Reproductive Factors, Heterogeneity, and Breast Tumor Subtypes in Women of Mexican Descent

Established risk factors for breast cancer include older age, family history, high mammographic density, and obesity in post- but not premenopausal disease (1). In addition, a number of reproductive factors, including earlier menarche, nulliparity, older age at first pregnancy, and use of hormone replacement therapy (HRT) containing synthetic progestins, have been associated with higher risk of developing breast cancer (1). Although data are scarce, longer menstruation duration has also been associated with higher risk (2, 3).

Differences in patient outcomes based on tumor hormone receptor status, estrogen receptor (ER) and progesterone receptor (PR), as well as the expression level of human epidermal receptor 2 (HER2) has resulted in routine clinical stratification of tumors in the treatment setting. Gene expression studies have confirmed the existence of distinct and reproducible breast cancer subtypes with molecular differences aligning on hormone receptor and HER2 status and at least four major disease subtypes (4). In parallel, epidemiologic studies provide evidence supporting differential effects of reproductive and genetic risk factors on the risk of developing hormone receptor-positive or -negative tumors, which further support etiologic heterogeneity among breast tumors.

As we have previously noted (5), case-only studies can serve as a useful initial step in understanding the extent of etiologic heterogeneity by identifying the correlations between risk factor and disease subtypes. Furthermore, much can be learned by assessing disease risk or etiologic heterogeneity in populations with unique risk factor distributions. Here, we report on reproductive differences by tumor subtype in a high-fertility patient series of breast cancer cases from women of Mexican descent, including a novel exploration of menstruation history.

The Ella Binational Breast Cancer Study is a case-only study of invasive breast cancer; details of the study have been previously described (6). Briefly, using the same protocol and data collection instruments, Mexican and Mexican-American women ages 18 years or more were recruited within 24 months of diagnosis. Recruitment sites included two in the United States (U.S.; the Arizona Cancer Center in Tucson, AZ, which recruited from throughout Arizona; and the M.D. Anderson Cancer Center in Houston, TX) and three in Mexico (the Universidad de Sonora in Hermosillo, Sonora; the Instituto Tecnológico de Sonora in Ciudad Obregón, Sonora; and the Universidad de Guadalajara in Guadalajara, Jalisco). All recruitment sites used a predominately clinic-based recruitment strategy. Recruitment took place from March 2007 through June 2011, with response rates ranging from 95% to 99% (6). Of 1,151 total eligible cases, 110 (9.6%) had unknown ER status, yielding a sample size of 1,041 for the present analysis (559 U.S. and 482 Mexico). The Institutional Review Board from each participating institution approved the study protocol, and all women provided informed consent.

Risk factor data were ascertained from an interview-administered questionnaire and included age at menarche, age at menopause, number of full-term pregnancies (pregnancies lasting more than 5 months regardless of outcome), age at first full-term pregnancy, breastfeeding history, HRT use, and hormone contraceptive use (including birth control pills, injections, implants, patches, and vaginal ring). To assess the association between menstruation history and tumor subtype, we derived three variables after excluding the women who reported irregular menstrual cycles (n = 62): (i) interval between age at menarche and age at first full-term pregnancy, (ii) duration of menstruation taking into account the number of pregnancies: [(age at menopause–age at menarche)–(0.75 years × number of pregnancies)], and (iii) duration of menstruation taking into account pregnancies and breastfeeding, derived as follows: [(age at menopause–age at menarche)–(0.75 years × number of pregnancies)]–(breastfeeding duration). In the last two variables, we substituted age at diagnosis for age at menopause for premenopausal women.

