The Impact of Sleeping Duration on the Risk of Breast Cancer: A systematic review and meta- analysis of population-based cohort studies

The impact of different sleeping categories on the risk of breast cancer has remained debatable. This paper sought to systematically synthesise the available literature on this relationship from population-based cohort studies using meta-analytic procedures. Studies about sleep duration and breast cancer were identified from the Cochrane Library database, EMBASE and PubMed databases for papers published up to February 2019. Identified studies were analysed for quality using the Newcastle-Ottawa Scale. Effect sizes were visualised using funnel plots. Study heterogeneity was quantified using I2 and visualised using Baujat plots. Publication prejudice was evaluated using Eggers regression model, with visualisations using funnel plots. Eight cohort studies met the inclusion criteria. Random-effects model revealed nonstatistically significant evidence of an association between short or long sleep and breast cancer Odds Ratio (OR) =0.90;(95% CI 0.79–1.02) and OR=0.95 (0.88–1.02) respectively. There was moderate to high heterogeneity I2 (95% CI)=74.40% (48.20– 87.40%) among studies examining short sleep and breast cancer, and low to moderate heterogeneity in studies for long sleep and breast cancer I2 (95% CI)=3.0% (0.00– 68.60%). This study found non-substantial evidence of associations between sleeping periods and breast cancer in women. Studies employing novel sleep-measurement methodologies should be carried out to examine the underlying relationship. Reinvention: an International Journal of Undergraduate Research 13:1 (2020)


Introduction
Globally, the burden of site-specific cancers has increased based on published global estimates (Bray et al., 2018). While most countries have implemented measures to reduce the effects of cancer -such as the implementation of national policies like the National Cancer Control Strategy in Kenya (Topazian et al., 2016) and the development of early detection and mammographic screening centres (Yip et al., 2008) -the disease still remains a significant cause of death with its burden expected to increase due to ageing, population growths and changing life regimens such as cigarette smoking, alcohol drinking and general body dormancy (Sanchis-Gomar et al., 2015;Jung et al., 2016). Breast tumours have remained a domineering cause of demise in the female population. In the USA alone, the rate of invasive breast cancer is approximated at 232,340 cases, with one in every eight US women anticipated to develop breast cancer during the course of her life (Ma and Jemal, 2013). Belgium and Luxembourg had the highest age-standardised rates of breast cancer in 2018 (113.2 and 109.3 per 100, 000 respectively) based on estimates from the American Institute for Cancer Research.
The underlying relationship between sleeping patterns and all cancers has been examined in depth elsewhere in case-control studies (Wang et al., 2015(Wang et al., , 2016Gu et al., 2016). Whereas sufficient sleep is necessary for healthy development, insufficient sleep has been previously linked with an increased incidence of cancer of the breast as a result of possible suppression of melatonin production (Stevens, 2005).
Sleeping patterns appear to be changing. In previous studies, the number of individuals who sleep for short durations of time has been reported to increase whereas the number of individuals who sleep for long durations of time (9+h of sleep per night) has been reported to decrease considerably (Jean-Louis et al., 2014).
While there exists ample literature on these relationships, as presented above, there exist no records that explicitly and systematically review, synthesise and evaluate the available research on the effect of both long and short sleeping patterns on the risk of breast cancer in both pre-and post-menopausal women of different ethnicities, a factor that is important in understanding variations in incidence by race. Moreover, the fund of knowledge concerning dose-response relationships of the various studies using self-reported sleeping categories from population-based cohort studies is deficient. Also, relations between sleeping patterns and breast tumours from analyses previously conducted is obscure (Markt et al., 2016;Heckman et al., 2017).
This systematic review and meta-analysis aimed to systematically evaluate existing literature and update the fund of knowledge on the relationship between both long and short sleep patterns on the risk of breast cancer using multi-ethnic populationbased cohort studies. We also aimed to explore the relationships performing subgroup analyses for publication biases and study heterogeneity.

Methodology
This systematic review and meta-analysis was conducted in accord with the 'Preferred Reporting Items for Systematic Reviews and Meta-Analyses' (PRISMA; Shamseer et al., 2015).

