Daijiworld Media Network – New Delhi
New Delhi, Jan 8: A poor night’s sleep may do more than leave one tired the next day — it could also signal serious health risks years in advance. Researchers from Stanford Medicine have developed a new artificial intelligence (AI) model that can analyse data from just one night’s sleep to predict a person’s likelihood of developing more than 100 diseases.
The model, named SleepFM, was trained on nearly 600,000 hours of sleep data collected from around 65,000 participants. The data was obtained through polysomnography, a comprehensive sleep study that records brain activity, heart rhythm, breathing, muscle movement, eye movement and other physiological signals during overnight monitoring.
Polysomnography is considered the gold standard for sleep assessment and is typically conducted in specialised sleep laboratories. Researchers noted that while these studies generate vast amounts of physiological data, only a small portion is currently used in routine sleep medicine.
“We record an incredible number of signals during sleep. It’s eight hours of detailed physiology from a subject who is completely monitored. It’s extremely data-rich,” said Dr Emmanuel Mignot, Professor of Sleep Medicine at Stanford and co-senior author of the study, published on January 6 in Nature Medicine.
Using advances in artificial intelligence, the research team built SleepFM as a “foundation model”, similar in concept to large language models, but trained to learn the “language of sleep”. The sleep data was divided into five-second segments, allowing the AI to identify complex relationships between multiple physiological signals.
The model was trained using a new technique called “leave-one-out contrastive learning”, in which one stream of data is hidden and the AI is challenged to reconstruct it using other available signals. This enabled SleepFM to integrate brain, heart, muscle and breathing data into a unified analytical framework.
After training, the model was tested on standard sleep medicine tasks such as sleep-stage classification and sleep apnea severity. Researchers said it matched or outperformed existing state-of-the-art models.
The team then paired sleep data with long-term health records from the Stanford Sleep Medicine Center, which has followed patients for decades. By analysing more than 1,000 disease categories, SleepFM was able to predict 130 conditions with notable accuracy, including cancers, heart disease, pregnancy complications and mental health disorders.
The model showed particularly strong predictive power for Parkinson’s disease, dementia, heart attacks, hypertensive heart disease, breast and prostate cancer, as well as overall mortality.
Researchers said the model’s performance suggests sleep data could serve as an early warning system for future health problems, even years before symptoms appear.
While the AI does not yet explain its predictions in simple terms, the team is developing tools to interpret what patterns the model identifies. They found that mismatches between body systems — such as the brain appearing asleep while the heart remains highly active — may be especially indicative of future disease risk.
The researchers plan to further improve SleepFM by incorporating data from wearable devices and refining its clinical applications. They noted that even models with lower predictive accuracy have already proven useful in medical decision-making.
The study involved collaborators from institutions including the Technical University of Denmark, University of Copenhagen, Harvard Medical School and others, and was supported by funding from the US National Institutes of Health and philanthropic research programmes.