Among women in the USA, breast cancer remains the most commonly diagnosed cancer, excluding skin cancers, and the second leading cause of cancer-related death 1. Breast cancer is characterized by its molecular and clinical heterogeneity. Studies using cDNA microarrays and immunohistochemical (IHC) markers 23456 have classified breast cancers into five distinct subtypes: luminal A (estrogen receptor (ER) positive and/or progesterone receptor (PR) positive, human epidermal growth factor receptor 2 (Her2) negative), luminal B (ER positive and/or PR positive, Her2 positive), Her2 overexpressing (ER negative, PR negative, Her2 positive), basal-like (ER negative, PR negative, Her2 negative, cytokeratin (CK) 5/6 positive and/or epidermal growth factor receptor (EGFR) positive) and normal breast-like tumors. Approximately 70% of 'triple-negative' breast cancers (ER negative, PR negative, Her2 negative) express basal markers 789, resulting in the triple-negative subtype commonly being used as a surrogate marker for the basal-like subtype.

Luminal tumors have been associated with the most favorable prognoses, while Her2-overexpressing and basal-like tumors, or their surrogate triple negative tumors, have been associated with the worst prognoses 2356101112131415161718. For triple-negative tumors, the peak risk of recurrence occurs within three years of diagnosis, and mortality rates are increased for five years after diagnosis 1419. The subtype accounts for approximately 15% of invasive breast cancers 24101120 and is commonly associated with African American race 21017202122, younger age at diagnosis 10111416171920222324, more advanced stage 101124, higher grade 2591114161719212425, high mitotic indices 21626, family history of breast cancer 27, and BRCA1 mutations 232829.

Although many studies have examined associations between common breast cancer risk factors, race 30313233343536 and hormone receptor status 363738394041, few studies have explored the relationship between common breast cancer risk factors and the molecular subtypes of breast cancer 22274243 [see Additional data file 1]. Therefore, we set out to describe breast tumor subtypes by race/ethnicity and common breast cancer risk factors and to determine correlates of breast cancer subtypes using baseline data from two large, prospective breast cancer survivorship studies of 2544 invasive breast cancer cases.

Demographic, reproductive, and lifestyle factors varied significantly by race/ethnicity in the combined studies (Table 1). Demographically, African Americans (mean age = 56.2 years) and Asians (mean age = 54.8 years) were more likely to be diagnosed at a younger age although whites were more likely to be diagnosed at an older age (mean age = 59.8 years). However, Asians (59.1%) were less likely to be post-menopausal than whites (75.6%), African Americans (71.4%), and other races/ethnicities (72.9%). A positive family history of breast cancer was more common among whites (22.6%) and other races/ethnicities (24.1%), than among the other groups. Reproductive history also differed markedly by race/ethnicity; African Americans and Hispanics had more biological children (42.6% and 44.2%, respectively) and were younger during their first pregnancy (73.4% and 60.7%) compared with the other races/ethnicities. Whites were more likely to have ever consumed alcohol (60.1%) while Asians (78.3%) were more likely to have never smoked. Among the hormonal factors, more whites had used HRT (76.2%) compared with the other races/ethnicities, while fewer Asians (44.1%) had used OCs compared with the other races/ethnicities. African Americans were more obese at diagnosis (49.0%) followed by other races/ethnicities (37.0%) and Hispanics (34.5%). Additional data file 2 shows the distribution of demographic, reproductive, and lifestyle factors in the LACE and Pathways Studies separately. Overall, the two study populations were similar demographically, yet non-similarities were apparent among reproductive and hormonal factors, likely to be due to differences in time periods of data collection.

The distribution of breast cancer subtypes by race/ethnicity in the combined studies is presented in Table 2. Among the 2544 invasive breast cancer cases, 1868 (73.4%) were classified as luminal A, 294 (11.6%) as luminal B, 288 (11.3%) as triple negative, and 94 (3.7%) as Her2-overexpressing. The distribution of race/ethnicity within each subtype compared with all other subtypes varied significantly (P < 0.05). The majority of the whites (75.3%), Asians (71.4%), Hispanics (68.5%), other (68.5%), and African Americans (59.4%) had the luminal A tumor subtype. The Her2-overexpressing subtype was least common among all races/ethnicities (whites 3.1%, African Americans 3.2%, Asians 6.4%, Hispanics 6.6%, other 5.5%). African Americans had the highest prevalence of the triple negative subtype (28.4%) compared with the other races/ethnicities (whites 10.5%, Asians 6.3%, Hispanics 10.7%, other 13.0%).

