Sleep Duration and Cancer Risk : a Systematic Review and Meta-analysis of Prospective Studies

To assess the risk of cancers associated with sleep duration using a meta-analysis of published cohort studies, we performed a comprehensive search using PubMed, Embase and Web of Science through October 2013. We combined hazard ratios (HRs) from individual studies using meta-analysis approaches. A random effect dose-response analysis was used to evaluate the relationship between sleep duration and cancer risk. Subgroup analyses and sensitivity analyses were also performed. Publication bias was evaluated using Funnel plots and Begg's test. A total of 13 cohorts from 12 studies were included in this meta-analysis, which included 723,337 participants with 15,156 reported cancer outcomes during a follow-up period ranging from 7.5 to 22 years. The pooled adjusted HRs were 1.06 (95% CI: 0.92, 1.23; P for heterogeneity=0.003) for short sleep duration, 0.91 (95% CI: 0.78, 1.07; P for heterogeneity <0.0001) for long sleep duration. In subgroup analyses stratified by cancer type, long duration of sleep showed an inverse relation with hormone-related cancer (HR=0.79; 95% CI: 0.65, 0.97; P for heterogeneity=0.009) and a greater risk of colorectal cancer (HR=1.29; 95% CI: 1.09, 1.52; P for heterogeneity=0.346). Further meta-analysis on dose-response relationships showed that the relative risks of cancer were 1.00 (95% CI: 0.99, 1.01; P for linear trend=0.9151) for one hour of sleep increment per day, and 1.00 (95% CI: 0.98, 1.01; P for linear trend=0.7749) for one hour of sleep increment per night. No significant dose-response relationship between sleep duration and cancer was found on non-linearity testing (P=0.5053). Our meta-analysis suggests a positive association between long sleep duration and colorectal cancer, and an inverse association with incidence of hormone related cancers like those in the breast. Studies with larger sample size, longer follow-up times, more cancer types and detailed measure of sleep duration are warranted to confirm these results.


Sleep Duration and Cancer Risk: a Systematic Review and Meta-analysis of Prospective Studies
Hao Zhao 1,2,3& , Jie-Yun Yin 2& , Wan-Shui Yang 1 , Qin Qin 2 , Ting-Ting Li 2 , Yun Shi 2 , Qin Deng 1 , Sheng Wei 2 , Li Liu 2 *, Xin Wang 1 *, Shao-Fa Nie 2 * al., 2008b;von Ruesten et al., 2012), or only found a null association (Verkasalo et al., 2005;McElroy et al., 2006;Pinheiro et al., 2006;Sturgeon et al., 2012;Luo et al., 2013).The mechanisms underlying the association between short or long sleep duration and cancer risk are not fully understood.Two potential biological mechanisms have been proposed to explain how short sleep duration directly influenced cancer incidence, including impaired immune function and metabolic pathways related to obesity (Knutson et al., 2007;Marshall et al., 2008).Moreover, the altered melatonin secretion has also been shown as a potential risk factor of cancer (Stevens et al., 2007;Benke et al., 2013).Melatonin is synthesized and secreted by the pineal gland, and could be regulated by the information of light/dark environment (Cutando et al., 2011), rather than by sleep per se.The concentration of melatonin is controlled by an endogenous circadian timing system and suppressed by light (Zeitzer et al., 2000).Usually, sleep has been viewed as a maker for "exposure to darkness", therefore may act as regulator of melatonin concentration, consequently, as a potential impact factor of cancer risk.To address this issue, we performed a meta-analysis to assess the relationship between short or long duration of sleep and cancers risk.The meta-analysis included prospective cohort studies with large sample size and long duration of follow-up, which ensured the higher statistical power compared to each individual study.

Literature Search
We developed a comprehensive search strategy to identify studies that reported the longitudinal association between duration of sleep and cancers risk.We searched the electronic databases PubMed, Embase and ISI Web of Science using the terms "sleep", "cancer", "tumor", "carcinoma", and "neoplasms".Furthermore, we reviewed reference lists of original and review articles to search for additional studies.Only those that were published as peerreviewed study were considered.No language restriction was applied.

