Relationship of Sleep Duration With All-Cause Mortality and Cardiovascular Events: A Systematic Review and Dose-Response Meta-Analysis of Prospective Cohort Studies.

BACKGROUND: Effects of extreme sleep duration on risk of mortality and cardiovascular outcomes remain controversial. We aimed to quantify the dose-response relationships of sleep duration with risk of all-cause mortality, total cardiovascular disease, coronary heart disease, and stroke. METHODS AND
RESULTS: PubMed and Embase were systematically searched for prospective cohort studies published before December 1, 2016, that examined the associations between sleep duration and at least 1 of the 4 outcomes in generally healthy populations. U-shaped associations were indicated between sleep duration and risk of all outcomes, with the lowest risk observed for ≈7-hour sleep duration per day, which was varied little by sex. For all-cause mortality, when sleep duration was <7 hours per day, the pooled relative risk (RR) was 1.06 (95% CI, 1.04-1.07) per 1-hour reduction; when sleep duration was >7 hours per day, the pooled RR was 1.13 (95% CI, 1.11-1.15) per 1-hour increment. For total cardiovascular disease, the pooled RR was 1.06 (95% CI, 1.03-1.08) per 1-hour reduction and 1.12 (95% CI, 1.08-1.16) per 1-hour increment of sleep duration. For coronary heart disease, the pooled RR was 1.07 (95% CI, 1.03-1.12) per 1-hour reduction and 1.05 (95% CI, 1.00-1.10) per 1-hour increment of sleep duration. For stroke, the pooled RR was 1.05 (95% CI, 1.01-1.09) per 1-hour reduction and 1.18 (95% CI, 1.14-1.21) per 1-hour increment of sleep duration.
CONCLUSIONS: Our findings indicate that both short and long sleep duration is associated with an increased risk of all-cause mortality and cardiovascular events.

Clinical Perspective

What Is New?

Uncertainty exists regarding the dose‐response relationship between sleep duration and the risk of all‐cause mortality and cardiovascular events.

In our systematic review and meta‐analysis, sleep duration that was either too short or too long was associated with higher risk of all‐cause mortality and cardiovascular events, with the lowest risk at sleep duration of ≈7 hours per day.

What Are the Clinical Implications?

The U‐shaped associations between sleep duration and adverse outcomes have clinical relevance with respect to recommendations for adequate sleep duration in routine clinical care as well as explicit suggestions for primary prevention in public health settings.

Introduction

According to the report of World Congress of Cardiology and Cardiovascular Health in 2016, cardiovascular diseases (CVDs) are the leading cause of death globally, with an estimate of >17 million deaths from total CVD. Of these deaths, >7 million were due to coronary heart disease (CHD) and >6 million were due to stroke. In <10 years, the premature deaths from CVDs could rise by a third.1 To reduce the risk of premature death from noncommunicable diseases by 25% by 2025, as a global target of the World Health Organization,2 it is imperative to identify modifiable lifestyle factors associated with lower occurrence of CVDs. Sleep is a complex set of brain processes that supports several physiological needs.3 Increased attention has been paid to understanding the extent of sleep duration problems at the population level and their associated negative effects on various health outcomes, such as metabolic syndrome, diabetes mellitus, and cancer.4, 5, 6 Previous publications suggest that the prevalence of short sleep duration (defined as <7 hours) may have gradually increased over past decades, whereas the prevalence of long sleep duration (defined as ≥9 hours) shows an opposite trend.7

In recent years, increasing evidence has suggested that extreme sleep duration is associated with the risk of mortality and cardiovascular outcomes; however, the results are not entirely consistent. Although several studies found that sleep duration that was either too short or too long was associated with increased risk of all‐cause mortality and cardiovascular events,8, 9, 10, 11, 12, 13 reverse associations were observed in other populations.14, 15 In addition, uncertainty exists about the dose‐response relationship between sleep duration and risk of the adverse outcomes because different quantitative categories of sleep duration were used in previous studies.8, 16, 17, 18 Two meta‐analyses reported the association between sleep duration and all‐cause mortality with dose‐response analysis, but the results were inconsistent.19, 20 A previous meta‐analysis published before 2011 reported the association between sleep duration and cardiovascular events21; however, without a dose‐response analysis, it remains unknown how many hours of habitual sleep are associated with the lowest risk of cardiovascular events. Since 2011, many more studies have been published and the number of prospective studies has nearly tripled, which allows quantitative analysis of the associations. Consequently, we conducted a comprehensive dose‐response meta‐analysis of prospective studies in generally healthy populations to determine the overall shape of the relationships and quantitative estimates between sleep duration and risk of all‐cause mortality, total CVD, CHD, and stroke.

Methods

Search Strategy

This study was conducted in accordance with the MOOSE (Meta‐Analysis of Observational Studies in Epidemiology) guidelines.22 We performed a literature search (up to December 1, 2016) of PubMed and Embase for prospective studies examining the association between sleep duration and risk of all‐cause mortality and selected cardiovascular outcomes (Data S1). In addition, we reviewed references from relevant original articles and review articles to identify further pertinent studies. Only articles published in the English language were considered.

