Sleep Health and Cardiovascular Risk

Evidence

This prospective observational cohort analysis used UK Biobank data to examine multidimensional sleep health in relation to cardiovascular events, cardiovascular mortality, and overall mortality. UK Biobank recruited participants aged 37–73 years in the United Kingdom between 2006 and 2010. At baseline, participants completed touchscreen questionnaires on lifestyle and health-related factors, including sleep characteristics; they also underwent physical assessments and provided biological samples. The cohort’s records were linked with hospital-admission records, primary-care data, and national death registries. The primary analysis included 257,509 participants, while 26,984 participants with available plasma proteomic measurements comprised the proteomic analysis. The primary sample resulted after exclusions for missing or invalid sleep-exposure data and missing key covariates; outcome-specific analyses also excluded people with the corresponding disease at baseline, so sample sizes differed by endpoint. (pmid:42924631)

Investigators calculated a seven-component sleep health score from baseline questionnaire data and separately derived data-driven sleep patterns using K-prototypes clustering. Multivariable Cox proportional-hazards models examined associations of the sleep health score and the sleep patterns with incident coronary heart disease, stroke, atrial fibrillation, and heart failure, as well as cardiovascular mortality and overall mortality. The study defined outcomes using UK Biobank first-occurrence and death-registry data fields. This structure allowed the authors to compare risk patterns across multiple cardiovascular and mortality outcomes rather than treating sleep as a single isolated characteristic. It also kept the score-based and cluster-based assessments analytically distinct: the former used an a priori multidimensional score, whereas the latter classified patterns from the available sleep data. (pmid:42924631)

For each one-point increase in the sleep health score, the reported hazard ratios were 0.92 (95% confidence interval [CI], 0.91–0.93) for coronary heart disease, 0.96 (95% CI, 0.94–0.98) for stroke, 0.97 (95% CI, 0.96–0.99) for atrial fibrillation, 0.91 (95% CI, 0.89–0.93) for heart failure, 0.96 (95% CI, 0.94–0.99) for cardiovascular mortality, and 0.95 (95% CI, 0.94–0.96) for overall mortality. These estimates compare successive one-point increments in the score; they are associations from adjusted time-to-event models, not evidence that changing a person’s sleep health score causes the reported outcomes to change. The authors reported that all of these associations remained statistically significant after false-discovery-rate correction. Among the listed associations, the heart-failure and coronary-heart-disease estimates had the greatest relative distance below 1.00. (pmid:42924631)

The pattern-based analysis added a second description of sleep exposure. Mixed sleep disturbance and non-restorative sleep patterns were associated with higher risks across multiple outcomes. The source does not provide effect estimates for those pattern-outcome comparisons in the supplied excerpts, so the magnitude and outcome-specific precision of those associations are not described here. Still, the presence of both score-based and cluster-derived analyses means the reported pattern was evaluated through more than one operationalization of multidimensional sleep. That consistency is descriptive within this study’s design; it does not remove the possibility that measured or unmeasured characteristics associated with sleep patterns also contributed to the observed outcome differences. (pmid:42924631)

The molecular component was conducted in the subset with plasma proteomic measurements. After excluding proteins with missing values in more than 20% of samples, investigators retained 2,911 proteins for downstream analyses. Remaining missing protein values were imputed with the K-nearest-neighbours method, and protein measurements were standardized before regression analysis. Linear regression identified sleep-health-score-related proteins; least absolute shrinkage and selection operator regression selected representative proteins, followed by mediation analyses and machine-learning approaches. The proteomic analyses identified 188 proteins related to the sleep health score, described as mainly enriched in inflammatory, immune, and chemotactic pathways. Exploratory mediation analyses identified 34 proteins with potential mediation signals, with hepatocyte growth factor, insulin-like growth factor 2 receptor, and adipocyte fatty acid-binding protein 4 described as recurrent cross-outcome signals. (pmid:42924631)

Analysis — Observed Risk Pattern

The central result is an observed dose-pattern: within this UK Biobank analysis, each higher point on a questionnaire-derived multidimensional sleep health score corresponded to lower modeled hazards for every specified cardiovascular, cardiovascular-mortality, and overall-mortality outcome. The effect estimates are modest in relative terms and have narrow reported confidence intervals, reflecting precision within the analyzed cohort and model, not clinical importance or causality. The strongest reported score associations were for heart failure and coronary heart disease, while associations for stroke, atrial fibrillation, cardiovascular mortality, and overall mortality were closer to the null value of 1.00. The parallel finding that mixed sleep disturbance and non-restorative sleep patterns were associated with higher risks across multiple outcomes is consistent with the score-based direction, but the supplied material does not quantify those comparisons. The proteomic work extends the study from epidemiologic association to candidate biology: identified proteins and potential mediation signals are plausible research leads within this measured plasma system, rather than confirmation that a protein pathway explains the observed associations. (pmid:42924631)

Limitations

The retrospective interpretation of these prospective observational data must remain limited: the retrospective design cannot establish causation. Although participants were followed after baseline assessment, sleep characteristics were collected from baseline questionnaire data, and the reported associations may reflect residual confounding, unmeasured confounding, or reverse causation. The authors also state that some potentially relevant factors were not included in the models and may have influenced the associations. Selection into the primary analysis depended on valid sleep-exposure information and key covariates, and the disease-free comparator differed by outcome because participants with the corresponding baseline disease were excluded for that endpoint. These features mean that each outcome analysis represents its own eligible comparison set rather than one identical population across all six outcomes. (pmid:42924631)

The molecular findings require a separate limitation. Proteomic analyses were restricted to participants with available plasma proteomic measurements, and protein preprocessing excluded measurements with more than 20% missingness before imputation and standardization of the remaining data. The identified 188 sleep-health-score-related proteins are associations with the score in that subset. Likewise, the 34 mediation results were explicitly exploratory potential mediation signals. They do not establish that these circulating proteins lie on a causal pathway between multidimensional sleep health and cardiovascular events or mortality. Finally, because outcome-specific analyses used differing sample sizes after disease-at-baseline exclusions, comparisons of apparent association strength across endpoints should not be treated as direct head-to-head tests. (pmid:42924631)