Association of Hormone-Related Characteristics and Breast Cancer Risk by Estrogen Receptor/Progesterone Receptor Status in the Shanghai Breast Cancer Study

Etiologic differences between subtypes of breast cancer deﬁned by estrogen receptor (ER) and progesterone receptor (PR) status are not well understood. The authors evaluated associations of hormone-related factors with breast cancer subtypes in a population-based case-control study involving 1,409 ER-positive (ER þ )/PR-positive (PR þ ) cases, 712 ER-negative (ER (cid:1) )/PR-negative (PR (cid:1) ) cases, 301 ER þ /PR (cid:1) cases, 254 ER (cid:1) /PR þ cases, and 3,474 controls aged 20–70 years in Shanghai, China (phase I, 1996–1998; phase II, 2002–2005). Polytomous logistic regression and Wald tests for heterogeneity across subtypes were conducted. Breast cancer risks associated with age at menarche, age at menopause, breastfeeding, age at ﬁrst livebirth, waist-to-hip ratio, and oral contraceptive use did not differ by hormone receptor status. Among postmenopausal women, higher parity ( (cid:2) 2 children vs. 1) was associated with reduced risk (odds ratio (OR) ¼ 0.69, 95% conﬁdence interval (CI): 0.52, 0.91) and higher body mass index (BMI; weight (kg)/height (m) 2 ) with increased risk (highest quartile: OR ¼ 2.40, 95% CI: 1.65, 3.47) of the ER þ /PR þ subtype but was unrelated to the ER (cid:1) /PR (cid:1) subtype (for parity, P heterogeneity ¼ 0.02; for BMI, P heterogeneity < 0.01). Hormone replacement therapy (OR ¼ 2.25, 95% CI: 1.40, 3.62) and alcohol consumption (OR ¼ 1.59, 95% CI: 1.01, 2.51) appeared to be preferentially associated with the ER þ /PR (cid:1) subtype. These ﬁndings indicate that BMI, parity, hormone replacement therapy, and alcohol consumption may play different roles in subtypes of breast cancer. More research is needed to better understand the etiology of 2 relatively rare subtypes, ER þ /PR (cid:1) tumors and ER (cid:1) /PR þ tumors. OR, waist-to-hip

Epidemiologic and biologic evidence suggests a pivotal role for estrogen and progesterone in the development of breast cancer. The effects of these hormones are mediated by their respective receptors, estrogen receptor (ER) and progesterone receptor (PR). ER and PR status are biologic markers considered to be a crucial factor in treatment recommendations (1). Furthermore, it has been hypothesized that hormone-related risk factors that reflect exposure to estrogen and progesterone may be predominantly associated with breast tumors that express ER and PR but not with tumors that are hormone receptor-negative (2)(3)(4). Given the evidence for differential associations, it is important to take into account the ER/PR status of breast tumors with respect to risk and prognosis prediction and therapeutic decisionmaking.
In a systematic review of publications on the etiology of hormone receptor-defined breast cancer, Althuis et al. (5) concluded in 2004 that nulliparity, later age at first livebirth, and early age at menarche were associated with increased risk of ER-positive (ERþ) tumors but not ER-negative (ERÀ) tumors. Althuis et al. also reported that postmenopausal obesity was more consistently associated with an increased risk of ERþ/PR-positive (PRþ) tumors compared with ERÀ/PR-negative (PRÀ) tumors. Another meta-analysis published in 2006 (6) suggested that each birth reduced the risk of ERþ/PRþ cancer by 11% (relative risk ¼ 0.89, 95% confidence interval (CI): 0.84, 0.94) and that women who were older at the time of their first birth had a greater risk of ERþ/PRþ cancer than women whose first livebirth occurred at a younger age (relative risk ¼ 1.27, 95% CI: 1.07, 1.50). Neither parity nor age at first birth was associated with the risk of ERÀ/PRÀ cancer (6). The protective effects of longer duration of breastfeeding and later age at menarche did not differ by ER/PR status (6). Recent studies have also found that the associations of some hormone-related factors with breast cancer risk, such as parity, age at first livebirth, age at menarche, and obesity, differed by ER/PR status (3,(7)(8)(9). However, inconsistent results have also been reported (10,11). Although most previous studies have included a large number of ERþ/PRþ breast cancer patients, only a few have included a sizable number (n ¼ 500) of ERÀ/PRÀ patients. Sample sizes for patients with ERþ/ PRÀ and ERÀ/PRþ cancer are even smaller, yielding limited evidence for these 2 subtypes of breast cancer.