Inclusion and Exclusion Criteria
For inclusion, studies had to meet the following criteria: (1) original article, (2) prospective cohort design, (3) assessment of duration of sleep as baseline exposure, (4) cancer recorded prospectively as outcome, (5) followup of at least 3 years, (6) adult population, (7) based on the same cohort but reported different cancers.If publications were duplicated, we only included the one with the most detailed and latest information for both exposure and outcome.Studies were excluded if they: (1) had a casecontrol study or cross-sectional design, (2) were review articles or meeting abstracts, or (3) were not conducted in humans.

Data Extraction
The eligibility of each full-text article was assessed independently in a standardized manner by two investigators, and differences were resolved by discussion and consensus.Information extracted included first author's surname, publication year, country, recruitment year, study design, number of participants, number of cases, sample characteristics (e.g., gender, age), duration of follow up, method of sleep data collection, reference category of sleep, category for "short" and "long" sleep, cancer type, RRs or HRs that reflected the greatest degree of adjustment for potentially confounding variables by both short and long sleep duration, corresponding 95% CI, and covariates adjusted in the statistical analysis.
For every study, the median or mean sleep duration for each category was assigned to each corresponding relative risk.When the median or mean sleep duration per category was not reported in the article, we assigned the midpoint of the upper and lower boundaries in each category as the average duration.If the upper boundary for the highest was not provided, we assumed that the boundary had the same amplitude as the adjacent category.When the lowest category was open-ended, it was then assigned by 80% of the lowest boundary.

Definition of "Short" and "Long" Sleep Duration
In most of the studies, the reference category of sleep duration was 7-8 h or 7 h per night or 24 h except, while three studies defined as ≤6 h.Short sleep duration was defined as ≤5 h or ≤6 h; long sleep duration was defined as ≥9 h for either nighttime sleep or 24 h sleep in the majority of studies.

Statistical Analysis
The hazard ratios (HRs) or relative risks (RRs) were extracted from the selected publications and were used to measure the relationship between sleep duration and cancer risk.Their standard errors were calculated from the respective confidence intervals.The value from each study and the corresponding standard error were transformed into their natural logarithms to stabilize the variances and to normalize the distribution.The study-specified HRs or RRs were pooled using the fixed-effect model if no or low heterogeneity was detected, or the random-effects model otherwise.Forest plots were produced to visually assess the HRs and corresponding 95% confidence intervals across studies.In dose-response analysis, we used the method proposed by Greenland and Longnecker (Greenland et al., 1992) and Orsini (Orsini et al., 2006) to compute the trend from the correlated Log HR estimates across categories of sleep duration.We examined a potential nonlinear doseresponse relationship between sleep durations and cancer risk by modeling sleep durations using restricted cubic splines with 3 knots at percentiles 25%, 50%, and 75% of the distribution (Harrell et al., 1988).Heterogeneity of HRs across studies was tested by Q-statistic (P<0.05 was considered indicative of statistically significant heterogeneity) and quantified by the I 2 statistic (values of 25%, 50%, and 75% were considered to represent low, medium, and high heterogeneity, respectively) (Higgins et al., 2003).P value for nonlinearity was calculated by testing the null hypothesis that the coefficient of the second spline is equal to zero.Funnel plot and Begg's test were used to detect the possibility of publication bias (Begg et al., 1994;Egger et al., 1997).We also conducted subgroup analyses by gender, sleep period, geographic location, occupational status, cancer type, menopause status, number of cases, and definition of short or long sleep duration and reference category.Moreover, sensitivity analyses were performed to evaluate the influences of included study and participant characteristics on study results.The dose-response analysis was conducted with SAS 9.2 (SAS Institute Inc., Cary, NC).Other statistical analyses were performed using STATA statistical software version 12.0 (StataCorp, College Station, Texas).P values were 2 sided with a significance level of 0.05.