Study Selection

Studies were included if they satisfied the following criteria: The study design was a prospective cohort study; the exposure of interest was sleep duration; the outcome was all‐cause mortality, CVD, CHD, or stroke; and the investigators reported relative risk (RR), hazard ratio, or odds ratio (OR) with 95% confidence intervals (CIs) for at least 3 quantitative categories of sleep duration. Given that primary prevention of CVD was the main focus of this work (rather than secondary prevention), we excluded studies if participants were not recruited from a generally healthy population (eg, those with diabetes mellitus or under regular dialysis therapy). In addition, we excluded reviews, editorials, nonhuman studies, and letters without sufficient data. Multiple reports from the same cohort study were reviewed, and only articles with the longest follow‐up for identical outcomes were included. If insufficient data were presented in the longer follow‐up study, we included the shorter follow‐up data. Study selection was conducted in 2 stages: an initial screening of titles and abstracts to identify potentially relevant articles, followed by screening of the full‐length articles. Two investigators (J.W.Y. and S.Z.L.) independently screened all studies by title or abstract and then by a full‐text evaluation. Any discrepancy between the 2 authors was solved by discussion with the senior investigator (X.L.J.).

Data Extraction and Quality Assessment

The extraction of data included authors, year of publication, study name, study location, years of follow‐up, sample size (number of participants and incident cases), participant characteristics (age and sex), measurement method of sleep duration (questionnaire and interview), types of sleep duration (24‐hour sleep, nighttime sleep), covariates adjusted in the multivariable analysis, and effect size (RR, hazard ratio, OR), with 95% CIs for all categories of sleep duration. When studies had several adjustment models, we extracted those that reflected the maximum extent of adjustment for potentially confounding variables.

Quality assessment was performed according to the Newcastle–Ottawa Quality Assessment Scale (NOS).23 Scores ranged from 0 to 9 points, with higher scores indicating higher study quality. We considered NOS scores of 0 to 3, 4 to 6, and 7 to 9 as low, medium, and high quality, respectively.

To evaluate potential dose‐response relationships, we further extracted numbers of cases, numbers of participants, and median sleep duration in each category. If the numbers of participants and cases were not provided, the corresponding authors were contacted for the data.

Data Synthesis and Analysis

In this meta‐analysis, the RR was used as the common measure of association across studies, and the hazard ratio was deemed equivalent to RR.24 If necessary, the OR was transformed into RR according to this formula: RR=OR/[(1−P0)+(P0×OR)], where P0 is the incidence of the outcome of interest in the nonexposed group.25 Any results stratified by sex were treated as 2 separate reports. Those articles reporting >1 outcome (eg, all‐cause mortality and total CVD) were also treated as separate reports and included in corresponding analyses. If the number of cases in each category was not available in 1 study and the authors did not give their reply, we used the method by Bekkering et al to provide approximate data.26

We recognized that sleeping 7 to 8 hours per night was treated as the reference category in the majority of studies. When the reference category was not 7 to 8 hours, we used the method proposed by Hamling and colleagues to convert risk estimates.27 We calculated pooled RRs and 95% CIs for the extreme categories of sleep duration versus the reference category of sleep duration. In addition, the reports with at least 3 quantitative categories of short or long sleep duration were included in dose‐response analyses. Potential nonlinear dose‐response relationships between sleep duration and all‐cause mortality and cardiovascular events were examined by using restricted cubic splines model with 4 knots at percentiles 5%, 35%, 65%, and 95% of the distribution.28, 29 We assigned the median or mean sleep duration in each category to the corresponding RR for each study. If the mean or median duration per category was not reported, the midpoint of the upper and lower boundaries in each category was assigned. When the shortest or the longest category was open‐ended, we assumed that the open‐ended interval length had the same length as the adjacent interval. The dose‐response curves are shown in the nonlinear figures. The RR estimates in the tables were based on the nonlinear figures but show RRs for selected sleep‐duration values. If a nonlinear shape association was observed, we treated the slope as 2 piecewise and conducted dose‐response analyses using the method by Greenland and Longnecker to calculate pooled RR and 95% CIs for 1‐hour increment or decrement compared with the reference category in sleep duration.30 We used a P value for curve linearity or nonlinearity to assess the difference between the linear and nonlinear models to test for nonlinearity.29 All pooled outcome measures were determined using random‐effects models, described by DerSimonian and Laird,31 to provide more conservative results than fixed‐effects models.

The heterogeneity among studies was estimated by the Cochran Q test (P≤0.1 to be indicative of statistically significant heterogeneity) and I2 statistic.32 We conducted subgroup and metaregression analyses stratified by sex, study location, number of participants, number of cases, duration of follow‐up, sleep assessment, sleep duration type, study quality, incidence or mortality (only in total CVD, stroke and CHD), and adjustment for confounders to investigate potential sources of heterogeneity between subgroups. Moreover, stratified analyses were performed to evaluate the influences of selected study and participant characteristics on the results. Publication bias was assessed by inspection of the funnel plots for asymmetry with the Egger test33 and Begg test.34 The Duval and Tweedie35 nonparametric trim‐and‐fill method was used to further assess the possible effect of publication bias. Additional sensitivity analyses were performed by omitting 1 study at each time to test the robustness of the results and the influence of an individual study on heterogeneity.36 All statistical analyses were performed with Stata version 12 (StataCorp LP), and all tests were 2‐sided with a significance level of 0.05 unless otherwise noted.