Results on breast cancer risk factors among subgroups defined by ER/PR status have been predominantly derived from studies of Western populations, and few studies have been conducted among other racial or ethnic groups. It has been suggested that ER/PR status varies significantly across racial/ethnic groups, and Chinese women are more often diagnosed with hormone receptor-negative tumors than Caucasian women (12,13). However, few studies have investigated etiologic differences for subtypes of breast cancer among Chinese women (14,15). In this study, we conducted a comprehensive evaluation of hormone-related breast cancer risk factors among breast cancer subgroups defined by ER/PR status, using data from the Shanghai Breast Cancer Study.

Study population
The Shanghai Breast Cancer Study is a large, populationbased case-control study of 3,443 cases and 3,474 controls being conducted in Shanghai, China. The study had 2 phases of recruitment, phase I (August 1996-March 1998) and phase II (April 2002-February 2005). The study design has been described in detail elsewhere (16). Briefly, eligible breast cancer patients were identified through a rapid case ascertainment system based on the population-based Shanghai Cancer Registry. All participants were permanent residents of urban Shanghai, were between 20 and 70 years of age, and had no prior history of cancer. Controls were randomly selected from the general female population using the Shanghai Resident Registry and were frequency-matched to cases on age distribution in 5-year groups. In phase I of the study, 1,455 (90.8%) cases and 1,556 (90.3%) controls were recruited, and in phase II, 1,988 (83.2%) cases and 1,918 (70.4%) controls were recruited. All participants in the study provided written informed consent, and the study protocol was approved by the institutional review boards of all participating study institutions.

Data collection
In-person interviews were conducted by trained interviewers using a structured questionnaire to collect information on demographic characteristics, family history of cancer, detailed menstrual and reproductive history, use of oral contraceptives, use of hormone replacement therapy (HRT), behavioral and dietary habits, and regular physical exercise during the 10 years preceding the interview. Current weight, waist and hip circumferences, and sitting and standing heights were measured. Body mass index (BMI; weight (kg)/ height (m) 2 ) and waist-to-hip ratio (WHR) were computed from these measurements.
We reviewed medical and pathology records from the hospital where patients were originally diagnosed to obtain information on the ER and PR status of cases, which was primarily measured by the immunohistochemical method. Among patients whose ER/PR status could not be obtained from medical charts, tumor slides were available for 299. For these 299 patients, ER and PR status was evaluated by the Vanderbilt Molecular Epidemiology Laboratory using a double immunohistochemical staining method. Of the total of 3,443 cases, data on ER status were available for 2,700 (78.4%) cases, PR status for 2,679 (77.8%) cases, and joint ER/PR status for 2,676 (77.7%) cases.

Data analysis
Patients were grouped into the following joint categories: ERþ/PRþ (receptor-positive), ERÀ/PRÀ (receptor-negative), ERþ/PRÀ, and ERÀ/PRþ. Analysis of variance and Pearson v 2 tests were used to compare differences in continuous and categorical variables in the frequency distributions between cases and controls and across case subgroups.
Multiple polytomous unconditional logistic regression analysis was used to calculate odds ratios and corresponding 95% confidence intervals for hormone-related factors in association with breast cancer case subtypes (ERþ/PRþ, ERÀ/PRÀ, ERþ/PRÀ, and ERÀ/PRþ). Included in the models were the following variables: age at diagnosis for cases and age at interview for controls (years; continuous variable), educational level (college or above, senior high school, junior high school, or elementary school or below), first-degree family history of breast cancer (yes, no), history of breast fibroadenoma (yes, no), regular physical exercise during the 10 years preceding the interview (at least twice a week; yes, no), BMI, WHR, years of menstruation, history of livebirth (never, ever), parity, and study phase (I, II). For premenopausal women, total years of menstruation were calculated by taking age at diagnosis (cases) or interview (controls) and subtracting age at menarche. For postmenopausal women, total years of menstruation were calculated by taking age at menopause and subtracting age at menarche. We also extended our analysis and stratified the results by the menopausal status of participants, because risk factor profiles for breast cancer may differ between pre-and postmenopausal women.