Age at diagnosis and tumor marker data for ER, PR, and HER2 were abstracted from the medical records. In the abstraction, priority was given to a numeric value for the percent of cells staining, where ER and PR positivity was based on 1% or more cell staining by immunohistochemistry (IHC). Cases were considered HER2⁺ if amplified as determined by FISH (ratio ≥ 2.2). If no FISH results were available, an IHC intensity score of 3/3+ was considered positive, 2/2+ equivocal, and 0/1/1+ negative. For HER2 classification, we excluded cases with an equivocal IHC intensity score and no FISH data from the analyses (n = 53). Cases were assigned to one of the three tumor marker categories: luminal A (ER⁺ and/or PR⁺ and HER2⁻), HER2⁺ (regardless of ER or PR status), and triple-negative breast cancer (TNBC; ER⁻, PR⁻, and HER2⁻). We classified HER2⁺ tumors independent of hormone receptor status based on data that HER2 mediates endocrine independence (7) and would likely be a shared etiologic factor. In a sensitivity analyses, we classified cases with equivocal IHC score and no FISH data for HER2 as HER2⁻ and included them in the analyses; results were unchanged when compared with those presented in the tables.

Descriptive statistics (mean/SD and proportions) were calculated separately for risk factor characteristics by tumor subtype and for tumor subtype by country of residence (U.S. or Mexico). Associations between reproductive factors and tumor subtypes were tested using multinomial logistic regression, considering luminal A tumors as the comparison group. Each model generated an OR and 95% confidence interval (CI). Because the primary objective was to quantify the associations between reproductive factors and tumor subtype as a means of understanding tumor heterogeneity rather than building risk models, the ORs were adjusted only for age at diagnosis (continuous) and recruitment country (U.S. or Mexico). We did, however, explore parity and breastfeeding together in one model given the interest in understanding the independent effect of each. Tests for trend were conducted by modeling risk factors as continuous variables. All statistical analyses were conducted using Stata 12.1 (StataCorp).

Table 1 presents the risk factor distributions for the study participants by tumor subtype. Participants with TNBC were younger at diagnosis than those with luminal A or HER2⁺ tumors. Duration of menstruation and the interval between menarche and first pregnancy were the shortest for patients with TNBC. Age at first full-term pregnancy was the lowest among women with TNBC. Parity was high in the total population (mean 3.6 births) and it was the highest among women with TNBC. Breastfeeding was more prevalent and longer in duration for women with TNBC. Women with TNBC had the youngest age at menopause and the lowest prevalence of hormone contraceptive use.

We observed variations in the prevalence of tumor markers by country (Table 2). A higher proportion of ER⁻ tumors was shown for women in Mexico compared with those in the U.S. Slightly higher proportions of luminal A and HER2⁺ tumors were observed in the U.S. versus Mexico. Prevalence of TNBC was 16.7% overall; however, the percentage was higher in cases in Mexico (19.5%) than in the U.S. (14.5%).

Reproductive risk factor associations for HER2 subtype and TNBC were conducted using luminal A tumors as the reference group (Table 3). Patients with 3 or more full-term pregnancies were more likely to have TNBC than luminal A tumors (OR, 1.68; 95% CI, 1.10–2.55). Women with a later age at first pregnancy were less likely to have TNBC than luminal A tumors (OR, 0.61; 95% CI, 0.39–0.95). When compared with cases who never breastfed, those who reported breastfeeding for more than 12 months were over twice as likely to have TNBC than luminal A tumors (OR, 2.14; 95% CI, 1.24–3.68); a similar association was observed for breastfeeding duration per birth. A longer interval between menarche and first pregnancy was significantly associated with lower odds of having TNBC than luminal A tumors (OR = 0.47; 95% CI, 0.29–0.77 for 13+ vs. <8 years). Likewise, TNBCs were less likely in women with longer menstruation duration. An inverse association between hormone contraceptive use and TNBC was observed but the point estimate was imprecise. Associations comparing HER2⁺ with luminal A tumors were weak or nonexistent, except for a higher likelihood of having HER2⁺ tumors among women whose interval between last full-term pregnancy and breast cancer diagnosis was within 10 years, which we have previously reported (8); no association was shown for TNBC. When we considered hormone receptor status of HER2⁺ tumors in the reported associations, we found no evidence of a difference by hormone receptor status (data not shown).

In an effort to elucidate the association between breastfeeding and TNBC, we explored confounding and effect modification by parity in this association. Nulliparous women were excluded from these analyses. We observed no significant interaction between parity and breastfeeding (P = 0.751). Associations between parity and TNBC were fairly consistent across breastfeeding categories, as were associations between breastfeeding and TNBC across parity categories (Table 4). When we included parity and breastfeeding in the same model, the ORs for TNBC and parity and breastfeeding were fairly consistent with the positive associations observed in the models that included these variables separately.

