A single night of sleep may reveal more about a person’s future health than previously thought. This was demonstrated by scientists at Stanford Medicine, who in January 2026 developed a new type of artificial intelligence that uses data obtained during one’s nighttime sleep to predict a person’s chances of developing more than 100 different diseases. The research suggests that sleep is not only important for short-term energy and focus, but may also serve as an early indicator of long-term health outcomes.
The model, SleepFM, was trained on almost 600,000 hours of sleep data from 65,000 participants. The data was collected through a process called polysomnography, a complex sleep test carried out over a night. While these tests are normally used to diagnose sleep disorders, the Stanford team realized they provide a complex picture of how different systems in the body behave together as the person sleeps.
According to Emmanuel Mignot, a Stanford professor of sleep medicine and co-author of the study, sleep studies give us a rare chance to look at the body’s functioning in a more or less stable and controlled environment. Traditionally, only a small portion of this information is used in medical practice. By applying artificial intelligence, researchers are now able to analyze far more of these signals and uncover patterns that may be linked to future disease.
Instead of focusing on a single signal, the SleepFM model examines relationships between brain activity, heart rhythm, breathing, and muscle movement. By analyzing these interactions in five-second segments, the system learns what typical coordination looks like and how subtle disruptions may signal underlying health risks.
Researchers have discovered that mismatches in activity patterns across different body systems can reveal important health implications. For example, if the brain signals deep sleep but the heart indicates a more alert state, this discordance may be associated with an increased risk of disease. Such patterns reflect underlying stress or dysfunction in the body that traditional medical screening techniques might have more difficulty detecting.
After training the model, researchers first tested SleepFM on standard sleep-related tasks, like identifying sleep stages and assessing the severity of sleep apnea. Then, the team linked the sleep data to long-term electronic health records of patients from Stanford’s Sleep Medicine Center. In some cases, the researchers had up to 25 years of follow-up health data on a patient, enabling them to compare early sleep patterns with later disease outcomes.
Using this combined dataset, SleepFM was able to predict risk for approximately 130 different health conditions with reasonable accuracy. Predictions for cancers, cardiovascular and circulatory diseases, mental health disorders, and complications regarding pregnancy came through the strongest. The model also demonstrated strong performance in predicting diseases such as Parkinson’s disease, dementia, heart attacks, and overall mortality risk.
However, it is worth noting that despite the impressive outcomes, researchers caution that the approach is still far from ready for widespread clinical use. This is mainly due to the need for further verification that the model performs accurately across diverse populations and healthcare settings. There are also important ethical considerations related to patient privacy, data security, and how predictive health information should be communicated.
The study, in essence, represents a new direction in which artificial intelligence can be utilized in medicine. By treating sleep as a rich source of physiological data, researchers may be able to identify early warning signs of disease years before symptoms appear. As this technology continues to develop, sleep data could become an important tool in preventive care, allowing doctors to move from reacting to illness toward identifying risk and intervening sooner.
