Study population: the EPIC-Potsdam study

The European Prospective Investigation into Cancer and Nutrition (EPIC) study is an ongoing large-scale cohort study on diet and chronic diseases, especially cancer, with over half a million participants across Europe. EPIC-Potsdam is one out of 23 centers of EPIC with 27 548 participants aged mainly 35–65 years that had been recruited between 1994 and 1998 from the general population of Potsdam and adjacent communities. The study instruments of the baseline examination included a computer-guided interview on lifestyle habits and medical history, self-administered questionnaires on food consumption and lifestyle as well as physical examinations performed by trained staff at the study centre [29]. The EPIC-Potsdam study has been approved by the Ethical Committee of the Federal State of Brandenburg. Written informed consent for measurements and inquiries including prospective data collection has been obtained from all participants before joining EPIC-Potsdam.

Every 2–3 years the participants received a mailed follow-up questionnaire to assess incident diseases, medication, and changes in diet and other lifestyle factors. Tracing of non-responders, reminder services, and computer programs controlling for completeness or implausible values contribute to the generation of valid and complete follow-up data and high response rates for each follow-up round (93–96%) [30].

Participants with prevalent diabetes mellitus, myocardial infarction, stroke, or cancer (all except non-melanoma skin cancer) (n = 3 130), missing follow-up time (n = 589), missing data on exposure or relevant covariates (n = 207) as well as those reporting not to sleep at all (n = 2) were excluded. Thus, 23 620 participants (14 497 women and 9 123 men) remained for the analyses. Study participants were followed up until 2007. The follow-up time (person-years) was derived from the age at recruitment and the age at one of the following end-points: date of diagnoses, date of death, or return of the last follow-up questionnaire, whichever came first.

Assessment of exposure

Participants were asked about their average duration of sleep (in hours) at night and at day during the last 12 months with two open questions in the baseline interview. Moreover, also sleep disorders were assessed in the baseline interview with the following questions: “Have you ever suffered from sleep disorders that were treated by a physician?” If this question was answered with “yes”, participants were further asked: “How long did you suffer from sleep disorders? Please total all periods in which you suffered from sleep disorders.”

Assessment of chronic diseases

Information about potential incident chronic diseases occurring during follow-up were regularly assessed from self-reports of the respective condition, medication use, or reasons for a reported change in diet in the follow-up questionnaires. All potential incident cases were verified using information of medical reports from the treating physician, cancer registries, or death certificates [30]. Diseases were coded based on the International Classification of Disease (ICD-10: I21 for myocardial infarction, I60, I61, I63, I64 for stroke, E11 for diabetes, and C00-C97 for cancer (except C44: non-melanoma skin cancer). In case of multiple diseases, only the first clinical event was considered for the analysis.

Statistical analyses

Sleep duration at baseline was considered as the main exposure variable. Based on the habitual sleep duration, participants are classified as short, long, or normal sleepers. Therefore, the sleep duration at night and at day was summed and participants were categorized into five groups according to their sleep duration following the distribution of sleep duration in this cohort and cut-offs used in former publications: <6, 6-<7, 7-<8, 8-<9 and ≥9 h per 24 h (<6 h is defined as short, 6-<7 h as shortened, 7-<8 h as normal/average, 8-<9 h as normal/prolonged and ≥9 h as long sleep duration). These groups were characterized by descriptive univariate statistics in order to examine the correlates of long and short sleep. Baseline characteristics of the study population are presented as arithmetic means and standard deviations (for continuous variables) and percentages (for categorical variables).

The risk of developing a chronic disease in different groups of sleep duration was analyzed by means of Cox proportional hazard regression analyses. Age- and sex- as well as multivariable-adjusted hazard ratios (HR) and 95% confidence intervals (CI) were calculated using 7-<8 h of sleep as the reference group. This reference was chosen for two reasons. First, 7 h of sleep had been shown to be associated with the lowest risk of mortality [31], [32], [33] and also morbidity e.g. cardiovascular disease [13] or diabetes [12]. Second, the average sleep duration was 7 h in this sample (mean: 7.3 h, median: 7 h).