Results

Literature Search

Figure 1 shows the results of literature research and selection. We identified 836 articles from PubMed and 837 articles from Embase before December 1, 2016. After exclusion of duplicates and studies that did not fulfill the inclusion criteria, 101 remaining articles seemed to be relevant for this meta‐analysis. After evaluating the full texts of these 101 publications and counting 1 study obtained by hand searching, the final meta‐analysis included 67 articles with 141 independent reports. Among these 67 articles, 43 articles with 57 reports provided statistical effects relevant to the meta‐analyses on all‐cause mortality, 26 articles with 37 reports on total CVD, 22 articles with 27 reports on CHD, and 16 articles with 20 reports on stroke (Data S1).

Figure 1

Flowchart of article selection. CHD indicates coronary heart disease; CVD, cardiovascular disease.

Study Characteristics

A summary of the study characteristics is shown in Tables S1 through S4. The sample sizes ranged from 724 to 1 116 936, with a total of 3 582 016 participants, including 241 107 cases of all‐cause mortality, 58 919 cases of total CVD, 22 511 cases of CHD, and 15 476 cases of stroke. The follow‐up periods ranged from 2.3 to 34 years. Among these 67 articles, most were conducted in Europe (n=22), the United States (n=16), and Asia (n=27); the others were done in Australia (n=2). Sleep duration was measured by self‐report questionnaires in 48 studies and by interview in 19 studies. The majority of the included studies had high quality, as indicated by the NOS score, and the mean study quality scores were 6.9 for all‐cause mortality, 7.0 for CVD, 7.0 for CHD, and 7.1 for stroke out of a maximum of 9 points (Tables S5 through S8).

Sleep Duration and Risk of All‐Cause Mortality

In total, 57 reports were included in the analysis of all‐cause mortality and extreme sleep duration. The pooled RR of the shortest and longest sleep duration versus reference sleep duration was 1.13 (95% CI, 1.09–1.17), with low to moderate heterogeneity (I2=37.5%, P<0.01), and 1.35 (95% CI, 1.29–1.41), with high heterogeneity (I2=76.2%, P<0.01), respectively (Table 1, Figure S1).

Table 1

Associations of Sleep Duration With All‐Cause Mortality, Total CVD, CHD, and Stroke

	n	Shortest vs Reference			Longest vs Reference
		RRa (95% CI)	I2	P Valueb	RRa (95% CI)	I2	P Valueb
All‐cause mortality	57	1.13 (1.10–1.17)	37.5	<0.01	1.35 (1.29–1.41)	76.2	<0.01
Total CVD	37	1.14 (1.09–1.20)	31.1	0.04	1.36 (1.26–1.48)	71.2	<0.01
CHD	27	1.22 (1.13–1.31)	39.6	0.02	1.21 (1.12–1.30)	37.4	0.03
Stroke	20	1.09 (0.99–1.19)	40.6	0.03	1.45 (1.30–1.62)	63.5	<0.01

CHD indicates coronary heart disease; CI, confidence interval; CVD, cardiovascular disease; RR, relative risk.

RR favors the analyses of shortest and longest vs reference sleep duration.

P for heterogeneity.

Reports with at least 3 quantitative categories of short or long sleep duration were included in dose‐response analysis. When using a restricted cubic splines model, we observed a U‐shape curvilinear association with the lowest risk of all‐cause mortality at a sleep duration of about 7 hours per day (P<0.01 for nonlinearity; Figure 2A). Both short and long sleep duration was associated with an increased risk of all‐cause mortality. Table 2 shows the RR estimates for selected sleep duration values, which were derived from the nonlinear figures. In the linear trend analyses for short sleep, no evidence of nonlinear association between short sleep duration and all‐cause mortality was found (P=0.12), and the pooled RR for all‐cause mortality was 1.06 (95% CI, 1.04–1.07) per 1‐hour reduction of sleep duration, with moderate to high heterogeneity (I2=55.5%, P<0.01; Figure 3A).8, 9, 10, 11, 12, 13, 18, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53 The heterogeneity was reduced when we excluded 2 reports9, 38 (I2=13.0%, P=0.26), but the association was not substantially altered (pooled RR: 1.06; 95% CI, 1.05–1.07). For long sleep, nonlinear association between long sleep duration and all‐cause mortality was found (P=0.02), and the pooled RR for all‐cause mortality was 1.13 (95% CI, 1.11–1.15) per 1‐hour increment of sleep duration, with high heterogeneity (I2=76.5%, P<0.01) (Figure 3B).1 The heterogeneity seemed to be mainly generated by 8 reports,8, 13, 40, 42, 44, 45, 53, 56 and when these were all excluded, the association still remained similar (RR: 1.12; 95% CI, 1.10–1.13) with low heterogeneity (I2=21.7%, P=0.15).

