A team of researchers from Stanford has designed a new artificial intelligence language model, allowing AI to interpret data collected from a single night’s sleep in order to determine a patient’s risk of developing some 130 health conditions. While SleepFM, as the model is called, doesn’t allow for causal relationships, it could facilitate the work of medical staff in the future.
Using almost 600,000 hours of sleep data collected from 65,000 sleepers – including brain activity, heart rate, respiratory signals, leg movements, and eye movements – complemented by patients’ individual health files, researchers from California’s Stanford University developed a large language artificial intelligence model. SleepFM, as the project is called, allows AI to predict future diseases.
“SleepFM is essentially learning the language of sleep”, stated James Zou, associate professor of biomedical data science at Stanford and co-author of the study.
According to the scientists, who published an article regarding their study in the journal Nature, SleepFM was able to predict whether or not a patient would develop Parkinson’s disease, Alzheimer’s disease, dementia, hypertensive heart disease, heart attack, prostate cancer, and breast cancer at least 80% of the time.
A patient’s death could be predicted in 84% of the cases, while it was capable of predicting chronic kidney disease, stroke, and arrhythmia in at least 78% of cases. Of the 1,000 possible diseases tested for, the AI model could be used for some 130 with medium to high accuracy.
“We record an amazing number of [health] signals when we study sleep. 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”, said Emmanuel Mignot, professor in sleep medicine at Stanford.
Promising but…
While SleepFM might well play an important role in the future of how science detects diseases of all kinds, the researchers point to the fact that the AI model should never be used on its own without a qualified medical team to interpret the data.
Firstly, because AI cannot, in most cases, successfully determine causal relationships. Secondly, because only a team of (specialised) doctors can next decide on which treatment to start in order to cure or prevent any diseases identified by SleepFM. Moreover, as the data used by the model result from analysis at specialised sleep institutions, only those who have access to such specialist centres will be able to have their data interpreted by the model, which isn’t the case for many who live in rural areas, for example.
I saw tons of overhyped news & tweets about this new @NatureMedicine SleepFM paper – AI that supposedly "predicts 130 diseases from one night of sleep". Time to pop this bubble. pic.twitter.com/lrshBOq64W
— Dvir Aran (@dvir_a) January 10, 2026
In the next step of their research, the Stanford scientists will be looking to integrate data from wearables into SleepFM’s database. This could further improve the model’s reliability and possibly open up the technique to those living further from specialised sleep centres.
