Association of Sleep Duration With All-Cause and Cardiovascular Mortality: A Prospective Cohort Study

Healthy sleep duration is increasingly recognized as an important public health issue. The American Academy of Sleep Medicine and the Sleep Research Society recommend seven or more hours per night for adults, yet short sleep (<7 h/day) has increased among US adults, while long sleep (≥9 h) shows an opposite trend. Experimental evidence links short sleep with adverse physiological and immunological outcomes, and several researchers suggest long sleep may be an even greater concern. Prior meta-analyses indicate both short and long sleep are associated with increased risks of all-cause mortality and cardiovascular events, though findings have been inconsistent across studies due to differences in design, populations, and reference sleep categories. Cardiovascular disease is the leading cause of death in the US, but few studies have translated sleep–mortality associations into population-level impact measures such as population attributable fraction (PAF). This study uses NHANES data linked to the National Death Index to clarify associations of sleep duration with all-cause and cardiovascular mortality and to estimate the PAF for 10-year CVD risk attributable to short/long sleep in US adults.

Several lines of evidence have examined sleep duration and health outcomes: (1) Meta-analyses report U-shaped associations where both short and long sleep are linked to higher all-cause mortality and cardiovascular events; (2) Other systematic reviews and flexible meta-regressions show non-linear relationships but with some heterogeneity and inconsistency; (3) Differences in study design (cross-sectional, case-control, prospective), target populations (general vs. those with metabolic disease), and selected reference categories (e.g., 6–8 h, 7–8 h, or 7–9 h) contribute to mixed findings; (4) Large cohort studies such as PURE and Chinese cohorts observed weaker associations for short sleep and stronger associations for long sleep with mortality; (5) Prior work indicated lowest predicted 10-year CVD risk among adults sleeping 7 h/night, but population-level attributable fractions for sleep duration had not been estimated. These gaps motivated a well-designed, nationally representative analysis with standardized covariate adjustment and translation of findings to contemporary population impact.

Study design and population: NHANES is an ongoing nationally representative cross-sectional survey (since 1999) of US civilians, administered by NCHS. Protocols were IRB-approved and participants provided written informed consent. For prospective mortality analyses, adults ≥20 years from five NHANES cycles (2005–2014) were included and linked to National Death Index (NDI) mortality data through December 31, 2015. For population impact estimation, sleep duration and CVD risk factors were taken from NHANES 2017–2018 to estimate 10-year CVD risk and PAF. Sleep assessment: In NHANES 2005–2014, sleep duration was self-reported via “How much sleep do you usually get at night on weekdays or workdays?”, recorded in hours as a continuous variable. In 2017–2018, habitual weekday sleep duration was derived from usual sleep and wake times. Mortality ascertainment: All-cause and CVD mortality were determined via linkage to NDI. CVD death was defined using ICD-10 codes I00–I09, I11, I13, I20–I51, and I60–I69. Follow-up time spanned from interview date to date of death or end of follow-up (December 31, 2015). Covariates: Demographics and lifestyle factors included age, sex, race/ethnicity, education, smoking status (≥100 lifetime cigarettes), alcohol consumption (≥12 drinks/year), physical activity (≥2.5 h/week classified as active), diet quality via HEI-2015, BMI, comorbidities (diagnosis and/or medication use), and family history (heart disease, diabetes). Statistical analysis: Sleep duration was categorized into five groups (≤5, 6, 7 [reference], 8, ≥9 h). Cox proportional hazards models estimated hazard ratios (HRs) and 95% CIs for all-cause and CVD mortality. Model 1 adjusted for age and sex; Model 2 additionally adjusted for race/ethnicity, BMI categories, education, physical activity, alcohol, smoking, and HEI-2015; Model 3 further adjusted for family history of diabetes and heart disease, and history of diabetes, heart disease, and cancer. Nonlinear associations were assessed using restricted cubic splines (3 knots at the 10th, 50th, and 90th percentiles), with likelihood ratio tests for nonlinearity. Subgroup analyses examined interactions by age (<65/≥65 years), sex, race/ethnicity (non-Hispanic white vs. other), education (less than college vs. college or above), BMI (<30 vs. ≥30 kg/m²), smoking (never vs. ever), alcohol (never vs. ever), and physical activity (inactive vs. active), using survey-weighted Wald F statistics and Bonferroni correction (significance P<0.005 for eight subgroups). Sensitivity analyses excluded participants with heart disease or cancer at baseline, and excluded deaths in the first year of follow-up. Population attributable fraction estimation: Under a causality assumption, 10-year CVD risk was predicted using ACC/AHA Pooled Cohort Equations (PCE) for adults 40–79 years, incorporating age, total cholesterol, HDL cholesterol, systolic blood pressure (treated/untreated), diabetes, and current smoking. For each participant, the predicted 10-year risk (Rp) was computed, then an altered risk (Rq) was derived by hypothetically setting sleep duration to 7 h using risk ratios per hour deviation from 7 h obtained from meta-analysis (per hour >7 h: HR 1.12; per hour <7 h: HR 1.06), with RR_x = exp(lnHR × x). Rq was calculated as Rp/RR_x. Attributable events were estimated as ∑(Rp − Rq) × population size; PAF was ∑(Rp − Rq)/∑(Rp). Analyses accounted for NHANES sample weights, stratification, and clustering. Software: SAS 9.4 and Stata 14. Two-sided P<0.05 indicated statistical significance.

