Introduction

Racial and ethnic differences in the incidence of breast cancer subtypes are well documented in the Surveillance, Epidemiology, and End Results (SEER) Program [1]. Among incident cases with known subtype defined by estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) [1], luminal A (ER-positive and/or PR-positive and HER2-negative) is the most common subtype, accounting for 72.7% of breast cancers, with the highest incidence among non-Hispanic White (NHW) women. Triple negative (TN) subtype (ER-negative and PR-negative and HER2-negative) accounts for 12.2% of breast cancers, and, among women diagnosed under age 50 years, the incidence is highest among African American and Hispanic women. Luminal B (ER-positive and/or PR-positive and HER2-positive) and HER2-enriched (ER-negative and PR-negative and HER2-positive) subtypes account for 4.6% and 10.3% of breast cancers, respectively. Racial and ethnic differences in the incidence of breast cancer subtypes suggest etiologic heterogeneity. Most epidemiologic studies, however, included NHW women only [2,3,4,5,6,7]. There is a need to better understand risk factors for breast cancer subtypes among racially and ethnically minoritized populations who have a greater burden of the clinically more aggressive subtypes that have poorer prognosis compared to luminal A subtype [8].

We investigated subtype-specific associations with reproductive characteristics which are well established risk factors for breast cancer [9, 10]. Heterogeneity by subtypes has been reported, although results are not consistent [2,3,4,5,6,7]. Furthermore, most findings on subtype-specific associations with reproductive factors are based on cohort and case-control studies [11,12,13,14,15,16,17,18,19] and pooled analyses [4, 6, 20, 21] that included mostly NHW women; few studies have been conducted among African American women [21,22,23,24,25], and subtype-specific analyses among Asian American or Hispanic women are lacking. We previously examined associations between reproductive factors and risk of breast cancer defined by joint ER/PR status in the Breast Cancer Etiology in Minorities (BEM) Study, a population-based pooled dataset with 90% of study participants who self-identified as African American, Asian American, or Hispanic [26, 27]. Building upon this previous work, the present analysis was based on a subset of women with breast cancer who had complete data on ER/PR/HER2 status. There is some evidence that age at diagnosis or menopausal status may modify some subtype-specific associations with reproductive factors, but findings are not consistent [6, 7, 17, 27,28,29,30,31,32]. Given that younger women are more likely to be diagnosed with more aggressive breast cancer subtypes compared with older women [1], an evaluation of menopause-specific associations with reproductive factors is warranted. To fill these gaps in knowledge, we conducted subtype-specific case-control analyses overall and by menopausal status and race and ethnicity.

Materials and methods

Study sample

The analysis was based on harmonized data from three population-based studies included in the BEM Study [26]: the Los Angeles County Asian American Breast Cancer Study (AABCS), a case-control study of Chinese, Japanese, and Filipina women [33]; the San Francisco Bay Area Breast Cancer Study (SFBCS), a case-control study of Hispanic, African American, and NHW women [34]; and the Northern California Breast Cancer Family Registry (NC-BCFR), a multiethnic family study that oversampled African American, Chinese, Filipina, Japanese, and Hispanic women and also included population controls [35] (Additional file 1: Table S1). Briefly, the three studies ascertained incident female breast cancer cases through regional population-based cancer registries that are part of the California Cancer Registry and the SEER Program. In AABCS, Chinese, Japanese, and Filipina cases aged 25–74 years, diagnosed with invasive breast cancer from 1995 to 2001 or 2003 to 2006, were ascertained through the Los Angeles County Cancer Surveillance Program. In SFBCS, African American, Hispanic and NHW women diagnosed with invasive breast cancer at age 35–79 years from 1995 to 1999 (all African American women and a 10% random sample of NHW women) or 1995 to 2002 (all Hispanic women) were ascertained through the Greater Bay Area Cancer Registry. In NC-BCFR, women diagnosed with invasive breast cancer at age 18–64 years were ascertained through the Greater Bay Area Cancer Registry (diagnoses 1995 to 2009) or the Sacramento and Sierra Cancer Registry (diagnoses 2005 to 2006). Details on the eligibility criteria and sampling in NC-BCFR are provided in Additional file 1: Table S1. Population controls were identified through random digit-dialing in SFBCS and NC-BCFR or neighborhood block-walking in AABCS, and frequency-matched to cases on race and ethnicity and age group. The Institutional Review Boards of the participating institutions approved the studies, and study participants provided signed informed consent.

