Body Size, Physical Activity, and Risk of Triple-Negative and Estrogen Receptor–Positive Breast Cancer

Study population

The WHI is a longitudinal study of postmenopausal women, encompassing multiple concurrent randomized clinical trials (CT) and an observational study (OS). WHI recruitment protocols and procedures are described in detail elsewhere (16, 17). Briefly, between October 1, 1993, and December 31, 1998, postmenopausal women ages 50 to 79 years were recruited from 40 clinical centers across the United States. Women with a medical condition with a predicted survival less than 3 years or who were unlikely to stay in the same geographic area for this period were excluded. Additional eligibility criteria related to competing risks and likely adherence were imposed for CT participation, including a baseline mammogram and/or clinical breast examination suspicious for breast cancer, or a history of breast cancer. Women not meeting CT eligibility criteria or not interested CT participation were given the opportunity to enroll in the WHI OS. At the time of enrollment, all women provided written informed consent for participation. Human Subjects Review Committees at all participating institutions approved the WHI study protocol.

In total, 161,808 women enrolled in the WHI (n = 93,676 in the OS; n = 68,132 in the CT). For this analysis, we excluded women with a history of breast cancer or mastectomy at baseline (n = 5,239) and women without follow-up information for breast cancer diagnoses (n = 846), leaving a study population of 155,723 women.

Exposure assessment

All women completed a series of self-administered questionnaires at the baseline screening visit from which detailed information on demographic factors, reproductive history, medical history, breast cancer family history, and physical activity were collected. Information on lifetime use of postmenopausal HT was collected through a structured in-person interview. Women were classified as current HT users if they reported using estrogen-containing pills or patches at baseline (CT or OS) or were randomly assigned to the active HT arms of one of the HT trials within the CT. Height, weight, and waist and hip circumferences were measured at baseline clinic visits. BMI was calculated as weight (kg) divided by squared height (m²). OS participants were asked to self-report their weight at ages 18, 35, and 50 years; BMI calculations for those ages were based on self-reported weight at that age and height at baseline. WHR was calculated as waist circumference (measured to the nearest 0.1 cm at the narrowest part of the torso) divided by hip circumference (measured to the nearest 0.1 cm at the maximal circumference).

At baseline, women were asked how often they currently exercised (never, 1, 2, 3, 4, ≥5 days per week) and how long they typically exercised per session (<20, 20–39, 40–59, ≥60 minutes). Questions were structured to inquire separately about strenuous-, moderate-, and low-intensity recreational exercise. Strenuous exercise was defined as exercise that led to sweating and increased heart rate (e.g., aerobics, swimming laps, jogging). Provided examples of moderate-intensity activity included biking outdoors, using an exercise machine, and calisthenics; provided examples of low-intensity activity included bowling and golf. Metabolic equivalent (MET) values, defined as the ratio of metabolic rate during a specific activity relative to metabolic rate at rest, were assigned for strenuous-, moderate-, and low-intensity activities (7, 4, and 3, respectively) and multiplied by the hours exercised at that intensity level per week to obtain composite recreational physical activity variables (MET-hours/week) for each intensity level and for all intensity levels combined (18).

Follow-up and outcome ascertainment

Medical history, including diagnosis of breast cancer, was updated annually (OS) or semiannually (CT) via mailed or telephone-administered questionnaires. Breast cancer diagnoses reported by participants were verified locally by WHI physician adjudicators. Medical records and pathology reports for locally confirmed breast cancers were sent to the WHI Clinical Coordinating Center for central adjudication and coding of ER, PR, and HER2 status.