The primary endpoint was defined as the first incident chronic disease – i.e. type 2 diabetes, myocardial infarction, stroke, or cancer. Additional events at a later date were not considered. The follow-up time was defined as time period between the age of recruitment and the age of exit (age of diagnoses, death, or censoring). To be less sensitive against violation of the proportional hazards assumption, all models were stratified by age. Relevant covariates, e.g. socioeconomic and lifestyle factors were considered by multivariable adjustment. Therefore, four models were carried out. The first model (model 1, reduced model) was adjusted for sex. The second model (model 2) additionally included sleeping disorders (yes/no), alcohol intake from beverages (non-consumers; men: 1. >0–12 g/d, 2. >12–24 g/d, 3. >24 g/d; women: 1. >0–6 g/d, 2. >6–12 g/d, 3. >12 g/d), smoking status (never, former, current), walking, cycling, sports (hours/week), employment status (employed vs. unemployed), and education (university of applied sciences or university degree vs. technical school or lower degree). Model 3 was further adjusted for potential intermediates, i.e. BMI (kg/m²), waist-to-hip ratio, prevalent hypertension at baseline (yes/no), and history of high blood lipid levels at baseline (yes/no). Model 4 furthermore included consumption of caffeinated beverages (coffee and black tea in g/day), satisfaction with life (4 levels: very satisfied, rather satisfied, rather dissatisfied, very dissatisfied), satisfaction with health (4 levels: very satisfied, rather satisfied, rather dissatisfied, very dissatisfied), and intake of antidepressants (yes/no). Covariables were defined based on a priori knowledge on risk factors for the investigated outcomes. In general, the same covariates were included in each model, except baseline prevalent hypertension and higher blood lipid levels, which were not included in the multivariable models for cancer.

In addition to sleep duration, we further investigated the effect of timing of sleep (daytime naps (yes/no) and night-time sleep (<6, 6-<7, 7-<8, 8-<9 and ≥9 h)) on the risk of chronic diseases using analogous multivariable Cox regression models.

Moreover, interactions of sleep duration (continuous) with sex, sleeping disorders (yes/no), age (<60 vs. ≥60 years), alcohol intake (category 2 and 3 vs. category 1), obesity (BMI ≥30 kg/m²: yes/no), and hypertension (yes/no) were investigated. Furthermore, interactions of these variables with daytime sleep were examined. Therefore, models with and without interaction terms were compared using the likelihood-ratio test. All p values were two-sided and a significance level of p<0.05 was applied.

All analyses were performed with SAS statistical software (release 9.2; SAS Institute Inc, Cary, North Carolina).

Short sleep and chronic diseases

Our results showed that, compared to average sleep (7-<8 h), short sleep (<6 h per day) was associated with a 30% increased risk of overall chronic disease, in particular stroke (2-fold increased risk) and overall cancer (more than 40% increased risk).

This analysis was based on a first incident event analysis which is a novel approach that enables to compare the effect of sleep on different endpoints such as diabetes, cardiovascular diseases, and cancer. There has been previous evidence that short sleep (<5–6 h) is a habit that is associated with increased risk of chronic diseases [3], [10], [11], [12], [13].

Many studies investigated the association between sleep duration and diabetes risk and some of them showed that short sleep is an independent risk factor for diabetes [14], [15], [16], [17], [18]. In contrast, our results showed no relation between short sleep duration and incidence of diabetes after adjustment for potential intermediates. Similar to us, Hayashino et al. [34] and Björkelund et al. [35] did not find a significant association of sleep duration and diabetes risk. A possible reason for this discrepancy is the role of BMI in the relationship of sleep duration and diabetes as a potential mediator. It is often described in the literature that sleep restriction may lead to obesity by a number of biological mechanisms, e.g. via increased insulin resistance or via altered secretion of hormones like ghrelin or cortisol, which in turn could lead to subsequent diabetes [3], [10], [36], [37]. Thus, our risk estimates for short sleep duration as well as those of Ayas et al. [25] were attenuated to non-significant risk estimates after adjustment for BMI. In addition, hypertension [38] is a further potential intermediate that contribute to attenuation of the association between sleep duration and diabetes.

