According to Stanford Medicine researchers, a new artificial intelligence system can identify long-term disease risks using data from a single overnight sleep study.
The model, called SleepFM, was trained on more than 585,000 hours of polysomnography recordings from about 65,000 participants, according to findings published Jan. 6, 2026, in Nature Medicine.
Polysomnography tracks brain activity, breathing patterns, heart rhythms and other physiological signals during sleep. Stanford’s team said the model was built to work across different clinical environments, including labs where certain monitoring channels are missing or recorded differently.
Using diagnosis codes from electronic health records to follow patients over time, SleepFM was evaluated across more than 1,000 disease categories and showed strong predictive performance for 130 conditions. Those included heart attack, heart failure, stroke, atrial fibrillation, chronic kidney disease and dementia, along with broader groupings such as cancers, pregnancy-related complications and mental health disorders.
A further study has been carried out by Stanford Medicine to understand predictions made by SleepFM and to see if wearable devices can one day make similar screenings possible, according to Stanford Medicine.
In a separate Nature Medicine study published the same week, European researchers reported progress on remote Alzheimer’s biomarker testing through the DROP-AD project. The study enrolled 337 participants across seven centers and analyzed finger-prick blood collected on dried plasma cards. According to a summary by NIHR, levels of the Alzheimer’s-linked marker p-tau217 closely matched standard blood draws and predicted cerebrospinal fluid biomarker positivity with an AUC of 0.864. The approach could increase testing availability, but more validation is needed, according to researchers.
A global media for the latest news, entertainment, music fashion, and more.
