Sir,

We read with interest the recent publication by Qian et al (2014). The authors examined the risk of incident breast cancer (BC) associated with sleep duration using data from Breast Cancer Detection Demonstration Project follow-up cohort, and found a null association between sleep hours and overall BC. They also reported risk estimates for BC according to different molecular subtypes of BC, and suggested a decreased risk for estrogen receptor (ER)+progesterone receptor (PR)+ BC with shorter sleep duration. The information provided is of interest as the relationship between sleep and BC is of increasing concern. However, we would like to raise several concerns related to this paper.

First, the validity of the sleep questionnaire used in the study is unclear. As self-reported sleep duration is potentially subject to misclassification, and the exposure variable (sleep hours) was categorical, even random misclassification may have led to bias in any direction (Rothman et al, 2013). A previous validation study (Girschik et al, 2012) concluded that a three-item sleep questionnaire that is similar to one used in the study by Qian et al (2014) and typically employed in other epidemiologic sleep studies exhibited a poor agreement with objective measures of sleep as assessed using actigraphy (kappa coefficients ranging from −0.19 to 0.14). Thus, the misclassification bias for exposure data in their study cannot be ruled out. Moreover, the data on sleeping habits in the analysis was about the information of most recent year at baseline, which may not reflect the long-term sleeping habits.

Second, the lack of adjustments for other sleep factors in the analysis could have confounded their results. A plausible biological model, that is, light exposure at night (LAN)–melatonin–BC (Stevens and Davis, 1996) may interpret how poor sleep can directly affect the development of BC. In this hypothesis (Stevens and Davis, 1996), LAN is deemed to be associated with an increased risk for incident BC by decreasing the melatonin release by pineal gland. However, melatonin release rely on the light/dark cycle (Blask, 2009) rather than on sleep duration only, and other sleep characteristics such as sleep quality, LAN, the use of sleeping pills, habitual timing of sleep, and night waking times may also influence the outcome for incident BC (Yang et al, 2014); therefore, the potential confounding bias may exist. In addition, as the exposure data collection is described in a concise manner, it is unclear whether the ‘sleep hours’ is a real sleep duration at night-time or hours spent in bed.

Third, the findings, particularly for ER+PR+ BC, seemed to somewhat contradict the current possible biological mechanism, that is, LAN–melatonin–BC (Stevens and Davis, 1996). Such a discrepancy, as acknowledged by authors, could be due to chance. However, other factors, as mentioned above, including poor quality of exposure data (see comment 1) and lack of consideration for other sleep factors (see comment 2) may partly explain this controversy, because if the LAN–melatonin–BC hypothesis is true, as mentioned above, sleep is not necessarily required for synchronisation of the endogenous circadian rhythm (Blask, 2009), although melatonin release depends on a stable 24-h light/dark cycle, other sleep patterns such as habitual timing of sleep, waking up frequency, night-time lighting conditions, and sleep quality may also affect melatonin release (Yang et al, 2014).

Altogether, although this cohort study provided new information on the relationship between sleep duration and BC risk, the quality of exposure assessment and other covariates relating to sleep should be considered when interpreting results. With this in mind, we now are establishing a large population-based case–control study to assess the risk of BC associated with sleeping factors and other potential risk factors in Jiujiang City, China. As for exposure assessments, we are systematically collecting sleeping factors including sleep quality, LAN, night/shift work, the use of sleeping pills, sleep hours, habitual timing of sleep, and frequency for night-time wakings using the self-made 22-item sleeping factors questionnaire (SFQ). We have conducted a pilot study to check the validity and reproducibility of SFQ used in our project. In the pilot study, the SFQ was interview-administered twice, ∼1 year apart, and participants were also asked to complete a ‘sleeping diary’ for 30 consecutive days every quarter over this same year accounting for seasonal effects. We examined the validation by comparing the average measures between two SFQs and four sleeping diaries, and examined the 1-year stability of SFQ by comparing the measures in the two SFQs.