Population and follow-up: Among 25,481 US adults (48.6% men; mean ages 46.3 years for men, 47.8 years for women), 146,484 person-years were accrued, with 2,033 all-cause deaths, including 378 due to heart disease and 461 due to cancer. Baseline characteristics varied by sleep category; those with ≥9 h or ≤5 h tended to be older, less educated, physically inactive, with higher BMI (≤5 h), and more comorbidity and family history of chronic disease. All-cause mortality: In fully adjusted models (Model 3), HRs by sleep category (≤5, 6, 7 [reference], 8, ≥9 h) were 1.40 (95% CI 1.14–1.71), 1.12 (0.91–1.38), 1.00, 1.35 (1.12–1.63), and 1.74 (1.42–2.12); cases by category: 340, 386, 406, 615, 286; person-years: 22,731; 34,409; 39,595; 38,858; 10,891. Restricted cubic splines indicated significant nonlinearity (P<0.001), consistent with a U-shaped association and lowest risk at 7 h/day. Cardiovascular mortality: In fully adjusted models (Model 3), HRs by sleep category (≤5, 6, 7 [reference], 8, ≥9 h) were 1.66 (95% CI 1.02–2.72), 1.15 (0.77–1.73), 1.00, 1.55 (1.05–2.29), and 1.81 (1.09–3.02); cases by category: 60, 66, 72, 121, 59. Nonlinearity was significant (P<0.01). Subgroups and sensitivity: Associations were generally consistent across subgroups with no significant interactions after Bonferroni correction; results were robust to excluding first-year deaths and participants with baseline heart disease or cancer. Population impact: Using NHANES 2017–2018 (1,805 individuals representing ~112.8 million US adults aged 40–79 years free of self-reported CVD), the absolute 10-year CVD event rate was 9.2% (~10.3 million events) from 2018–2028. Estimated attributable events: short sleep, 187,000 (PAF 1.8%, 0.9%–2.3%); long sleep, 947,000 (PAF 9.2%, 6.4%–11.6%). Stratified PAFs suggested higher short sleep impact among adults <60 years, men, and non-Hispanic Black individuals, and higher long sleep impact among adults 60–79 years, women, and other races.

Sleep duration demonstrated U-shaped associations with all-cause and cardiovascular mortality, with the lowest risk at 7 h/day. Although associations for short sleep were relatively weaker, both short (≤5 h) and long (≥9 h) sleep were linked to elevated risks after multivariable adjustment and in sensitivity analyses, suggesting sleep duration itself may contribute to mortality risk beyond confounding by overt disease. Translating these findings to population impact, habitual sleep duration—particularly long sleep—may be associated with approximately one million excess CVD events over 10 years in the US under a causality assumption. These results support public health efforts to optimize sleep duration for primary CVD prevention. Potential mechanisms for short sleep include endocrine and metabolic dysregulation (lower testosterone and melatonin; reduced leptin and increased ghrelin), which may influence cardiometabolic pathways and mortality. Mechanisms for long sleep are less clear; hypotheses include undiagnosed subclinical illness, sleep fragmentation linked to vascular pathology, and fatigue/lethargy cycles that impede adequate recovery. The robustness of findings after excluding early deaths and prevalent heart disease or cancer strengthens the inference that sleep duration may be an independent risk factor, though residual confounding cannot be fully excluded.

Sleep duration shows a U-shaped association with all-cause and cardiovascular mortality, with 7 hours/day associated with the lowest risk. Under a causality assumption, optimizing sleep duration may prevent a substantial number of CVD events in the US, highlighting sleep as a modifiable factor for primary prevention.

1) Sleep duration was self-reported, introducing potential misclassification; however, subjective estimates correlate reasonably with objective measures in large studies. 2) A single baseline measurement may not capture changes in sleep habits over time. 3) Residual or unmeasured confounding may remain despite extensive adjustment. 4) Population attributable fractions were estimated under a causality assumption using observational data; randomized trials manipulating sleep duration are scarce and challenging, limiting causal inference.