Details of the methods used in the Women’s CARE Study have been published previously (8
), and the relation between established reproductive risk factors for breast cancer and the risk of different breast cancer subtypes has also been reported previously (9
). Briefly, a population-based case-control study was conducted at five metropolitan sites in the United States (Atlanta, Georgia; Detroit, Michigan; Los Angeles, California; Philadelphia, Pennsylvania; and Seattle, Washington). White and African-American women aged 35–64 years with no history of in situ or invasive breast cancer who received a diagnosis of invasive breast cancer from July 1994 through April 1998 were eligible as cases. Women of other races/ethnicities were not recruited into this study. These women were ascertained by Surveillance, Epidemiology, and End Results Registry staff at four sites (Atlanta, Detroit, Los Angeles, and Seattle) and by field-center staff at one site (Philadelphia). Controls were women with no history of breast cancer who were identified using random digit dialing and were frequency-matched to cases within strata of site, race, and age. Of the women identified, 76.5 percent of eligible cases and 78.6 percent of eligible controls participated in the study, and it included a total of 4,575 cases and 4,682 controls.
Of the 4,575 cases, 2,089 women were premenopausal and 1,924 were postmenopausal. Of the 4,682 controls, 2,040 women were premenopausal and 2,029 were postmenopausal. Menopausal status was defined on the basis of self-report. For this analysis, premenopausal women were those who reported that they had not experienced menopause and were younger than 55 years of age. Postmenopausal women were those whose menstrual periods had stopped naturally, who had undergone a bilateral oophorectomy (with or without a hysterectomy), who had used menopausal hormone therapy in the past 12 months, or who were 55 years of age or older. The 562 cases and 613 controls whose menopausal status could not be defined using these definitions were excluded. All results were stratified by menopausal status, because risk factors for breast cancer are known to vary by menopausal status (10
The study protocol was approved by an institutional review board at each study site. Written informed consent was obtained from all study subjects. All interviews were conducted in person, typically in subjects’ homes, using standardized procedures. The Women’s CARE Study was principally designed to assess the relation between oral contraceptive use and breast cancer risk, but self-reported information on other possible risk factors for breast cancer was also obtained. The following additional information was collected for the period before each participant’s reference date: detailed reproductive history data, family history of cancer, anthropometric characteristics, use of menopausal hormone therapy, and lifetime histories of physical activity, smoking, and alcohol use. The reference date used for cases was the date of breast cancer diagnosis; for controls, it was the date on which the woman was first contacted by random digit dialing. Control identification was timed so that control reference dates would approximate the distribution of case reference dates.
The estrogen receptor (ER) and progesterone receptor (PR) status and histology of each breast cancer case were ascertained from local cancer registry data, which are based on reviews of pathology reports (except in Philadelphia, where these data were collected by staff at the field center). No statistical differences were observed in the risks associated with ER+/PR+, ER+/PR−, and ER−/PR+ breast cancer. As a result, analyses by hormone-receptor status were stratified into two groups: women with ER+ or PR+ (hormone-receptor-positive) tumors (n
= 2,371) and women with ER−/PR− (hormone-receptor-negative) tumors (n
= 938). For analyses by histology, cases were grouped on the basis of the International Classification of Diseases for Oncology
(ICD-O) morphology codes assigned to their tumors; ICD-O morphology code 8500 was used to define ductal cases (n
= 3,039), code 8520 was used to define lobular cases (n
= 244), and code 8522 was used to define ductal-lobular cases (n
= 240). No statistically significant differences in the risks associated with lobular and ductal-lobular cancers for the exposures of interest were observed. Accordingly, similarly to the way in which these histologic types have been treated in other epidemiologic studies, they were grouped together and defined as lobular carcinomas in all analyses (11
). Cases with other ICD-O morphology codes (n
= 490) were excluded from our histology-specific analyses.
We compared cases with controls using unconditional logistic regression. We compared cases of each histologic type and each hormone-receptor profile to controls separately using multinomial polytomous logistic regression (13
). All analyses were conducted using Stata SE, version 9.0 (Stata Corporation, College Station, Texas). Odds ratios and associated 95 percent confidence intervals were calculated as estimates of the relative risk. All analyses were adjusted for age and study site because controls were frequency-matched to cases on these variables. Race was found to be a statistically significant effect modifier of the relations of both age at first full-term birth and interval between ages at menarche and first full-term birth with premenopausal breast cancer, but not with postmenopausal breast cancer. Thus, all analyses of premenopausal breast cancer were stratified by race, while those of postmenopausal breast cancer were not. Multiple variables, including oral contraceptive use, socioeconomic factors, anthropometric characteristics, family history of breast cancer, and use of menopausal hormones, were evaluated separately as potential confounders of each of the associations we evaluated. Inclusion of these factors individually in the statistical model did not alter the odds ratios of interest by more than 10 percent, so none was added to the final models (14
We also adjusted the main effects of our three exposures of interest—age at menarche, age at first full-term birth (a pregnancy of >26 weeks’ gestation), and the interval between these two ages (as categorical variables, with the categories used throughout the tables)—for each other in order to assess their relative contributions to breast cancer risk. These three exposures were also highly correlated with each other: The correlation coefficients for each combination of these three variables all had p values less than 0.0001. However, age at menarche did not confound either the association of age at first-full term birth with breast cancer risk or the association of interval between menarche and first-full term birth with breast cancer risk, so it was not included in the final statistical models. In contrast, age at first-full term birth and the interval between menarche and first-full term birth did confound the relation of each factor with breast cancer risk to some extent. Given that these two factors were highly correlated, we present results both with and without these adjustments.