The association between night shift work and breast cancer risk in the Finnish twins cohort

The exclusively female base population for the current analyses was nested within the Older Finnish Twin Cohort, – a long-running prospective cohort study, which started in 1975 by mailing a baseline questionnaire to all Finnish same-sex twin pairs that were born before 1958 and where both co-twins were alive in 1975 (89% response rate) [20]. Briefly, to establish the cohort, twin pairs were selected from the Central Population Registry of Finland in 1974, and twin zygosity was determined by a validated questionnaire shown to accurately classify 93% of twin pairs as monozygotic (MZ) or dizygotic (DZ) [21]. Two follow-up health surveys were conducted, mailed to all participants in 1981 and to twins born between 1930 and 1957 in 1990. The response rate to the 1981 questionnaire was 84%, and to the 1990 questionnaire 77%. The surveys contained questions on work schedules including whether the women’s work schedule comprised night work, and if so, the type of rotation; as well as sleep patterns, and chronotype. Additionally, the survey queried detailed information on socio-demographic, psychosocial, health, and lifestyle factors. Further details of the cohort have been described elsewhere [20]. Of the 11,713 female twin individuals born 1930–1957 in the cohort, the present study included all 6,804 female twin individuals who had replied to the 1990 survey. After excluding prevalent breast cancer cases (n = 43) and women residing abroad (n = 3), there were 6,758 women, among them 474 incident breast cancer cases from 1990 to 2018. After further exclusion of women who did not answer the questions on night shift work (n = 239), and those with missing data on key covariates (n = 738) including the 53 women who reported that they had never worked in their entire life (28 (53%) of these were homemakers, and 18 (34%) on disability pension), 5,781 women remained including 407 incident breast cancer cases. The study population includes 1,474 MZ and 2,568 DZ twins from pairs in which both sisters met the inclusion criteria. In addition, there were 411 MZ and 1,062 DZ twins without their co-twin and 266 twins of uncertain zygosity included in the study. The mean age (± standard deviation, SD) of the participants in the final analysis sample at the time of study entry was 43.6 years (± 7.6). The study was approved by the ethics committee of the Hjelt Institute, Faculty of Medicine, University of Helsinki. Permission for linkage of the cancer registry data was provided by the Finnish Institute for Health and Welfare, Helsinki, Finland. Informed consent was obtained from all individuals.

Our primary exposure of interest was a woman’s self-reported (1990) shift work pattern based on current or latest work type. Rotating-shift work referred to work that rotated through morning, evening, or night shifts in either a two-shift or three-shift pattern. Shift work information was queried by assessing the respondent’s current or latest work type. The question “The present work or the work you last did (mainly) is regular day work, regular night work, two-shift work without a night shift, two-shift work with a night shift, three-shift work, or never worked”. We classified these data into 3 categories (fixed days only, rotating 2-shifts without night work, and rotating 2- or 3-shifts with night work or fixed nights). Further, night work history was also assessed in 1975 and 1981 but did not distinguish between 2-shift work with versus without night work; as well as in 2011, using the same question as posed in 1990 but only among the subset of women born in 1945–1957 who were presumably still in the workforce in 2011. We used these additional night work assessments for sensitivity analyses (1) creating a stricter definition of the reference group “never night work” (requiring never night work both at baseline 1990, as well as in earlier reports); and (2) using the 2011 night work assessment as baseline.

In addition to night/shift work exposure, information on sleep duration and quality was queried in 1990, while chronotype was assessed in 1981. Specifically, usual sleep duration over a 24-hour period, and hours of sleep needed during the night to be alert the next day [22] was assessed (nine response categories: ≤6, 6.5, 7, 7.5, 8, 8.5, 9, 9.5, and ≥ 10 h, which we collapsed into three categories: <7, 7–8, and > 8 h). Following our previous approach we considered a difference of one hour or more (usual versus needed hours of sleep duration) to indicate insufficient sleep [22]. Further, sleep quality was elicited by asking “Do you usually sleep well?” (5 response categories: well, fairly well, fairly poorly, poorly, and cannot say, which we collapsed into 3 categories: well, fairly well, and fairly poorly/poorly; “Cannot say” responses were incorporated into a missing data category for this variable). Chronotype was assessed by a question according to the Diurnal Type Scale [23] “Will you try to estimate to what extent your being ‘a morning or an evening people?” [24]. It is akin to item 19 on the Horne and Østberg morningness-eveningness questionnaire (MEQ) [25], which has previously been shown to correlate well with the overall score [25]. Based on the possible response categories, we defined four chronotypes: definite morning type, somewhat morning type, somewhat evening type, and definite evening type.

