A Prospective Study on Habitual Duration of Sleep and Incidence of Breast Cancer in a Large Cohort of Women

Both long and short habitual sleep may be associated with increased risk of coronary heart disease (1) and symptomatic diabetes (2). Duration of sleep has also been associated with increased all-cause mortality in several studies (3–5), although only among men in one study (5). There is some evidence indicating that sleep duration may affect women and men differently. In a study by Amagai et al. (6), short sleep was associated with increased all-cause mortality only among men whereas long sleep was associated with increased all-cause mortality only among women. Data on sleep duration and cancer risk are sparse and inconclusive. An association between melatonin, a hormone closely related to sleep (7), and breast cancer has been suggested (8), raising the question of whether sleep duration may also affect breast cancer risk.

Sleep pattern disruptions related to night-shift work, a surrogate for light exposure at night, may increase breast cancer risk (9), possibly through decreased levels of melatonin (8). Melatonin is a hormone secreted primarily by the pineal gland during the dark phase of the light-dark cycle (10); exposure to light suppresses melatonin production (11). Melatonin may inhibit breast tumorigenesis directly by inhibiting breast cell proliferation and invasiveness, and indirectly by suppressing the mitotic activity of endogenous hormones such as 17β-estradiol (8). Decreased first morning urine levels of 6-sulfatoxymelatonin were recently associated with increased premenopausal breast cancer risk (12). First morning measurements of 6-sulfatoxymelatonin, the major metabolite of melatonin (7), seem to closely reflect nighttime melatonin levels (13).

Melatonin levels may be closely related to sleep duration. Serum melatonin concentrations have shown to be lower in habitual short sleepers (<6 h/night) than in long sleepers (>9 h/night; ref. 14), and in elderly insomniacs than in age-matched controls (15). Under experimental conditions, increasing the length of the night (participants were confined to a dark environment 14 hours per day for four weeks) was associated with both longer sleep duration and increased nocturnal secretion of melatonin (16). Sleep duration may influence melatonin levels by determining length of light exposure, thereby affecting breast cancer risk.

Sleep duration may also influence breast cancer risk by affecting immune functioning (17). Although it is unclear whether decreased immune function plays direct role in breast cancer, the importance of immune responses in antineoplastic activities is established (18). In this study, we evaluated whether self-reported habitual sleep duration influences breast cancer incidence in a large cohort of women.

We prospectively examined the relation between self-reported habitual sleep duration and breast cancer incidence among participants of the Nurses' Health Study. The Nurses' Health Study is composed of 121,700 U.S. female registered nurses, ages 30 to 55 years at entry, who responded to a mailed questionnaire in 1976 (19). Self-reported information on several covariates including weight, reproductive events, physical activity, and smoking has been updated every 2 years via mailed questionnaires; information on diet and family history of breast cancer has been updated every 4 years. Information on height, age at menarche, and age at first birth was obtained in 1976. Information on sleep duration was obtained in 1986 and again in 2000. In 1986, participants were asked to specify the total hours of sleep in a 24-hour period; possible choices included ≤5, 6, 7, 8, 9, 10, and ≥11 hours. Information on timing and quality of sleep was not collected from participants.

In 1986, 83,027 participants provided information on habitual sleep duration. We excluded participants with a diagnosis of cancer (except non-melanoma skin cancer) before baseline, leaving 77,418 women for this analysis. Incident invasive breast cancer cases, diagnosed between June 1, 1986 and May 31, 2002, were identified through self-report and confirmed by blinded medical chart review or by the participants. Most breast cancer deaths were identified by the next of kin or through National Death Index searches.

We used Cox proportional regression models to estimate breast cancer hazard ratios (HR) and 95% confidence intervals (95% CI) by sleep duration [≤5, 6, 7 (reference), 8, and ≥9 hours per 24 hours] adjusting for established risk factors for breast cancer (most updated biennially), and likelihood ratio tests to evaluate whether the association between sleep duration and breast cancer was modified by menopause status, history of night-shift work, or smoking status. Only sleep information obtained in 1986 was used in our main analyses. Additional analyses restricted to participants who consistently reported same sleep duration in both 1986 and 2000 questionnaires were conducted.