Figure 2

Nonlinear dose‐response analyses of sleep duration and risk of all‐cause mortality (A), total CVD (B), CHD (C), and stroke (D). CHD indicates coronary heart disease; CVD, cardiovascular disease.

Table 2

Association Between Sleep Duration and All‐Cause Mortality, Total CVD, CHD and Stroke From Non‐Linear Dose‐Response Analysis

Sleep Duration	All‐Cause Mortality (n=40a)	Total CVD (n=26a)	CHD (n=20a)	Stroke (n=17a)
3 h	1.12 (1.10–1.14)	1.14 (1.09–1.19)	···	···
4 h	1.08 (1.06–1.09)	1.09 (1.06–1.13)	1.16 (1.09–1.23)	1.05 (0.96–1.15)
5 h	1.04 (1.03–1.05)	1.05 (1.03–1.08)	1.11 (1.06–1.16)	1.02 (0.96–1.08)
6 h	1.01 (1.00–1.01)	1.02 (1.00–1.03)	1.05 (1.03–1.08)	0.99 (0.96–1.03)
7 h	1.00	1.00	1.00	1.00
8 h	1.04 (1.04–1.05)	1.03 (1.02–1.05)	1.01 (0.99–1.03)	1.08 (1.06–1.11)
9 h	1.15 (1.14–1.16)	1.16 (1.13–1.19)	1.14 (1.08–1.20)	1.30 (1.24–1.37)
10 h	1.32 (1.29–1.35)	1.37 (1.29–1.45)	1.34 (1.20–1.50)	1.64 (1.47–1.82)
11 h	1.53 (1.47–1.59)	···	···	···

CHD indicates coronary heart disease; CVD, cardiovascular disease.

n denotes number of risk estimates.

Figure 3

The forest plots between sleep duration (per hour) and risk of all‐cause mortality for short sleep (A) and long sleep (B). CI indicates confidence interval.

Sleep Duration and Risk of Total CVD

Overall, 37 reports were included in the analysis of total CVD and extreme sleep duration. A U‐shaped association was observed with the lowest risk of total CVD at a sleep duration of ≈7 hours per day (P<0.01 for nonlinearity; Figure 2B, Table 2). Both short and long sleep duration was associated with an increased risk of total CVD.

For short sleep, the pooled RR of the shortest sleep duration versus the reference sleep duration was 1.14 (95% CI, 1.09–1.20), with low to moderate heterogeneity (I2=31.1%, P=0.04; Table 1, Figure S2). We found no evidence of nonlinear association between short sleep duration and total CVD (P=0.19), and the pooled RR was 1.06 (95% CI, 1.03–1.08) per 1‐hour reduction of sleep duration, with moderate heterogeneity (I2=52.0%, P<0.01; Figure 4A).2 The heterogeneity was reduced when we excluded 1 report9 (I2=24.8%, P=0.63), and the association remained similar (pooled RR: 1.04; 95% CI, 1.02–1.06).

Figure 4

The forest plots between sleep duration (per hour) and risk of total cardiovascular disease for short sleep (A) and long sleep (B). CI indicates confidence interval.

For long sleep, the pooled RR of the longest sleep duration versus the reference sleep duration was 1.36 (95% CI, 1.26–1.48), with high heterogeneity (I2=71.2%, P<0.01; Table 1, Figure S2). A nonlinear association between long sleep duration and total CVD was found (P=0.02), and the pooled RR was 1.12 (95% CI, 1.08–1.16) per 1‐hour increment of sleep duration, with high heterogeneity (I2=75.3%, P<0.01; Figure 4B).3 The heterogeneity seemed to be generated mainly by 4 reports, and when those were all excluded, the association not substantially altered (RR: 1.13; 95% CI, 1.11–1.16) with low heterogeneity (I2=14.6%, P=0.28).

Sleep Duration and Risk of CHD

In total, 27 reports were included in the analysis of CHD and extreme sleep duration. A U‐shaped association was observed, with the lowest risk of CHD at a sleep duration of ≈7 hours per day (P<0.01 for nonlinearity; Figure 2C, Table 2). Both short and long sleep duration was associated with an increased risk of CHD.

For short sleep, the pooled RR of the shortest sleep duration versus the reference sleep duration was 1.22 (95% CI, 1.13–1.31), with low to moderate heterogeneity (I2=39.6%, P=0.02; Table 1, Figure S3). In the linear trend analyses for short sleep, a nonlinear association was noted between short sleep duration and CHD (P=0.02), and the pooled RR was 1.07 (95% CI, 1.03–1.12) per 1‐hour reduction of sleep duration, with moderate to high heterogeneity (I2=59.3%, P<0.01) (Figure 5A).4 The heterogeneity was reduced when we excluded 2 reports13, 66 (I2=23.2%, P=0.19), and the association remained similar (pooled RR: 1.04; 95% CI, 1.01–1.08).

