Introduction

Breast cancer (BC) is a heterogeneous disease, with multiple intrinsic molecular subtypes [1]. Subtypes are assessed by immunohistochemistry (IHC) and grouped, based on estrogen (Er) and progesterone (Pr) receptors, HER2 and Ki67 status, into luminal (A, BH− , BH+), HER2+ and triple negative [2]. Although BC subtypes vary in biological features, clinical and prognostic implications [3], it is still unclear if subtypes are etiologically distinct [4]. Data from epidemiologic studies suggest a different effect of risk factors on subtypes: some common risk factors may be restricted to certain subtypes, while others may be shared across subtypes [5, 6]. Thus, established BC risk factors may reflect the risk related to the most common subtypes (such as luminal A) [7].

The aim of this study is to compare, using a case-case approach, selected risk factors with each BC subtype, by comparing luminal A with other subtypes, using data from the Piedmont Cancer Registry (Registro Tumori Piemonte—RTP). Clarifying and quantifying which BC risk factors are subtype specific is particularly relevant for less frequent and more clinically aggressive subtypes (i.e. HER2+ and triple negative) [8]. This may have implications to improve subtype-targeted prevention.

Materials and methods

Study population and data collection

Using data from RTP, we identified a series of invasive female BCs (n = 1332) (International Classification of Disease for Oncology, 3rd edition, (ICD-O-3) site codes C50.0–50.9 [9]), diagnosed between January 2008 and December 2014 and treated at AOU (Azienda Ospedaliera Universitaria) Città della Salute e della Scienza, in Turin, Italy.

We retrieved information from RTP and clinical records from the hospital discharge form and reports. We carried out a deterministic record linkage based on unique patient-based variables.

For each cancer case included in this study, we collected age at diagnosis, education (we defined primary and middle school as low education, and high school and college as high education), age at menarche, parity (defined as number of births), age at menopause, BC family history in first- or second-degree relatives.

From pre-anaesthesia examination, additional data on tobacco smoking habits, as well as on weight and height, were obtained. Body mass index (BMI) was defined according to World Health Organization (WHO) criteria [10]. In addition, information on anti-diabetes drugs was considered.

Breast density (BD) was assessed from the preoperative mammogram report closest to the time of diagnosis. Density measurement was performed by a single radiologist from diagnostic digital mammograms of the unaffected breast. BD is routinely classified according to the Breast Imaging Reporting and Data System (BI-RADS) 5th edition into four categories: almost entirely fat, scattered areas of fibroglandular density, heterogeneously dense, and extremely dense [11]. In this study, we re-categorized BI-RADS densities into low BD (BI-RADS density 1), medium BD (BI-RADS density 2) and high BD (BI-RADS density 3 and 4).

From pathology reports, information on Er, Pr, HER2 and Ki67 status (assessed by IHC) was extracted and classified on the basis of St. Gallen criteria [2] and ASCO-CAP guidelines [12, 13]. In particular, Er and Pr status are positive for a nuclear staining in at least 1% of tumour cells. HER2 positivity (IHC result 3 +) is defined as a complete, intense and in at least 10% of tumour cells membrane staining. HER2 is negative (IHC score 0 and 1 +) if the membrane staining is incomplete and faint perceptible or if no staining is observed. In case of an equivocal IHC score of 2 + (weak membrane staining with circumferential distribution in at least 10% of cells), an amplification test (Fluorescence in situ hybridization—FISH), which overruled results of IHC, was considered. The Ki67 index represents the percentage of positively staining cells among the total number of invasive cells in the scored area [14]. A cut-off of 20% was used to dichotomize (low versus high) Ki67 score. Considering Er, Pr, HER2 and Ki67 status together, we defined molecular subtypes as: luminal A (Er + and/or Pr + , HER2−, low Ki67), luminal BH- (Er + and/or Pr + , HER2− , Ki67 high), luminal BH + (Er + and/or Pr + , HER2+), HER2+ (Er − , Pr − , HER2+), triple negative (Er − , Pr − , HER2−). Moreover, pathology reports included information on tumour features categorized according to AJCC Cancer Staging Manual criteria [15]. In particular, we retrieved information on histotype and pathological Tumour-Node-Metastasis (pTNM) stage.