Search Results
Based on our selection criteria, 11547 citations were attained via the initial database search.After the first  round of screening based on titles and abstracts, 21 articles needed for further assessment.After examining those articles in more detail, 9 papers were excluded for reasons shown in Figure 1.One cohort study which was published in two different journals reported the association between sleep duration and breast cancer in different time, the latest one was finally included (Wu et al., 2008;Wu et al., 2013).We were unable to include one study due to lack of data (Odegaard et al., 2013).Finally, 12 articles fulfilling the inclusion criteria were included in the meta-analysis.Among these 12 articles, five studies reported breast cancer (Verkasalo et al., 2005;Pinheiro et al., 2006;Kakizaki et al., 2008b;Vogtmann et al., 2013;Wu et al., 2013), two studies on colorectal cancer (Jiao et al., 2013;Zhang et al., 2013), one each study on prostate cancer (Kakizaki et al., 2008a), endometrial cancer (Sturgeon et al., 2012), epithelial ovarian cancer (Weiderpass et al., 2012), thyroid cancer (Luo et al., 2013), and all type cancers (von Ruesten et al., 2012).

Study Characteristics
Characteristics of 12 selected studies including 13 cohorts are shown in Table 1.Overall, the total number of participants included was 723,337, with 15,156 reported cancer outcomes.Six studies were performed in USA, three in Japan (Kakizaki et al., 2008b;Weiderpass et al., 2012), and one each from German (von Ruesten et al., 2012), Finland (Verkasalo et al., 2005), and Singapore (Wu et al., 2008).The study samples ranged from 12, 222 to 142, 933, and the median of the follow-up durations is 11.3 years.Nine studies focused on the cancer of women, one only on men (Kakizaki et al., 2008a), one on men and women combined (von Ruesten et al., 2012), and one reporting men and women separately according to different cohorts (Zhang et al., 2013).Most of studies reported the sleep duration of 24 hours period, while four studies investigated typical night sleep duration (Sturgeon et al., 2012;Jiao et al., 2013;Luo et al., 2013;Vogtmann et al., 2013).In most of the studies, cancer was assessed by self-reported and confirmed by adjudicators via medical record review, four studies identified cancers through cancer registry data (Verkasalo et al., 2005 When results were reported for men and women separately, it was entered into meta-analysis as separate studies.Overall, there were 10 cohorts available for the relationship between short sleep duration and cancer risk and 13 cohorts available for the long sleep duration.

Quantitative Analyses
Figure 2 shows the forest plot of relationship between short sleep duration and risk of overall cancers.Short sleep duration significantly associated with all cancer risk (HR=1.06;95% CI: 0.92-1.23),with no evidence of publication bias (Begg's test, P=0.531; Egger's test, P=0.373; Figure 4a).The heterogeneity was high (I 2 =63.8%,P=0.003).The sensitivity analysis showed that omission of anyone of these studies did not change the quantitative relationship between short sleep duration and cancer risk (all P>0.05).
Figure 3 presents the pooled results of the relationship between long sleep duration and cancers risk.Long sleep duration was not significantly related to all cancer risk (HR=0.91;95% CI: 0.78, 1.07), with no evidence of publication bias (Begg's test, P=0.050; Egger's test, P=0.072; Figure 4b) and high heterogeneity between studies (I 2 =67.6%,P<0.001).Meanwhile, after omission each study one by one and recalculating the combined estimates on remaining studies., the results did not notably alter the main results (all P>0.05).
In dose-response analysis, we found no evidence on a nonlinear association between sleep duration and cancer risk (P for nonlinearity =0.5053) (Figure 5).Compared with 7.5 hours of sleep duration, the summary relative risk of cancer for per hour increase of sleep per day was 1.00 (95% CI: 0.99, 1.01; P for linear trend=0.9151),and one hour of sleep per night was 1.00 (95% CI: 0.98, 1.01; P for linear trend=0.7749).
As shown in Table 2, long sleep duration was found to be associated with cancer risk in Asian population (HR=0.49;95% CI: 0.27, 0.90).In terms of cancer types, the associations were significant between long sleep  doi.org/10.7314/APJCP.2013.14.12.7509 Sleep Duration and Cancer: a Systematic Review and Meta-analysis of Prospective Studies duration and hormone related cancer (HR=0.71;95% CI: 0.55, 0.92) and colorectal cancer (HR=1.29;95% CI: 1.09, 1.52).In addition, no significant associations were observed when stratified by gender or sleep period.

