Hormone-related risk factors for breast cancer in women under age 50 years by estrogen and progesterone receptor status: results from a case–control and a case–case comparison

Case patients were identified through the Los Angeles County Cancer Surveillance Program (CSP), the population-based cancer registry that is part of the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) cancer registry program. Eligible case patients were US-born English-speaking, white (including Hispanic) or African-American, female residents of Los Angeles County aged 20 to 49 years when diagnosed with histologically confirmed first primary invasive breast cancer. We identified 2,882 eligible case patients (2,534 white and 348 African-American). White patients were diagnosed between February 1998 and May 2003 and African-American patients were diagnosed between January 2000 and May 2003. We were unable to interview 1,088 of the 2,882 eligible case patients (38%) because of patient refusal (n = 428), no longer living in Los Angeles County (n = 37), inability to be located (n = 88), death (n = 38), serious illness or disability (n = 18), physician refusal (n = 50), or inability to schedule the interview within 18 months of diagnosis (n = 429). We successfully interviewed 1,794 (62%) eligible case patients (1,585 white, 209 African-American).

Because this study was originally designed as a case–case study to examine genetic risk factors for breast cancer, we did not collect control subjects for all case patients. The control subjects who were recruited were matched by race and age (within 5 years and aged 20 to 49 years) to a subset of case patients who were diagnosed between July 2000 and March 2003. Control subjects were US-born English-speaking white or African-American women who had never been diagnosed with invasive or in situ breast cancer. They were identified by using a neighborhood walk algorithm that we had used in previous case–control studies [23, 24]. Field staff conducted walks according to a predefined pattern in the neighborhoods where case patients lived at the time of their diagnoses. The houses on the immediate blocks surrounding the home of the case patients were excluded from the walk. Residences were visited sequentially and information on potentially eligible women was obtained. If no one was home at the time of the visit, we left a request for information at the door, and we sought further information from neighbors so that we could contact the residents later. If we received no response to our written request, we sent additional letters until we were able to determine whether an eligible woman lived at the address in question. Detailed records were maintained to determine the number of housing units contacted in order to identify and interview a control subject. By the end of the study we had identified 603 eligible control subjects for the 1,108 case patients (1,018 white, 90 African-American). We were unable to interview 159 of the 603 control subjects (26%) as a result of subject refusal (n = 77), no longer living in Los Angeles County (n = 18), death (n = 1), serious illness (n = 2), or inability to schedule the interview within 18 months from the date of initial household contact (n = 61). We successfully interviewed 444 (74%) of eligible control subjects (409 white, 35 African-American). On average, 20 houses were canvassed to find an eligible control subject who agreed to be interviewed.

All participants were interviewed in person with the use of a structured questionnaire, which was a modified version of the questionnaire used for the Women's Contraceptive and Reproductive Experiences (CARE) Study [25]. Our questionnaire included reproductive history (including breastfeeding), detailed histories of oral contraceptive use and alcohol consumption, family breast cancer history, demographics, and other factors. Information was recorded up to a predetermined reference date for each participant. The reference date was the date of diagnosis for case patients and the date of initial household contact for control subjects. All participants signed informed consent documents before interview. The study protocol was approved by the federally approved Institutional Review Board at the University of Southern California Medical Center.

Information on ER and PR status for interviewed case patients was obtained from the CSP. Among the 1,794 case patients, 1,510 (84%) had information on both ER and PR status; 91 of these were reported as weakly or borderline positive (84 cases) or undecided (7 cases) for either ER or PR. Among the other 1,419 case patients, 881 (62%) were ER⁺PR⁺, 92 (6%) were ER⁺PR^-, 41 (3%) were ER^- PR⁺, and 405 (29%) were ER^- PR^- .

We compared demographic and hormone-related risk factors among case patients with known ER and PR information, borderline positive or undecided results, and patients without ER or PR information, with the use of F tests for differences in means and Pearson χ² tests for differences in frequency distributions. When the two-sided p value comparing all three groups was less than 0.05, we also performed pairwise comparisons by using Bonferroni t tests or Pearson χ² tests, imposing a Bonferroni correction to the p value, restricting the overall type I error to 5% by setting as statistically significant only two-sided p values less than 0.017 for each pairwise comparison [26].

