The participants for this analysis include women from Detroit and Los Angeles (LA), two of the five participating sites (Atlanta, Detroit, LA, Philadelphia, Seattle) in the Women’s CARE Study 39. The Women’s CARE Study, which was supported by National Institute of Child Health and Human Development (NICHD), was a population-based, case–control study designed to examine risk factors for invasive breast cancer among US-born white women and black women 39. The age distribution and participant response rates by study site, case–control status and race have been published 39. Tissue collection, as part of the Women’s CARE Study, was supported by NICHD for the Detroit and LA study sites, as advised by the Women’s CARE Steering Committee 39.

Case participants in the Women’s CARE Study had no prior diagnosis of invasive or in situ breast cancer and were diagnosed with their first primary invasive breast cancer (International Classification of Diseases for Oncology codes C50.0–C50.9) between July 1994 and April 1998. Control participants were women with no history of invasive or in situ breast cancer who were identified by random digit dialing. Control participants were frequency matched to the expected distribution of cases in strata defined by 5 year age groups, ethnicity (white or black), and residence located in the same geographic (study) region. The Women’s CARE Study recruited 1921 case and 2034 control participants from Detroit and LA. The interview response rates were 74.7% for cases in Detroit, 74.1% for controls in Detroit, 73.3% for cases in LA, and 73.7% for controls in LA. All participants for this study provided written informed consent and the study protocol was approved by the Institutional Review Boards at the University of Southern California (USC), the Karmanos Cancer Institute Center, the Centers for Disease Control and Prevention, and the City of Hope.

The Women’s CARE Study collected demographic characteristics, detailed information about current and past recreational physical activity, menstrual and reproductive history, family history of breast cancer, body size measures including height and weight, history of oral contraceptive use, and information pertaining to other factors from each participant during an in-person interview conducted from August 1994 through December 1998. Information was recorded up to a predetermined reference date for each participant. The reference date was the date of diagnosis for women with breast cancer or the date of the initial telephone screening of the household for control participants.

Details regarding the measures of recreational physical activity in the Women’s CARE Study have been published elsewhere 10. Briefly, the Women’s CARE Study documented all episodes of exercise activity in which a participant engaged throughout her lifetime up to her reference date, and recorded details of activities in chronologic order starting with the first activity recalled by the respondent. The information collected for each activity episode included the type of activity, the age at which the woman started and stopped the activity, the number of months per year of participation in the activity, and average duration in hours per week. The activities reported included any organized sports activities, such as school sports or teams, and individual activities, such as walking, jogging, running, hiking, bicycling, aerobics, swimming, and dancing. The details as to the extent of the activity were also recorded. For example, for swimming, we collected the types of swimming including recreational swimming, snorkeling, swimming laps, or training for competitive swimming.

The average number of hours of exercise activity per week for each year of age for each participant was estimated. Women were considered to be inactive at any given age if they reported no activity for that age or if their average number of hours per week of activity for that age was less than 0.67 h (i.e., equivalent to less than 2 h/week for 4 months). The metabolic equivalents of energy expenditure (MET)-hours per week for each age were estimated by multiplying together the number of hours per week a woman spent in a particular activity, the proportion of the year spent in that activity, and the estimated MET score for the activity based on the Compendium of Physical Activity 40. A measure of lifetime activity was defined as average annual exercise activity from age 10 years to the woman’s age on her reference date in hours per week and in MET-hours per week. The average number of MET-hours per week was also assessed for the following specific times: the first 10 years after menarche, ages 10–19 years, ages 20–34 years, and the 10 years before each woman’s reference date.

Paraffin-embedded tumor blocks were obtained from pathology laboratories where diagnoses were made for 1333 participating breast cancer cases (Detroit: 414, LA: 919). Approximately 80% of the blocks requested were received. Tumor blocks were carefully reviewed and processed in the centralized pathology laboratory of Dr. Michael F. Press at USC.

We excluded 113 tumor samples because the tumor blocks contained either no tumor tissue (n = 46), insufficient tissue for the laboratory assays (n = 3), only carcinoma in situ (n = 56), or only hematoxylin and eosin-stained tissue sections (n = 8); we also excluded 14 samples that had other problems that made evaluation of the tumor ER, PR, HER2, or p53 difficult. Expression of ER, PR, HER2, and p53 was determined for 1206 samples (Detroit: 367, LA: 839).

ER and PR expression was determined using previously published immunohistochemistry (IHC) methods 41,42. Immunostaining results for ER and PR expression were interpreted in a blind fashion and scored semi-quantitatively on the basis of the visually estimated percentage of positively stained tumor cell nuclei. At least 100 tumor cells were examined for each specimen and samples with ≥1% of immunostained tumor cell nuclei were considered positive for ER and PR 43.