In this case-only study in women of Mexican descent, several reproductive factors, including age at first full-term pregnancy, parity, interval between menarche and first pregnancy, duration of menstruation, and breastfeeding, differed in their distribution by tumor subtype. Despite the vast amount of published reports on breast cancer risk factors, relatively few have investigated associations by tumor subtypes. With few exceptions (8–11), most of these study populations comprise predominately or exclusively non-Hispanic White (NHW) women. Consistent with a recent report of Hispanic women (12), the reproductive pattern of participants in our study with breast cancer would classify them as low risk (i.e., high parity, early age at first pregnancy, high breastfeeding rates, and low HRT use).

Our findings add to the growing evidence that reproductive risk factors have divergent effects on breast tumors according to subtype. These differences likely contribute to population-specific disease patterns, which, given the effect of subtype on prognosis, have direct impact on population disease outcomes. For example, the observation that patients with high parity are more likely to present with poor prognosis TNBC than with hormone-responsive luminal A tumors could partly explain the higher cancer-specific mortality in Mexican women compared with lower parity NHWs (13). These results suggest that the differences in reproductive patterns across racial/ethnic groups are the key contributors to population differences in breast phenotype and burden.

Emerging patterns of breast cancer subtype-specific risk associated with reproductive history are complex. It has been hypothesized that breast epithelial cells undergo differentiation following pregnancy, making them less susceptible to the effect of carcinogens (14), which suggests that the period of relevant exposure is that before the initiation of pregnancy (15). However, results from other studies show that both number of menstrual cycles before first full-term pregnancy and total menstrual cycles are positively associated with risk of breast cancer (2, 3). A more recent case–control study (16) showed that the interval between menarche and first pregnancy was inversely associated with the risk of TNBC but not of ER⁺ tumors. Our results are consistent with this degree of heterogeneity, which applies to both duration of menstruation before first pregnancy, as well as that extending to menopause.

Case–control and cohort studies have shown positive associations between parity and risk of ER⁻ tumors or TNBC (17–25). Studies extending outcome to include the intrinsic subtypes also find basal-like tumors associated with higher parity (9, 17, 26). Our results are in agreement with four case–case analyses that reported positive associations between parity and TNBC compared with hormone receptor-positive or luminal A tumors (10, 18, 19, 26, 27); one of these was conducted in Mexican women (10). The literature on age at first pregnancy and tumor subtype is mixed. Our results show that women with later age at first full-term pregnancy (≥ 25 years) have a lower odds of having TNBC than luminal A tumors, consistent with the data from a case–control study of younger women (16) and a large pooled analysis (17) but not with others (9, 11, 21, 25, 26, 28, 29). Differences in the findings across the studies could be due in part to variation in the age cutoff used to define later age at first pregnancy, as well as whether nulliparous or women with younger age at first birth is used as the referent group.

A pooled analysis of 47 studies showed that the relative risk of breast cancer decreased by 4.3% for every 12 months of breastfeeding (30). This protective effect could be due to breastfeeding's induction of final differentiation of the terminal duct epithelium, making it less sensitive to hormonal stimulation (31), or through the delay in return of ovulation, reducing the cumulative number of menstrual cycles and exposure to ovarian hormones. Relatively fewer studies have reported associations between breastfeeding and tumor subtype. Although not all associations are statistically significant, several studies have reported a lower risk of ER⁻, TNBC, or basal-like breast cancer in parous women who breastfeed compared with those who do not (9, 10, 16, 20–22, 28, 29, 32); one exception is a study among younger women (33). Three studies found a protective effect of breastfeeding but no difference in risk by tumor subtype (23, 24, 34). Likewise, case-only studies have shown lower odds of having TNBC or ER⁻ breast cancer than luminal A tumors associated with breastfeeding (18, 19, 27). However, a clinic-based study from Turkey reported a positive association between breastfeeding and TNBC compared with luminal subtype, although the OR was imprecise (35). Our results show a higher likelihood of TNBC than luminal A tumors associated with breastfeeding for more than 12 months. Reasons for this opposing observation are unclear. Although some have explored the independent risk of parity and breastfeeding (11, 19), challenges occur due to their natural co-occurrence. This is even more challenging in the Ella Study population because highly parous women tend to also breastfeed. Our stratified analyses support the positive association for parity and TNBC because the ORs among women who never breastfed are similar to those who breastfed for more than 12 months. Likewise, our stratified analyses for breastfeeding are consistent with a positive association with TNBC, which is also supported in the multivariate analyses that adjusted for parity. Given the higher parity and breastfeeding duration in the Ella Study, future studies among populations with similar exposures should assess these associations. As we have previously noted (5), we must emphasize that our case–case ORs can only point to the degree of heterogeneity by tumor subtype, and it is not possible for us to assess the risk by subtype.

