Multidimensional sleep health and risks of all-cause and cardiovascular mortality: A MAIHDA-based prospective cohort study.

Objectives

The combined and interactive effects of multiple sleep dimensions on mortality remain unclear. We applied Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) to examine how multidimensional sleep health relates to all-cause and cardiovascular mortality.

Methods

The analytic sample included 4340 adults for all‑cause mortality and 3790 for cardiovascular mortality from the Sleep Heart Health Study. Six sleep dimensions were dichotomized into healthy versus unhealthy categories and combined to form 64 unique strata representing all possible sleep health profiles. Individuals were nested within strata in multilevel Cox MAIHDA models. Three models were estimated: model 1 (null), model 2 (sleep-domain fixed effects), and model 3 (sleep-domain fixed effects plus covariates).

Results

In null models, sleep‑health strata explained 8.51% of the variance in all‑cause mortality and 15.01% in cardiovascular mortality, indicating meaningful heterogeneity across sleep profiles. After accounting for sleep-domain main effects and covariates, most between‑strata variance was explained (PCV = 99.95% for all-cause mortality and 78.32% for cardiovascular mortality), indicating that mortality risk was predominantly additive rather than driven by unique non‑additive combinations. Healthier continuity and timing were associated with lower all-cause mortality, while timing was the most consistent sleep feature for cardiovascular mortality; absence of sleep disorders showed a weaker association mainly for all-cause mortality. Strata with multiple unhealthy sleep dimensions consistently had the highest marginal standardized predicted 10-year mortality risks.

Conclusions

MAIHDA maps mortality risk across multidimensional sleep profiles and quantifies additive versus non-additive effects. Mortality risk was explained primarily by the additive accumulation of sleep-health attributes.