Multimodal Sleep Foundation Model Can Predict Risk for 130 Conditions

FRIDAY, Jan. 9, 2026 (HealthDay News) -- A multimodal sleep foundation model based on polysomnography data can predict the risk for multiple conditions, including death, dementia, and myocardial infarction (MI), according to a study published online Jan. 6 in Nature Medicine.Rahul Thapa, from Stanford University in California, and colleagues curated more than 585,000 hours of polysomnography data from about 65,000 participants across multiple cohorts to develop a multimodal sleep foundation model (SleepFM), which was trained with a novel contrastive learning approach accommodating a multiple polysomnography montage.The researchers found that the model produces sleep embeddings that allow accurate prediction of future disease risk. SleepFM achieved a C-Index and area under the receiver operating characteristic curve (AUROC) of at least 0.75 for 130 conditions, including death, dementia, MI, heart failure, chronic kidney disease (CKD), stroke, and atrial fibrillation (0.84, 0.85, 0.81, 0.80, 0.79, 0.78, and 0.78, respectively). Strong transfer learning performance was demonstrated on the Sleep Heart Health Study dataset that was excluded from pretraining. Competitive performance was also seen for SleepFM on common sleep analysis tasks, achieving mean F1 scores of 0.70 to 0.78 for sleep staging, while accuracy was 0.69 and 0.87 for classification of the severity and presence of sleep apnea, respectively."We were pleasantly surprised that for a pretty diverse set of conditions, the model is able to make informative predictions," co-senior author James Zou, Ph.D., also from Stanford University, said in a statement.One author disclosed ties to Beacon Biosignals.Abstract/Full Text.Sign up for our weekly HealthDay newsletter