Tests for trend were conducted by fitting ordinal values corresponding to exposure categories. We used multivariable polytomous unconditional logistic regression to conduct Wald tests to evaluate the heterogeneity of the associations across breast cancer subtypes. All statistical tests were based on 2-sided probability with a significance level of 0.05 and were performed using SAS, version 9.2 (SAS Institute Inc., Cary, North Carolina). Table 1 shows the frequency distributions of sociodemographic and study variables for cases by joint ER/PR status and for controls. Controls tended to have a lower educational level, later age at menarche, younger age at first livebirth, a greater percentage of breastfeeding, lower BMI, less HRT use, more regular exercise, and less frequent family history of breast cancer than cases and were less likely to have a personal history of breast fibroadenoma than cases. Among cases with known receptor status, 1,409 (52.7%) were ERþ/PRþ, 712 (26.6%) were ERÀ/PRÀ, 301 (11.2%) were ERþ/PRÀ, and 254 (9.5%) were ERÀ/PRþ. Age at breast cancer diagnosis (F test: P < 0.01), menopausal status (v 2 test: P < 0.01), duration of oral contraceptive use (F test: P ¼ 0.03), and HRT use (v 2 test: P v2 < 0.01) differed across the 4 case groups with known ER/PR information. Overall, we had information on joint ER/PR status for 2,676 (77.7%) breast cancer cases. Cases with missing data on ER/PR status were similar to those with known ER/PR status with respect to most known breast cancer risk factors, with a few exceptions. Cases with known ER/PR status appeared to have a slightly lower percentage of livebirths (93.22% vs. 95.44%) and lower BMI (23.33 vs. 23.74) and WHR (0.817 vs. 0.824) than those with unknown ER/PR status (see Web Table 1, which appears on the Journal's Web site (http://aje.oxfordjournals. org/)). We adjusted for these variables throughout subsequent analyses. Table 2 presents associations between menstrual factors and breast cancer by joint ER/PR status. No differences in risk were found across the 4 ER/PR breast cancer subtypes for age at menarche, age at menopause, or years of menstruation. Later age at menarche was inversely associated with risk for the ERþ/PRþ and ERÀ/PRÀ subtypes of breast cancer (P-trend < 0.01 for both). Compared with women who were younger at menarche (age 13 years), there was a 34% reduction (odds ratio (OR) ¼ 0.66, 95% CI: 0.53, 0.82) in risk of ERþ/PRþ cancer and a 38% reduction (OR ¼ 0.62, 95% CI: 0.46, 0.83) in risk of ERÀ/PRÀ cancer for women who were older at menarche (age !17 years). Later age at menarche was associated with a statistically nonsignificantly reduced risk of ERþ/PRÀ and ERÀ/PRþ cancers. Women in the highest quartile of years of menstruation had increased risk of all 4 subtypes of breast cancer, although trend tests were significant only for the ERþ/PRþ and ERÀ/PRÀ subtypes. However, the association with menopausal status differed between subtypes defined by ER/PR status (heterogeneity test: P < 0.01), with menopausal status being associated with a reduced risk of ERþ/PRþ and ERÀ/PRþ breast cancer.

RESULTS
Associations of breast cancer risk with never having a livebirth were confined to women with ERþ breast cancer, although the tests for heterogeneity across subtypes of breast cancer were not significant (Table 3). Later age at first livebirth was associated with both ERþ/PRþ and ERÀ/PRþ breast cancer. Having multiple children was significantly related to a reduced risk of ERþ/PRþ breast cancer (P-trend ¼ 0.02). Longer duration of breastfeeding was significantly and inversely related to the ERþ/PRþ subtype (P-trend ¼ 0.01). However, the test for heterogeneity was not significant for any of these factors.
As shown in Table 4, we found no association between oral contraceptive use (ever use or duration of use) and breast cancer for any subtype defined by ER and PR status. Risk was higher for women with ERþ/PRÀ tumors who had used HRT (OR ¼ 2.25, 95% CI: 1.40, 3.62), and associations differed statistically across the 4 subtypes (heterogeneity test: P ¼ 0.01), mainly between the ERþ/PRÀ and ERÀ/PRþ subgroups.
Alcohol intake was not associated with risk of the ERþ/ PRþ, ERÀ/PRÀ, or ERÀ/PRþ subtypes of breast cancer but was related to increased risk of ERþ/PRÀ breast cancer (OR ¼ 1.59, 95% CI: 1.01, 2.51) ( Table 4). However, the heterogeneity test did not reach statistical significance (P ¼ 0.12). The percentage of alcohol intake (5.09%) was low in our study population.