The associations between various demographic, reproductive, and lifestyle factors within each subtype (luminal B, triple negative, and Her2-overexpressing) compared with luminal A are shown in Tables 3 and 4. All case-only ORs were adjusted for age, race/ethnicity, and Pathways/LACE study origin except when either age at diagnosis or race/ethnicity were the main predictors in the logistic regression model. Compared with luminal A cases, luminal B cases were more likely to be younger at diagnosis (OR for < 50 years = 1.83, 95% CI = 1.32 to 2.55; P = 0.0001) and were less likely to consume alcohol (OR = 0.74, 95% CI = 0.56 to 0.98), use HRT (OR = 0.66, 95% CI = 0.46 to 0.94), and OC (OR = 0.73, 95% CI = 0.55 to 0.96). Compared with luminal A cases, triple negative cases tended to be younger at diagnosis (OR for < 50 years = 2.78, 95% CI = 1.99 to 3.90; P ≤ 0.0001) and African American (OR = 3.14, 95% CI = 2.12 to 4.16). Breastfeeding for at least four months was associated with being less likely to have a triple negative tumor, yet this association was of borderline significance (OR = 0.78, 95% CI = 0.59 to 1.03). Compared with luminal A cases, Her2-overexpressing cases were more likely to be younger at diagnosis (P = 0.03), similar to luminal B and triple negative cases, and less likely to use HRT (OR = 0.49, 95% CI = 0.26 to 0.79), similar to luminal B cases. Furthermore, Her2-overexpressing cases were more likely to be Hispanic (OR = 2.19, 95% CI = 1.16 to 4.13) and Asian (OR = 2.02, 95% CI = 1.05 to 3.88). The associations between the risk factors and tumor subtypes for the individual studies are presented in Additional data files 3 and 4. The majority of the case-only ORs were in the same direction as observed in the combined analysis, except for the association between BMI and the triple-negative subtype. In the LACE Study, triple-negative cases were more likely to have higher BMI while in the Pathways Study the opposite trend was observed, although these effect measures were not statistically significant.

For the subgroup analyses, among non-breastfeeding cases, parity of at least three children was associated with a statistically significant increased likelihood of having a triple-negative tumor (OR = 1.68; 95% CI = 1.00 to 2.81) and a Her2-overexpressing tumor (OR = 3.03; 95% CI = 1.27 to 7.23) compared with luminal A. In contrast, no differential associations of breastfeeding and parity were observed among luminal B cases. All P values for interaction of parity by breastfeeding were not statistically significant. As for the effect of BMI by menopausal status, premenopausal triple-negative and Her2-overexpressing cases were more likely to be overweight (triple negative: OR = 1.82, 95% CI = 1.03 to 3.24; Her2-overexpressing: OR = 2.15, 95% CI = 0.70 to 6.58) or obese (triple negative: OR = 1.97, 95% CI = 1.03 to 3.77; Her2-overexpressing: OR = 2.51, 95% CI = 0.74 to 8.51) at diagnosis, yet the effect measures for Her2-overexpressing tumors were not statistically significant. Among luminal A cases, these associations were not observed. All P values for interaction of BMI by menopausal status were not statistically significant.

In a pooled analysis of 2544 breast cancer cases using data from two prospective cohort studies housed within a large health maintenance organization, associations between breast cancer subtypes and various demographic, reproductive, and lifestyle factors were examined. In case-case analyses with the luminal A cases as the reference group, luminal B cases were more likely to be younger at diagnosis and were less likely to consume alcohol, use HRT, and OCs. Triple-negative cases tended to be younger at diagnosis and African American, and were more likely to be overweight and/or obese at diagnosis if premenopausal. Women with triple-negative tumors were also less likely to breastfeed for longer periods, and were more likely to not breastfeed if they had at least three children. Her2-overexpressing cases were more likely to be younger at diagnosis and Hispanic or Asian, and less likely to use HRT. We also found that these cases were more likely to be women with at least three children and no history of breastfeeding. These case-case observations suggest that heterogeneity in associations with traditional breast cancer risk factors exists by tumor subtype.

Several studies have assessed risk factor profiles of tumor subtypes, including the Carolina Breast Cancer Study (CBCS; n = 1424 in situ and invasive cases) 22, the Polish Breast Cancer Study (PBCS; 804 invasive cases) 27, and a pooled study of two Washington State (WS) case-control studies (n = 1023 invasive cases) 4243 [see Additional data file 1]. The CBCS and PBCS were able to classify their triple-negative cases into basal-like and unclassified using CK5/6 and EGFR IHC expression data while the WS study did not do so. The CBCS performed case-case and case-control analyses while the PBCS and WS study conducted case-control analyses only. Although we were unable to further classify triple-negative cases into basal-like and unclassified, similar to the results of the CBCS (case-case analysis) and PBCS (case-control analysis) for basal-like cases, our triple-negative cases were more likely to be younger at diagnosis and African American. We also observed that premenopausal triple-negative cases tended to have higher BMI, which was in agreement with the basal-like cases in the CBCS but not the PBCS, the latter of which found no association. Interestingly, the WS study (case-control analysis) reported a suggestive increased risk of triple-negative tumors with increasing BMI among women currently using hormone therapy 42, yet we did not see any such association in our study. The WS study (case-control analysis) also reported that breastfeeding for at least six months was related to a reduced risk of triple-negative tumors 43. Similarly, both the CBCS (case-case analysis) and our study found suggestive associations of shorter duration of breastfeeding (less than four months) with being more likely to have a triple-negative tumor. Furthermore, both studies observed a strong positive association for triple-negative cases (basal-like cases for CBCS) among women who had higher parity and never breastfed; the CBCS reported a case-only OR for parity of at least three children and no breastfeeding as 1.9 (95% CI = 1.1 to 3.4) compared with luminal A cases. The PBCS (case-control analysis) did not present data on the impact of breastfeeding on tumor subtypes.

