Sleep Duration and Risks of Incident Cardiovascular Disease and Mortality Among People With Type 2 Diabetes

Cardiovascular disease (CVD) is the leading cause of morbidity and mortality in people with diabetes (1). Evidence-based and integrated approaches involving lifestyle modification and metabolic control are essential for risk management (2), while identifying additional modifiable risk factors for screening and intervention is also needed.

Insufficient and prolonged sleep durations are more common among people with diabetes than those without diabetes (3) and have been linked to worse metabolic risk control (4–6) and a higher prevalence of microvascular complications (7,8). Besides, diabetes medications (e.g., insulin treatment) might also disturb normal sleep by various mechanisms (e.g., nocturnal hypoglycemia) (9,10). These observations suggest that the potential impact of inappropriate (short or long) sleep duration on the long-term health of people with diabetes, especially CVD risk, warrants further investigation. Several studies reported that inappropriate sleep duration was associated with increased total and CVD mortality in this high-risk population (11,12), but evidence on sleep duration and incident CVD is scarce and limited to cross-sectional designs with inconsistent findings (13–15). In addition, such studies need to carefully account for the potential influence of metabolic control and diabetes severity in light of their associations with sleep duration and CVD risk.

In the current study, we evaluated the associations of habitual sleep duration with long-term risks of incident atherosclerotic cardiovascular disease (ASCVD), including coronary artery disease (CAD), ischemic stroke, and peripheral artery disease (PAD), and CVD mortality among people with type 2 diabetes from the UK Biobank (UKB). We also investigated whether factors related to progression and management of diabetes modified these associations.

The UKB is an ongoing, prospective, population-based cohort study that enrolled >500,000 participants aged 40–69 years during 2006–2010. Participants completed a touchscreen questionnaire and a verbal interview, took physical measurements, and provided biological samples in 1 of the 22 assessment centers throughout England, Scotland, and Wales, as reported in detail previously (16). The UKB received ethics approval from the North West Multi-Center Research Ethics Committee (REC reference no. 11/NW/03820). All participants provided written informed consent for the study.

In the current analysis, we included 26,839 participants with type 2 diabetes at baseline who were identified through self-reported medical history and medication information, hospital inpatient records (ICD-9 codes 250.00, 250.10, 250.20, and 250.90 and ICD-10 code E11), and abnormal glucose levels (random glucose ≥11.1 mmol/L or glycated hemoglobin [HbA_1c] ≥48 mmol/mol [6.5%]) (17). Participants who had abnormal glucose levels but no diagnosis of diabetes were defined as undiagnosed patients and were assumed to have type 2 diabetes (n = 3,610). After excluding those who had missing or implausible values of sleep duration (i.e., <4 or >12 h/day) or had prevalent CVD (including heart failure) or cancer at baseline, a total of 18,876 participants with type 2 diabetes remained (Supplementary Fig. 1).

Information regarding hospital admission and diagnoses were obtained through linked inpatient data from Health Episode Statistics (England and Wales) and the Scottish Morbidity Records (Scotland) up to 30 September 2021 (England), 31 July 2021 (Scotland), and 28 February 2018 (Wales); date and cause of death were obtained from death certificates held within the National Health Service Information Centre (England and Wales) and the National Health Service Central Register Scotland (Scotland) up to 30 September 2021 (England and Wales) and 31 October 2021 (Scotland). Incident ASCVD (including CAD, ischemic stroke, and PAD) and CVD mortality were defined based on the ICD and Office of Population Censuses and Surveys Classification of Interventions and Procedures codes (Supplementary Table 1). Person-time was calculated from the date when participants answered the touchscreen questionnaire to either the date of death or diagnosis (for the incident ASCVD analysis) or the end of follow-up, whichever came first.

Habitual sleep duration was self-reported through the baseline questionnaire with the following question: “About how many hours sleep do you get in every 24 hours? (please include naps).” Responses to the question were in hourly increments. We grouped participants into six categories (≤5, 6, 7, 8, 9, and ≥10 h/day) after merging those who reported extreme values and used 7 h/day as the reference group. Nearby groups with similar CVD risks were further combined as <7 (short), 7–9 (intermediate), and >9 h/day (long) in the stratified and joint analyses to account for the limited sample size in each subgroup.

