The combination of increased parity and shorter breastfeeding duration might increase the odds of the least differentiated triple-negative breast cancer (BC) phenotype, theoretically because an expanded progenitor cell population from each pregnancy would incompletely differentiate postpartum.
Subjects consisted of a consecutive case series of 2473 women treated for invasive breast cancer between 2001 and 2006. Breast cancer phenotype (triple-negative BC, vs non–triple-negative BC) was compared with reproductive and demographic information. Odds ratios (OR) with 95% confidence intervals (CIs) for the association of breastfeeding duration (months per child) and parity with triple-negative BC were calculated after adjusting for ethnicity, age at menarche, family history, and age at diagnosis.
Compared with non–triple-negative BC, triple-negative BC was associated with shorter duration of breastfeeding per child (OR, 0.93; 95% CI, 0.90-0.97) and with higher parity (OR, 1.12; 95% CI, 1.06-1.20). By using multivariate logistic regression, triple-negative BC was independently associated with higher parity (OR, 2.76 [95% CI, 1.86-4.08] if ≥3 live births; OR, 1.89 [95% CI, 1.30-2.74] if ≤2 live births vs nulliparae), breastfeeding duration (OR, 0.55 [95% CI, 0.41-0.74] if >2 mo/child and OR, 0.58 [95% CI, 0.42-0.82] if ≤2 mo/child vs none), African American ethnicity (OR, 2.10; 95% CI, 1.52-2.92), and younger age at diagnosis (OR, 3.02 [95% CI, 2.03-4.47] if ≤40 years vs >60 years).
Triple-negative breast cancers (BCs) are characterized by the absence of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2), and have an aggressive natural history with fewer systemic treatment options.1, 2 Triple-negative BC and basal-like categories of BC overlap significantly, and either provides an acceptable definition of undifferentiated phenotype.3-5
Undifferentiated BC phenotype is more prevalent in carriers of BRCA1 germ-line mutation, and among premenopausal African American women.4-7 There is also a reported association with shorter lifetime duration of breastfeeding.7-9 However, both parity and breastfeeding duration might contribute in combination to risk by altering the population of progenitor cells that are retained within breast parenchyma, a cell type that known to have enhanced survival potential under physiological or pathologic stress.8, 10-15 Indeed, recruitment and expansion of progenitor cells within breast terminal ducts sustains epithelial expansion during pregnancy and lactation, and this is followed by epithelial differentiation or apoptosis during the course of weaning.16 However, it is not known what minimum duration of breastfeeding might induce differentiation of breast progenitor cells, or whether a combination of higher parity with repeated short or absent breastfeeding might augment the pathogenesis of BC with undifferentiated phenotype. Therefore, we hypothesized that if malignant transformation of retained progenitor cells might lead to undifferentiated BC (triple-negative BC), then absent or short duration of breastfeeding per child might increase the probability of undifferentiated (triple-negative BC) BC phenotype.17 It is currently not possible to reliably identify and quantify progenitor cells within annotated normal postpartum breast tissues. Therefore, to examine this hypothesis, we examined parity, breastfeeding history, and tumor phenotype in a consecutive case series of women who presented with newly diagnosed BC. The primary objective of the study was to identify an association between parity and breastfeeding duration with the frequency of undifferentiated (triple-negative BC) phenotype. The secondary objective was to examine these associations by ethnicity after adjustment for other potential risk factors for triple-negative BC, including age at diagnosis, age at menarche, and family history of breast or ovarian cancer in a first-degree relative.
MATERIALS AND METHODS
This consecutive case series includes all female patients newly diagnosed with invasive BC who received some form of primary surgical treatment at The University of Texas M. D. Anderson Cancer Center from 2001 through 2006 and whose molecular marker results (including ER and HER2) were reviewed and reported by a pathologist at The University of Texas M. D. Anderson Cancer Center. We retrospectively identified 2511 female patients meeting those criteria and reviewed their pathological and clinical records (institutional review board protocols RCR01-182, LAB03-0432, and RCR06-0891). The patients were referred from the following regions as follows: greater Houston area, 55%; other counties in the state of Texas, 24%; other states, 17%; and other countries, 4%. Ethnicities were self-reported by patients as part of a routine clinical evaluation form at the time of initial clinic visit for all new patients, and were grouped as non-Hispanic white, Hispanic, African American, or Asian. This clinical evaluation form, used at our center since before 2001 and entered into electronic medical records, includes questions about parity, age at menarche, age at menopause, duration of breastfeeding, hormone replacement therapy, and family history. Data from clinical risk factors (described below), histological diagnosis, and receptor status were manually abstracted from the electronic medical records and then deidentified before statistical analysis. A small number of subjects with missing or other ethnic designations or missing receptor and clinical information (n = 38) were excluded from the analyses. The institutional review board of The University of Texas M. D. Anderson Cancer Center approved this research with waiver of consent.