The present analysis included women with a first primary invasive breast cancer defined by joint ER/PR/HER2 status obtained from the regional cancer registries at each study site. Reporting of HER2 expression was not required before 1999 in California. Thus, HER2 data were available for only a subset of cases diagnosed during the early years of case ascertainment in the three studies. For 108 NC-BCFR cases diagnosed from 1995 to 1998 with data on ER/PR status, stored tumor slides were used to determine HER2 expression status by immunohistochemistry (by T.L.). Of 5,243 available controls, 20% were NHW, compared to 10% NHW cases. To achieve a more balanced pooled dataset for NHW women, we selected a random sample of available NHW controls frequency-matched to NHW cases at a 1:1.5 case-control ratio by 1-year age group. The current study sample comprised 2,840 cases and 4,653 controls, of whom 90% self-identified as non-Hispanic African American, non-Hispanic Asian American, or Hispanic (White or Black).

Data collection and harmonization

The three studies collected information on breast cancer risk factors using structured questionnaires that were administered in English, Spanish, Cantonese or Mandarin by trained staff in home visits. Risk factors were assessed up to the reference year which was defined as the calendar year before diagnosis for cases or before the interview for controls in AABCS and NC-BCFR or the calendar year before selection into the study for controls in SFBCS. Height and weight during the reference year were assessed by self-report in the three studies, and height and weight were measured at the interview in AABCS and SFBCS.

Questionnaire data were harmonized according to common definitions [26]. Race and ethnicity were based on self-report and categorized as non-Hispanic African American, non-Hispanic Asian American, Hispanic (White or Black), or NHW. Parity was defined as the number of full-term pregnancies (FTP). Lifetime duration of breast-feeding was calculated by summing duration of breast-feeding reported as a continuous measure for each live birth, except for NC-BCFR. In that study, breast-feeding was assessed as a categorical measure (0, < 1, 1–5, 6–11, 12–24, ≥ 25 months) for each pregnancy, and the midpoint of the reported category was used, or 0.5 and 30 months for the categories  < 1 month and ≥ 25 months, respectively, to calculate lifetime duration of breast-feeding. To assess the joint association of breast-feeding and parity, we generated a composite variable (1–2 FTP/never breast-fed; 1–2 FTP/ever breast-fed; ≥3 FTP/never breast-fed; ≥3 FTP/ever breast-fed) that we and others have used previously [18, 27, 36,37,38]. Given that the lower breast cancer risk associated with higher parity is apparent only about 10 years after the last FTP [6], we also used a composite variable to assess the impact of time since last FTP on parity (< 10 years/1–2 FTP; <10 years/≥3 FTP; ≥10 years/1–2 FTP; ≥10 years/≥3 FTP). Women who still had menstrual periods or were pregnant, breast-feeding or perimenopausal during the reference year, and under age 55 years were classified as premenopausal. Women who reported that their periods had stopped naturally or due to surgery, medical treatment, or other reasons prior to the reference year were classified as postmenopausal. Women who still had periods when they started using menopausal hormone therapy were classified as postmenopausal if they were ≥ 55 years of age; otherwise, their menopausal status was classified as unknown. Body mass index (BMI) was calculated as self-reported weight (kg) in the reference year divided by measured or self-reported height (m) squared. If self-reported weight in the reference year was missing, measured weight was used. If measured height was missing, self-reported height was used.