Covariates obtained through questionnaires or verbal interviews at baseline included age, sex, ethnicity, education, family history of CVD, lifestyle, medication use, self-rated health, and other sleep traits (insomnia, napping, chronotype, daytime sleepiness, and snoring). A higher education level was defined as a college or university degree or other professional qualifications. According to American Heart Association recommendations (18), regular physical activity was defined as ≥150 min/week of moderate activity, ≥75 min/week of vigorous activity, or an equivalent combination. A healthy diet was defined as an adequate intake of ≥5 of 10 recommended food groups for cardiometabolic health according to a previous UKB study (19). Townsend deprivation index was calculated using national census data and assigned using postal codes of residence, with a lower score indicating a higher area level of socioeconomic status (20). Depression was identified by a self-reported medical history, an inpatient-recorded diagnosis of clinical depression, or a score of ≥3 on the two-item Patient Health Questionnaire tool (depression scale) (21). Physical measurements, including height, weight, and systolic and diastolic blood pressure (BP), were performed by qualified personnel at baseline. The standard protocols for collecting and handling baseline blood and urine samples have been previously described (22), including measurements of HbA_1c (Variant II Turbo Hemoglobin Testing System; Bio-Rad), LDL cholesterol, serum creatinine (AU5800 clinical chemistry analyzer; Beckman Coulter), and urine microalbumin and creatinine (AU5400 clinical chemistry analyzer; Beckman Coulter). Complications were defined as any of the following: an inpatient-recorded diagnosis of microvascular disease (Supplementary Table 2), urine albumin-to-creatinine ratio ≥30 mg/g, or estimated glomerular filtration rate <60 mL/min/1.73 m² (23). Diabetes duration was defined as years between the first diagnosis of diabetes and UKB baseline assessment for diagnosed patients or 0 years for undiagnosed patients. Diabetes severity status was defined by the sum of factors reflecting higher diabetes severity, including diabetes duration ≥5 years, HbA_1c ≥53 mmol/mol (7.0%), prevalent diabetes complications, and insulin use. Metabolic control included HbA_1c (individualized targets), BP (target range <140/90 mmHg), and LDL cholesterol (target range <100 mg/dL) (24). Metabolic control status was defined by the sum of metabolic risk factors outside the target range.

For variables with a missing rate of ≥5%, missing data were coded as an independent category; otherwise, missing data were replaced with median values for continuous variables or the mode for categorical variables. Detailed information on missing covariates is shown in Supplementary Table 3.

Cox proportional hazards models were applied to examine the associations of baseline sleep duration (≤5, 6, 7, 8, 9, and ≥10 h/day) with the risk of total ASCVD, CAD, ischemic stroke, PAD incidence, and CVD mortality, using sleep duration of 7 h/day as the reference. Hazard ratios (HRs) with 95% CIs were estimated from four models: model 1 was adjusted for age (continuous) and sex (women or men); model 2 was further adjusted for ethnicity (White British or other), education (higher or other), Townsend deprivation index (continuous), and family history of CVD (yes or no) based on model 1; model 3 was further adjusted for regular physical activity (yes, no, or missing), healthy diet (yes, no, or missing), smoking (never, previous, or current), alcohol consumption (never, previous, or current), insomnia (usually or other), and nap (never or ever) based on model 2; and model 4 was further adjusted for BMI (<25, 25–30, or ≥30 kg/m²), depression (yes or no), diabetes duration (<5 or ≥5 years), diabetes medication (oral antidiabetic drugs only, insulin, or neither), complications (yes or no), individualized HbA_1c target (yes, no, or missing), BP (≥140/90 mmHg; yes or no), antihypertensive medication (yes or no), LDL cholesterol (≥100 mg/dL; yes, no, or missing), and cholesterol-lowering medication (yes or no) based on model 3. We performed tests for the linear trend by entering the median value of each category of sleep duration as a continuous variable in the models. The proportional hazard assumption was examined using Schoenfeld residuals, and models were stratified on the variable that was found to violate proportionality.