Figure 5

The forest plots between sleep duration (per hour) and risk of coronary heart disease for short sleep (A) and long sleep (B). CI indicates confidence interval.

For long sleep, the pooled RR of the longest sleep duration versus the reference sleep duration was 1.21 (95% CI, 1.12–1.30), with low to moderate heterogeneity (I2=37.4%, P=0.03; Table 1, Figure S3). A nonlinear association was noted between long sleep duration and CHD (P<0.01), and the pooled RR was 1.05 (95% CI, 1.00–1.10) per 1‐hour increment of sleep duration, with moderate to high heterogeneity (I2=64.2%, P<0.01; Figure 5B).5 The heterogeneity was reduced when we excluded 2 reports15, 66 (I2=4.0%, P=0.41), and the association remained similar (pooled RR: 1.06; 95% CI, 1.03–1.09).

Sleep Duration and Risk of Stroke

Twenty reports were included in the analysis of stroke and extreme sleep duration. An approximate U‐shape curvilinear association was observed, with the lowest risk of stroke at a sleep duration of ≈6 to 7 hours per day (P<0.01 for nonlinearity; Figure 2D, Table 2). Both short and long sleep duration was associated with an increased risk of stroke.

For short sleep, the pooled RR of the shortest sleep duration versus the reference sleep duration was 1.09 (95% CI, 0.99–1.19), with low to moderate heterogeneity (I2=40.6%, P=0.03; Table 1, Figure S4). In the linear trend analyses for short sleep, we found no evidence of nonlinear association between short sleep duration and stroke (P=0.23), and the pooled RR for stroke was 1.05 (95% CI, 1.01–1.09) per 1‐hour reduction of sleep duration, with no significant heterogeneity (I2=0.0%, P=0.55) (Figure 6A).6

Figure 6

The forest plots between sleep duration (per hour) and risk of stroke for short sleep (A) and long sleep (B). CI indicates confidence interval.

For long sleep, the pooled RR of the longest sleep duration versus the reference sleep duration was 1.45 (95% CI, 1.30–1.62), with moderate to high heterogeneity (I2=63.5%, P<0.01; Table 1, Figure S4). No evidence of nonlinear dose‐response relationship was detected (P=0.13), and the pooled RR for stroke was 1.18 (95% CI, 1.14–1.21) per 1‐hour increment of sleep duration, with low heterogeneity (I2=4.9%, P=0.40; Figure 6B).7

Publication Bias

For the shortest or longest sleep duration versus the reference sleep duration, the publication bias was found between longest sleep duration and total CVD. The Begg rank correlation test indicated no publication bias (P=0.41), but the Egger linear regression test indicated possible publication bias for the association (P=0.01). We used the trim‐and‐fill method to recalculate our pooled risk estimate, and 13 missing studies were imputed to produce a symmetrical funnel plot (Figure S5). The analysis suggested that the imputed risk estimate was 1.22 (95% CI, 1.12–1.32), which is slightly decreased in risk but still identical to our original risk estimate. No significant publication bias was observed for other outcomes.

For the dose‐response analysis, we analyzed the publication bias of short sleep duration and all‐cause mortality and found that the Begg rank correlation test indicated no publication bias (P=0.59), but the Egger linear regression test indicated possible publication bias for the association (P=0.01). The trim‐and‐fill method was used to recalculate our pooled risk estimate, and 10 missing studies were imputed to produce a symmetrical funnel plot (Figure S6). The analysis suggested that the imputed risk estimate was 1.04 (95% CI, 1.03–1.06), which is identical to our original risk estimate. No significant publication bias was observed for other outcomes.

Subgroup, Metaregression, and Sensitivity Analyses

Tables S9 through S12 shows the different subgroup analyses of studies on all‐cause mortality, total CVD, CHD, and stroke. To explore potential sources of heterogeneity between subgroups, we carried out metaregression analyses of prespecified moderator variables. In the analyses of all‐cause mortality, the association between sleep duration and risk were not substantially changed in most subgroups. There was indication of heterogeneity (P=0.01) when we stratified studies by sleep duration type, and the pooled RRs for 1‐hour increment in long sleep duration were 1.16 (95% CI, 1.13–1.18; n=24) and 1.11 (95% CI, 1.10–1.13; n=13) for nighttime and 24‐hour sleep duration, respectively. In the nonlinear dose‐response analysis, slight variations in the risk estimates from the nonlinear dose‐response analyses were observed (Figure S7).

In the analyses of total CVD, the associations between sleep duration and risk were not substantially changed in most subgroups. Heterogeneity was indicated (P<0.01) when we stratified studies by incidence or mortality, and the pooled RRs for 1‐hour increment in long sleep duration were 1.00 (95% CI, 0.97–1.03; n=6) and 1.15 (95% CI, 1.12–1.16; n=16) for incidence and mortality, respectively. In the nonlinear analysis restricted to studies that reported the incidence of total CVD, there was no significantly increased risk of total CVD at the extreme sleep duration, whereas the U‐shaped association was more pronounced among the studies that reported mortality of total CVD (Figure S8). There was evidence of heterogeneity by study location in the linear dose‐response analysis of all participants (P=0.01), and the lowest RR was observed at 8‐hour sleep duration in Europe (Figure S9).