Fewer studies were conducted on sleep deficiency in relation to stroke and myocardial infarction. Chen et al. reported associations of short sleep duration with ischemic stroke in postmenopausal women (HR = 1.14, 95% CI: 0.97–1.33 for ≤6 h compared to 7 h of sleep), which became significant after exclusion of those with baseline comorbidities like cardiovascular disease or diabetes mellitus (HR = 1.22, 95% CI: 1.03–1.44) [21]. In contrast, Amagai et al. found no association of sleep duration and stroke after multivariable adjustment [22] which might be due to a relative small number of cases incorporated into this analysis. Regarding myocardial infarction we could not identify a significant relation with short sleep duration which is in contrast to the results of some previous studies, e.g. the Monitoring Trends and Determinants on Cardiovascular disease (MONICA) study [20], or the Japanese Jichi Medical Cohort Study [22], that reported on increased risk of myocardial infarction among women or men with short sleep. In our analyses, the association of sleep duration and myocardial infarction was significant in the sex- and age-adjusted models but inclusion of further covariates, especially smoking, but also alcohol drinking, hypertension, and satisfaction with health, removed the effect of sleep.

To our knowledge, this is the first study analyzing the relationship between sleep duration and overall cancer risk thereby observing an increased risk of developing cancer among participants reporting <6 h of sleep. Only a few studies examined the association between short sleep duration and breast cancer showing varying results. Kakizaki et al. found that women with ≤6 h sleep are at increased risk of developing incident breast cancer (HR = 1.62, 95% CI: 1.05–2.50; p for trend = 0.03) [24] whereas results by other researchers indicate that a long sleep duration of 9 h or more might have protective effects for breast cancer risk [39], [40]. On the contrary, data from the Nurses' Health Study does not support an inverse association of long sleep duration and breast cancer risk [41]. Furthermore, these authors called upon more research on the specific cancer sites and their relation to sleep quantity or quality.

Biological mechanisms

A number of possible explanations for the association between decreased sleep duration and chronic diseases can be discussed. Short-term sleep deprivation induces inflammatory processes as well as an activation of the sympathetic nervous system, increased blood pressure, higher evening cortisol levels, and impaired glucose tolerance [42], [43], [44], [45], [46], [47] thus leading to an elevated risk of cardiovascular diseases and diabetes. Moreover, short sleep was shown to be related to increased levels of ghrelin, an appetite stimulant, and decreased levels of leptin, a satiety factor, which promotes appetite and food intake thereby leading to an increased risk of obesity [10]. However, total energy intake or food intake did not substantially differ across sleep categories in this cohort (see Table 1). More research is necessary to examine the relationship between sleep duration and food intake or meal patterns and meals contents.

Concerning cancer, current experimental and epidemiological evidence supports a link between melatonin production and risk of the development of cancer, especially breast cancer. Melatonin is a pineal hormone that is involved in the circadian regulation and facilitation of sleep, the inhibition of cancer development and growth as well as the enhancement of immune function. Individuals with short night sleep experience a suppression of melatonin secretion. Thus, these individuals are more likely to suffer from sleep disturbances, immune suppression and moreover are at a higher risk of developing a number of different cancers [11]. In addition, other mechanisms like inflammatory processes or changes in the autonomous tone could possibly mediate the link between sleep deficiency and cancer risk. Still, further research is necessary to get advanced insight into the relationship between sleep deficiency and cancer.

Long sleep and chronic diseases

In agreement with others, our study indicates that prolongation of sleep above 9 h might be also not the best option. We found that risk of stroke was also increased by 65% in persons reporting long sleep duration (≥9 h). Similar observations were made by Chen et al. who found not only short sleep but also long sleep duration to be associated with ischemic stroke in postmenopausal women (HR = 1.24, 95% CI: 1.04–1.47 and HR = 1.70, 95% CI: 1.32–2.21 for women reporting 8 and ≥9 h of sleep) [21]. In addition, an increased risk of stroke was observed only for those sleeping more than 8 h in the First National Health and Nutrition Examination Survey (NHANES-I) [27]. Previous studies suggest that long sleep is further associated with an increased risk of developing incident diabetes [15], [16], [25], [26] which was not confirmed by our study. Of note, these studies were less stringent in excluding participants with chronic diseases at baseline which could affect sleeping hours.