Data sources
Studies were expansively and intensively identified through searches in the academic journal databases Cochrane Library database, PubMed and EMBASE, and they included studies up to February 2019 that evaluated the relationship between the duration of sleep and the risk of acquiring breast cancer. Study identification and extraction involved use of keywords including 'sleep' or 'sleep duration' and 'tumor' or 'cancer' or 'breast cancer'. Additional pertinent cohort studies were acquired through scanning the bibliographies of studies previously identified.

Study selection and eligibility criteria
Studies were initially screened for relevance using their abstracts and titles. Those studies that bore potential of inclusion into the review were identified and their full versions obtained. Inclusion of articles was based on their novelty, cohort studies, or because the paper explicitly provided estimates for calculation of Odds Ratio (OR) and 95% Confidence Intervals (CI). Studies were included that allowed calculation of these estimates through their literature, were published in English and were based on adult populations. On the other hand, review articles, case-control studies, studies with duplicated data, that did not provide estimates for OR and CI, were clinical trials, cellrelated or did not mention breast tumours were excluded.

Data extraction
Spreadsheets which were designed in MS Excel specifically for this study. Where papers met the inclusion criteria, their details were checked and entered into the spreadsheet. For all included studies, details of the authors, year of publication, the country where the studies were conducted, types of cancers reported and the number of cancer cases reported at the end of the evaluation period were extracted (see Table   1). Also, the sample sizes and published estimates (Relative Risk (RR), Odds Ratio (OR), Hazard Ratio (HR) and 95% CI), including the author conclusions, were obtained.
Unadjusted Odds Ratios were calculated using the published study estimates. The number of events in the experimental group (Ee), the total number of persons in the experimental group (Ne) and the number of events and total individuals in the reference group (Ec and Nc respectively) were extracted for analyses.   Meta-analytical procedures were carried out using the inverse variance method and DerSimonian-Laird Estimator for tau 2 . Both fixed-and random-effects models were used in pooling the risk estimates from individual studies. The extracted associations from each study were used in evaluating the relationship between the risk of breast cancer and the extent of sleep (RR, HR and OR 95% CI). These estimates were deemed equivalent to each other due to the rareness of breast cancer as an outcome (Shamseer et al., 2015).
The OR for each included study were calculated and 95% CIs estimated for both short and long sleepers in comparison to the reference category. A standard reference category of 7h of sleep per night was employed to allow uniformity, comparability and to reduce biases.
Between study, heterogeneity was guesstimated using the Q test (conveyed with a pvalue) and quantified using I 2 statistic. The Q-statistic was used to assess the null hypothesis that all the studies included in the analysis are investigating a similar effect. A statistically significant Q-statistic was suggestive of a non-similar effect size among the studies (Quintana, 2015). In quantifying heterogeneity, I 2 values from 25%, 50% and 75% were viewed as evidence of low, modest and high heterogeneity respectively (Quintana, 2015).
Sub-group analyses were performed by sleep duration as either short or long sleep duration against the developed standard reference category (7h). Funnel plots and Baujat plots were used to inspect the included studies for possibilities of publication bias and each studies contribution to overall heterogeneity respectively. Eggers regression was used to test for publication bias (Egger et al., 1997). Forest plots were used to visually present the pooled effects of the studies end to end with the calculated summary effect sizes (Quintana, 2015). All analytical procedures used R programming environment, version 3.6.0, and employed the R Package 'Meta' (Schwarzer et al., 2015). P-values smaller than 0.05 were deemed statistically significant.

Search results and selection
Database searches identified 5288 records with 3418 from EMBASE, 1321 from PubMed and 549 articles from the Cochrane Library (see Figure 1). Article screening excluded a total of 4236 articles from the analysis since they were either reviews, meta-analyses, reports or showed obvious irrelevance. From the remainder, 95 articles were then screened and 70 articles were subsequently excluded since they did not provide relevant data. The 25 remaining articles were further evaluated based on the inclusion and exclusion criteria to remain with eight population-based cohort studies after exclusion of ten articles with no mention of breast cancer and four case-control studies (excluded because they do not clearly indicate the temporal sequence between exposure and outcome).   hours of sleep per night was used instead for uniformity, consistency and to allow comparability of the findings. The probability that a short sleeper develops breast cancer was 0.92 times higher than for the reference group. The influence of short sleep on the risk of breast cancer was statistically significant (Figure 2).
The amount of heterogeneity was I 2 =74.40% (95% CI 48.20-87.40), suggestive of moderate to high heterogeneity among the included studies. The Q-statistic was Q=27.36 (p=0.00). Eggers regression revealed non-statistically significant evidence for funnel plot asymmetry (p=0.74). This was supported by the funnel plot for the assessment of publication bias, which also illustrated symmetry ( Figure 3). Study 7 (Hurley et al., 2015) accounted the most towards overall heterogeneity as illustrated by Figure 5.  The pooled-effect OR (95% CI) for the included studies was OR=0.95 (0.89-1.02) for the cohort studies included in the analysis based on the fixed-effects model. The probability that a long sleeper (≥8h) develops breast cancer was 0.95 times higher than the probability that a person in the reference group (7h) develops breast cancer. The influence of extended sleep on the risk of breast cancer was statistically insignificant (p=0.21) (Figure 4).