In the analyses of CHD, the pooled RRs for 1‐hour increment in long sleep duration were 0.89 (95% CI, 0.82–0.97; n=4) for Europe with indication of heterogeneity (P=0.02) by study location, which was inconsistent with other results. There was indication of heterogeneity (P=0.02) when we stratified studies by incidence or mortality, and the pooled RRs for 1‐hour increment in long sleep duration were 1.01 (95% CI, 0.97–1.07; n=12) and 1.13 (95% CI, 1.06–1.20; n=7) for incidence and mortality, respectively. There was no significantly increased risk of CHD at the extreme sleep duration; the U‐shaped association was more pronounced among the studies that reported mortality of CHD (Figure S10).

In the analyses of stroke, the association between sleep duration and risk was not substantially changed in most subgroups. There was indication of heterogeneity (P=0.01) when we stratified studies by duration of follow‐up, with a weaker association among studies with increasing durations of follow‐up (Figure S11).

To further confirm the robustness of the results, the dose‐response analyses were repeated using a fixed‐effects model; the pooled estimates were consistent for short and long sleep duration in relation to risk of all‐cause mortality and cardiovascular events. Sensitivity analyses omitting 1 study at a time did not substantially alter the pooled results for both short and long sleep duration and all‐cause mortality, total CVD, and CHD. For stroke, when we excluded 1 study,72 there was a statistically significant association in the analysis of the shortest versus reference sleep duration, and short sleep duration was associated with an increased risk of stroke (Figures S12 and S13).

Discussion

To our knowledge, the present work is the largest and most comprehensive study on the association of sleep duration with all‐cause mortality and cardiovascular events. Our study demonstrated U‐shaped associations between sleep duration and risk of all‐cause mortality, total CVD, CHD, and stroke, with the lowest risk observed with ≈7 hours of sleep duration. Sleep duration that was too short or too long was significantly associated with elevated risks of all‐cause mortality, total CVD, CHD, and stroke. Compared with 7 hours per day, a 1‐hour decrease was associated with 6%, 6%, 7%, and 5% increased risk of all‐cause mortality, total CVD, CHD, and stroke, respectively, and a 1‐hour increase in sleep duration was associated with 13%, 12%, 5%, and 18% increased risk, respectively.

To date, association between extreme sleep duration and increased risk of all‐cause mortality was reported previously in studies with large sample sizes and high quality,8, 9, 10, 11, 12, 13 which was consistent with our results. Heslop and colleagues,14 however, analyzed data from a workplace‐based study of Scottish men and women who were followed over a 25‐year period and found that long sleep was associated with decreased all‐cause mortality in men. But this study reported RRs with only 3 quantitative categories of sleep duration; meanwhile, long sleep duration was defined as >8 hours, which may result in inaccurate assessment of extreme long sleep. Recently, 2 systematic reviews,19, 20 both exploring the association between all‐cause mortality and sleep duration (separate analysis of 24‐hour sleep duration and nighttime sleep duration), observed markedly inconsistent results for short sleep duration. Results from Liu et al20 showed that short sleep duration was not associated with higher risk of all‐cause mortality in nighttime sleep duration. Nevertheless, results from Shen et al19 showed that for both 24‐hour and nighttime sleep duration, U‐shaped relationships were found, and the lowest risk of all‐cause mortality was observed with 7 hours per day of sleep duration, in line with our results; however, in the study by Shen et al, 1 cohort study74 was included twice in analysis. Moreover, the linear associations on the 2 sides of 7‐hour sleep duration were not detected.

Some studies have found an adverse association between extreme sleep duration and cardiovascular events. In our study, both short and long sleep duration was indicated to be associated with an increased risk of total CVD, which was inconsistent with a previous systematic review21 in 2011. In that study, short duration of sleep was not significantly associated with a greater risk of total CVD, possibly because of limited included studies. Nineteen prospective cohort studies (26 reports) have been published since 2011 and were included in our study to describe the dose‐response relationship between sleep duration and risk of total CVD. To our surprise, the findings from our subgroup analyses showed a decreased risk of CHD with long sleep duration in Europe, which should be interpreted carefully, given limited included studies. The association disappeared when we omitted the MOGEN study.15 This research showed that long sleep duration tended to be protective for CHD; however, U‐shaped associations were observed in the subgroup analysis of sleep quality in participants with available data. Notably, the proportion of women among long sleepers was significantly higher than that of men in the baseline population, whereas higher mortality rates and risks of CHD were observed in men than in women in published studies.75 This may lead to the different result. Moreover, our subgroup analyses for total CVD and CHD showed indications of heterogeneity when we stratified studies by incidence and mortality. The U‐shaped association was more pronounced among the studies that reported the mortality of total CVD or CHD compared with those that reported the incidence of total CVD or CHD. The association between cardiovascular events and sleep duration might be enhanced in the process through which patients tended to go from the occurrence of disease to death. It may also indicate that appropriate sleep duration is particularly important for delaying death among those people with chronic CVDs, and this needs to be identified further in additional studies. In our study, the adverse effect of short sleep for stroke was not observed in the shortest sleep duration versus reference analysis, whereas short sleep duration was associated with a higher risk of stroke in the dose‐response analysis. By sensitivity analysis, we found that 1 study72 had an obvious influence on the result of the shortest sleep duration versus reference analysis. The research indicated that a decreased risk of mortality from stroke was associated with short duration of sleep. Nonetheless, the small number of participants with short sleep duration limited the ability to separately analyze the effect of ≤5 and 6 hours of sleep, and the study was not included in the dose‐response analysis because it had too few categories of short sleep. After omitting the studies with <3 categories of short sleep, the pooled RR of the shortest versus reference sleep duration was 1.16 (95% CI, 1.03–1.31), which was in line with the dose‐response analysis.