Discussion
Our studies provided a comprehensive review of the literature and quantitative estimates of longitudinal associations between short or duration and risk of cancers among cohort studies of adults around the world.Although, our results did not found short or long sleep duration associated with all cancer risk, long sleep duration among Asians presented a protect role in cancer initiation.Interestingly, in terms of cancer subtypes, long sleep duration was found to be associated with increased risk for colorectal cancer and decreased risk for hormone related cancer in the stratified analysis.
The protected effect of long sleep duration on hormonerelated cancer was observed.Although, the mechanisms underlying are not fully understood, melatonin might exert its ant-cancer role via inducing tumor cells apoptosis and anti-proliferation and anti-angiogenesis (Di Bella et al., 2013).Melatonin has been described to be involved in inhibitory influences on sex hormone levels (Cohen et al., 1978), which have been reported to be associated with cancers of breast, endometrium, ovary, prostate, and thyroid (Henderson et al., 2000).In addition, Melatonin has shown dose-dependent anti-oxidative effect, providing protection against damage from carcinogenic substance, hence acting as a free radical scavenger (Reiter et al., 2008).Melatonin level was suggested to be positively related to sleep duration (Aeschbach et al., 2003).Therefore, longer sleepers may be possessed of higher melatonin concentration and subsequent protective effect in hormone-related cancers.
Conversely, a positive association between long sleep hours and colorectal cancer was observed.To our knowledge, the mechanism explained the effect of long sleep duration on cancer risk is still obscure.The association between long sleep duration and cancer may be explained by comorbidities (Knutson et al., 2006) and residual confounding.For instance, some other mental or physiologic disorder, low socioeconomic status, low level of physical activity, undiagnosed chronic comorbid conditions have been suggest to be correlated with long sleep duration and can confound the association with cancer incidence (Stranges et al., 2008b).The pooled results stratified by geographic location indicated that the effects for both short and long sleep duration were significant in studies performed in Asia, predominantly in Japan.The biologic difference between Asian and Caucasian study populations regarding melatonin suppression could be an explanation for some conflicting results (Girschik et al., 2010).Nonetheless, this subgroup finding of a positive association between sleep duration and cancer risk was based on small number of studies and thus further studies should be warranted.Several limitations of this meta-analysis should be considered.First, since the data were from observational studies, the confounding bias from included study per se cannot be excluded, though we made an attempt to include adjusted estimates from multivariate models from each contributing all of the studies included in this meta-analysis assessed sleep duration using written questionnaires by self-reported which may not obtain actual information of sleep duration.Although correlations between subjective estimates of sleep duration and the more direct assessments have been found (Signal et al., 2005), this might still attenuate the effect of our study.Third, the studies included used different reference categories and definitions of short and long sleep duration, hence precluding our ability to provide sleep duration recommendations in public health practice.However, the pooled results in subgroup analyses did not changed significantly (as shown in Table 2).Fourth, sleep duration was assessed at one point in time in most of studies, and it might not accurately reflect the sustained effects of sleep duration over time when relating them to long-term development of cancer.Fifth, the cancer types reported in the majority of studies were restricted in breast, prostate, endometrial, thyroid,colorectal and epithelial ovarian cancers, except one study reported all types from German (von Ruesten et al., 2012).Therefore, it can only be representative of the cancer types that have been included and are unable to provide a representative inference of all cancer types.
In summary, our meta-analysis suggests a positive association between long sleep duration and colorectal cancer, and an inverse association between long sleep duration and incidence of hormone related cancer.Studies with larger sample size, longer follow-up times, more cancer types and detailed measure of sleep duration are warranted to confirm these results.

Figure 2 .
Figure 2. Forest Plot of the Association Between Short Sleep Duration and Cancer Risk.Results are expressed as hazard ratio(HR) and 95% confidence intervals(95% CI) Figure 1.Flow Chart of the Study Selection Process

Figure 4
Figure 4. a. Funnel Plot for Meta-analyses of Short Duration of Sleep and Cancer Risk; b.Funnel Plot for Meta-analyses of Long Duration of Sleep and Cancer Risk

Table 2 . Stratified Analyses of Hazard Ratio (HR) of Cancer Risk
a represent the short sleep duration group; b represent the long sleep duration group