HER2 expression was determined by IHC using the 10H8 monoclonal antibody 44 to assess HER2 membrane protein immunostaining. Immunostaining results for HER2 were categorized as no (0) or weak (1+), moderate (2+), and strong (3+) membrane immunostaining. No (0) or weak (1+) membrane immunostaining was classified as low HER2 expression (HER2−) whereas moderate (2+) or strong membrane immunostaining (3+) was classified as HER2 overexpression (HER2+). This was based on previous validation results from the same pathology laboratory, indicating that the agreement between 10H8-IHC and fluorescent in situ hybridization (FISH) analysis was 92%; discordant results were found for 5.7% of tumor samples, which scored as 0 or 1+ by 10H8-IHC, but showed HER2 gene amplification; and 2.1% of tumor samples, which scored as 2+ or 3+ by 10H8-IHC, but showed no HER2 gene amplification 44.

The expression of p53 protein was determined by IHC using the monoclonal mouse antibody DO7 (Oncogene Science, Inc. Cambridge, MA) and BP 53-12-1 (Biogenex, San Ramon, CA) to measure p53 nuclear protein immunostaining. Based on findings from previous studies, comparing p53 mutations in exons 2–11 with p53 protein expression levels 45,46, ≥10% nuclear staining for p53 protein was deemed positive 47.

We used Pearson Chi-squared tests to compare frequency distributions of categorical variables. Because of the non-normal distributions of age at reference date and body mass index (BMI) 5 years before the reference date, we conducted the nonparametric Wilcoxon test to evaluate differences in these two variables between case participants and control participants.

For case–control comparisons, we fit multivariable polychotomous unconditional logistic regression models 48 to data to estimate the odds ratios (ORs) and corresponding 95% confidence intervals (CIs) of breast cancer associated with lifetime recreational physical activity (1) by the expression status of each individual receptor for all women, premenopausal women, and postmenopausal women, (2) by various combinations of ER, PR and HER2 status including two common subtypes (TN and ER/PR+/HER2−), which were further stratified by p53 status, and (3) by three levels of HER2 expression (none/weak, moderate, strong expression). We also examined the association between time-period-specific or age-specific recreational physical activity and breast cancer risk according to HER2 status. Moreover, since differential recall of detailed physical activity history between cases and controls might occur, we conducted case–case comparisons for ER− versus ER+, PR− versus PR+, HER2− versus HER2+, and p53− versus p53+ patients using a multivariable unconditional logistic regression approach 48.

We used previously published categories of average MET-hours per week of physical activity (less than or equal to 2.2, 2.3 to 6.6, 6.7 to 15.1, or at least 15.2 annual MET hour/week), which were generated according to approximate quartiles of the distribution of all Women’s CARE Study control participants classified as active 10. We included the following factors, selected a priori, as potential confounders in all multivariable logistic regression models: study site (Detroit or LA), race (white or black), education (high school graduate or a lower level of education, attended technical school or college, but did not graduate, or college graduate), age (in 5 year age groups from 35–39 to 60–64), family history of breast cancer [first-degree (mother, sister, or daughter); no first-degree family history including 4% of participants with uncertain answers], age at menarche (less than or equal to 11, 12, 13, or at least 14 years), parity (nulligravid, pregnant but no full-term pregnancy, or parity 1, 2, 3, or 4+), a four-category variable combining menopausal status and hormone therapy (HT) use (premenopausal, postmenopausal and never HT use, postmenopausal and ever HT use, or unknown menopausal status), BMI five years before the reference date (continuous variable, kg/m²), duration of OC use (never, less than 1 year, 1–4 years, 5–9 years, or at least 10 years). When we examined the association between time-period-specific or age-specific recreational physical activity and breast cancer risk, we did not mutually adjust time-period-specific and age-specific physical activity as some of the periods overlap and, in addition, physical activity measures are highly correlated (e.g., the Spearman correlation coefficient for average-annual MET-hours/wk of physical activity at ages 10–19 years and 20–34 years was 0.69); we included women who engaged in recreational physical activity only in other time periods or other age groups as a separate category.

Tests for trend were conducted by fitting ordinal values corresponding to categories of recreational physical activity in our models and testing whether the coefficient (slope of the dose response) differed from zero. When conducting tests for trend for time-period-specific or age-specific physical activity variables, we excluded women who engaged in recreational physical activity only in other time periods or other age groups. We also conducted Wald chi-square tests for homogeneity of the associations with recreational physical activity across different subtypes of breast cancer by fitting a model using ordinal values.