To examine whether metabolic control and diabetes severity modified the associations of sleep duration with cardiovascular outcomes, we performed stratified and joint analyses using Cox proportional hazards models. In the analysis for metabolic control status, baseline sleep duration of 7–9 h/day, no metabolic risk factors outside the target range, or the combination of both was set as the reference if applicable. Models were adjusted as for model 4 except for HbA_1c, BP, and LDL cholesterol. Similarly, the independent and joint associations of sleep duration and diabetes severity with CVD risks were examined with baseline sleep duration of 7–9 h/day, no severity factors, or the combination of both as the reference if applicable. Models were adjusted as for model 4 except for HbA_1c, diabetes duration, complications, and diabetes medication. The statistical significance of interaction was tested using a likelihood ratio test to compare models with and without cross-product terms between the sleep duration category (<7, 7–9, >9 h/day) and the number of metabolic risk factors or diabetes severity factors (continuous).

We also conducted other stratified analyses by age (≤60 or >60 years), sex, Townsend deprivation index (median), smoking, diet, BMI (<30 or ≥30 kg/m²), HbA_1c, diabetes duration, prevalent diabetes complications, and insulin use. The likelihood ratio test was used to examine the statistical significance of interactions by comparing models with and without cross-product terms between the sleep duration category (<7, 7–9, >9 h/day) and the stratification variables.

Several sensitivity analyses were performed to test the robustness of our results. First, we excluded participants with undiagnosed diabetes to minimize the potential misclassification of diabetes type. Second, we further excluded cases occurring within the 1st year of the recruitment to reduce reverse causation. Third, we excluded participants with severe diabetes or those self-rating as having poor health to test the potential influence of suboptimal health status on sleep duration. Fourth, we further adjusted for other sleep traits (i.e., chronotype, snoring, daytime sleepiness), sleep apnea, sleep medication use, and baseline work schedule to explore potential sources of confounding. Fifth, we restricted the analysis to participants who were not engaged in night shifts at baseline, as night shifts might influence circadian rhythms and were independently associated with higher CVD risk (25). Sixth, we constructed competing risk proportional subdistribution hazards regression models to account for the possible competing relationships among ASCVD subtypes. Seventh, we examined the associations without exclusion of participants reporting implausible sleep durations. Eighth, we restricted the analysis to participants with complete covariate data or used multiple imputations with chained equations to assign missing values in covariates to test whether missingness influenced the results.

All statistical analyses were performed using R version 3.6.0 software. A two-sided P < 0.05 was considered to be statistically significant.

Of the 18,876 people with type 2 diabetes, 7.3%, 20.8%, 32.5%, 28.2%, 7.6%, and 3.6% reported sleeping for ≤5, 6, 7, 8, 9, and ≥10 h/day, respectively. Compared with those who slept for 7 h/day, participants with a shorter or longer sleep duration were of lower socioeconomic status and education. Besides, they were more likely to be women and noncurrent drinkers, have prevalent depression, take antihypertensive or insulin treatment, and have poor self-rated health (Table 1).

We documented 2,570 incident ASCVD cases (1,460 CAD, 555 ischemic stroke, and 596 PAD; mean ± SD follow-up 11.3 ± 2.8 years) and 598 deaths from CVD (mean ± SD follow-up 12.1 ± 2.0 years). Divergence from 7–9 h/day of habitual sleep duration was associated with an increased risk of incident ASCVD and CVD mortality among people with type 2 diabetes, independent of sociodemographic and lifestyle factors (model 3) (Table 2). After further adjustment for BMI, depression, factors reflecting diabetes severity, and metabolic control, these associations were slightly attenuated (model 4). Compared with sleeping for 7 h/day, the multivariable-adjusted HRs of incident ASCVD were 1.26 (95% CI 1.08, 1.48) for ≤5 h/day, 1.21 (1.09, 1.35) for 6 h/day, and 1.41 (1.16, 1.70) for ≥10 h/day, and HRs of CVD mortality were 1.42 (1.02, 1.97) for ≤5 h/day, 1.36 (1.07, 1.71) for 6 h/day, and 1.85 (1.30, 2.64) for ≥10 h/day in the final model. For the incidence of specific ASCVDs, sleep duration of ≤5 h/day was associated with a 70% higher risk for ischemic stroke (1.70 [1.23, 2.35]); participants sleeping for ≥10 h/day were at an increased risk of ischemic stroke (2.08 [1.44, 3.01]) and PAD (1.45 [1.01, 2.10]).