Data on risk factor associations with breast tumors overexpressing HER2 are sparse. Phipps and colleagues (21) reported a higher risk for HER2⁺ tumors associated with a later age at first birth (compared with younger women), whereas Kwan and colleagues (18) showed that a younger age at first pregnancy (compared with nulliparous women) was associated with higher odds of HER2⁺ tumors than luminal A breast cancers. Two case–control studies (22, 24) showed a lower risk for HER2⁺ tumors associated with breastfeeding. Our results do not support any associations for the reproductive risk factors assessed, with the exception of time since last birth, which we have previously reported (8).

Strengths of our study relate to the large sample of women of Mexican descent residing in the U.S. or Mexico and the well-characterized risk factor data and tumor phenotypes. Though recruitment was not population based, participation was very high (95%–99%), which is a challenge in population-based studies. The value of a case-only study design continues to be appreciated, particularly for understanding the etiologic heterogeneity (5, 17, 36). Recognizing the limitations of this design, we were careful not to interpret the associations as indicators of risk. Finally, despite the value in assessing the unique distribution of risk factors by tumor subtype, we were unable to assess these in the context of nulliparity due to its low prevalence in the Ella Study.

In summary, unique reproductive risk factor distributions in a population of Hispanic women indicate substantial heterogeneity in associations between reproductive risk factors and luminal A and TNBC breast cancers. Such heterogeneity was less evident for HER2⁺ tumors, with the exception of the interval between last birth and breast cancer diagnosis. Given the case-only design, these results must be replicated in populations including nondiseased groups. Identification of distinct breast tumor subtypes with discrete natural histories will continue to be helpful in identifying alternate mechanisms of etiopathogenesis for specific tumor phenotypes. This, in turn, can aid in identifying target populations for optimal prevention, diagnosis, and treatment.

M.L. Bondy is a consultant/advisory board member of Susan G. Komen for the Cure. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.E. Martinez, M.L. Bondy, P. Thompson

Development of methodology: M.E. Martinez, L. Natarajan, M.L. Bondy, A. Daneri-Navarro, M.M. Meza-Montenegro, A.M. Brewster, P. Thompson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.E. Martinez, M.L. Bondy, A. Daneri-Navarro, M.M. Meza-Montenegro, L.E. Gutierrez-Millan, A.M. Brewster, I.K. Komenaka, P. Thompson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.E. Martinez, B.C. Wertheim, L. Natarajan, A. Daneri-Navarro, P. Thompson

Writing, review, and/or revision of the manuscript: M.E. Martinez, L. Natarajan, R. Schwab, M.L. Bondy, A. Daneri-Navarro, M.M. Meza-Montenegro, L.E. Gutierrez-Millan, A.M. Brewster, I.K. Komenaka, P. Thompson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.E. Martinez, M.L. Bondy, M.M. Meza-Montenegro, L.E. Gutierrez-Millan, A.M. Brewster, P. Thompson

Study supervision: M.E. Martinez, M.L. Bondy, M.M. Meza-Montenegro, P. Thompson

The authors thank Erin Ashbeck, Julie Buckmeier, Carole Kepler, and Fang Wang for their contribution.

This work was supported by NIH/NCI grants UO1CA153086, CA023074-2953, CA116199-02S1; the Avon Foundation; and the Susan G. Komen for the Cure (KG090934).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.