Researchers at Stanford Medicine have developed an artificial intelligence model capable of predicting a person’s risk of developing more than 100 diseases using data from just one night’s sleep.
The AI system, known as SleepFM, was trained on nearly 600,000 hours of sleep data collected from about 65,000 participants, according to a study published in Nature Medicine. The dataset includes a wide range of physiological signals recorded during sleep, such as brain activity, heart rate, breathing patterns, eye movements, and muscle activity.
Stanford Medicine said the model is among the first AI systems designed to analyse sleep as a comprehensive indicator of long-term health risk, rather than focusing on individual organs or conditions.
“From an AI perspective, sleep is relatively understudied,” said Dr James Zou, associate professor of biomedical data science at Stanford and a co-author of the study. “Despite sleep being such a fundamental part of life, most AI research has focused on pathology or cardiology rather than sleep.”
The researchers trained SleepFM using polysomnography data collected at Stanford’s sleep clinic between 1999 and 2024 from patients aged two to 96. The data was paired with electronic health records and broken into five-second segments, similar to how large language models process text.
In tests, SleepFM performed well on standard sleep-analysis tasks such as identifying sleep stages and diagnosing sleep apnea. More significantly, it was able to predict 130 diseases with reasonable accuracy, including cancers, cardiovascular conditions, mental disorders, and neurodegenerative diseases.
The model achieved particularly strong results in predicting Parkinson’s disease, dementia, heart disease, several cancers, and mortality, with concordance index scores exceeding 0.8, a threshold considered clinically useful by researchers.
The study comes as major technology companies expand into healthcare AI, with OpenAI and Anthropic recently launching specialised health-focused tools. However, concerns remain around data privacy and the risk of AI systems generating inaccurate medical advice.
Stanford researchers said further work is needed to improve SleepFM’s accuracy and interpretability, including incorporating data from consumer wearables.