BMI was associated with a significantly increased risk of ERþ/PRþ breast cancer (P-trend < 0.01), and the test for heterogeneity across the 4 subtypes of breast cancer was statistically significant (P ¼ 0.01) ( Table 5). Increasing WHR substantially increased the risk of all 4 tumor subtypes. Compared with women in the lowest quartile of WHR, women in the highest quartile (!0.843) had more than double the risk of breast cancer for all subtypes (OR range ¼ 2.20-2.72).
We further analyzed associations of hormone-related factors with ERþ/PRþ and ERÀ/PRÀ breast cancer by menopausal status. Results for WHR and other factors that had significant heterogeneity are presented in Table 6. Because the numbers of patients with ERþ/PRÀ and ERÀ/PRþ tumors were too small to provide stable estimates, we did not include them in the stratified analyses. We found that parity was significantly and inversely associated with both ERþ/ PRþ and ERÀ/PRÀ tumors among premenopausal women (heterogeneity test: P ¼ 0.56). Among postmenopausal women, the heterogeneity test for parity between ERþ/ PRþ and ERÀ/PRÀ was significant (P ¼ 0.02). Compared with women who had 1 livebirth, having 2 or more livebirths was associated with a substantially lower risk of ERþ/PRþ breast cancer in postmenopausal women (OR ¼ 0.69, 95% CI: 0.52, 0.91). The inverse association of breastfeeding with ERþ/PRþ and ERÀ/PRÀ breast cancer was mainly confined to postmenopausal women. BMI was positively associated with ERþ/PRþ breast cancer in postmenopausal women (P-trend < 0.01), and the odds ratio for the highest quartile was 2.40 (95% CI: 1.65, 3.47). Statistically significant differences for ERþ/PRþ and ERÀ/PRÀ subtypes were found for BMI in postmenopausal women (heterogeneity test: P < 0.01). WHR was positively associated with ERþ/PRþ and ERÀ/PRÀ tumors in both pre-and postmenopausal women. There were no statistically significant ERþ/ PRþ and ERÀ/PRÀ subgroup differences for age at menarche, age at first livebirth, oral contraceptive use, or alcohol   consumption among pre-or postmenopausal women (data not shown).

DISCUSSION
In this large study of 2,676 breast cancer cases with hormone receptor information and 3,474 controls from the Shanghai Breast Cancer Study, we found that parity and BMI were differently associated with ERþ/PRþ breast cancers but not ERÀ/ERÀ breast cancers among postmenopausal Chinese women (heterogeneity test: P < 0.05) and that HRT and alcohol consumption were associated only with ERþ/PRÀ breast cancer. There were no differences in associations between the 4 breast cancer subtypes and age at menarche, age at menopause, breastfeeding, age at first livebirth, oral contraceptive use, or WHR.
Many previous studies have shown that menstrual and reproductive factors, including age at menarche, age at first livebirth, and parity, are mainly associated with hormone receptor-positive tumors and not with hormone receptornegative tumors (5,8). Some investigators have also reported no significant differences in risk associated with delayed first livebirth (2,17) or early menarche (2,6,18) for ERþ/PRþ and ERÀ/PRÀ breast cancer. In our study, it was only among postmenopausal women that we observed significant differ-ences in the risk of subtypes of breast cancer associated with parity. The associations of other reproductive factors did not appear to differ by the ER/PR status of the tumor. Consistent with previous reviews (5, 6), we found that the inverse association with long-duration breastfeeding did not vary according to the receptor status of the tumor. This finding supports the notion that, in addition to hormonal mechanisms, breastfeeding may act through other, nonhormonal mechanisms-such as differentiation of breast epithelial cells induced by lactation-to reduce the risk of breast cancer (19).
Evidence has shown that the association between obesity and breast cancer risk may vary by menopausal status (20) and hormone receptor status (3,5). Multiple studies have found that excess endogenous estrogen due to obesity contributes to an increased risk of ERþ/PRþ breast cancer in postmenopausal women (8,9,21). In a recent meta-analysis that included 9 cohort studies and 22 case-control studies, Suzuki et al. (9) concluded that excessive BMI was associated with differences in risk of developing ERþ/PRþ tumors according to menopausal status, with a 10% (95% CI: À18, À1) reduction in risk for every 5-unit increase in BMI among premenopausal women and a 33% (95% CI: 20, 48) increase in risk for every 5-unit increase in BMI among postmenopausal women. Consistent with these results, we found that BMI was more strongly associated with receptorpositive tumors among postmenopausal women. However, BMI was unrelated to either ERþ/PRþ or ERÀ/PRÀ breast cancer among premenopausal women. We observed no heterogeneity across tumor subtypes for associations with WHR in analyses with and without stratification by menopausal status. Our findings are supported by 2 (10, 22) of 3 other published studies on WHR (10, 22, 23) that took into account joint ER/PR status. Our finding that higher BMI was differentially associated with the risk of ERþ/PRþ and ERÀ/PRÀ breast tumors among postmenopausal women while the WHR associations were observed across subtypes of breast cancer suggests that different biologic mechanisms may be involved. It has been suggested that higher BMI increases levels of circulating steroids (21,24) and reduces levels of sex hormone-binding globulin among postmenopausal women (25,26). This may increase the overall level of bioavailable estrogen, which binds estrogen receptors to promote the development of hormone receptor-positive breast cancer. In addition to being inversely correlated with sex hormone-binding globulin (26), WHR is also associated with increased insulin levels and insulin-like growth factors (27), which may stimulate the growth of breast cancer cells, an effect that does not depend on ER/PR status.