Analyses were conducted to assess the association between breast cancer and the following factors: age at menarche, number of full-term (greater than 26-week gestation) pregnancies, age at first full-term pregnancy (defined for each woman as the age at which that pregnancy ended), duration of breastfeeding, body mass index (BMI, kg/m²) one year before the participant's reference date, duration of combined oral contraceptive (COC) use, alcohol drinking status during reference age (never, former, and current – that is, drinking alcohol during reference age), and the average number of alcoholic drinks per week in the 5-year period that ended 2 years before the reference age. One alcoholic drink was defined as 12 ounces (355 ml) of beer, 4 ounces (118 ml) of wine, or 1.5 ounces (44 ml) of liquor.

We conducted case–control comparisons for overall, ER⁺PR⁺, and ER^- PR^- case patients with control subjects, and also compared ER⁺PR⁺ with ER^- PR^- case patients. We used polytomous logistic regression [27] to compare ER⁺PR⁺ and ER^- PR^- case patients simultaneously with control subjects. We used a multivariable unconditional logistic regression approach [27] for the comparisons of all case patients with controls, and ER⁺PR⁺ with ER^- PR^- case patients. We estimated multivariable adjusted odds ratios (ORs) and 95% confidence intervals (CIs). Tests for trend were conducted by fitting ordinal values corresponding to categories of exposure in our models and testing whether the coefficient (slope of the dose response) differed from zero.

Adjustment was made for race (white, African-American), age (less than 30, 30 to 34, 35 to 39, 40 to 44, 45 to 49 years), and education (high school or lower, technical school or some college, college graduate) in all our models. Additionally, multivariable models included variables selected a priori as potential confounders: first-degree family history of breast cancer (no first-degree family history, mother or sister with breast cancer, unknown first-degree family history), age at menarche (11 or less, 12, 13, at least 14 years), gravidity (never pregnant, ever pregnant), number of full-term pregnancies (never full-term pregnant, one, two, three, at least four full-term pregnancies), BMI 1 year before reference date (less than 25, 25 to 29, 30 to 34, at least 35 kg/m²), COC use (never, less than 1, 1 to 4, 5 to 9, at least 10 years), the average number of alcoholic drinks per week in the recent 5 years (never, less than 3, 3 to 5, 6 to 11, at least 12 drinks, drinking alcohol but not within the 5 years of interest), and a three-category variable combining menopausal status and hormone therapy use (premenopausal or – among postmenopausal women – never used hormone therapy, or had used estrogen therapy or estrogen plus progestin therapy). When estimating the effects of parity or restricting analyses to parous women, we did not include gravidity in our models. A single model was fitted to assess the joint effects of age at first full-term pregnancy (less than 22, 22 to 27, 28 to 31, at least 32 years) and breastfeeding duration (0, less than 1, 1 to 6, 7 to 23, at least 24 months) among parous women. All variables were included as categorical variables in the models. In reporting the results of trend tests, we considered a two-sided p value of less than 0.05 as statistically significant. All analyses were performed with the SAS statistical package (Version 9.0, SAS Institute, Cary, NC, USA).

To maintain a constant sample size for all analyses, we excluded 69 case patients and 4 control subjects for the following reasons: missing information on educational attainment (15 cases and 1 control), age at menarche (3 cases), parity (4 cases and 1 control), duration of breastfeeding (6 cases), BMI (4 cases), duration of COC use (17 cases and 1 control), recent alcohol consumption (14 cases); missing information on two or more of these factors (6 cases and 1 control). This resulted in 1,725 case patients and 440 control subjects available for the overall case–control analyses. Among the 1,725 case patients, 1,449 (84%) had information on both ER and PR status; 83 of these were reported as weakly, borderline positive or undecided for either ER or PR and were excluded from the analyses by receptor subtypes. Among the other 1,366 case patients, 63% were ER⁺PR⁺, 6% were ER⁺PR^-, 3% were ER^-PR⁺, and 28% were ER^- PR^- . The frequency distribution across receptor subtypes was similar to those reported by previous studies conducted within the SEER registries [21, 28]. In addition, our numbers are very similar to those that we observed among the white Los Angeles cases in the Women's CARE Study. This may be the most appropriate comparison, because 88% of the cases in the present study were white. We excluded ER⁺PR^- and ER^- PR⁺ subtypes from the analyses by receptor status because they were rare. There were therefore 1,239 remaining for the analyses by ER and PR status (854 ER⁺PR⁺ and 385 ER^- PR^-).