Participants sleeping ≤5 hours were more likely to be postmenopausal, nulliparous, and to have worked rotating night shift, and less likely to be married or to report current postmenopausal hormone use or history of benign breast disease. Alcohol intake was lowest among short sleepers and highest among long sleepers. Both groups were more likely than normal sleepers to report depressive symptoms (Table 1).

Our analysis included 4,223 incident cases of invasive breast cancer newly diagnosed during 1,153,077 person-years of follow-up between 1986 and 2002. We did not find any significant difference in breast cancer incidence across categories of sleep length in age-adjusted analyses (Table 2). Additional adjustments for potential confounding variables including caffeine consumption, depression, and stress did not alter our results. We did not adjust for night-shift work because the effects of shift work on breast cancer may be mediated through changes in sleep duration; however, adjusting for shift work in our main analyses did not affect our results. Secondary analyses restricted to participants who reported the same number of hours of sleep in 1986 and in 2000 (n = 26,346) provided similar results although there was a moderate trend in risk increase in breast cancer among women sleeping >7 hours (Table 3).

We also conducted analyses in the entire cohort stratifying on menopause status or night-shift work; results were essentially unchanged from our main analyses (Table 4). Additional analyses excluding breast cancer cases diagnosed within 4 years after baseline in 1986 (to minimize the possibility of undiagnosed breast cancer affecting sleep duration in 1986) did not alter results [multivariate-adjusted HR and 95% CI for those sleeping ≤5, 6, 8, and ≥9 hours compared with 7 hours were 0.94 (0.79-1.12), 0.94 (0.86-1.03), 1.04 (0.95-1.14), and 0.90 (0.76-1.08), respectively]. Secondary analyses including both in situ and invasive breast tumors were also conducted but results were essentially unchanged (data not shown).

Although plasma melatonin may be inversely related to tumor estrogen receptor concentrations (20), habitual sleep duration was not associated with estrogen-positive or estrogen-negative tumors. Moreover, in spite of recent evidence suggesting that plasma melatonin levels may be affected by smoking (21), stratifying our analyses by smoking status produced essentially unaltered results (data not shown).

The effect of sleep duration on health has only begun to be elucidated. Sleep restriction has been associated with detrimental physiologic effects, including impairment of glucose tolerance (22) and inflammatory responses (23). Both long and short habitual sleep have been implicated in various health outcomes. Long sleep was associated with increased risk of coronary heart disease (1) and symptomatic diabetes (2). Short sleep was also associated with increased risk of coronary heart disease (1) and diabetes, although the association with the latter was not statistically significant after adjustment for body mass index in one study (24), perhaps suggesting that sleep restriction may increase diabetes risk through changes in body mall index. Duration of sleep has also been associated with increased all-cause mortality in several studies although their findings are not entirely consistent. Both short and long sleep duration were associated with increased all-cause mortality among participants of the Nurses' Health Study (3) and among participants of the Japanese Collaborative Cohort Study (4). Kojima et al. (5) reported an association between both short and long sleep duration and increased all-cause mortality among men but not among women. Indeed, there is some evidence indicating that sleep duration may affect men and women differently. In a study by Amagai et al. (6), short sleep was associated with increased all-cause mortality among men, but not among women, whereas long sleep was associated with increased all-cause mortality only among women. In our study, sleep duration was not associated with breast cancer incidence although a moderate trend in risk increase towards those sleeping >7 hours was suggested when analyses were restricted to participants who consistently reported same sleep duration in 1986 and 2000.