Data have accumulated linking use of HRT solely to ERþ/ PRþ tumors (8,18,(28)(29)(30)(31), although null findings have also been observed (17,23,32,33). Slanger et al. (30) suggested that associations between HRT use and histologic type could potentially be attributed to ER/PR status. In our study, interestingly, the increased risk for HRT use was associated with ERþ/PRÀ tumors. This finding is consistent with the French E3N cohort study (34), which found that the use of some HRT was more markedly associated with the risk of ERþ/PRÀ tumors than with ERþ/PRþ tumors and suggested that progestogens might have increased the potency of growth factors and hence preferentially affected the risk of ERþ/PRÀ tumors. Potter et al. (32) also suggested that there were differences in risk associated with HRT for ERþ/ PRÀ tumors as compared with the other 3 ER/PR categories in the Iowa Women's Health Study.
There is increasing evidence that ERþ/PRÀ tumors may be a distinct subgroup of breast cancers that occur more frequently than ERþ/PRþ tumors in older patients and patients with a higher frequency of overexpression of human epidermal growth factor receptors 1 and 2 (35). Our finding that alcohol consumption increased the risk of ERþ/PRÀ breast cancer supports the notion that the ERþ/PRÀ subtype may be etiologically distinct. However, all previous studies, including the current study, have had relatively small sample sizes for a comprehensive evaluation of the etiology of ERþ/PRÀ breast cancer tumors. The same applies to ERÀ/PRþ breast cancer. Further studies with larger sample sizes are needed to confirm our findings and to further elucidate the risk profiles of the 2 mixed ER/PR subgroups and the potential mechanisms underlying them.
Several potential limitations of this study should be noted. Hormone receptor information for patients was collected primarily from medical records. Hence, misclassification of hormone receptor status was unavoidable. The percentage of ER/PR positivity previously reported in Chinese populations has varied. The percentage of ERþ breast cancer in our   (38). To further evaluate the possible bias introduced by misclassification of ER/PR status, we compared ER/PR status distributions between cases diagnosed by teaching hospitals or municipal hospitals (third-tier hospitals, the highest quality) and those diagnosed by district hospitals (second-tier hospitals) and found no major differences (Web Table 2). We also analyzed associations for hormone-related factors and breast cancer subgroups by including ER/PR information on cases from only the teaching or municipal hospitals and observed similar association patterns (Web Table 3). Another limitation was missing data on ER/PR status for 22% of our cases. Although the differences between cases with and without known data on receptor status were small (Web Table 1), potential bias could not be completely ruled out. Last, statistical power was limited for our ERþ/PRÀ and ERÀ/PRþ analyses. However, compared with previous published reports, our study had the largest number of cases for these 2 subgroups.
Our study is one of the largest to date with the capacity to conduct an in-depth investigation of the association between hormonal risk factors and breast cancer risk characterized by ER/PR status. The comprehensively collected exposure information also allowed for adjustment for multiple potential confounders during the analysis. The high response rate minimized selection bias. Additionally, over 98% of Chinese women living in Shanghai belong to a single ethnic group (Han Chinese), thus greatly reducing potential confounding by unknown and/or unmeasured factors related to ethnic differences in culture, heritage, lifestyle, etc.
In summary, our study found that associations of BMI, parity, HRT use, and alcohol consumption with breast cancer risk differed by ER/PR status and by menopausal status. However, the association of other hormone-related factors with breast cancer did not vary by ER/PR status. More research is needed to better understand the etiologic differences and underlying mechanisms for subtypes of breast cancer, particularly the 2 relatively rare subtypes, ERþ/PRÀ and ERÀ/PRþ.