Multidimensional sleep health and all-cause mortality: Counterfactual risk estimation and component-specific contributions in the sleep heart health study.

BACKGROUND: Multidimensional sleep health has been linked to mortality, but most evidence is based on conventional association models and provides limited information on population-level risk under hypothetical sleep-health improvement. METHODS: We included 4382 adults from the Sleep Heart Health Study with baseline polysomnography, questionnaire-based sleep assessments, covariates, and mortality follow-up. Six sleep-health domains were assessed: duration, continuity, timing, sleepiness/alertness, architecture, and sleep disorders. A composite sleep health score ranging from 0 to 6 was calculated, with higher scores indicating better sleep health. Cox model-based g-computation estimated 1-, 5-, and 10-year population standardized mortality risks. Pragmatic scenarios increased each participant's observed score by up to 1 or 2 points, capped at 6; component-specific analyses set individual domains to the healthy level while other domains remained observed. RESULTS: During a median follow-up of 4327 days, 926 deaths occurred. Each 1-point higher score was associated with lower all-cause mortality (HR, 0.924; 95% CI, 0.875-0.977). The standardized 10-year risk under observed scores was 16.78%. Increasing scores by up to 1 and 2 points yielded risks of 15.82% (RD, -0.96 percentage points; 95% CI, -1.63 to -0.34) and 15.06% (RD, -1.72 percentage points; 95% CI, -2.92 to -0.61), respectively. Flexible modeling indicated modest nonlinearity and attenuated incremental estimates. Continuity and timing had the largest isolated component-specific risk differences, but neither remained significant after Holm correction. CONCLUSIONS: Better multidimensional sleep health was associated with lower mortality. Pragmatic hypothetical improvements were associated with modest reductions in standardized 10-year mortality risk.