When the analysis was stratified by metabolic control status, we observed similar associations of sleep duration with incident ASCVD and CVD mortality in each stratum (both P for interaction >0.1) (Supplementary Table 4). In the joint analysis that set participants who slept for 7–9 h/day with no metabolic risk factors outside the target range as the reference, short and intermediate sleepers with worse metabolic control were at higher ASCVD risk, while long sleepers had a similarly high risk across the different metabolic control subgroups (Fig. 1A and Supplementary Table 5). For long sleepers (>9 h/day), the HR of incident ASCVD was 2.18 (95% CI 1.33, 3.57) for those with no metabolic risk factors outside the target range, 2.11 (1.49, 2.98) for those with one factor, and 2.01 (1.46, 2.77) for those with two to three factors. No significant association between metabolic control and CVD mortality among participants with varying sleep durations was observed (Fig. 1B and Supplementary Table 6).

When the analysis was stratified by diabetes severity status, we observed similar associations of sleep duration with CVD mortality in each stratum (P for interaction > 0.1), while the association with incident ASCVD tended to be more prominent among participants with higher diabetes severity (P for interaction = 0.03) (Supplementary Table 7). In the joint analysis that set those who slept for 7–9 h/day and had no severity factors as the reference, higher diabetes severity was associated with increased risk of incident ASCVD and CVD mortality among participants with different sleep durations (Fig. 1C and D and Supplementary Tables 8 and 9).

The association between sleep duration and incident ASCVD was stronger among participants with longer diabetes duration (P for interaction = 0.03). Long sleepers who were of higher socioeconomic status or had insulin treatment seemed to have higher ASCVD risk than other subgroups (both P for interaction <0.05). Consistent results were observed in other analyses stratified by age, sex, smoking, diet, BMI, HbA_1c, and complications for ASCVD incidence and CVD mortality (all P for interaction >0.05) (Supplementary Tables 10 and 11).

The results were not materially changed in most sensitivity analyses (Table 3 and Supplementary Tables 12–16), while the association between sleep duration and PAD risk turned null when the analysis was restricted to participants with complete information of all covariates (Supplementary Table 17).

Overall, we found that both short and long sleep durations were associated with higher risks of ASCVD incidence, especially ischemic stroke, and CVD mortality in a large U.K. population of people with type 2 diabetes. Similar results were observed after adjustment for potential confounders and in most sensitivity analyses that considered the health status of participants. However, the association between sleep duration and incident ASCVD tended to be more prominent among those with higher diabetes severity.

The J-shaped association between sleep duration and CVD mortality we found is consistent with what has been shown in recent prospective studies among people with type 2 diabetes (11,12). A Chinese study showed that compared with sleeping for 7 h/day, sleep duration of <4 and >10 h/day was associated with a 54% (HR 1.54 [95% CI 1.04, 2.28]) and 88% (1.88 [1.49, 2.37]) increased risk of CVD mortality, respectively (12). The current study is among the first to report that short and long sleep durations are associated with a higher risk of incident ASCVD in this high-risk population. Our findings reflect similar results of existing data among people with diabetes that linked sleep duration and CVD risk directly or indirectly; that is, short and long sleep durations were associated with poor glycemic control (4), worse cardiometabolic risk profiles (5,6), and a higher prevalence of CVD (13–15). For example, a meta-analysis of seven studies showed a U-shaped association of sleep duration with HbA_1c (4). A Japanese study of 4,402 people with type 2 diabetes reported that those with insufficient or prolonged sleep durations were more likely to have metabolic syndrome and insulin resistance (5). However, unlike the observed U-shaped relationship between sleep duration and stroke prevalence (13,14), data from the U.S. National Health Interview Survey study showed that only long sleep duration was associated with a higher prevalence of coronary heart disease in women (15), which may probably be due to the limited sample size. Of note, these cross-sectional studies were more susceptible to reverse causation. Besides, given that diabetes is associated with increased risks of several comorbidities (26,27), including insomnia, obstructive sleep apnea, depression, and hypertension, all of which might lead to divergence from intermediate sleep durations and be linked to cardiovascular health, the potential confounding by participants’ health status should be carefully accounted for. In our study, similar associations between sleep duration and cardiovascular outcomes were largely observed in various analyses that aimed to address these issues, suggesting an independent role of intermediate sleep duration in the long-term cardiovascular health of people with type 2 diabetes.