The association of long sleep with cardiovascular diseases like stroke could be explained by poor sleep quality or poor health at baseline, e.g. chronic inflammation, unknown sleep apnea that could lead to subsequent hypertension, or circulatory disturbances in the brain, factors for which we could not control for in the statistical analysis. Long sleep may therefore represent an epiphenomenon of co-morbidity [48]. The positive association of a long overall sleep duration per day with stroke is also reflected by the association of daytime sleep and stroke among hypertensives which might be responsible for this overall effect.

Daytime naps and chronic diseases

It is important to note that there had been effect modification by prevalent hypertension at baseline regarding daytime sleep. Among hypertensive persons reporting daytime naps the risk for overall chronic disease, in particular stroke, tend to be increased; whereas among non-hypertensive participants that stated to sleep during the day the risk of overall chronic disease, primarily cancer, is decreased.

This observation could be explained by a worse health condition or relevant symptomatic sleep-related breathing disorder (SRBD) among persons with hypertension reporting daytime sleep. It has been shown that persons with poor health, pain, depression, or those with poorer quality of sleep at night, e.g. by fragmented sleep with elevated number of arousals, in general tend to have daytime naps more frequently [49], [50], [51]. Therefore taking long naps was found to be associated with higher morbidity and mortality, especially among the elderly [17], [28], [52], [53], which is in line with our findings.

In contrast, short daytime naps among healthy persons might be a part of good lifestyle or a result of good sleep hygiene with stress relieving effects like it is making siesta in southern Europe [52], [54], [55]. A nap of less than 30 min duration during the day promotes wakefulness and alertness, reduces a sleep deficit and enhances performance and learning ability. Thus, daytime sleeping habits in general can reflect different concepts of lifestyle and health.

Strengths and limitations of the EPIC-Potsdam study

The present study has a number of strengths including its prospective nature, the large sample size, high rate of follow-up, and verification of the self-reported information on disease through medical reports [30].

Though, with a recruitment proportion of 22.7% of the originally inquired persons, the EPIC-Potsdam cohort does not present a representative sample of the general population. In most studies, hence also in EPIC-Potsdam, the participating people are more health-conscious than non-participating people. Comparison to data from a representative health survey in eastern Germany showed that study participants in EPIC-Potsdam are characterized by a higher socio-economic status, higher health consciousness, and a better health status than the general population in East Germany [56].

One important methodological limitation of this study is the fact that sleep duration was self-reported and measured only at study recruitment. The Coronary Artery Risk Development in Young Adults (CARDIA) study showed that subjective reports of habitual sleep may be biased by systematic over-reporting among participants with short sleep durations compared to actigraph-measured sleep [57]. However, other comparisons of subjective sleep duration to the gold standard, polysomnography, or other more direct assessments like actigraphy showed good correlations [58], [59].

To increase validity of the results, it would be preferable to rely on repeated objective assessment of sleep quantity and quality, such as actigraphy [60] or even polysomnography [61], although feasibility of such methods might be questionable in large-scale cohort studies.

Another drawback of the EPIC study is that information on shift work or individual sleep requirement is missing. Furthermore, our question about the presence of any sleep disorder treated by a physician can not be regarded as a valid medical diagnosis and might underestimate the total prevalence of sleep disturbances. We also do not have further information about sleep quality or specific sleep disorders, like for example insomnia, sleep-disordered breathing (snoring, sleep apnea), restless legs syndrome, hypersomnia, or parasomnia, which might mediate the relationship of sleep and chronic diseases.

A further limitation is that all types of cancer were combined to one endpoint due to small number of cases for single cancer sites in this sample. However, in view of public health messages it is most relevant to ask whether the overall risk of cancer is impacted by sleep deprivation, independent of the type of cancer.

Conclusion

In conclusion, short sleep duration is a strong risk factor for development of chronic diseases and thus is of significant public health relevance as sleep habits are amenable to behavioral interventions. This may be especially important in the context of the reduced sleeping hours observed during the last decades. However, compared to other lifestyle factors, e. g. nutrition, smoking, or physical activity, sleep is less apparent to the public to be a modifiable risk factor for chronic diseases. Therefore, it is important to increase awareness of the importance of sleep among health professionals and the lay public. Since short sleep is not only a personal choice also the roots of sleep deprivation, for instance shift work and prolonged working hours, should be addressed by making changes in work patterns and access to sleep recovery [7], [8]. Future studies should explore the efficacy of sleep interventions or rather changes in sleep habits for prevention of chronic disease.