Sleep Temporal Entropy as a Digital Biomarker of Sleep Fragmentation for Cardiometabolic and Mortality Risk
Journal:
medRxiv
Published Date:
Jun 9, 2026
Abstract
Background Sleep fragmentation is increasingly recognized as a risk factor for cardiometabolic disease and mortality. However, existing measures primarily capture sleep wake transitions and do not account for fragmentation within specific sleep stages, limiting their clinical utility. Methods We developed Sleep Temporal Entropy (STE), an entropy-based metric derived from hypnogram data to quantify overall and stage-specific sleep fragmentation. We evaluated its performance in two independent cohorts: a clinical cohort of 3,219 adults and a community-based cohort of 4,862 adults. Machine learning models and survival analyses were used to assess associations with cardiometabolic outcomes and mortality. Results STE outperformed conventional fragmentation metrics in predicting cardiometabolic conditions, including hypertension, diabetes, and hyperlipidemia. Across outcomes, STE-derived features consistently ranked among the top contributors in machine learning models. In longitudinal analyses, both low and high levels of STE were associated with increased mortality risk, forming a U-shaped relationship (p for nonlinearity = 0.025). This pattern was most pronounced for rapid eye movement (REM) sleep, where individuals in the lowest quintile of REM STE had increased risk of all-cause mortality (HR 1.58, 95% CI 1.16 to 2.15) and cardiovascular mortality (HR 2.83, 95% CI 1.66 to 4.80) compared with the reference group. Conclusions STE captures stage-specific sleep fragmentation and reveals its non-linear associations with health outcomes. These findings support the potential of STE as a scalable and interpretable biomarker for sleep health assessment and risk stratification.