The observed associations of short sleep duration with total ASCVD incidence, ischemic stroke in particular, and CVD mortality in our study seemed to be stronger than those found in the general population. According to a meta-analysis of 20–37 prospective studies on sleep duration and incident CVD (28), the pooled relative risk of the shortest sleep duration compared with the 7–8 h/day group was 1.14 (95% CI 1.09, 1.20) for total CVD, 1.22 (1.13, 1.31) for coronary heart disease, and 1.09 (0.99, 1.19) for stroke. Another meta-analysis of prospective studies that excluded baseline CVD showed that sleeping for <7 h/day compared with 7–8 h/day was associated with an 8–21% higher risk of CVD mortality (29). Multiple mechanisms have been proposed to explain the elevated CVD risk associated with insufficient sleep duration, such as the hyperactivation of the sympathetic nervous system and increased oxidative stress (30). Because insufficient sleep duration has been related to clustered metabolic abnormalities (e.g., general and abdominal obesity [31], hypertension, dyslipidemia) that are common in people with type 2 diabetes (32), short sleepers in this population might likely have faster deterioration in metabolic metrics, which probably triggers the subsequent CVD risk. In a recent Mendelian randomization analysis (33), short sleep duration showed casual associations with CAD, while evidence regarding stroke was less clear and limited to specific subtypes (i.e., large artery stroke) (34). However, the association between sleeping for ≤5 h/day and incident ischemic stroke was stronger than that of CAD in our study, which clearly requires confirmation by more prospective studies within this population. Inconsistent with previous findings from the general population that insufficient sleep duration was associated with higher PAD prevalence (35), we report a null association among people with type 2 diabetes. Such inconsistency is probably due to the difference in study design or mechanism related to the study population, which warrants further investigation.

Although long sleep duration was associated with most CVD outcomes in our study and previous studies among the general population (36), evidence on the causal relationship between prolonged sleep duration and CVD remains lacking (33). It has been speculated that long sleep duration may indicate undiagnosed illness or worse health conditions (37). In our study, the association between sleeping for ≥10 h/day and CAD was not statistically significant either after multivariable adjustment or in most sensitivity analyses, while the association of sleeping for ≥10 h/day with a higher PAD risk was not significant in the complete case analysis. Despite these, findings for total ASCVD, ischemic stroke, and CVD mortality, were not materially changed, suggesting that the results might not be fully explained by the broad-spectrum risk proxy hypothesis in the context of diabetes.

The independent associations of sleep duration and CVD events among people with type 2 diabetes were further supported by joint analyses of sleep duration with metabolic control or diabetes severity, as higher CVD risks in the groups of inappropriate sleep duration were observed in most subgroups. Besides, the association of short and long sleep durations with increased ASCVD risk seemed to be more apparent among people with higher diabetes severity, such as those having longer diabetes duration or being on insulin treatment. Existing evidence suggests that inappropriate sleep duration might accelerate disease progression (38) and lead to poor glycemic control (4) or diabetes complications (7,8). Conversely, certain diabetes medications, such as insulin, might also disturb normal sleep (9,10). Moreover, diabetes severity was associated with increased CVD risk (39). Thus, it is speculated that the influence of inappropriate sleep duration and diabetes progression might be bidirectional, which is likely associated with subsequent higher ASCVD risk.