Statistical analyses

We used unconditional logistic regression models to calculate odds ratios (OR) as estimates of relative risks, in accordance with the rare disease assumption, particularly for breast cancer subtypes. We calculated OR and 95% confidence intervals (CI) for associations of breast cancer subtypes with parity, lifetime duration of breast-feeding, a composite parity/breast-feeding variable, age at menarche, age at first FTP, interval between age at menarche and first FTP, interval between last FTP and diagnosis, and a composite variable of interval between last FTP and diagnosis/parity. Because of smaller sample sizes, analyses for luminal B, TN, and HER2-enriched subtypes were based on broader exposure categories. Regression models were adjusted for race and ethnicity, study, age, education, first-degree family history of breast cancer, personal history of benign breast disease, history of oral contraceptive use, BMI in the reference year, and alcohol consumption in the reference year. Categories of the covariates are shown in the footnotes of the tables. Because the association between BMI and breast cancer risk differs by menopausal status [39], regression models for all women combined were additionally adjusted for a composite variable of menopausal status/BMI (premenopausal BMI < 25 kg/m2, premenopausal BMI 25-29.9, premenopausal BMI ≥ 30, postmenopausal BMI < 25, postmenopausal BMI 25-29.9, postmenopausal BMI ≥ 30, unknown menopausal status).

Among premenopausal women, we also adjusted the parity analyses for interval between last FTP and diagnosis. The OR estimates changed very minimally (results not shown) and we did not adjust for years since last FTP in the multivariable models presented in the tables. Linear trends were assessed across ordinal values of categorical variables. Separate analyses were performed for premenopausal and postmenopausal women. For comparison of findings with other studies, most of which did not stratify the analyses by menopausal status or age, we also performed analyses for all women combined that included those with unknown menopausal status. To assess heterogeneity in associations by subtype, we used polytomous regression models, and tested for differences in subtype-specific ORs using a Wald statistic p value. We tested for heterogeneity by menopausal status by including interaction terms for reproductive factors and menopausal status in unconditional logistic regression models, excluding women with unknown menopausal status. To test for heterogeneity by race and ethnicity, we included an interaction term of each exposure variable with race and ethnicity, and tested for heterogeneity using a Wald statistic p value. Among all women combined, we evaluated between-study heterogeneity in subtype-specific associations, separately for premenopausal and postmenopausal women, by including interaction terms for reproductive factors and study. We excluded 46 cases and 74 controls with missing covariate data, leaving 2,794 cases and 4,579 controls in the analytic dataset. NHW cases were only included in the TN analyses as there were only a small number of NHW cases with information on all three markers (84 luminal A, 14 luminal B, 10 HER2-enriched cases). However, because NC-BCFR recruited all TN cases diagnosed from 2007 to 2009 (see Additional file 1: Table S1), the TN case group included 165 NHW cases and analyses were stratified by the four racial and ethnic groups. Counts of controls and cases by subtype, menopausal status, race and ethnicity, and parity status are shown in Additional file 2: Table S2. Two-sided p values were used for tests of trend, with a p < 0.05 considered statistically significant. Statistical analyses were conducted using SAS version 9.4 software (SAS Institute, Inc., Cary, NC).

Results

Of 2,794 breast cancer cases in the analysis, 17% self-identified as African American, 39% Asian American, 34% Hispanic, and 10% NHW (Table 1). Hispanic cases were mostly White; only 17 Hispanic cases self-identified as Black. Compared to controls, higher proportions of cases had a higher education, a first-degree family history of breast cancer, nulliparity or low parity, older age at first FTP, no breast-feeding or for ≤ 12 months, premenopausal status, and higher alcohol consumption. Distributions of reproductive factors among controls varied widely by race and ethnicity (all p < 0.05) (Additional file 3: Table S3). Among premenopausal controls, proportions ranged from 6 to 30% for ≥ 4 FTP, 6 to 26% for breast-feeding ≥ 24 months, 4 to 34% for first FTP at age < 20 years; and 20 to 55% for ≥ 15-year interval between menarche and first FTP.

