Statistical analysis

Differences in time spent sitting by demographic, body composition, medical and reproductive history and lifestyle habits were evaluated using Wald X² tests. Proportional hazards regression was used calculate hazard ratios (HR) and 95% confidence intervals (95% CI) for associations between time spent sitting at work (hours/day), time spent sitting watching television (hours/day), and total time spent sitting (sum of time spent sitting at work and watching television) and breast cancer incidence. Participants contributed person-time from baseline until the occurrence of breast cancer, loss to follow-up, death, or the end of follow-up in 2013. Average follow-up time between 1995 through 2013 was 15.4 years.

Time spent sitting at work and watching television were evaluated as four-level categorical variables (<1, 1–2, 3–4, or ≥5 hours/day). Combined time spent sitting was also evaluated as a four-level variable (<5, 5–<7, 7–<10, or ≥10 hours/day). To assess trend, total time spent sitting was additionally evaluated as a continuous variable. Analyses were conducted using baseline questionnaire data only and updated using follow-up questionnaire data to account for changes in behavior over time. The Andersen-Gill data structure was used to update the score variables and covariates at each time point where new data was available [23]. To evaluate differences by hormone receptor subtypes, all models were additionally run with breast cancer subtypes as the outcome (subtypes evaluated: HR+, HR− or triple negative). All analyses were repeated stratified by menopausal status.

Covariates in the adjusted model were chosen a priori based on factors known or suspected to be associated with breast cancer risk and with sedentary behavior. All adjusted models included age (continuous), geographic region of residence (Northeast, South, Midwest, West), daily caloric intake (continuous), smoking (never, former, <15 cigarettes/day, ≥15 cigarettes/day), BMI (<25, 25–<30, 30–<35, ≥35 kg/m²), physical activity level (high: ≥3–4 hours/week vigorous activity or ≥5–6 hours/week walking for exercise, moderate: 1–2 hours/week vigorous exercise or 1–4 hours/week walking for exercise, low: <1 hour/week vigorous exercise or walking for exercise), family history of breast cancer (yes/no), education (≤12, 13–15, ≥16 years), menopausal status (premenopausal, postmenopausal), duration of postmenopausal female hormone supplement use (none, <5 years, ≥ 5 years), duration of oral contraceptive use (never, <5 years, ≥5 years), mammography usage (ever, never) and parity (none, 1–2, ≥3). Categorical variables were modeled as indicator variables. Participants with missing data on covariates were included as a missing data indicator variable. BMI was included as a covariate due to the relationship between sedentary time and body composition; however, body composition was not associated with breast cancer risk in a previous analysis in the BWHS [24]. Alcohol intake in this study population is low (>94% consumed an average of <7 drinks/week) and was not associated with breast cancer incidence or time spent sitting; therefore, it was not included as a covariate.

To evaluate whether the associations between time spent sitting and breast cancer differed by measures of body composition or physical activity level, analyses were additionally repeated stratified by total physical activity level (low, moderate, high), vigorous physical activity (<1, 1–<3, ≥3 hours/week), walking for exercise (<2, 2–<5, ≥5 hours/week), current BMI (<25, 25–<30, ≥30 kg/m²), BMI at 18 years (<20, 20–<25, ≥25 kg/m²), adult weight gain (<10 kg, 10–19 kg, ≥20 kg), waist-to-hip ratio (WHR) (<0.75, 0.75–<0.85, ≥0.85), and waist circumference (<80, 80–88, >88 centimeters (cm)). Body measurement cut-points were chosen a priori based on previously reported health standards, when possible. BMI cut-points were based on standard BMI cut-points for adults [25]. WHR and waist circumference categories were based on cut-points set for risk of metabolic complications defined by the World Health Organization [26]. BMI at 18 years was based on previous work in the BWHS [24]. Tests for interaction by menopausal status (premenopausal vs. postmenopausal), body measurements and physical activity levels were evaluated using likelihood ratio tests. Additionally, total sitting time combined with physical activity levels and body measurements and breast cancer incidence were evaluated in relation to breast cancer incidence to assess the joint effects. Test for heterogeneity by breast cancer subtype were conducted using the Lunn and McNeil method adapted by Wang et al for assessing disease heterogeneity by subtype [27,28]. Heterogeneity was assessed comparing HR+ to HR− and comparing HR+ vs. HR−/HER2+ vs. triple negative. Results did not differ by approach. Presented p-values are from the assessment of all three subtypes.

Sensitivity analyses were completed excluding participants diagnosed with breast cancer within 4 years of baseline. Results did not differ; therefore, analyses including cases diagnosed since baseline are presented. Two-sided P-values are reported. Analyses were conducted using SAS 9.3 (SAS Institute, Inc.).