As described above, our control subjects were identified through matching to a subset of the cases. An alternative to conditional logistic regression in matched studies with disproportionate numbers of cases and controls is to break the match and conduct unconditional logistic regression with detailed adjustment for the matching factors [29]. We conducted both detailed stratified analyses, with strata defined by race, age categories (five-year categories) and education as a proxy of socioeconomic status/neighborhood as well as standard unconditional logistic regression with adjustment for the same factors. Because the results remained the same, we chose to use multivariable unconditional logistic regression for all our analyses. We also repeated all the analyses with the subset of the cases used to identify the controls. Again, the results were essentially identical with the overall analyses; we therefore present results based on all case patients.

Age at menarche was negatively associated with breast cancer risk regardless of ER/PR status (all p _trend≤ 0.008; Table 2). In comparison with women who had menarche before the age of 12 years, later age at menarche (at least 14 years) was associated with an approximately 40% reduced risk of breast cancer among the ER⁺PR⁺ case patients, ER^- PR^- case patients, and all case patients combined. The associations with age at menarche did not differ between ER⁺PR⁺ and ER^- PR^- case patients (p _trend = 0.85).

A protective effect of parity was confined to women with ER⁺PR⁺ breast cancer (Table 2). The OR of ER⁺PR⁺ breast cancer decreased with increasing number of full-term pregnancies (p _trend = 0.005). Parous women who had four or more full-term pregnancies had an approximately 50% reduction in the risk of ER⁺PR⁺ breast cancer compared with women who never had a full-term pregnancy. ER⁺PR⁺ case patients were less likely to have had many full-term pregnancies than ER^- PR^- case patients (p _trend = 0.09).

A slight increase in risk for ER⁺PR⁺ breast cancer and a reduced risk for ER^- PR^- breast cancer was observed with increasing age at first full-term pregnancy, but none of the confidence limits for the risk estimates excluded 1.0 and no linear trend in risk was observed for either cancer type (ER⁺PR⁺, p _trend = 0.49; ER^- PR^-, p _trend = 0.08; Table 2). ER⁺PR⁺ case patients were more likely to have had a late first full-term pregnancy than ER^-PR^- case patients (p _trend = 0.009).

Duration of breastfeeding was negatively associated with breast cancer risk regardless of ER/PR status (all p _trend≤ 0.03; Table 2). In the case–case analysis, duration of breastfeeding was not associated with ER/PR status (p _trend = 0.63).

Increasing BMI was associated with a non-statistically significant decreasing risk of ER⁺PR⁺ breast cancer (p _trend = 0.20) but was not associated with ER^- PR^- breast cancer (Table 2). Moreover, among premenopausal women, increasing BMI was marginally statistically significantly associated with a decreasing risk of ER⁺PR⁺ breast cancer (p _trend = 0.08) but not ER^-PR^- breast cancer (p _trend = 0.54). In comparison with premenopausal women who had a low BMI (less than 25 kg/m²), the OR among premenopausal women in the highest BMI category (at least 35 kg/m²) was 0.58 (95% CI 0.34 to 1.00) for ER⁺PR⁺ breast cancer and 1.07 (95% CI 0.58 to 1.97) for ER^- PR^- breast cancer. ER⁺PR⁺ case patients were less likely to have had a higher BMI than ER^- PR^- case patients (p _trend = 0.005).

COC use was not associated with risk of either ER⁺PR⁺ or ER^-PR^- breast cancer (Table 2). Women who had used COC for 10 years or longer had a slightly higher OR of ER^- PR^- breast cancer (OR 1.27, 95% CI 0.75 to 2.14) but a lower OR of ER⁺PR⁺ breast cancer (OR 0.76, 95% CI 0.49 to 1.18) compared with never users. ER⁺PR⁺ case patients were less likely to have had longer duration of COC use than ER^- PR^- case patients (p _trend = 0.008).

Alcohol drinking status during reference age was not associated with breast cancer risk (Table 2). However, the average number of alcoholic drinks per week in the recent 5 years was positively associated with ER⁺PR⁺ breast cancer (p _trend = 0.03), weakly associated with all types of cancer together (p _trend = 0.12), and not associated with ER^- PR^- breast cancer (p _trend = 0.42). Overall, ER⁺PR⁺ case patients seemed more likely to have drunk larger quantities of alcohol than ER^- PR^- case patients, but the difference was not statistically significant (p _trend = 0.23).