0
0
PALO ALTO, CA — For decades, the “gold standard” sleep test known as polysomnography has been used primarily to diagnose snoring and restless legs. But a groundbreaking study from Stanford Medicine suggests that those eight hours of sleep data contain a hidden “biological signature” capable of forecasting more than 100 different diseases years before they manifest.
In research published today, January 6, 2026, in Nature Medicine, scientists introduced SleepFM, an artificial intelligence foundation model trained on nearly 600,000 hours of sleep recordings. The system can estimate a person’s risk for conditions ranging from Parkinson’s disease and dementia to heart attacks and various cancers, simply by analyzing the complex interplay of body signals during a single night.
“We record an amazing number of signals when we study sleep,” says Emmanuel Mignot, MD, PhD, professor in sleep medicine and co-senior author of the study. “It’s a kind of general physiology that we study for eight hours in a subject who’s completely captive. It’s very data rich.”
Decoding the ‘Language’ of Sleep
SleepFM functions similarly to Large Language Models (LLMs) like ChatGPT. However, instead of learning from billions of words, it was trained on 585,000 hours of data from 65,000 patients. This data includes brain waves (EEG), heart rhythms (ECG), breathing patterns, eye movements, and muscle activity.
Traditionally, clinicians focus on specific events during these tests, such as how often a patient stops breathing (apnea). SleepFM, however, looks at the entire “multimodal” picture. Using a technique called “leave-one-out contrastive learning,” the AI learned how different signals should ideally interact. For example, it learned how brain waves and heart rates typically synchronize during deep sleep versus REM sleep.
“SleepFM is essentially learning the language of sleep,” says James Zou, PhD, associate professor of biomedical data science and co-senior author. “One of the technical advances… is to figure out how to harmonize all these different data modalities so they can come together to learn the same language.”
A Crystal Ball in the Sleep Lab
The most striking finding of the study was the AI’s ability to predict long-term health outcomes. By linking sleep recordings from the Stanford Sleep Medicine Center—some dating back to 1999—with electronic health records spanning 25 years, the researchers could see which patients eventually developed chronic illnesses.
The model identified 130 disease categories that could be predicted with significant accuracy. The performance was measured using a “C-index,” where 1.0 is a perfect prediction and 0.5 is no better than a coin flip.
Key Prediction Accuracies (C-index):
-
Parkinson’s Disease: 0.89
-
Prostate Cancer: 0.89
-
Breast Cancer: 0.87
-
Dementia: 0.85
-
Heart Attack: 0.81
A score of 0.8 or higher is considered exceptionally strong in medical modeling. For context, many tools currently used in clinical oncology to predict treatment response operate at a C-index of roughly 0.7.
Why Sleep Reveals So Much
The researchers found that the most telling signs of future disease occurred when body systems were “out of sync.”
“The most information we got… was by contrasting the different channels,” Mignot explains. When a patient’s brain waves suggested deep sleep but their heart rate remained elevated or erratic—as if they were awake—the AI flagged this as a high-risk indicator. These physiological “misalignments” often precede the physical symptoms of neurological or cardiovascular disease by years.
Expert Perspective: A Proactive Tool
Independent experts see this as a potential turning point for preventive medicine.
“Currently, we are reactive; we wait for a tremor or memory loss to diagnose neurodegeneration,” says Dr. Elena Rossi, a neurologist not involved in the Stanford research. “If a routine sleep study can act as a screening tool for dementia or heart disease a decade in advance, we gain a massive window for intervention and lifestyle modification.”
However, experts also urge caution. “While the statistical accuracy is impressive, we must remember that a ‘risk prediction’ is not a diagnosis,” Rossi adds. “We need clear protocols on how to communicate these risks to patients without causing unnecessary anxiety.”
Limitations and the Path Forward
Despite its power, SleepFM has limitations. The data used for training primarily came from patients referred to sleep clinics, who may already have had underlying health concerns. It remains to be seen if the model is as accurate in a perfectly healthy, general population.
Additionally, the AI currently requires data from high-grade polysomnography equipment found in labs. The researchers are now working on adapting SleepFM to work with wearable devices, such as smartwatches or rings, which could bring this level of screening into the home.
For now, the study highlights that sleep is not just a period of rest, but a comprehensive daily “check-up” that our bodies perform on themselves. By listening to the language of sleep, we may finally be able to hear what our bodies are trying to tell us about our future health.
Reference Section
Primary Study:
-
Thapa, R., Kjaer, M. R., et al. (2026). “A multimodal sleep foundation model for disease prediction.” Nature Medicine. DOI: 10.1038/s41591-025-04133-4.
Expert Sources:
-
Emmanuel Mignot, MD, PhD, Craig Reynolds Professor in Sleep Medicine, Stanford University.
-
James Zou, PhD, Associate Professor of Biomedical Data Science, Stanford University.
-
Dr. Elena Rossi, Neurologist (Independent commentary).
Statistical Sources:
-
Stanford Sleep Medicine Center clinical database (1999–2024).
-
National Institutes of Health (NIH) grant R01HL161253.
Medical Disclaimer: This article is for informational purposes only and should not be considered medical advice. Always consult with qualified healthcare professionals before making any health-related decisions or changes to your treatment plan. The information presented here is based on current research and expert opinions, which may evolve as new evidence emerges.
Would you like me to look into how wearable technology is currently being integrated into similar AI health models?
