SleepJEPA: Learning the latent world of sleep with at-home sleep data to estimate disease risk

Sleep disturbances lead to risk for cardiovascular (CV), metabolic, and neurological diseases. While in-lab polysomnography (PSG) is the gold standard for measuring sleep disturbances, at-home PSG (hPSG) is increasingly being used and collects a similar wealth of information. However, the link between hPSG measurements and future disease risk are not well understood. We present SleepJEPA, a foundational sleep study representation model trained via a joint embedding predictive architecture that learns full night, multichannel sleep representations using hPSGs in the latent space. SleepJEPA was trained, validated, and tested with 422,035 hours of sleep signal data from 55,518 sleep studies. It accurately estimates 1- to 15-year cumulative risk using a discrete hazard loss function for 10 conditions, including angina (integrated area under the receiver operating characteristic curve at 15 years [iAUC15] = 0.74), CV disease death (iAUC15 = 0.81), congestive heart failure (iAUC15 = 0.83), coronary heart disease death (iAUC15 = 0.86), incident cognitive decline (iAUC10 = 0.79), diabetes (iAUC10 = 0.82), hypertension (iAUC10 = 0.79), obstructive sleep apnea (iAUC5 = 0.86), myocardial infarction (iAUC15 = 0.82), and stroke (iAUC15 = 0.78). We also show that these outcomes can be reliably predicted in independent cohorts, including CV disease death (iAUC10 = 0.80), coronary heart disease death (iAUC10 = 0.76), obstructive sleep apnea (iAUC5 = 0.77), and myocardial infarction (iAUC10 = 0.62). We report increased or matched performance as compared to recent sleep foundation models, such as SleepFM. Through correlational analyses and explainability approaches, we illustrate what features are most informative for risk at different horizons. Finally, we demonstrate SleepJEPA can effectively estimate sleep stages with high accuracy (F1 = 0.77 [95% CI: 0.76 - 0.77]), objective daytime sleepiness with modest performance (AUC = 0.64 [0.57 - 0.71]), and type 1 narcolepsy (AUC = 0.88 [0.68 - 0.97]), allowing for comprehensive labeling and disease risk assessment from hPSG signals.