Statistical analysis

To evaluate the association between the five BC subtypes and selected risk factors, we simultaneously estimated the odds ratios (ORs), and their corresponding 95% confidence intervals (CIs), using a multinomial (baseline category) logistic regression model. We considered luminal A subtype as the reference category of the polytomous response variable. The model included terms for age at diagnosis (modelled as a continuous variable), histotype (invasive ductal carcinoma (CDI)/other), pTNM stage (1/2–3), family history of BC (Yes/No), parity (Yes/No), age at menarche (< 13/ ≥ 13), age at menopause (< 45/45–54/ ≥ 55), BMI (< 25/ ≥ 25 kg/m2), diabetes (Yes/No), BD (1/2/3 + 4), education level (Low/High) and tobacco smoking habits (Yes/No). We applied the same adjusted multinomial model stratifying by age at diagnosis (< 50/50–69/ ≥ 70). Likelihood ratio test and the Akaike’s Information Criterion statistics were used to assess the goodness of fit of the models. Pearson Chi-Square test was also used to evaluate association between categorical variable.

The analyses were performed using SAS software, 9.4 version.

Results

In the study, we included 1321 invasive BCs, after exclusion of 7 cases with missing data for Er, Pr, HER2 or Ki67 status and 4 cases treated with only systemic therapy (no surgery performed).

Table 1 shows sociodemographic, reproductive and clinical characteristics of patients by molecular subtype of BC. Of the 1321 cases, 729 (55.2%) were luminal A, 317 (24%) luminal BH − , 115 (8.7%) luminal BH + , 56 (4.2%) HER2+ and 104 (7.9%) triple negative. The median age was 61 in luminal A, 65 in luminal BH − , 59 in luminal BH + , 62 in HER2+ and 58 in triple negative. Regarding tumour characteristics, 89.3% of HER2+ and 78.8% of triple-negative subtypes were CDI, compared to only 58.8% of luminal A. PTNM 1 was the most common stage in all subtypes, except for triple negative (48.3%). Median age at menarche was around 13 years in all subtypes, except for triple negative. Parous women were less frequent in luminal BH + . About 30% of women in each luminal subtype had positive BC family history, over 35% in HER2+ and more than 40% in triple negative. Shifting from BI-RADS 1 and BI-RADS 2, percentages quadrupled in luminal BH + and did not modify in triple negative. All subtypes showed higher percentages for low education, except for triple negative.

Table 1 Baseline characteristics and selected risk factors of 1321 breast cancer patients according to molecular subtypes
Full size table

Table 2 compares selected risk factors with cancer subtypes and gives the multivariate ORs and the corresponding 95% CIs. We considered luminal A as reference category of the response variable. With comparison to a negative BC family history, the OR for a positive BC family history was 1.83 (95% CI 1.13–2.97) for triple negative (versus luminal A). The OR for BI-RADS 2 (versus BI-RADS 1) was 0.56 (95% CI 0.27–1.14) and the OR for BI-RADS 3–4 was 0.37 (95% CI 0.15–0.88) for triple negative (versus luminal A). With reference to BI-RADS 1, the OR for BI-RADS 2 was 2.36 (95% CI 1.08–5.11) for luminal BH + (versus luminal A).Compared to a lower education, the OR for a higher education was 1.78 (1.03–3.07) for triple negative (versus luminal A). The OR for late menarche (versus early) was 1.69 (1.02–2.81) for triple negative (versus luminal A).Compared to nulliparous women, the OR for parous women was 0.56 (95% CI 0.34–0.92) for luminal BH + (versus luminal A). Considering number of births, with reference to nulliparity, for luminal BH + (versus luminal A) the OR for one birth was 0.59 (95% CI 0.33–1.05) and the OR for at least 2 births was 0.55 (95% CI 0.32–0.93) (data not shown). When we stratified for age at diagnosis, for luminal BH + (versus luminal A) the OR for parous women w as 0.36 (95% CI 0.12–1.03) in the < 50 stratum, the OR was 0.68 (95% CI 0.34–1.36) in the 50–69 stratum, and 0.99 (95% CI 0.20–5.01) in the ≥ 70 stratum (data not shown).