Sex and age are important variables in risk of death and CVDs; this was generally accepted. In light of previous studies, the association between sleep duration and mortality8, 57, 58 and cardiovascular events16, 67 varies by sex; however, in our subgroup analyses, extreme sleep durations were significantly associated with elevated risks of all‐cause mortality, total CVD, CHD, and stroke in both men and women. Our metaregression analyses further demonstrated that there was no potential source of heterogeneity from the sex variable; therefore, a sex difference in the association of sleep duration with death and CVDs must be interpreted with caution. In addition, several studies found a stronger U‐shaped association between sleep duration and CVDs in older adults compared with younger adults (cutoff at age 65 years).10, 16 Nevertheless, the result in a study including 60 000 Chinese participants (cutoff at age 60 years) was not entirely consistent.66 Considering that the age range of the study population varied widely and the length of follow‐up was different among the included studies, we did not conduct subgroup analyses stratified by age. Further studies concentrated on sleep duration and adverse outcomes among different age groups are warranted in the future.

Short and long sleep duration may share some relevant mechanisms in relation to all‐cause mortality and cardiovascular events. As elucidated in published articles, extreme sleep duration on both sides was associated with elevated C‐reactive protein.76 As widely accepted, however, distinctive mechanisms with their own characteristics may operate at either end of the distribution of sleep duration.77

Several potential mechanisms may contribute to the relationship between short sleep duration and adverse outcomes. First, sleep restriction during the night has multiple effects on endocrine and metabolic function such as decreases of testosterone78 and melatonin secretion,79 which also may be implicated with mortality or cardiovascular events.80, 81 Second, observational studies also found that short duration of sleep was associated with vascular damage, such as coronary artery calcification.82 Third, short duration of sleep was associated with reduced levels of leptin and elevated levels of ghrelin.83, 84 The serum leptin and ghrelin levels are independent predictors of cardiovascular morbidity and mortality.85, 86 Finally, individuals with sleep deprivation, especially shift workers, have irregular sleep schedules, resulting in circadian misalignment, which may aggravate CVD in humans.87

The potential mechanisms underlying the association between long sleep duration and adverse outcomes are considered more speculative. Some insisted that the elevated risk of long sleep duration most likely represented the confounding effects of subhealthy status or uncontrolled chronic illness, such as obstructive sleep apnea, a known cause of increased need for sleep and an identified risk factor for mortality and cardiovascular events.88 As mentioned, changes in inflammatory markers and vascular health come with long sleep duration, as shown by new evidence in recent years. First, long sleep duration may be associated with an increased risk of atherosclerosis.82 Second, excessive time in bed has been linked to increased sleep fragmentation,89 which was considered to be associated with more severe arteriolosclerosis and subcortical macroscopic infarcts. These were independent risk factors of CVD and several medical comorbidities.90 Third, long sleep duration has been linked with feelings of fatigue and lethargy, which in turn would cause sleep extension. These states may fail to provide sufficient restoration against stress and disease and then lead to increased mortality.91 Finally, long duration of sleep was associated with depressive symptoms, low socioeconomic status, unemployment, low household income, low level of education, and other risk factors for mortality and cardiovascular events.92 Further experimental studies are warranted to explore the potential effects of sleep extension on health outcomes.

This meta‐analysis has several strengths. All studies included in our meta‐analysis used a prospective design, thus the differential misclassification of sleep duration attributable to recall bias was minimized. The majority of the included studies had relatively high quality. Moreover, we investigated a dose‐response relationship between sleep duration and the outcomes, allowing us to examine the shape of this possible association. Linear and nonlinear relationships were also tested to assess the dose‐response relationship.