Table 1 Characteristics of controls and breast cancer cases by molecular subtype
Full size table

Associations between reproductive factors and breast cancer subtypes among all women

Among all women combined, heterogeneity in associations with parity status, parity, and age at first FTP was observed across subtypes (p < 0.05) (Table 2). For luminal A and luminal B subtypes, parity vs. nulliparity (OR = 0.64 and 0.68) and ≥ 4 vs. 1 FTP (OR = 0.55 and 0.46) were associated with lower risk. Longer breast-feeding (> 12 vs. 0 months) was associated with lower risk of luminal A (OR = 0.69) and HER2-enriched (OR = 0.60) subtypes. For the composite of parity/breast-feeding, lower risks were observed for women with ≥ 3 FTP and a history of breast-feeding compared to those with lower parity who never breast-fed, for all subtypes, with ORs ranging from 0.55 to 0.76 and all 95% CIs excluded the null except for TN subtype. Age at menarche was not associated with risk of any subtype. Higher risk of luminal A subtype was associated with older age at first FTP (OR per year = 1.02, p-heterogeneity by subtype = 0.02).

Table 2 Associations between reproductive characteristics and breast cancer subtypes among all women combined
Full size table

In analyses stratified by menopausal status (Table 3; Additional files 7–10: Figures S1-S4), associations of parity with risk of luminal A and luminal B subtypes were consistent by menopausal status. Parity was associated with lower risk of TN subtype among postmenopausal women only. Longer breast-feeding was associated with lower risk of both premenopausal (OR = 0.64, p trend = 0.02) and postmenopausal (OR = 0.76, p trend = 0.02) luminal A subtype and lower risk of HER2-enriched subtype among postmenopausal women only (OR = 0.54, p trend = 0.05). Among premenopausal women, the composite ≥ 3 FTP/ever breast-fed (vs. 1–2 FTP/never breast-fed) was associated with lower risk of luminal A subtype only (OR = 0.66), whereas among postmenopausal women, lower risks were associated with all subtypes, with ORs ranging from 0.46 to 0.64, although of borderline statistical significance for TN subtype.

Table 3 Associations between reproductive characteristics and breast cancer subtypes, by menopausal status
Full size table

Associations with timing of reproductive events were limited to luminal A subtype among premenopausal women, although heterogeneity by menopausal status did not reach statistical significance. Younger age at menarche was associated with higher risk of all subtypes, with ORs per year ranging from 1.06 to 1.10, although the p trend reached statistical significance only for luminal A subtype. Two-fold elevated risks were associated with older age at first FTP (≥ 30 vs. <20 years: OR = 2.09, p-heterogeneity by subtype = 0.01), longer interval between menarche and first FTP (≥ 15 vs. <10 years: OR = 2.41, p-heterogeneity by subtype = 0.04), and shorter interval since last FTP (< 10 vs. ≥20 years: OR = 1.74).

The assessment of between-study variation in subtype-specific associations, separately for premenopausal and postmenopausal women, showed no significant heterogeneity by study.

Associations between reproductive characteristics and breast cancer subtypes by menopausal status and race and ethnicity

Luminal A subtype (African American, Asian American, and Hispanic women)

Premenopausal women. Associations of parity status, parity, and the composite parity/breast-feeding history with risk of luminal A subtype were generally of similar magnitude across Asian American and Hispanic participant groups (Table 4; Fig. 1). Risk of luminal A subtype was not associated with age at menarche among premenopausal African American women, whereas for Asian American and Hispanic women, OR per year were 1.10 and 1.16, respectively. Higher risks were associated with older age at first FTP, longer interval between menarche and first FTP, and shorter interval since last FTP across the three racial and ethnic groups, with estimates of OR per year generally of similar magnitude. For the composite < 10 years since last FTP/1–2 FTP (vs. ≥10 years/≥3 1FTP), suggestive higher risks were observed among Asian American (OR = 1.85, 95% CI = 0.99–3.46) and Hispanic (OR = 2.36, 95% CI = 1.00-5.57) women, with no association among African American women.