Receptor status (ER, PR, and HER2) was obtained from The University of Texas M. D. Anderson Cancer Center pathology reports. ER-positive and PR-positive status were defined as an immunohistochemical (IHC) staining of ≥10% of cancer cells. HER2-positive status was defined an IHC score of 3+ membranous staining and/or fluorescence in situ hybridization ratio >2.0 (ratio of c-erb-B2/centromere 17 copy number). Combined negative ER, PR, and HER2 status was classified as triple-negative BC, and any positive receptor status was considered non–triple-negative BC.
Clinical Risk Factors
We recorded breastfeeding (categorized as ever vs never, and also as average duration per child), ethnicity, age at menarche, age at first full-term pregnancy, parity (live births), menopausal status (defined as cessation of menstrual cycles, or bilateral oophorectomy), use of hormone replacement therapy (HRT, ever user vs never user), and any family history of BC or ovarian cancer in a first-degree relative.
Breastfeeding duration as months per child was calculated by dividing lifetime duration of lactation by parity. For nulliparous women, this variable was assigned the value of zero. We observed that the distribution of breastfeeding duration per child was skewed (skewness = 3.39 and Kurtosis = 15.53). Therefore, for the purpose of further analysis, this variable was log-transformed and categorized. The strength and direction of associations of various non–triple-negative BC phenotypes (HER2+, ER+/HER2−, and PR+ alone) with the demographic and reproductive risk factors were similar, when compared with triple-negative BC using multivariate logistic regression analysis (data not shown); therefore, we compared all non–triple-negative BC with triple-negative BC for all of the analyses. We used the chi-square test for categorical variables, and Student t test for independent samples for continuous variables, to determine whether risk factors were associated with triple-negative BC. Crude odds ratios (ORs) with 95% confidence intervals (CIs) were calculated for each potential risk factor using univariate logistic regression. These analyses were repeated within the subset of parous women. A multivariate logistic regression model was constructed to evaluate the effect of breastfeeding on triple-negative BC phenotype, adjusted for ethnicity, parity, age at diagnosis, age at menarche, and family history of breast or ovarian cancer in a first-degree relative. This was repeated in subset analyses of each ethnic group. We used SPSS software (version 15.0, SPSS Inc., Chicago, Ill) to perform these analyses (2-tailed P < .05 was considered statistically significant), and used STATA software (version 9.1, STATA Corp., College Station, Tex) to plot graphs.
Information on ER, PR, and HER2 receptor status was retrieved for 2473 of 2511 patients, representing 98.5% of subjects. The frequencies of phenotypic category were: 18.9% triple-negative BC (ER−/PR−/HER2−), 15% HER2+ (any ER or PR status), 63.8% ER+/HER2− (ER+/PR+/HER2− or ER+/PR−/HER2−), and 2.2% PR+ alone (ER−/PR+/HER2−).
Women with triple-negative BC had younger ages at menarche, at diagnosis, and at first full-term pregnancy (Table 1). Specifically, the mean and standard deviation (SD) for ages at menarche were 12.58 ± 1.51 compared with 12.75 ± 1.53 for triple-negative BC and non–triple-negative BC (P = .03, Table 1), respectively; the mean ages at diagnosis were 53 years (SD, 12 years) for triple-negative BC, compared with 56 years (SD, 12 years) for non–triple-negative BC (P < .001, Table 1). Mean age at first live birth was 22 years (SD, 5 years) for triple-negative BC, compared with 24 years (SD, 5 years) for non–triple-negative BC (P < .001, Table 1).
