Short and long sleep duration is an emerging risk factor for mortality [1], and the association between sleep duration and various health outcomes is a significant public health concern [2]. Many epidemiological and experimental studies have demonstrated that sleep duration is associated with circulatory diseases [3,4], diabetes mellitus [5], impaired immune function, and neuropsychiatric disorders [6,7]. In addition, existing studies have reported that both short and long sleep durations (generally defined as <5–7 and >8 or 9 hours, respectively) are associated with an increased risk of all-cause [2,8], cardiovascular [9], and respiratory mortalities [10].

Despite sufficient evidence from previous studies, several critical points need to be considered: First, numerous studies assessing the relationship between sleep duration and all-cause mortality are available; however, relatively few studies have examined the impact of sleep duration on cause-specific mortalities, especially respiratory mortality [10]. Although biological evidence needs to be further demonstrated, some studies have revealed that sleep disorders impact respiratory symptoms [11,12], lung function [13], and chronic obstructive pulmonary disease (COPD) [14], all identified as risk factors for increased mortality in large cohort studies [15,16]. Given that chronic respiratory diseases are the third leading cause of death accounting for 4.0 million deaths worldwide [17], the association between sleep duration and respiratory mortality should be addressed. Secondly, additional risk factors affecting the association between sleep duration and mortality should be examined. Previous studies revealed that the relationship between sleep duration and mortality varies by age and sex [18,19]. However, body mass index (BMI) and income status have been less studied as effect modifiers [1], although these variables are closely related to sleep duration [20] and mortality [21-23]. Therefore, diverse risk factors, including the aforementioned two variables, should be assessed to identify diverse and precise high-risk populations. Third, previous studies exhibited heterogeneity in their designs, follow-up periods, confounders/covariates, and operational definitions of “short” and “long” sleep duration. Thus, evaluating high-risk populations under consistent conditions and analytical frameworks is necessary.

Therefore, this study aimed to assess the heterogeneous associations between sleep duration and mortality by cause of death (all-cause, cardiovascular, and respiratory mortalities), age, sex, race/ethnicity, income status, and BMI categories (underweight, normal weight, and overweight) using a prospective cohort design based on the National Health and Nutrition Examination Survey (NHANES) and the National Death Index (NDI) in the United States between 2009 and 2018.

Table 1 presents descriptive information on the study population. A total of 13,947 NHANES participants aged 50 years were included in the study. A total of 804 all-cause deaths were observed within the 3-year follow-up period. Among them, 396 and 41 deaths were associated with cardiovascular and respiratory diseases, respectively. Supplementary Table 1 (in the online-only Data Supplement) displays the descriptive characteristics of the study population, considering death within a 5-year follow-up period.

A U-shaped association between sleep duration and all-cause and cardiovascular mortalities was observed (Fig. 1). For each cause of death, the optimal sleep duration associated with the lowest mortality was approximately 6–7 hours. Respiratory mortality exhibited a sharper increase in HR at short sleep durations (less than 7 hours) compared to that observed with cardiovascular mortality, although it exhibited a decreasing pattern at 3 hours or lower with wide confidence intervals (CI). In contrast, increased risks of all-cause and cardiovascular deaths were associated with long sleep duration (>8 hours), and respiratory mortality demonstrated a statistically unstable association with long sleep duration.

The HR estimates based on categorized sleep durations are illustrated in Fig. 2. In the total population, for all-cause mortality, sleep duration ≤5 hours (HR: 1.86; 95% CI: 1.29, 2.68; p<0.01) and >8 hours (HR: 1.32; 95% CI: 1.03, 1.70; p=0.029) demonstrated a greater association compared to that noted for sleep duration ≥7 and ≤8 hours (normal sleep). The HR for sleep duration >5 and <7 hours (moderate short sleep) were not statistically significant (HR: 1.04; 95% CI: 0.75, 1.43; p=0.826). The increased risk of cardiovascular mortality was associated with sleep duration ≤5 hours (HR: 1.70; 95% CI: 0.98, 2.95; p=0.057); however, the association between sleep duration >5 and <7 hours (moderate sleep) or >8 hours (long sleep) was not statistically evident. However, sleep duration >5 and <7 hours (moderate sleep) was marginally related to an increase in respiratory mortality risk (HR: 1.97; 95% CI: 0.82, 4.75; p=0.132), and we could not identify an association between respiratory mortality and sleep duration ≤5 hours (severe short sleep; p=0.678).