Fig. 1
figure 1

Luminal A breast cancer: Associations with reproductive characteristics among premenopausal women, by race and ethnicity

Full size image
Table 4 Luminal A breast cancer: Associations with reproductive characteristics, by menopausal status and race and ethnicity a
Full size table

Postmenopausal women. For parity status, parity, and breast-feeding, no heterogeneity by race and ethnicity was observed (Fig. 2). Higher parity (≥ 3 vs. 1 FTP) was associated with lower risk of luminal A subtype across racial and ethnic groups, with ORs ranging from 0.48 to 0.59. Lower risk was associated with the composite of higher parity with breast-feeding (vs. low parity without breast-feeding) across groups, with OR estimates ranging from 0.39 to 0.56. For age at menarche, we observed heterogeneity by race and ethnicity (p < 0.01). Earlier menarche (< 12 vs. ≥14 years) was associated with higher risk of luminal A subtype among postmenopausal Hispanic women only (OR = 2.00); no association was observed among African American women, whereas among Asian American women, there was an inverse association (OR = 0.52).

Fig. 2
figure 2

Luminal A breast cancer: Associations with reproductive characteristics among postmenopausal women, by race and ethnicity

Full size image

Luminal B subtype (African American, Asian American, and Hispanic women)

Few reproductive factors were associated with risk of luminal B subtype (Table 5). Among premenopausal women, heterogeneity by race and ethnicity was observed for parity (p = 0.04), breast-feeding history (p < 0.01), and interval between last FTP and diagnosis (p = 0.03). Higher parity was associated with lower risk among premenopausal Asian American (OR = 0.45) and Hispanic (OR = 0.33) women, but not among premenopausal African American women. Among postmenopausal women, higher parity (≥ 3 vs. 1–2 FTP) was associated with lower risk overall (OR = 0.57), with OR estimates of similar magnitude across the three racial and ethnic groups, ranging from 0.56 to 0.66. Lower risk was associated with older age at first FTP among Hispanic women and earlier menarche among Asian American women.

Table 5 Luminal B breast cancer: Associations with reproductive characteristics, by menopausal status and race and ethnicity a
Full size table

Triple-negative subtype (African American, Asian American, Hispanic women, and NHW women)

No significant heterogeneity in associations by race and ethnicity was observed among premenopausal women (Table 6; Fig. 3); however, patterns of association were different with respect to TN subtype among premenopausal African American women. Higher parity was associated with higher risk of TN subtype (≥ 3 vs. 1 FTP: OR = 5.75, 95% CI = 1.39–23.8), and an even higher OR for the composite of higher parity without breast-feeding (OR = 16.1, 95% CI = 2.64–97.8). While the OR was attenuated for the composite of higher parity with breast-feeding, it remained elevated (OR = 4.58, 95% CI = 1.02–20.5).

Fig. 3
figure 3

Triple-negative breast cancer: Associations with reproductive characteristics among premenopausal women, by race and ethnicity

Full size image
Table 6 Triple-negative breast cancer: Associations with reproductive characteristics, by menopausal status and race and ethnicity
Full size table

Among postmenopausal women, the composite of higher parity with breast-feeding was associated with lower risk of TN subtype, although the association was statistically significant among Asian American women only (OR = 0.38) (Fig. 4). Heterogeneity by race and ethnicity was observed for the interval between menarche and first FTP (p = 0.01), with a higher risk associated with longer interval observed among Asian American women only (≥ 11 vs. <11 years: OR = 2.31).

Fig. 4
figure 4

Triple-negative breast cancer: Associations with reproductive characteristics among postmenopausal women, by race and ethnicity

Full size image

HER2-enriched subtype (African American, Asian American, and Hispanic women)

Analyses of HER2-enriched subtype stratified by menopausal status and race and ethnicity were based on small sample sizes (Table 7). Among premenopausal Hispanic women, lower risk was associated with parity vs. nulliparity (OR = 0.19, p-heterogeneity by race and ethnicity < 0.01), and higher risk was associated with longer interval between menarche and first FTP (≥ 11 vs. <11 years: OR = 4.87). Among African American women, higher risk was associated with parity vs. nulliparity, higher parity, and a breast-feeding history, but OR estimates were based on very small case counts. Among postmenopausal women, higher parity was associated with lower risk among African American women (≥ 3 vs. 1–2 FTP: OR = 0.23), and younger age at menarche was associated with higher risk among Hispanic women (< 13 vs. ≥13 years: OR = 2.26).