We excluded 11 case participants and 22 control participants with missing information on physical activity (2 cases, 3 controls), parity (1 case, 4 controls), BMI (4 cases, 9 controls), or OC use (4 cases, 6 controls). This resulted in 1195 cases (581 premenopausal, 497 postmenopausal, and 117 with unknown menopausal status) and 2012 controls available for the current analysis (929 premenopausal, 831 postmenopausal, and 252 with unknown menopausal status). Among 1328 postmenopausal women, 827 women (307 cases, 520 controls) reported having ever used HT.

When reporting the results of univariate comparisons between case participants and control participants, trend tests, or homogeneity tests, we considered a two-sided P value less than 0.05 as statistically significant. We did not adjust P values for multiple comparisons as these analyses were considered as exploratory 49. All analyses were performed using the SAS statistical package (Version 9.2, SAS Institute, Cary, NC).

Overall, case participants were more likely than control participants to be better educated (P_χ2 = 0.01), to have a first-degree breast cancer family history (P_χ2 < 0.0001), and to never have been pregnant (P_χ2 = 0.02) (Table1). The case–control differences in education and pregnancy history were restricted to LA women, whereas the difference in first-degree breast cancer family history was observed for both LA and Detroit women.

As previously reported among all participants of the Women’s CARE Study 50, lifetime recreational physical activity was associated with a decreased risk of ER− and ER+ breast cancer, but only the result for ER+ disease was statistically significant in our sample of LA and Detroit women (P_trend = 0.67 for ER− vs. P_trend = 0.03 for ER+, Table2). Analyses by HER2 status showed that the ORs of HER2− breast cancer declined with increasing lifetime MET-hours of physical activity (P_trend = 0.04), whereas no trend was observed for HER2+ breast cancer (P_trend = 0.93). Homogeneity tests of trends neither between ER− and ER+ nor between HER2− vs. HER2+ was statistically significant (both P_{homogeneity of trends} ≥ 0.19). Our data showed no evidence of an association between recreational physical activity and breast cancer risk that varied according to PR or p53 protein status. Although HER2− cases did not differ statistically from HER2+, HER2− cases were less likely to have engaged in recreational physical activity. Moreover, the results observed for all participants are likely driven by those of premenopausal women since we did not observe any association among postmenopausal women (Table S1).

Analyses by the status of ER and HER2 jointly showed that lifetime MET-hours of physical activity were associated with decreased risks for the HER2− subtypes (ER−/HER2− and ER+/HER2− breast cancers), but only the result for ER+/HER2− was statistically significant (Table3, P_trend = 0.01 for ER+/HER2−). Our data did not provide any evidence that lifetime MET-hours of physical activity was associated with a reduced risk for HER2+ subtypes including ER−/HER2+ and ER+/HER2+ breast cancer (both P_trend ≥ 0.88). However, we found no difference in the trends across subtypes defined by ER and HER2 (test for homogeneity of trends: P = 0.38).

Analyses combining ER, PR, and HER2 also demonstrated that lifetime MET-hours of physical activity were inversely associated with the risk for HER2− subtypes, especially for ER/PR+/HER2− subtype (P_trend = 0.02), but were not associated with HER2+ subtypes (HER2-enriched, ER/PR+/HER2+; both P_trend ≥ 0.84). The difference in trends across the four subtypes defined by ER/PR/HER2 was not statistically significant (test for homogeneity of trends: P = 0.52). Subclassification of two common subtypes (TN and ER/PR+/HER2−) by p53 status did not further differentiate the associations of these subtypes with recreational physical activity (test for homogeneity of trends: P = 0.44, results not shown). We assessed whether lifetime recreational physical activity was associated with breast cancer with each level of HER2 expression (negative/weakly positive, moderately positive, strongly positive; Table S2). This analysis showed that the ORs of HER2− breast cancer declined with increasing lifetime MET-hours of physical activity (P_trend = 0.04), but no association was observed for breast cancers that were either moderate or strong expressers of HER2 (both P_trend ≥ 0.76).

HER2–breast cancer was inversely associated with the MET-hours of physical activity for each specific time period of life that we examined, although some tests for trend were only marginally statistically significant (Table4, all P_trend ≤ 0.06). No associations were found for HER2+ breast cancers (all P_trend ≥ 0.25). The difference in trends between HER2− and HER2+ breast cancer was statistically significant for recreational physical activity only in the first 10 years after menarche (test for homogeneity of trends: P = 0.03) and was marginally statistically significant for physical activity in which the woman engaged at ages 10–19 years (test for homogeneity of trends: P = 0.06), whereas the trends in risk did not differ statistically for activity at ages 20–34 or 10 years before reference date (both tests for homogeneity of trends: P ≥ 0.12).

When we compared HER2− cases with HER2+ cases, ORs decreased with increasing MET-hours of physical activity for all specific time periods. However, the association was statistically significant for physical activity in the first 10 years after menarche (P_trend = 0.02), but not for physical activity in other time periods (all P_trend ≥ 0.05).