Several limitations of our study should also be acknowledged. First, nearly all studies relied on sleep duration that was self‐reported by questionnaire or interview; 1 study93 provided the RRs between all‐cause mortality and both subjective and objective sleep duration, but no substantial difference was observed. Meanwhile, in the big data era, the widespread availability and acceptance of electronic wearable devices, such as consumer‐level activity monitors, may allow accurate, reliable, and scalable objective sleep‐duration assessment in large epidemiological studies.94 Second, sleep duration is a dynamic biological process. A single measure of exposure may not fully capture the sustained effects of sleep duration over time when related to long‐term disease incidence. One included study95 addressed this issue by measuring changes in sleep duration twice, several years apart, and found that stable short and stable long sleep was associated with a significantly increased risk of mortality; moreover, moving to either shorter of longer sleep from average sleep was also associated with increased mortality. This finding was in line with our result that appropriate sleep duration was important for the delay or prevention of premature mortality. Third, we cannot rule out the possibility of residual or unmeasured confounding, even though we have taken into consideration major confounding factors by using adjusted risk estimates from multivariate models from each contributing study. Finally, sleep quality affected by factors like sleep apnea is an independent predictor of risk of adverse outcomes96 but was not assessed in our study. Despite the limitations, at this stage, results from prospective cohort studies are still the best evidence available to assess the longitudinal effect of sleep duration on all‐cause mortality and cardiovascular events.

Conclusions

In summary, our dose‐response meta‐analysis of prospective studies provides further evidence that sleep duration that is either too short or too long is associated with higher risk of all‐cause mortality and cardiovascular events, with the lowest risk with ≈7 hours per day of sleep duration. Longer term randomized controlled trials are needed to establish causality and to elucidate the underlying mechanisms.

Author Contributions

Yin, Shan, Chen, and Liu conceived the study. Yin searched the databases, checked them according to the eligible criteria and exclusion criteria, extracted and analyzed the data, and wrote the draft of the article. S.Z. Li and Jin helped extract quantitative data from some articles and contributed to writing, reviewing, or revising the article. Huang, P.Y. Li, Shan, Bao, Yang, X.B. Peng, Z. Peng and Yu critically reviewed and revised for important intellectual content. Shan and Bao provided advice on meta‐analysis methodology and contributed to reviewing, or revising the article. Liu is the guarantor and had full access to all the data and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Sources of Funding

This work was funded by the National Natural Science Foundation of China (NSFC 81472978), the National Science and Technology Support Program (2012BAI02B02) and China Postdoctoral Science Foundation (2016M602314). Integrated Innovative Team for Major Human Diseases Program of Tongji Medical College, HUST. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article.

Disclosures

None.

Data S1.

Table S1. Sleep Duration and All‐Cause Mortality

Table S2. Sleep Duration and Total Cardiovascular Disease

Table S3. Sleep Duration and Coronary Heart Disease

Table S4. Sleep Duration and Stroke

Table S5. Study Quality of Studies Included in the Analysis of Sleep Duration and All‐Cause Mortality

Table S6. Study Quality of Studies Included in the Analysis of Sleep Duration and Total Cardiovascular Disease

Table S7. Study Quality of Studies Included in the Analysis of Sleep Duration and Coronary Heart Disease

Table S8. Study Quality of Studies Included in the Analysis of Sleep Duration and Stroke

Table S9. Subgroup Analyses of Sleep Duration and All‐Cause Mortality Per Hour, Per Day

Table S10. Subgroup Analyses of Sleep Duration and Total Cardiovascular Disease Per Hour, Per Day

Table S11. Subgroup Analyses of Sleep Duration and Coronary Heart Disease Per Hour, Per Day

Table S12. Subgroup Analyses of Sleep Duration and Stroke Per Hour, Per Day

Figure S1. Sleep duration and all‐cause mortality, shortest and longest vs reference analysis.

Figure S2. Sleep duration and total cardiovascular disease, shortest and longest vs reference analysis.

Figure S3. Sleep duration and coronary heart disease, shortest and longest vs reference analysis.

Figure S4. Sleep duration and stroke, shortest and longest vs reference analysis.

Figure S5. Trim‐and‐fill correction for publication bias for total cardiovascular disease, longest vs reference analysis.

Figure S6. Trim‐and‐fill correction for publication bias for all‐cause mortality, dose‐response analysis for short sleep.

Figure S7. Nonlinear dose‐response analysis of sleep duration and all‐cause mortality by nighttime sleep duration (A) and 24‐hour sleep duration (B).

Figure S8. Nonlinear dose‐response analysis of sleep duration and total cardiovascular disease by incidence (A) and mortality (B).

Figure S9. Nonlinear dose‐response analysis of sleep duration and total cardiovascular disease by Asia (A), Europe (B), and the United States (C).

Figure S10. Nonlinear dose‐response analysis of sleep duration and coronary heart disease by incidence (A) and mortality (B).

Figure S11. Nonlinear dose‐response analysis of sleep duration and stroke by follow‐up duration <10 years (A) and follow‐up duration ≥10 years (B).

Figure S12. Sensitive analysis of stroke and sleep duration, shortest vs reference analysis.

Figure S13. Sensitive analysis of stroke and sleep duration after excluding the study of Kawachi (2016), shortest vs reference analysis.

Click here for additional data file.