To our knowledge, only one other recent study (also prospective in nature) has explored the association between sleep duration and breast cancer risk, in which a protective effect of long sleep duration was suggested (25). However, in the Finish study, sleep duration reported in 1975 and 1981 were used to predict breast cancer from 1976 to 1981 and from 1982 to 1996, respectively. Thus, undiagnosed breast cancer may have affected duration of sleep, as it is possible that those who reported longer sleep hours were less likely to have had breast cancer. Moreover, the proportion of short (≤6 hours) and long sleepers (≥9 hours) in the study by Verkasalo et al. (∼10% and 16%, respectively) varied considerably from the proportion of short and long sleepers in our study (∼30% and 5%, respectively); intrinsic differences between the two populations may explain, at least in part, the differences in findings. Finally, because their findings were based on a limited number of breast cancer cases [14 (≤6 hours), 125 (7-8 hours), and 7 (≥9 hours) cases], their results may be associated with greater uncertainty.

Exposure to light at night may be associated with decreased melatonin production (26), which is implicated in increased breast cancer risk (12, 27, 28). Melatonin production may be closely related to sleep duration (14–16). Short sleepers are probably more exposed to light by staying up later or by waking up earlier, potentially lowering melatonin production. Our results, however, are not compatible with the hypothesized detrimental effects of decreased sleep duration on breast cancer incidence through decreased melatonin levels.

We were unable to capture whether most of the reported sleep occurred at night, which is of particular importance to our cohort of nurses, as many have worked night shifts. Because melatonin is produced mainly at night, the effects of sleep duration on breast cancer may be attenuated among participants who sleep mostly during daytime. However, analyses restricted to those who never worked night shift produced results similar to our main analyses. Urine concentrations of melatonin have been associated with premenopausal breast cancer risk (12) and decreased melatonin levels may primarily affect breast cancer diagnosed before menopause. However, our results were unaltered by excluding postmenopausal women from analyses.

Our study has some limitations. Although its prospective nature eliminates recall bias, nondifferential misclassification of exposure cannot be ruled out, as we relied on self-reported data. Because our exposure has multiple categories, even random misclassification could have biased our results in any direction (29). However, self-reporting of habitual sleep duration was strongly correlated to 1-week sleep diaries in our population (3). Moreover, in light of the association between melatonin levels and breast cancer risk, it might have been interesting to consider timing of sleep in our analyses. However, we accounted for rotating shift work, which is likely to affect sleeping pattern and the times that sleep occurred. Additionally, information on quality of sleep was not obtained from our participants. Though it is possible that quality of sleep is affected among women working rotating shifts, this is likely to affect women regardless of duration of sleep. Although we used information on sleep duration obtained only at baseline in our main analyses, restricting our analyses to those who consistently reported same amount of sleep hours in 1986 and 2000 (thus minimizing misclassification of exposure) produced similar results, though a moderate trend in risk increase towards those sleeping >7 hours was suggested. We cannot rule out residual confounding by characteristics such as healthy lifestyle, although we controlled for closely related covariates (e.g., diet and physical activity). Short and long sleepers may differ from each other in many ways, thus confounding by unmeasured factors cannot be discarded.

Our study has several strengths, foremost its large size, prospective nature, and long duration of follow-up. Information on most of covariates was updated every 2 years. Although measurement error is inevitable, particularly in studies using self-reporting instruments, using repeated measurements over the follow-up period may reduce within-person random error. The detailed nature of our questionnaires allowed for control of many potential confounders, including most established breast cancer risk factors.

To our knowledge, this is one of the first studies to explore the effects of sleep duration in breast cancer. Our study provides no convincing evidence for an association between sleep duration and breast cancer. Our findings suggest that unknown sleep-related factors may counterbalance the effects of melatonin on breast cancer, and raise the possibility of mechanisms beyond decreased melatonin levels to explain the relation between shift work and breast cancer risk.

Grant support: National Cancer Institute, NIH research grant CA87969 (The Nurses' Health Study).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Sue Malspeis for technical assistance and especially the participants of the Nurses Health Study for their continuing cooperation.