This analysis was performed using data from the Collaborative Breast Cancer Studies, a series of four population-based case-control studies of invasive breast cancer conducted continuously between 1992 and 2007 by investigators at the University of Wisconsin-Madison, the Harvard School of Public Health, and Dartmouth Medical School. Participant identification, enrollment, and data collection procedures were maintained across studies with the specific intent for future data pooling. Each study was conducted according to institutionally-approved protocols (19
All case participants had an incident diagnosis of invasive breast cancer and date of diagnosis reported to the statewide cancer registry at the study site. During 1992–2001 cases were identified in three states: Wisconsin, Massachusetts, and New Hampshire; during 2002–2007 cases were enrolled in Wisconsin only. In total, 21,713 eligible breast cancer cases were identified. Of the identified cases, physicians refused contact with 359 (1.7%), 743 (3.4%) were deceased, 624 (2.9%) could not be located and 2,794 (12.9%) refused to participate. Physician refusals indicated that an eligible participant should not be contacted to participate in a 20–30 minute telephone interview. Interviews were conducted for 17,193 (79.2%) eligible case women. Thirty-eight cases were considered to have provided unreliable information by the interviewers, leaving 17,155 case interviews available for analysis.
Population controls were identified in each state from lists of licensed drivers (<65 years) and Medicare beneficiaries (65–79 years). Controls were randomly selected within 5-year age strata to yield an age distribution similar to the cases enrolled in each state and were required to have no personal history of breast cancer. Of the 26,269 potential controls identified, 316 (1.2%) were deceased, 1,198 (4.6%) could not be located, and 5,446 (20.7%) refused to participate. Interviews were obtained for 19,309 (73.5%) women. Thirty-seven control interviews were considered unreliable by the interviewer, leaving 19,268 control interviews available for analysis.
All case and control participants provided information on medical history, hormone use, lifestyle and demographic factors during a structured telephone interview. Questions pertaining to postmenopausal hormone use queried information on formulation, routes of administration, age started, frequency of each episode of use, total duration, and time since last use. Study participants also reported whether they had undergone surgery to remove the uterus or ovaries, the type of surgery (hysterectomy and/or oophorectomy, including number of ovaries removed), and age at surgery. Information about personal and first-degree family history of cancer was obtained at the end of the interview to maintain interviewer blinding of case-control status.
For each case, a reference date was defined as the registry-reported date of invasive breast cancer diagnosis. For comparability, the control subjects interviewed contemporaneously with cases were assigned an individual reference date corresponding to the average time from diagnosis to interview for the case group. Reference age was defined as age at diagnosis for cases or on the reference date given to controls. Only exposures that occurred prior to the assigned reference date were included in analyses.
In three of the four pooled studies (1992–1995; 1997–2001; 2001–2004) natural menopause was defined as the absence of menses for 6 consecutive months not due to surgery, chemotherapy, radiation, or other reasons. In the most recent study (2004–2007), menopause was defined as 12 consecutive months without menses not due to surgery, chemotherapy, radiation, or other reasons. In all studies, women who reported bilateral oophorectomy before the reference date were categorized as postmenopausal. Women who reported hysterectomy without bilateral oophorectomy were categorized as premenopausal if their reference age was in the first decile of age at natural menopause among controls (<42 years of age for current smokers and <43 years of age for nonsmokers), to be postmenopausal if the reference age was in the highest decile for age at natural menopause among controls (>55 years of age), and otherwise to have an unknown age at menopause. Participants who had started postmenopausal hormone use before cessation of menses were categorized as postmenopausal with unknown age at menopause.
The pooled analysis was limited to women ages 50 and older (N=13,253 cases, 14,900 controls). We further excluded women who were premenopausal (N=1,352 cases, 1,462 controls) or had unknown menopausal status (N=620 cases, 661 controls), or who had a previous history of cancer (except non-melanoma skin cancer) (N=614 cases, 726 controls). Women with discordant ages at bilateral oophorectomy and hysterectomy (93 cases, 155 controls) or missing ages for both procedures (41 cases, 45 controls) were also excluded. Finally, 84 cases and 64 controls who reported bilateral oophorectomy and/or hysterectomy at the same age or older compared to the reference age were excluded. After these exclusions, 10,449 cases and 11,787 controls contributed information to our analyses.
Odds ratios (OR) and 95% confidence intervals (CI) for breast cancer were calculated using multivariable logistic regression models. The reference group was comprised of women with an intact uterus and ovaries who reported undergoing a natural menopause for estimates of breast cancer odds after TAHBSO. For estimates of breast cancer odds according to estrogen therapy use after TAHBSO, the reference group was additionally restricted to never users of hormone therapy. This allowed us to jointly assess the protective effect of early ovarian removal on breast cancer risk and potential variation according to estrogen therapy use after surgery. In sensitivity analyses, we additionally required that women in the reference group reported ages at menopause between 50–51 years (the median age of menopause among controls). Results were similar when age at menopause was and was not specified for the reference group; therefore we present results with the more expansive definition that did not specify age at menopause to maximize the sample size for the reference group.
Covariates in multivariable models were selected a priori as factors conceptually related to both gynecologic surgery and breast cancer risk. Preliminary multivariable models were adjusted for age (5-year groups), study enrollment period (1992–1995; 1997–2001; 2001–2004; 2004–2007) and study site; final models additionally included the following covariates: age at menarche (<12 years, 12, 13, ≥14, unknown), age at first birth (<20 years, 20–24, 25–29, ≥30, unknown), parity (0–1 livebirths, 2–3, ≥4, unknown), postmenopausal hormone use (never, estrogen only, estrogen + progestin only, combination of estrogen and estrogen + progestin, other/unknown), first-degree family history of breast cancer (yes, no, unknown), mammography screening (yes, no, unknown), and body mass index (underweight, normal, overweight, obese, unknown). P-values ≤0.05 were considered to be statistically significant. P-trends represent P-values from the Wald test for the categorical variable included as an ordinal term in regression models. All analyses were performed using SAS version 9.2 software (SAS Institute, Inc., Cary, NC).
We conducted two additional sensitivity analyses to evaluate whether the definitional change of natural menopause from 6 to 12 months of amenorrhea during 2004–2007 and whether the enrollment of participants in Wisconsin only during 2001–2007 influenced our results. In analyses restricted to participants who were defined as menopausal after 6 months amenorrhea (N=9,051 cases; 10,216 controls) and those residing in Wisconsin (N=7,310 cases; 7,892 controls), our findings were essentially unchanged compared to the overall analysis.