Table 7 HER2-enriched breast cancer: Associations with reproductive characteristics, by menopausal status and race and ethnicity a
Full size table

Discussion

To our knowledge, this is the only U.S. pooled study of breast cancer subtypes enriched with African American, Asian American, and Hispanic women. In the pooled dataset that comprised over 2,700 women with breast cancer, subtype-specific associations with reproductive factors were generally of similar magnitude across racial and ethnic groups and consistent with associations reported for NHW women. For luminal A subtype, lower risk associated with higher parity combined with a breast-feeding history was observed, regardless of menopausal status, with one exception. Among premenopausal African American women, higher parity without a breast-feeding history was associated with a higher risk of luminal A and TN subtypes; these higher risks, however, were attenuated by breast-feeding. For luminal A subtype among premenopausal women only, higher risk was associated with older age at first FTP, longer interval between menarche and first FTP, and shorter interval since last FTP, with similar OR estimates across the three racial and ethnic groups.

The two largest pooled analyses of breast cancer subtypes include an NCI Cohort Consortium analysis by Gaudet et al. (11,741 cases) [4] and an analysis of the Breast Cancer Association Consortium (BCAC) by Jung et al. (23,353 cases, 71,072 controls) [6]. Neither study presented racial- and ethnic-specific subtype results. Data are sparse for African American women on associations of reproductive factors with specific subtypes [21, 24, 25] or TN subtype [22, 23, 38]. The largest study for African American women to date is the African American Breast Cancer and Risk (AMBER) consortium (1,128 cases, 2,932 controls) [24]. To our knowledge, no prior studies have evaluated case-control associations with subtypes defined by joint ER/PR/HER2 status among Asian American and U.S. Hispanic women. Due to the diversity of the study sample (90% African American, Asian American, or Hispanic) and the over-sampling of TN cases in NC-BCFR, the proportions of women with luminal B (16%) and TN (21%) subtypes were higher in our study compared to U.S. population estimates [1].

For all women combined, the present findings of lower risk associated with parous status and higher parity (luminal A and luminal B) and longer breast-feeding (luminal A, HER2-enriched subtype, and TN of borderline statistical significance), and higher risk associated with older age at first FTP (luminal A subtype) were generally consistent with other studies [2, 4, 6, 7]. While some studies of breast cancer subtypes included only younger [12, 16] or older [13, 20] women, only a few studies stratified the analysis by menopausal status [17] or age [4, 6, 11, 21] for select reproductive factors. The present findings of heterogeneity by menopausal status for some reproductive variables highlight its importance, as associations could be masked without stratification. Among premenopausal African American women, we found no evidence of benefit associated with being parous or higher parity; in fact, higher ORs associated with higher parity were observed for all four subtypes, and the OR was statistically significant for TN subtype. For African American women overall, some studies found no evidence of higher risk of luminal A subtype associated with higher parity [21, 24], whereas other studies observed a higher risk of TN or basal-like subtypes [37, 38], likely reflecting the higher risk among premenopausal women only, since we found a strong inverse association with parity among postmenopausal African American women.

Although breast-feeding has been associated with lower risk of breast cancer, regardless of menopausal status [36], associations with breast cancer subtypes have not been consistent [3, 6, 40]. Some studies found similar risk reductions for luminal A and TN subtypes [21], or associations that were stronger for or limited to TN or basal-like subtypes [6, 12, 17, 24, 37]. Notably, in BCAC, a clear inverse association with breast-feeding was observed for TN subtype only [6]. In the present study, longer breast-feeding was associated with lower risk of luminal A, TN (borderline statistical significance), and HER2-enriched subtypes, although in analyses by race and ethnicity, none of the associations reached statistical significance. In agreement with a large pooled analysis of breast cancer overall [36], the risk reduction associated with higher parity was greater in the presence of a breast-feeding history among postmenopausal women for all four subtypes and among premenopausal women for luminal A and luminal B subtypes. Importantly, for luminal A, the most common subtype, this added benefit of breast-feeding was observed among all racial and ethnic and menopausal groups.

