large language model designed to interpret multiple biological signals recorded during sleep. These include brain activity, heart rate, breathing patterns, and leg movements. By analysing these signals together, the AI can identify patterns linked to disease development.
The findings were published in Nature, following years of data analysis. Researchers trained SleepFM on more than 580,000 hours of sleep recordings from approximately 65,000 patients collected between 1999 and 2004. The data came from sleep clinics that routinely monitor patients overnight.
To train the model, scientists divided sleep recordings into five-second segments. These segments functioned like words in a language model, allowing the AI to learn complex sleep patterns over time. Study co-author James Zou said the system is effectively “learning the language of sleep.”
🚨 Stanford’s new AI predicts over 100 diseases from just one night’s sleep.
Stanford Medicine researchers have developed an artificial intelligence system, SleepFM, that can estimate a person’s risk of developing more than 100 diseases using data from just one night of… pic.twitter.com/lkRuZNZ3ab
— Shining Science (@ShiningScience) January 11, 2026Researchers combined sleep data with patients’ clinical health records. This approach enabled SleepFM to link sleep patterns to future medical outcomes. Using this method, the AI achieved an accuracy rate of at least 80% in predicting several major conditions.
SleepFM successfully identified elevated risk for diseases such as Parkinson’s, Alzheimer’s, dementia, hypertensive heart disease, cardiovascular disease, and both prostate and breast cancer. These results suggest sleep data alone may hold powerful clues about long-term health.
The model showed lower accuracy for some conditions, including chronic kidney disease, stroke, and cardiac arrhythmia. In these cases, detection rates fell to around 78.5%, which researchers say highlights areas for further improvement.
The research team cautioned that the study focused on patients who had already been referred to sleep clinics. As a result, the findings may not fully represent how well the model would perform in the general population. Larger and more diverse studies will be needed before widespread clinical use.
Despite these limitations, researchers believe SleepFM marks a major step forward. By using sleep as a diagnostic window, AI tools like this could one day support earlier detection, lower healthcare costs, and more personalised medical care.