Discussion
This study comprehensively reviewed literature related to the impact of both long and short categories of sleeping on the risk of developing breast cancer in females using population-based cohort studies that met the inclusion criteria. Whereas this analysis found no statistically significant links between long sleeping duration and the risk of breast cancer based on both fixed-and random-effects models, the study did find statistically significant evidence of influence of short sleeping patterns on the risk of breast cancer based on the fixed-effects model. However, on the other hand, the random-effects model revealed non-statistically significant evidence of an association between short sleep and cancer of the breast.
Comparing these findings (random-effects model results) to previously conducted individual researches on the same topic, consistencies were with a study that reported that sleeping durations (both short and long) had a non-significant statistical impact or association with an increase in cancer risk (Zhao et al., 2013). Besides, an analysis of prospective cohort studies by Lu et al. (2013) did not find substantial evidence of an association between the duration of sleep and an increased risk of all cancers (Lu et al., 2013).

Possible explanations of associations between breast cancer and sleep
While cancer as a disease has a multi-faceted etiology, different studies have tried to explain the link or pathways behind the relationship between breast cancer (and all cancers) and the duration of sleep. In previous studies, decreased melatonin levels have been linked to short sleeping durations; in addition, melatonin has been proposed as a suppressor of the initial phase in tumour generation, a process that inhibits propagation of cancer cells in humans (Blask, 2009;Hill et al., 2015).
Moreover, studies have also tried to explain the association between breast cancer using sex hormones (Germain, 2011). Melatonin might have a modulating effect on the production of sex hormones through its interactions with estrogen-signalling pathways using a variety of mechanisms (Alvarez-García et al., 2013). Bovbjerg (2003) linked impairments in immune functioning in his explanation of sleep and regulation of the immune system. Sleep deficiency studies have found that changes in duration of sleep might suppress the functioning of the immune system, resulting in shifts in the production of cytokine (Chen et al., 2018). Also, circadian physiology disruptions have also been used in explaining sleep disturbances with respect to all cancers (Hill et al., 2015).

Study strengths and limitations
This study has comprehensively used up-to-date literature on sleep and breast cancer, predominantly cohort studies to explain whether there exist associations between long and short categories of sleep and risk of breast cancer in women. The studies synthesised had considerably large sample sizes, which improved the statistical power.
Moreover, the pooled effects of the studies included in the analysis were analysed using both fixed-and random-effects models, allowing comparison of the findings.
Sub-group analysis was performed separately using both short and long durations of sleep, a situation that allowed an analysis of the effect of each paying attention to breast cancer. Besides, data for the pooled effect was obtained from the primary cohort studies in which individuals were followed prospectively. The methodology employed in the evaluation of the eminence of the cohort studies encompassed in this analysis strengthens the validity of the findings of this paper.
On the other hand, this meta-analysis did not use adjusted estimates from the previously conducted studies, a situation that might have allowed effects of confounding variables, which might obscure the true associations between variables.
Studies synthesised in this paper had different classifications of sleeping categories in addition to the measures of sleep being self-reported. Studies reported different measures of association (RR, OR and HRs); we, however, used the OR calculated from the estimates obtained from various studies (see Script).

Conclusions
While individuals continue to sleep less due to stress, anxiety, depression and change in lifestyle behaviours, findings from this analysis suggest non-statistically significant relations between durations of sleep and breast cancer. Future research studies should be focused on establishing the links and mechanisms suggesting relationships between sleep and breast cancer using more reliable measures of sleep durations to enhance the quality of the sleep measures obtained.    List of tables