Fig. 3-6 present the results of the subgroup analyses of cause-specific mortalities. First, we informed the participants that all results of the subgroup analyses were interpreted based on point estimates. The association between mortality and both severe short sleep (≤5 hours) and long sleep (>8 hours) was more pronounced in individuals aged 50–64 years (p=0.106) or 75 years or older (p=0.012) than in those aged 65–74 years (p=0.497), although statistical evidence for an association with sleep durations >8 hours was weak across all age groups. Males exhibited stronger associations with severe short (≤5 hours; p<0.01) and long sleep (>8 hours; p=0.013) duration compared to females. Non-Hispanic White individuals (p=0.001) and those with high income (p=0.056) had the strongest relationship with severe short sleep (≤5 hours) than other races/ethnicities and other income groups, respectively. People with a BMI ≥23.0 kg/m² (overweight) had a stronger relationship with severe short sleep (≤5 hours; p=0.001) compared to the relationship observed with other BMI groups. However, individuals with normal weight (BMI: ≥18.5 and <23.0 kg/m²) demonstrated a stronger association with long sleep (>8 hours; p=0.005) than those with a BMI ≥ 23.0 kg/m 2 (overweight). Underweight individuals with a BMI <18.5 kg/m² (underweight) exhibited a similar HR for long sleep (>8 hours) as those with normal weight, although the 95% CI was wider for underweight individuals. People with “Hypertension status=Yes” demonstrated a stronger association with sleep duration ≤5 hours (p<0.01) and >8 hours (p=0.044) than the association exhibited by individuals with “Hypertension status=No”. People with “High cholesterol level status=Yes” also displayed a higher association between sleep duration ≤5 hours (p<0.01) and total mortality, compared to people with “High cholesterol level status=No” (p=0.160).

Additionally, for cardiovascular death, we could not identify a statistically evident association with sleep duration >5 hours and <7 hours (moderate short sleep) or >8 hours (long sleep) across all subpopulations. Moreover, the association with severe short sleep (≤5 hours) was observed in certain subgroups. The association with severe short sleep was higher in individuals aged 75 years or older (p=0.043) and females (p=0.028) than in other age groups and males, respectively. Furthermore, individuals with “Diabetes status=Yes” (p=0.074), “Hypertension status=Yes” (p=0.046), and “High cholesterol level status= Yes” (p=0.046) demonstrated more statistically evident association with sleep duration ≤5 hours than their counterparts (each status=“No”). Second, for respiratory mortality, although most subgroups had a convergence issue (HRs could not be estimated), males (p=0.177) and people with a BMI ≥23 kg/m² (overweight) exhibited a weak linkage with sleep duration >5 and <7 hours (moderate short sleep; p=0.092).

Sensitivity analyses (Supplementary Figs. 1-5 in the online-only Data Supplement) using mortality records from the 5-year follow-up period generally exhibited similar patterns to the main analysis. However, notable differences were observed in respiratory mortality: specifically, a marginal association was noted between severe short sleep (≤5 hours) and increased respiratory mortality. Second, in respiratory death, low- and middle-income individuals demonstrated higher HRs than those in high-income individuals. This reverse pattern contrasts with the associations observed for other causes of mortality.

This study evaluated the association between sleep duration and cause-specific mortality using population-representative NHANES data between 2009 and 2018. We identified a U-shaped association between sleep duration and mortality for all-cause and cardiovascular diseases, with minimum mortality sleep duration between 6 to 7 hours. For all-cause death, people aged 50–64 years, males, White individuals, high-income groups, and overweight people had a significant association between mortality and severe short sleep (≤5 hours). Although the statistical evidence was weak, males and overweight individuals demonstrated a strong association between respiratory death and severe short sleep. Regarding cardiovascular disease, the relationship with severely short sleep was stronger in individuals aged 75 years or older and females than in other age groups and males.

Our results are generally consistent with existing findings, although previous studies were performed in diverse regions. A large cohort study in East Asian countries reported that the association between sleep duration and all-cause mortality was U- or J-shaped in both males and females however, this study did not reveal a significant effect modification by BMI [1]. A review of 23 studies reported that both short and long sleep durations were related to an increased risk of all-cause and cardiovascular-related mortalities; however, they did not address respiratory mortality [2]. Another review that included 40 cohorts also reported U- or J-shaped associations between sleep duration and all-cause mortality [18]. However, a United States study reported that the link between short sleep and mortality was not observed in middle-aged individuals and was only detected in older participants [19]. This observation was different from our findings demonstrating that people aged 50–64 years had a higher risk of short sleep than other older age groups, although the study period (1982–1992) was different.