Over follow-up (median, 7.9 years), invasive breast cancer was identified in 5,194 women. Information on ER, PR, and HER2 status was available for 4,677 (90%), 4,600 (89%), and 3,139 (60%) cases, respectively. In light of variability in HER2 testing practices over the study period and across institutions, ER⁺ cases with unknown HER2 status (n = 1,334) were excluded from the ER⁺ case group to enable greater comparability to the triple-negative group. Cases with missing HER2 data were diagnosed, on average, earlier during the study period than cases with known HER2 status (mean duration of follow-up = 2.9 versus 5.0 years in cases with unknown versus known HER2 status) and were more likely to have been enrolled in the OS (59.0% versus 55.7%). The distribution of exposure variables, however, was similar in cases with unknown versus known HER2 status (26.5% versus 27.2% in the uppermost quartile of baseline BMI; 25.7% versus 23.0% in the uppermost quartile of total recreational physical activity). Of the 3,116 cases with complete tumor marker data, 307 (10%) were triple-negative and 2,610 (84%) were ER⁺. Remaining cases were ER⁻/PR⁻/HER2⁺ (n = 154), ER⁻/PR⁺/HER2⁻ (n = 31), or ER⁻/PR⁺/HER2⁺ (n = 14).

Statistical analyses

We used Cox regression to assess associations between WHR, waist and hip circumferences, BMI at different ages (baseline and ages 50, 35, and 18), baseline recreational physical activity level (total, strenuous intensity, moderate/low intensity), and subtype-specific breast cancer risk. We constructed separate models for triple-negative and ER⁺ subtypes, with the time axis defined as the time (in days) since randomization (CT) or study enrollment (OS). Women diagnosed with in situ breast cancer or an invasive breast cancer other than the model-specific outcome were censored at the time of diagnosis. Proportional hazards assumptions were verified by testing for a nonzero slope of the scaled Schoenfeld residuals on ranked failure times and on the log of analysis time (19).

Analyses were adjusted for a common set of confounders selected a priori, including study arm, race, education level, household income, family history of breast cancer in first-degree relatives, receipt of mammography during follow-up, and history of mammography within the 2 years prior to baseline. Analyses were adjusted for age both explicitly, through inclusion of a linear adjustment term, and implicitly, via stratification of the baseline hazards (10-year categories). Analyses of baseline recreational physical activity were further adjusted for baseline BMI; analyses of baseline body size characteristics were further adjusted for total recreational physical activity at baseline. We also assessed confounding by parity, ages at first birth and menopause, alcohol consumption, and smoking; however, because adjustment for these factors altered effect estimates for both subtypes by less than 10%, none were included in final multivariate models. We evaluated interaction by current HT use with respect to physical activity and baseline body size characteristics; where interaction was found to be statistically significant, we replicated analyses among current nonusers of HT.

Anthropometric variables were categorized into quartiles based on distributions in the overall study population. We also conducted analyses of BMI at baseline using categories based on the National Heart Blood and Lung Institute definitions for normal weight (BMI < 25.0 kg/m²), overweight (BMI = 25.0–29.9 kg/m²), and obese (BMI ≥ 30 kg/m²; ref. 20). For analyses of total recreational physical activity, we grouped women reporting any activity into 3 categories based on the tertile distribution of MET-hours/week among active women, with a separate (referent) category for women reporting no activity. For analyses of strenuous-intensity, and moderate/low-intensity activity, we dichotomized women reporting any activity according to MET-hours/week and, again, used a referent category of women reporting no activity of the intensity level of interest; we mutually adjusted analyses for activity at these intensity levels. For all exposures, we compared subtype-specific HR estimates using competing risks partial likelihood methods (21). All analyses were carries out by STATA SE version 11.

Missing exposure, covariate, and tumor marker data were handled by a complete-case approach, excluding observations with missing exposure or covariate data and censoring case observations with missing tumor marker data at the time of diagnosis. We conducted sensitivity analyses using multiple imputation (22) to assess the impact of missing tumor marker, exposure, and covariate data. Subtype-specific HRs obtained in our multiple-imputation analyses (not shown) were almost identical to those in the primary analysis. Thus, only the results from complete-case analyses are presented here.