Table 1. Patient Characteristics by Triple-Negative Status
The majority of women were parous (85%, n = 2112). Lactation history (ever vs never breastfed a child) was available in 84% (n = 2079, Table 1), and lifetime duration of lactation in months was available in 82% (n = 2019, Table 1) of the study population. Overall, 37% of women with triple-negative BC had ever breastfed, as compared with 49% of women with non–triple-negative BC, including nulliparous women (P < .001, Table 1). For parous women, the frequency of breastfeeding was 43% in those with triple-negative BC, versus 59% in those with non–triple-negative BC (P < .001). By ethnicity, non-Hispanic white women constituted the majority of the study population (73%), followed by Hispanic (12%), African American (10%), and Asian women (5%). The frequency of triple-negative BC was 35% for African American women, significantly higher than 17% for non-Hispanic white women, 17% for Hispanic women, and 11% for Asian women (P < .001, Table 2). We observed that 31% of parous African American women had breastfed, significantly lower than 58% of non-Hispanic white women, 59% of Hispanic women, and 70% of Asian women (P < .001, data not shown). The mean number of live births for parous African American women was 2.96, similar to 3.02 births for Hispanic women, but significantly higher than 2.46 births for non-Hispanic white women (P < .001) and 2.59 births for Asian women (P = .029).
Table 2. Ethnicity and Breast Cancer Phenotype
TNBC indicates triple-negative breast cancer; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2.
The average lifetime duration of breastfeeding per child, as a continuous log-transformed variable, was inversely associated with the odds of triple-negative BC (OR, 0.90; 95% CI, 0.87-0.94) with a linear relationship, and without an obvious cutoff to recommend a minimum duration for breastfeeding (Fig. 1A). However, we selected a threshold of breastfeeding 2 mo/child for subsequent categorical analyses because the upper limit of 95% CI of the estimated probability of triple-negative BC phenotype after breastfeeding for at least 2 mo/child was less than the estimated probability of triple-negative BC phenotype in parous women who had never breastfed (Fig. 1A). When 3 categories of breastfeeding duration (none, ≤2 mo/child, and >2 mo/child) were evaluated in parous women, we observed that either category of breastfeeding was similarly associated with lower probability of triple-negative BC phenotype, compared with not breastfeeding at all among the study population (Table 3).
Table 3. Crude ORs (with 95% CI) for TNBC Phenotype in All Women and in Parous Women
The probability of triple-negative BC phenotype increased by 12% with each live birth, such that women with BC who delivered ≥3 children were 1.53× more likely to be diagnosed with triple-negative BC phenotype (OR, 1.53; 95% CI, 1.10-2.12) than nulliparous women with BC. We observed particularly strong associations between triple-negative BC phenotype and shorter duration of breastfeeding per child, and between triple-negative BC phenotype and higher parity, in Hispanic women (Figs. 2 and 3). Similar trends (nonsignificant) were observed in non-Hispanic white and African American women.
Triple-negative BC phenotype continued to be significantly associated with shorter duration of breastfeeding per child, higher parity, younger age at diagnosis, and African American ethnicity, even after adjustment for age at menarche and family history of breast or ovarian cancer in a first-degree relative (Table 4). Similar findings were observed in models with all women, and parous women only. Age at first pregnancy was excluded from these models because it was significantly correlated with parity (r = −0.35, P < .001). Premenopausal status (r = −0.693, P < .001) and HRT duration (r = 0.428, P < .001) were excluded from the model, as they were correlated with age at diagnosis. We computed another set of multivariate analyses (data not shown) by categorizing women who never breastfed into nulliparous women and parous women. Of note, parous women who never breastfed were 2.18× more likely to be diagnosed with triple-negative BC phenotype as compared with nulliparous women (OR, 2.18; 95% CI, 1.52-3.12). Because ER+/HER2− was the most dominant group among non–triple-negative BC, we compared this phenotype to triple-negative BC. Similar associations were found to hold true for comparison between these 2 groups.
Table 4. Adjusted ORsa for TNBC Phenotype in All Women and in Parous Women
ORs indicates odds ratios; TNBC, triple-negative breast cancer; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; CI, confidence interval.
Adjusted for ethnicity, age at menarche, age at diagnosis, and family history of breast or ovarian cancer in a first-degree relative.
n represents the number of cases included in the analysis after excluding the cases with missing data.