This study has a limitation in addressing the possible biological mechanisms explaining the association between sleep duration and mortality. However, several plausible mechanisms have been proposed. One plausible mechanism is that short sleep duration is related to reduced leptin levels and elevated ghrelin levels [29]. This upregulation of appetite can partly explain the association between short sleep duration and the incidence of obesity and diabetes [1,30]. Existing studies have also reported that restricted sleep might be associated with impaired glucose tolerance, alterations in sympathetic nervous system activities, increased inflammatory markers (including high-sensitivity C-reactive protein and interleukin-6), and elevated cortisol levels, which can increase mortality risk [2]. Furthermore, previous studies have demonstrated that dysregulation of melatonin secretion may be related to hypertension and diabetes. The suprachiasmatic nucleus (SCN) in the hypothalamus affects melatonin secretion, and the SCN could influence the activity and rest cycle of peripheral organs, hormone secretion, sensitivity of target organs through neural mechanisms (e.g., insulin secretion from the pancreas), and glucose secretion from the liver [31]. In addition, research demonstrated that extremely short sleep disrupts hormonal balance, leading to decreased morning cortisol levels. This stress from insufficient sleep can increase heart rate, decrease heart rate variability, raise blood pressure, and elevate catecholamine secretion—all well-known risk factors for coronary artery disease [32]. However, the association between long sleep duration and mortality remains uncertain and is related to residual confounders and underlying diseases [1,3]. Individuals with long sleep duration may already have undiagnosed comorbidities and inflammatory biomarkers. Additionally, both high-sensitivity C-reactive protein and interleukin-6, are related to long sleep duration [33].

Assessing the association between sleep duration and respiratory mortality, which has been examined less frequently than all-cause or cardiovascular mortality, is a novel finding of this study. Several studies have revealed that various sleep disorders are associated with respiratory symptoms and diseases including asthma and COPD [34]. Furthermore, experimental studies have demonstrated that sleep loss can adversely influence the components of the immune system critical for host resistance to respiratory infectious illnesses. Additionally, short sleep duration and sleep disorders have been suggested as risk factors for increased susceptibility to upper respiratory infections [35]. Moreover, sleep can exaggerate instability in breathing patterns, which can affect hypoxia and pulmonary congestion, impair the upper airway reflex dilator response, and induce hypotonia in the upper airways, resulting in decreased and delayed responses to negative pressures [36]. Although the related statistical evidence was weak due to the insufficient sample size of respiratory deaths (41 deaths during the study period), our study discovered that respiratory mortality was associated with moderately short sleep (>5 and <7 hours), and males and overweight individuals had a strong relationship between moderately short sleep and respiratory death compared to that in the total population. Sensitivity analysis with a 5-year follow-up period demonstrated a marginal association between severe short-term sleep and respiratory death. Given that the prevalence of chronic respiratory diseases has increased with aging in recent decades [17], further studies evaluating the impact of sleep duration on respiratory diseases and mortality should be conducted.

Another novel finding of this study was that effect modifications were based on BMI and income status. In particular, this study identified that overweight individuals had a stronger association between short sleep duration and both all-cause and respiratory mortality than that observed in the total population. Furthermore, the relationship between severely short sleep duration and increased risk of all-cause mortality was strongest in overweight individuals; however, this association was not observed in underweight individuals. As previous studies have identified that an elevated BMI is related to increased risks of sleep disorders, cardiovascular and respiratory diseases, cancer, and mortality [1,20,37], our results on BMI and sleep mortality risk could provide epidemiological evidence for identifying high-risk target populations. Additionally, this study discovered that the high-income group demonstrated the strongest association between short sleep duration and all-cause mortality. To our knowledge, this concept has not been widely addressed. Although some studies have reported that people with high income likely have improved sleep duration [38], several studies have reported that certain types of high-pressure work might be related to high-paying jobs [39]. In other words, a high workload and the resultant physical and mental stress can result in inadequate sleep duration and an increased risk of sleep disorders. Degenerated sleep quality could be associated with an increased risk of mortality. However, the association between long sleep duration and all-cause mortality was significantly pronounced in the low-income group. Although the explanation for this study is limited, previous research has suggested an association between long sleep duration and underlying diseases [1,3], and it is well-established that socially marginalized people have poor health conditions [40]. These findings can provide evidence for future studies identifying high-risk labor related to poor sleep and health conditions, and future studies with suitable study designs and variables should be performed to examine the effect modifications based on income and BMI in depth.