A major strength of our study was the use of data from a prospective study with a large sample size and extensive phenotype measurement, which enabled detailed analyses of ASCVD subtypes and stratification by potential risk factors. Several limitations warrant comments. First, sleep duration was self-reported once at baseline; therefore, misclassification was inevitable, and changes or a long-term pattern of sleep duration could not be captured. Second, sleep quality is an important confounding factor of our findings regarding sleep quantity, but these data are not available in the UKB. However, we adjusted for other self-reported sleep phenotypes, including insomnia, nap, chronotype, daytime sleepiness, and snoring, to reflect sleep quality, and the findings remained robust. Third, cases of undiagnosed sleep apnea might confound our findings, although results were consistent when we adjusted for prevalent cases identified by self-reported data and hospital inpatient records. Fourth, the possibility of reverse causation could not be eliminated. To address this issue, we conducted several sensitivity analyses, such as excluding cases that occurred within the 1st year of follow-up, the results of which were largely unchanged. Fifth, the UKB represents a relatively healthy population in the U.K. (40) among whom short or long sleepers could be less common, which might underestimate the association between inappropriate sleep duration and cardiovascular outcomes. Sixth, residual confounding, such as genetic susceptibility, is possible but was not evaluated in this study. For example, the MTNR1B diabetes risk variant (rs10830963G) related to altered circadian rhythms was associated with an increased risk of myocardial infarction among people with type 2 diabetes, although sleep duration did not statistically differ between carriers and noncarriers (41). Finally, because of the limitation of the observational study design, causality between sleep duration and CVD among people with type 2 diabetes cannot be established.

The findings from our research provide an opportunity for advancement in targeted screening initiatives, modifications, and interventions in both a clinical and public health setting. Those with inappropriate sleep duration should be prioritized for intervention efforts because they are a high-risk group for CVD incidence and mortality. Aiming to intentionally modify behavior-driven lifestyle factors, such as sedentary behavior, food intake, and substance use, could provide a starting point for improving sleep duration and overall outcomes (42). Additionally, it is imperative to further assess possible interventions to aid in improved sleep duration and quality to reduce the burden of CVD. Based on the findings of our study, there are several future research direction recommendations. Further assessing sleep duration in a clinical setting would allow for causal associations to be drawn, allowing for effective and appropriate measures to be taken in improving sleep duration among people with type 2 diabetes. We also recommend conducting a study that could expound on the current study’s results and understand how intervening at various points of life could affect their probability of getting CVD. Moreover, understanding factors affecting sleep duration among people with diabetes is integral to providing comprehensive evidence-based approaches to reducing CVD.

In conclusion, we found that both short and long sleep durations were significantly associated with higher risks of CVD incidence and mortality among people with type 2 diabetes. Our findings highlight an independent role of sleeping for 7–9 h/day in diabetes care and call for effective and safe strategies for sleeping issues among this population.

Acknowledgments. The authors thank the participants and staff of the UK Biobank for their dedication and contribution to the research. This research was conducted using the UKB resource under application no. 55005.

Funding. This study was supported by the National Science Fund for Excellent Young Scholars (81922060), the Talent Introduction Programme of Chinese Academy of Sciences, the Science and Technology Commission Fund of Hongkou District (2102-08), and the Discipline Promotion Programme of Shanghai Fourth People’s Hospital (SY-XKZT-2020-1010).

The funding sources had no role in the study design; collection, analysis, and interpretation of data; writing of the report; or decision to submit the article for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. H.H. performed the statistical analysis. H.H., Y.W., T.L., C.F., C.K., Y.S., Y.W., J.Z., L.W., and G.Z. participated in the interpretation of the results and critical revision of the manuscript. H.H., Y.W., and L.W. drafted the manuscript. H.H., Y.W., L.W., and G.Z. participated in the study concept and design. L.W. and G.Z. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the American Heart Association Scientific Sessions, Chicago, IL, 5–7 November 2022.