As for luminal B and Her2-overexpressing cases, our results are in agreement with those of the CBCS that luminal B and Her2-overexpressing cases tended to be younger than luminal A cases. In contrast to the CBCS results, we observed that these cases were less likely to use HRT although luminal B cases were less likely to consume alcohol. No associations with these factors were observed in the PBCS, and the WS study did not examine luminal tumors separately by luminal A and luminal B subtype. We found that Her2-overexpressing cases were more likely to be Hispanic or Asian, but not African American, an observation which was not seen in any of the other studies. In fact, the CBCS, comprised of only whites and African Americans, reported that Her2-overexpressing cases were slightly more likely to be African American. Finally, we observed that Her2-overexpressing cases were more likely to be women who had at least three children and had not breastfed, an association not seen in the CBCS.

Although our results tend to be in agreement with those of other studies, limitations of our study should be discussed. Only case-case comparisons were conducted, and it must be emphasized that the associations reported here are all in reference to risk of having a luminal A tumor subtype and should not be extended to risk of having invasive breast cancer. Case-case analyses among tumor subtypes are a useful exploratory tool to examine etiologic heterogeneity between the subtypes 53. As previously mentioned, we have no data on CK5/6 and EGFR tumor markers to further classify triple-negative tumors into basal-like and unclassified. However, with additional funding, we plan to conduct these additional IHC assays in triple negative cases. Also, as there were a limited number of Her2-overexpressing tumors (n = 94; 3.7%), results concerning this subtype should be interpreted with caution. Finally, although our large study population of 2544 women diagnosed with invasive breast cancer was more ethnically diverse (76.6% white, 6.1% African American, 7.8% Hispanic, 7.4% Asian, 2.1% other) than other studies that have examined breast cancer risk factors among tumor subtypes, unlike the CBCS, we were unable to further examine risk factors by white and African American race/ethnicity due to limited numbers. Our findings, especially those regarding Hispanic and Asian differences, should be replicated in other population-based studies.

In summary, using a case-case analysis to assess the associations between traditional breast cancer risk factors and breast cancer subtypes (luminal A, luminal B, triple negative, and Her2-overexpressing), we observed significant heterogeneity of associations by tumor subtype. These varying associations by subtype lend further support to the growing evidence base that breast cancer is a heterogeneous disease defined by ER, PR, and Her2 expression with distinct etiologic pathways and prognoses. Future research should focus on refinement of tumor subtypes into more homogenous subgroups in order to best elucidate how risk factors may vary by subtype. Important modifiable factors that may be related to the development of specific tumor subtypes include obesity and possibly breastfeeding (triple negative) and alcohol consumption (luminal B), yet no clear modifiable risk factor profile was apparent for Her2-overexpressing subtypes due to a limited sample size. Given this information, public health programs aimed towards achieving a healthy weight and promoting breastfeeding might reduce the number of poor prognostic triple negative tumors among all breast cancer cases, especially the high-risk African American group.

BMI: body mass index; CBCS: Carolina Breast Cancer Study; CI: confidence interval; CK: cytokeratin; EGFR: epidermal growth factor receptor; ER: estrogen receptor; FISH: fluorescence in situ hybridization; Her2: human epidermal growth factor receptor 2; HRT: hormone replacement therapy; IHC: immunohistochemical; IRB: institutional review board; KPNC: Kaiser Permanente Northern California; LACE: Life After Cancer Epidemiology; OC: oral contraceptive; OR: odds ratio; PBCS: Polish Breast Cancer Study; PR: progesterone receptor; WS: Washington State.

The authors declare that they have no competing interests.

MLK contributed to study conception and design, and acquisition, analysis, and interpretation of data; and drafted and revised the manuscript critically for important intellectual content. LHK and BJC contributed to study conception and design, and acquisition, analysis, and interpretation of data; and revised the manuscript critically for important intellectual content. EW contributed to analysis and interpretation of data; and revised the manuscript critically for important intellectual content. BM contributed to study conception and design; and revised the manuscript critically for important intellectual content. SEK contributed to analysis and interpretation of data; and revised the manuscript critically for important intellectual content. RSF contributed to acquisition of data; and revised the manuscript critically for important intellectual content. MML contributed to analysis and interpretation of data; and revised the manuscript critically for important intellectual content. CBA contributed to analysis and interpretation of data; and revised the manuscript critically for important intellectual content. All authors read and approved the final manuscript.