Furthermore, because this study utilized a self-report questionnaire regarding sleep duration collected by the NHANES, previous findings on the validity of self-reported sleep duration in epidemiological studies should be carefully addressed. The validation results of the self-reported sleep duration questionnaire were mixed. Several studies have documented that self-reported and actigraphy sleep durations were strongly associated; thus, self-reported sleep duration could be used to confirm actual sleep duration [41,42]. Although different studies exhibited that self-reported sleep duration was moderately correlated (ρ~0.5) [43], discrepancies between the two are not uncommon (in general, self-reported sleep duration questionnaires might overestimate the prevalence of people with short sleep durations) [44,45]. To reduce the systematic biases of self-reported sleep duration, previous studies have consistently suggested assessing multiple self-reported questionnaires on sleep duration, performing sensitivity analyses with various sleep duration variables, and collecting a multitude of validated instruments (e.g., the Morningness Eveningness Questionnaire, Munich ChronoType Questionnaire, Pittsburgh Sleep Quality Index, British Sleep Survey, Karolinska Sleep Diary, Insomnia Severity Index, etc.) [46], which could examine sleep quality more comprehensively than a single sleep duration. Particularly, due to the limited availability of NHANES data, this study could not use validated sleep quality indices. Future studies should survey these instruments and examine the association between sleep quality and mortality or morbidities in depth to improve relevant evidence.

This study has several limitations. First, in addition to the self-reported sleep duration questionnaire, the NHANES only measured sleep duration at a single point (when people participated in the survey), which did not allow us to consider changes in sleep duration during the study period. Third, although we restricted the follow-up period to 3 years to reduce the potential estimation biases from the time-varying covariates and exposures, such as the sleep duration variable, other covariates were also measured at one baseline point. Therefore, we could not address the time-varying impacts of covariates on the association between sleep duration and mortality. Numerous time-varying confounders may impact both sleep and mortality, including high-risk alcohol consumption, mental health, physical activity, and medications. In particular, as the NHANES was designed as a cross-sectional study to investigate health and nutritional status at a certain time point, collecting these time-varying confounders was systemically impossible. However, if large prospective cohort studies to assess the association between sleep and mortality can be performed in the future, these time-variant variables should be carefully selected and widely collected, and the corresponding improved statistical analytic frameworks (e.g., a Cox regression with time-varying variables) should be used to provide less-biased estimates than studies with single baseline covariates. Fourth, although analyzing the association between sleep duration and respiratory death is one of the strengths of our study, the related estimation results should be interpreted carefully if the corresponding statistical evidence (p-values or CI) remains weak. Numerous existing epidemiological studies do not recommend interpreting estimates using statistical values as criteria of significance [47]. This statistical evidence depends on the sample size, and the NHANES was not designed to examine the hypotheses of this study. Thus, the post-hoc p-value or CI could not achieve “statistical significance.” [47] Nevertheless, these statistics provide evidence for the reliability of the association and reflect the possibility of sampling and information errors. Thus, estimates with weak statistical evidence should be carefully interpreted, and these findings should be cautiously regarded as indicating the possibility of no association or a degree of directionality of insufficient association. Fifth, although we operationally defined disease status (diabetes, hypertension, and high cholesterol levels) based on self-report questionnaires, these definitions were limited in reflecting the actual disease status. Although several studies have demonstrated that self-reported data on cardiovascular diseases and diabetes are reliable [48,49], the related results should be interpreted carefully due to the possibility of errors related to self-reporting. Finally, as mentioned earlier, the main disadvantage of this study is the inability to consider standardized indicators for measuring sleep deprivation or sleep disorders, such as the Insomnia Severity Index, Pittsburgh Sleep Quality Index, and polysomnography, which could provide suitable and accurate sleep measurements. Furthermore, our sleep duration measurement did not address wakefulness during sleep or the duration of rapid eye movement sleep, which negatively affects sleep quality and related health [50].

The results of this study suggest that sleep duration is associated with the risk of mortality. In particular, we determined that short sleep duration was related to an increased risk of all-cause, cardiovascular, and respiratory deaths. Moreover, these relationships were modified by age, sex, race/ethnicity, household income, and BMI. Our results provide scientific evidence for the establishment of targeted action plans to improve sleep and health.