Our findings add to the growing evidence that breast-feeding may mitigate the higher risk of TN or ER-negative subtypes associated with higher parity [6, 18, 24, 37, 41]. It has been suggested that the mitigating effect of breast-feeding is more difficult to detect in populations with a high prevalence of breast-feeding [42]. We observed a mitigating effect among premenopausal African American women only who had the lowest prevalence of breast-feeding (48%) compared with 80% among premenopausal Hispanic control women. Pregnancy-associated breast cancer has been attributed to changes in pregnancy-related hormones, as well as immune factors and inflammatory processes triggered during postpartum involution that resemble the pro-tumorigenic process of wound healing. Specifically, the tissue microenvironment of involution, which includes the influx of immune cells, activated fibroblasts, extracellular matrix deposition, elevated matrix metalloproteinase levels, and bioactive matrix fragments, promotes tumorigenesis [43, 44].

We found that early menarche was associated with higher risk of luminal A subtype only and limited to premenopausal women, in agreement with two other pooled analyses that observed an association among younger women only [6, 21]. In contrast, early menarche was also associated with higher risk of non-luminal A subtypes, and in particular with TN subtype among younger women in BCAC [6]. Unlike some studies that observed a higher risk of luminal A subtype associated with earlier menarche among African American women [21, 24, 25], we found no association among African American women, although a longer interval between menarche and first FTP was associated with a suggestive higher risk of borderline statistical significance. The positive associations with luminal A subtype observed among Asian American and Hispanic women are consistent with other studies of NHW women [4, 17].

The exposure measure integrating two early reproductive events (age at menarche, age at first FTP) may be a more relevant risk factor for luminal A subtype, as this represents a window of increased susceptibility when breast tissue undergoes rapid cellular proliferation and rapid accumulation of risk until terminal differentiation occurs during a first pregnancy [45, 46]. The more than two-fold higher risk of premenopausal luminal A subtype associated with ≥ 15 vs. <10 years between menarche and first FTP is of particular concern given trends of delayed childbearing. We did not have data on exposures during this critical time window to further explore what factors might underlie this association, but additional research is warranted.

Pregnancy is associated with a transient increase in breast cancer risk that follows an FTP, wanes over time, and then shifts to a long-term reduction in breast cancer risk [47, 48], about 10 years after a last birth [6]. Consistent with these observations and the large BCAC analysis [6], a shorter interval (< 10 years) between last FTP and diagnosis was associated with a higher risk of luminal A subtype among premenopausal women. The overall OR estimate of 1.03 per year was the same across the three racial and ethnic groups, but reached statistical significance only for women overall.

Comparisons across different subtype classifications

In analyses of mostly NHW women, associations with reproductive factors were generally of similar magnitude for subtypes defined by joint ER/PR/HER2 status or joint ER/PR status [4, 6, 18], and for ER-negative and TN subtypes [4, 6, 22]. Similarly, in our earlier BEM Study analysis [27], associations for ER/PR-positive breast cancer were similar to those for luminal A subtype in the present study, particularly for Asian American and Hispanic women. Larger studies will need to confirm the distinct associations we observed for luminal A vs. luminal B subtypes (e.g., breast-feeding among premenopausal women) and for TN vs. HER2-enriched subtypes (e.g., parity among postmenopausal women). In BCAC, associations with reproductive factors differed primarily between TN subtype and the other subtypes [6].

Racial and ethnic differences in reproductive risk factors

Subtype-specific associations with reproductive factors among premenopausal and postmenopausal women were in the same direction and generally of similar magnitude across racial and ethnic groups, except for parity and breast-feeding among premenopausal African American women. Variation in OR estimates and very wide confidence intervals were likely due to small numbers, particularly among premenopausal women. Distributions of reproductive factors varied considerably across racial and ethnic groups which may contribute to racial and ethnic differences in the incidence of specific breast cancer subtypes. Palmer [22, 49] and Ambrosone [50] suggested that the higher prevalence of high parity, absence of breast-feeding, and young age at first FTP contributes to the higher incidence of early-onset ER-negative breast cancer among African American women. This constellation of factors may also contribute to the higher incidence of TN subtype among premenopausal African American women.

