Study Design
This was a secondary analysis using data from the WHI. The WHI is a longitudinal study investigating healthy aging among more than 161,000 postmenopausal women. The WHI began in 1993 with 40 clinical centers nationwide and recruited women aged 50-79 years into either one or more of three randomized clinical trials (CT) or an observational study (OS)[14]. The design and selection criteria of the WHI has been described previously [15]. Data were collected at routine clinical examinations and using questionnaires. Institutional Review Boards at all participating WHI institutions approved the study protocols and all participants provided informed consent prior to study participation.
Inclusion / Exclusion Criteria
From the WHI CT and OS, we included women who developed breast cancer between year 1 and year 3, had sleep duration and sleep quality reported at baseline, and had weight collected at baseline, pre-diagnosis, and year 3. We then excluded women who had a personal history of breast cancer before baseline or were missing data for the covariates, with the exception of income for which we created an indicator variable for missing income.
Breast cancer diagnosis was collected via questionnaires mailed to participants every six months. Women who did not send in the completed questionnaire were contacted by trained study outcomes coordinators to complete the information by telephone. If women were not able to complete the questionnaire, proxy contacts, provided by the participants, were reached to collect the information. All reported breast cancer diagnoses were adjudicated by centrally trained physicians using medical records including cancer pathology reports [16]. Between years one and three, 1,595 WHI participants were diagnosed with breast cancer.
Exposures of Interest
Sleep Duration: Baseline sleep duration was collected via the self-report WHI Lifestyle Questionnaire according to the question: “About how many hours of sleep did you get on a typical night during the past 4 weeks?”, with possible responses of: ≤5 hours, 6 hours, 7 hours, 8 hours, 9 hours, and ≥10 hours. To be consistent with previous WHI studies [17, 18], we collapsed these categories into four levels: short sleep (5 or less hours), 6 hours, long sleep (9 hours or more) [19] and the reference group of optimal sleep (7 or 8 hours)[19, 20].
Sleep Quality: Baseline sleep quality was collected via the self-report WHI Lifestyle Questionnaire and was assessed using the question: “Overall, was your typical night's sleep during the past 4 weeks”, with possible responses of: “very sound or restful, sound or restful, average quality, restless, or very restless?”. We collapsed these categories into three levels: poor sleep quality (“very restless” or “restless”), average sleep quality (average quality) and the reference group of above average sleep quality (very sound or restful).
Primary Outcome
Clinical examinations were completed for participants at routine study clinic visits according to corresponding study arm schedule to collect weight (kg). All women attending the clinic visit completed a measurement of height and body weight, following a standardized WHI protocol, using a beam scale, with height recorded to the nearest 0.1 centimeter and weight recorded to the nearest 0.1 kilogram. All weights were assessed by trained clinic staff. Weight change in kilograms (kg) was calculated as the change in weight from baseline to year 3 and pre-diagnosis to year three. Pre-diagnosis weight was the weight recorded closest to but preceding diagnosis. For participants of the OS, weight was collected at baseline and year 3, and for CT participants, weight was collected at baseline, year 1, 2, and 3. Thus, weight change from baseline to year 3 and pre-diagnosis to year 3 was the same for participants of the OS.
A priori power calculations showed that a weight change of 7kg, 2kg, and 5kg would be detected for the ≤5 hours, 6 hours, and ≥9 hours sleep duration group at 80% power when compared to the 7-8 hour sleep duration group. Compared to the above average sleep quality group, a weight change of 4kg would be detected for the poor sleep group and a weight change of 3kg would be detected for the average sleep quality group at 80% power.
Covariates
Baseline covariates were derived from the literature and include age, race/ethnicity, marital status, education, employment status, family income, BMI, smoking status, alcohol consumption, caffeine consumption, total physical activity in metabolic equivalent (MET) -hours/week, dietary energy (kcals/day), pain construct score, depression construct score, prevalent diabetes, prevalent arthritis, prevalent hypertension, use of sleep medication, and breast cancer factors (cancer stage at diagnosis, years from baseline to diagnosis), use of HRT at baseline, and use of anti-neoplastic hormonal medications post-diagnosis [21-23]. Additionally, participation in the hormone replacement therapy (HRT), dietary modification (DM), or calcium and vitamin D (CaD) intervention was controlled for because women who were eligible for these clinical trials were on average healthier than those who were in the observational study, and because the interventions may have impacted weight and/or sleep. Self-reported race and ethnicity of American Indian or Alaskan Native, Asian or Pacific Islander, Hispanic/Latino, and other were aggregated due to small sample sizes to create a single category. Self-reported race and ethnicity of white and black or African American were also reported.
Statistical Methods
Descriptive statistics at baseline were stratified by sleep duration (≤5 hours, 6 hours, 7-8 hours, ≥9 hours). The relationship between baseline sleep duration and sleep quality with weight change was described using mean and 95% confidence interval (CI) and plotted for visual description.
We used linear regression to evaluate the relationship between sleep duration at baseline and weight change, as well as sleep quality at baseline and weight change. We fit both unadjusted and adjusted models adjusting for age, race/ethnicity, marital status, education, employment status, family income, BMI, smoking status, alcohol consumption, caffeine consumption, total physical activity in metabolic equivalent (MET)-hours/week, dietary energy (kcals/day), pain construct score, depression construct score, prevalent diabetes, prevalent arthritis, prevalent hypertension, use of sleep medication, and breast cancer factors (cancer stage at diagnosis, years from baseline to diagnosis), study arm, use of HRT at baseline, and use of anti-neoplastic hormonal medications post-diagnosis.
We performed sensitivity analyses to examine the potential influence of the WHI study arm on weight change. In this analysis, women who participated in the DM study arm were removed, but the analyses were otherwise identical to the primary adjusted and unadjusted models.
All analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC, USA), tests were two-sided with a specified type I error rate of α = 0.05.