Data on breast cancer incidence (ICD-10 code 174) were obtained through record linkage (using unique personal identity codes assigned to every permanent resident of Finland) to the Finnish Cancer Registry, where 100% of registered cases are histologically verified. The Finnish Cancer Registry is a nationwide database with information on all cancers diagnosed in Finland since 1953 [26]. Cancer reporting became mandatory starting in 1961. Data on emigration and vital status were obtained through linkage to the Population Register Center of Finland. Through follow-up (1990–2018; mean 25.9 years), 407 incident cases of breast cancer were accrued among women with shiftwork exposure data and with no missing data on key covariates. Information on histologic characteristics was used to explore differences by type of breast cancer (ductal versus other histological type of invasive breast cancer).

After all exclusions, 5,781 female twin individuals who had replied to the 1990 survey that included questions on night shift work and were free of breast cancer at that time formed the base population for these analyses. Women contributed person-time from the return of the 1990 questionnaire and were censored at first report of any cancer, the date of diagnosis of breast cancer, date of death, or the end of follow-up (December 31, 2018), whichever came first. During this follow-up period (1990–2018), 5,781 women contributed 149,756 person-years at risk for breast cancer. Cox proportional hazard models (with age (in years) as the underlying time metameter) were used to calculate age-adjusted hazard ratios and 95% confidence intervals in each night work exposure category compared with the reference category. Due to the dependent nature of this sample of twin pairs, standard errors and CIs were adjusted for possible within-pair correlations using robust variance estimators. In multivariable analyses, we further adjusted for risk factors for breast cancer, including smoking status (never, occasional, former, current), current (1990) body mass index (BMI, kg/m²) and BMI at age 20 (queried on the 1990 questionnaire), physical activity (leisure-time metabolic equivalents METs, quintiles; assessed in 1975 and 1981), use of oral contraceptives (yes/no; assessed in 1981), alcohol consumption (number of drinks per day on average, with one standard drink defined as 12 g of alcohol, based on reported weekly or monthly consumption of beer, wines or spirits), educational status (< 6 years, 6 years, middle school, high school or more), socioeconomic status i.e. social class (upper white collar, lower white collar, skilled worker, unskilled worker, farmer, other), and zygosity (MZ, DZ, XZ). Information on age at first child-birth and number of biological children was available for women born between 1950 and 1957 (nulliparous, 1, 2, 3, 4 or more children); in sensitivity analyses restricting to these women and excluding childless women, we explored confounding by parity. Log–log plots of survival curves of the shift work exposure categories were used to verify that the proportional hazards assumption was not violated. We created indicators for missing values of categorical variables.

Secondarily, we performed a stratified analysis to explore whether chronotype or sleep quality/duration had a modifying effect on the association between rotating night shifts and breast cancer. The P value for interaction was calculated using the likelihood ratio test, which compares the models with and without the interaction term of rotating night shift work and chronotype/sleep along with the same covariates.

Lastly, a co-twin analysis to assess the association between night shift work exposure and breast cancer risk within twin pairs discordant for breast cancer was performed, though with limited power. In these analyses, we stratified Cox models on twin pairs, allowing each twin pair to have its own baseline hazard [27]. This serves as a powerful approach to account for potential familial confounding (genetics and shared family environment) when assessing twins discordant for night shift work and breast cancer outcomes. All statistical analyses were performed using Stata version 17.0 (Stata Corporation, College Station, TX, USA).