Study limitations and strengths

The subtype-specific analyses were limited by sample size, especially for analyses of the less common subtypes stratified by menopausal status. Subtype was based on readily available cancer registry records, similar to other pooled analyses where subtype was based on medical records, pathology reports, or cancer registry data [4, 6]. The lack of centralized subtyping, as done in some studies [11, 12, 15, 17, 18, 24, 37], might have introduced some misclassification, but it is unlikely that such misclassification would be differential by reproductive characteristics. The small numbers of luminal A, luminal B, and HER2-enriched cases among NHW women precluded subtype-specific analyses in NHW women for comparison with published data from other studies. Not all eligible women with breast cancer and control women in the parent studies participated in the study interviews, which could have introduced selection bias. Reproductive characteristics were based on self-report, therefore subject to inaccurate recall. Non-differential recall bias could result in exposure misclassification which would bias the OR estimates towards the null. There is the possibility that recall is differential between cases and controls, although that may apply to a lesser extent for reproductive factors. Nevertheless, the associations for luminal A subtype in our study were generally consistent with the literature on breast cancer risk factors, providing support to the validity of our findings.

Study strengths include the population-based design of the three studies that were pooled, and case ascertainment through the regional population-based cancer registries which increases the generalizability of our study findings. The diversity of the study sample and use of harmonized exposure variables allowed the direct comparison of OR estimates for African American, Asian American, and Hispanic women. Detailed information was collected on pregnancy and breast-feeding histories and other risk factors. Lastly, we performed analyses stratified by menopausal status that revealed some important differences in associations.

Implications for breast cancer prevention and risk reduction

Breast-feeding is likely the only reproductive risk factor for breast cancer that is potentially modifiable. Efforts focused on improving knowledge on the benefits of breast-feeding and creating a more supportive environment that facilitates breast-feeding could have major impact on lowering breast cancer risk for all subtypes, particularly among premenopausal African American women who are at higher risk. Breast-feeding disparities are tied at multiple levels to social determinants of health that impose barriers to breast-feeding, particularly among African American women (e.g., shorter parental leave; differential access to breast-feeding programs and lactation support; limited accommodations for pumping and storing breast milk at work; and historical and cultural factors [51,52,53,54]. Effective primary breast cancer prevention efforts focused on increasing breast-feeding need to address these barriers among African American women and implement tailored approaches that overcome them [54, 55]. The interval between menarche and first FTP may be a risk factor of increasing importance, given trends of earlier menarche [56, 57] and delayed childbearing [58]. Consistent with these trends, we saw a higher prevalence of longer mean interval between menarche and first FTP and a higher proportion of women with a first FTP at age ≥ 30 years among premenopausal compared to postmenopausal women. These findings warrant studies focused on identifying etiologic factors during this critical time window. The finding of a higher risk of luminal A subtype after a full-term pregnancy suggests that increased surveillance for breast cancer after a full-term pregnancy may be an important strategy to detect breast cancers at an early stage when they are easier to treat and have better survival.

Conclusions

The higher incidence of TN and HER2-enriched breast cancer in some racial and ethnic groups [1], the worse prognosis for these subtypes [8], and the limited knowledge about risk factors warrant research focused on these less common subtypes. Foremost, larger studies and/or pooled analyses in racially and ethnically diverse populations are needed to evaluate reproductive and other risk factors for breast cancer subtypes with greater precision. The distinct associations with parity and breast-feeding among premenopausal African American women, as well as rising incidence rates of distant-stage breast cancer among women under age 40 years [59] underscore the importance of identifying risk factors for breast cancer subtypes among younger women. Centralized subtyping would minimize potential misclassification, and tumor expression data may further facilitate the detection of etiologic heterogeneity for more refined subtypes. A deeper understanding of subtype-specific risk factors, based on both menopausal status and race and ethnicity, is critical for prevention efforts aimed at reducing breast cancer risk and improving survival.