Breastfeeding duration per child, mo
Age at diagnosis, y
≤60 to >40
1-2 full-term pregnancies
≥3 full-term pregnancies
We also constructed multivariate models for parous women by ethnicity (Table 5). Shorter duration of breastfeeding and higher parity were both independently associated with triple-negative BC phenotype in non-Hispanic white and Hispanic women, but not in African American or Asian women. In parous non-Hispanic white women, triple-negative BC was strongly associated with shorter breastfeeding duration (OR, 0.55 [95% CI, 0.39-0.78] if >2 mo/child; and OR, 0.55 [95% CI, 0.36-0.83] if ≤2 mo/child vs none), higher parity (OR, 1.41 [95% CI, 1.03-1.93] if ≥3 live births vs 1-2 live births), and age ≤40 years at diagnosis (OR, 2.52 [95% CI, 1.42-4.46] vs >60 years). In parous Hispanic women, triple-negative BC was similarly strongly associated with shorter breastfeeding duration (OR, 0.27 [95% CI, 0.11-0.67] if >2 mo/child and OR, 0.34 [95% CI, 0.13-0.89] if ≤2 mo/child vs none), higher parity (OR, 2.43 [95% CI, 1.11-5.33] if ≥3 live births vs 1-2 live births), and age ≤40 years at diagnosis (OR, 3.91 [95% CI, 1.34-11.42] vs age >60 years).
Table 5. Adjusted ORsa for TNBC Phenotype by Ethnicity of Parous Women
The observed association of increased parity and shorter breastfeeding duration per child with triple-negative BC phenotype among women with BC supports the a priori hypothesis of this study, that triple-negative BC would be associated with short or absent breastfeeding per child.8, 17 This could theoretically be from repeated failure of an expanded progenitor cell population in breast tissue to naturally undergo differentiation and apoptosis from prolonged breastfeeding, producing a persistent pool of cells with survival capability and at potential risk for carcinogenesis and subsequent development of undifferentiated tumor phenotype (such as triple-negative BC).8, 10-15 However, this hypothetical stochastic model might not predict the interval until cancer is detected, because of variable effects from each pregnancy and lactation, the timing and potency of subsequent carcinogenic events, and differing growth rates according to phenotype. We stress that we did not study progenitor cells or normal postpartum breast tissues from this patient cohort, and we did not directly study parity and breastfeeding as risk factors for future development of triple-negative BC. Therefore, this study only provides indirect support as an association, and pathogenesis has not been tested at a biological level. However, these and other published data provide a strong clinical rationale for further scientific and epidemiologic research to directly evaluate any pathogenetic role for progenitor cells retained after cessation or weaning.7, 8
We observed a linear relationship between duration of breastfeeding per child and triple-negative BC phenotype (Fig. 2), without an obvious inflection point that would identify an optimal prevention target for breastfeeding duration. Perhaps the physiology associated with initiation of breastfeeding creates an environment conducive for further differentiation of progenitor cells. However, the duration of breastfeeding also seems to have a contributory effect, with >2 mo/child associated with a significant reduction in triple-negative BC phenotype (Fig. 2). Because only 43% of parous women with triple-negative BC in our study ever breastfed, at least several months of breastfeeding might become a lifestyle intervention in prospective prevention studies.
We also observed that increased parity was associated with triple-negative BC phenotype. In general terms, a long duration of breastfeeding, higher parity, and younger age at first full-term pregnancy are known to reduce BC risk.18-21 These all reduce lifetime estrogen exposure and could protect from hormone receptor-positive BC, which accounts for the majority of BCs.22-26 Conversely, estrogen-independent phenotypes of BC might be associated with endocrine risk factors that are not related to lifetime estrogen exposure.8, 14, 15
The study cohort was large, with an adequate proportion with triple-negative BC (19%), and we adjusted for recognized covariates and stratified by ethnicity. However, the lack of a cancer-free control group may limit our ability to generalize these results to the general population. Also, the generalizability of results from our single institution could be questioned, although this cohort did include many patients from other states or countries. However, HER2-positive cancer (15%) was slightly under-represented in our population.
