Sleep-Stage Dynamics Predict Current Sleep-Disordered Breathing and Future Cardiovascular Risk

Sleep-disordered breathing (SDB) is a major contributor to cardiovascular morbidity and disrupts both the macrostructure and dynamics of sleep stages (W, N1, N2, N3, REM). While specific alterations in sleep macrostructure, such as reduced durations of N3 and REM, have been linked to cardiovascular risk, the predictive value of sleep-stage dynamics remains unexplored. Using data from the prospective Sleep Heart Health Study, we applied a flexible forest-based modelling approach to a carefully selected cohort of 2579 subjects free from prior cardiovascular events and sleep-altering medications to minimize confounding. First, we demonstrate that a random forest classifier reliably identifies moderate-to-severe SDB (apnea–hypopnea index; AHI >15), achieving AUROC=76.1%, from sleep-stage architecture, dynamics, and common risk factors (demographics, BMI, smoking status) alone, without direct respiratory measurements. This highlights a dependency chain in which SDB correlates with altered sleep patterns that, in turn, encode cardiovascular risk. Second, a random survival forest robustly predicted future cardiovascular events (concordance-index=73.3%) over >10 years follow-up. Comparable results with and without including AHI as a predictor indicate that sleep patterns encode cardiovascular risk independently of direct SDB measurement. Partial dependence analyses revealed monotonic SDB risk profiles and predominantly U-shaped associations for cardiovascular risk, identifying ranges of total sleep time, wake after sleep onset, and REM/N3 continuity linked to minimal or elevated risk. Notably, rare transitions such as N3 →N1 or REM →N3, even occurring once per night, emerged as sensitive markers of cardiovascular vulnerability, increasing risk by up to 10%. Our findings extend prior evidence on linear associations between sleep macrostructure and cardiovascular outcomes, revealing non-linear patterns and positioning sleep dynamics as promising non-invasive biomarkers for diagnostics and early risk stratification.