Extraction of breastfeeding data from individual medical records in this study might be subject to potential inconsistency of history taking and data recording at the time, and did not allow for more detailed analyses of breastfeeding duration for each postpartum period. The exact duration of breastfeeding for each child was not available. We estimated the breastfeeding duration per child by dividing lifetime duration of breastfeeding months by parity and thereby assumed that women who reported breastfeeding actually breastfed every child. This assumption is supported by the observation made by O'Campo et al among women who were pregnant from March 1985 to August 1986 (approximately the same time when some of the women in our study population would have been pregnant); notably, that women with higher parity were more likely to breastfeed than women with their first child.27 Although parity was a major factor influencing the probability of breastfeeding and its duration, more recent data by others no longer support this pattern. Specifically, Kruse et al reported that breastfeeding declined across successive children among women who delivered from 1996 to 2001.28
Also, the breastfeeding data were based on self-reported information collected years later. In general, breastfeeding history has been shown to be accurate and reliable.29, 30 However, Promislow et al examined maternal recall of breastfeeding duration by elderly women and reported significant misclassification in breastfeeding duration recalled 34 to 50 years after child birth; their finding demonstrated that misclassification would appreciably attenuate the true strength of association between breastfeeding and later health event under study.31 Thus, even if there is misclassification for length of (or exclusivity of) breastfeeding, the true association between breastfeeding duration and triple-negative BC would be stronger than observed in this study.
Conversely, our findings are consistent with those of 2 recent reports. The prospective Nurses' Health Study reported a significant reduction in risk for premenopausal BC risk from any breastfeeding duration, after adjusting for other covariates.32 This was more pronounced in women with a family history of BC,32 possibly influenced by predominance of triple-negative BC in women who carry BRCA-1 germ-line mutation.4 These investigators also found that anovulation did not mediate the observed association between lactation and risk.32 Comparatively, that cohort included fewer (n = 608) incident BCs and was more uniformly a white population (90%) who had a higher rate of breastfeeding (87%) than the subjects in our cohort.32 The Kaiser Permanente Group in California have reported a case-case comparison of phenotype and breastfeeding in 2544 newly diagnosed BC of stage I-IIIA, finding that patients with triple-negative BC were more likely than patients with ER+/PR+/HER2− BC to have not breastfed if their parity was at least 3 births.9 Comparatively, that cohort included a lower frequency of triple-negative BC (11% vs 19%) and excluded patients with locally advanced BC (stages IIB-IIIC), whereas our study did not.9 Although these studies (including ours) have identified a significant association of triple-negative BC (or premenopausal BC with family history) with higher parity and short duration or absent breastfeeding, we think there is still a need to include detailed prospective or retrospective collection of this information in future studies to confirm the association and understand what aspects of breastfeeding practice might reduce future risk of triple-negative BC.
There also seems to be some overlap in the demographics associated with triple-negative BC phenotype,33 and those with lower rates and shorter duration of breastfeeding.34-37 In our study, the reduction in odds of triple-negative BC phenotype was most pronounced in Hispanic subjects (Fig. 2B). African American ethnicity was an independent risk factor for triple-negative BC, but breastfeeding and parity were not associated with triple-negative BC in African American women. However, detection of any significant association in our study may have been hindered by the relatively low rate of breastfeeding (31% of parous African American women), short duration in those who breastfed (see Fig. 2), and relatively high parity in the African American population. We note that the incidence of hormone receptor negative tumors (ER and PR negative) has increased among Hispanics by 26.8% and among African Americans by 26.1% between 2001 and 2004.38 Interestingly, this increase was preceded by a period when the popularity and prevalence of breastfeeding also declined. Recently, there has been a relative increase in breastfeeding across ethnic and age subgroups, although the absolute rates and duration of breastfeeding are still modest.35 This raises the tantalizing possibility that programs educating mothers about the benefits of even a few months of breastfeeding per child, including African American and Hispanic women, might reduce the incidence of triple-negative BC in the mother in addition to the known benefits to the child. Large and prospective epidemiological studies that include BC free controls would be required to address this possibility. In support of undertaking prospective studies, 3 different cohort analyses have now identified this association, there is a reasonable biological rationale, and the potential for public education and implementation could be significant if at least short-term breastfeeding were proven to reduce the risk of triple-negative BC.
CONFLICT OF INTEREST DISCLOSURES
Supported by a research grant BCTR0600984 to W.F.S. from Susan G. Komen for the Cure, and by the Nellie B. Connally Breast Center at The University of Texas M. D. Anderson Cancer Center.