Table 2 Odds ratios (ORs) and 95% confidence intervals (CIs) of breast cancer subtypes as compared to luminal A, according to selected risk factors
Full size table

No associations were detected between tumour subtypes and BMI, age at menopause, tobacco smoking habits, diabetes (Supplementary Table S1). Results of multinomial model adjusted for age, family history, parity, age at menarche, age at menopause, education, density BI-RADS are shown in Supplementary Table S2.

Discussion

This case-only study showed that triple negative, compared to luminal A, was negatively associated with higher BD, while it was positively associated with positive family history of BC, higher education and late age at menarche. Further, this study suggested that luminal BH+ , compared to luminal A, was positively associated with higher BD, whereas it was negatively associated with parity.

As regards to family history, one Asian case-case analysis showed an inverse association between family history and triple negative (OR 0.52, 95% CI 0.32–0.82) [16], but several case-only studies did not find any association [17,18,19,20,21,22,23,24]. Notably, most of these studies did not include Ki67 status, thus no distinction between luminal A and luminal BH− was possible. Further, different criteria were used to define positive family history: Edwards et al. [17] and Song et al. [24] included women with BC history in first- or second-degree relatives (as in our study), others restricted to first-degree relatives [21,22,23]. Only one study included women with a family history of breast and/or ovarian cancer [19], and some studies did not report degree-related information [16, 18, 20].

According to our results, Anderson et al. [25] showed, in a cohort of 1150 BC women, that the risk of triple negative (versus no triple negative) increased for women with a family history of BC in a first- or second-degree relative (OR 2.04, 95% CI 1.40–2.98). Moreover, Jiang et al. [26], analysing 645 BC cases, suggested that women with a BC family history in a first- or second-degree relative (versus without) were more likely to develop hormone receptor-positive tumours (OR 1.43, 95% CI 0.91–2.26). This association was significant when limited to cancers diagnosed before age 50 (OR 2.79, 95% CI 1.34–5.81). Our study provided evidence that triple-negative patients have a familial background of breast cancer, underlying that genetic susceptibility may play a stronger role in triple-negative carcinogenesis as compared to other subtypes.

As regards BD, a meta-analysis of case-only studies [27] showed no differences in risk for estrogen receptor-negative versus estrogen receptor-positive tumours across BD categories, although significant heterogeneity across studies was observed. No BD subtypes association was also reported in two other smaller case-only studies [28, 29]. However, in line with our findings on triple negative-BD inverse association, Arora et al. [30] observed in a cohort of women with BCs (1323) a lower risk of triple negative (versus, according to our subtypes classification, luminal A and luminal BH- considered together due to the lack of information on Ki67 status) in patients with extremely dense breast (OR 0.39, 95% CI 0.20–0.77 for BI-RADS 4 versus others); Kim et al. [31], in a case-only study including 178 BC cases, showed that women with high BD (BI-RADS 3 and 4 versus BI-RADS 1 and 2) were less likely to have triple negative compared to luminal A and BH- (OR 0.23, 95% CI 0.07–0.76). Further, using a quantitative method of BD assessment, Shaikh et al. [32] showed in a small study, including 123 BC cases, that triple negative had a significant lower percent density compared with non-triple negative and Conroy et al. [33] estimated a 26% higher risk of developing hormone receptor-positive cancers per 10% increase in density. Moreover, two other case-only studies suggested an increase in risk for HER2-enriched subtypes in dense breast. In particular, Edwards et al. [17] reported an OR 2.98 (95% CI 1.14–7.83) for extremely dense breast (versus fibroglandular) in HER2+ tumours, compared to luminal A (remarkably, this subtype was assigned by hormone receptor status, tumour grade, and mitotic score); Li et al. [34], in a Chinese study including 2001 BC cases, considering fat and fibroglandular as reference category, showed an increased risk of HER2+ (OR 1.81, 95% CI 1.08–3.06) in extremely dense breast and an increased risk for luminal BH + (OR 1.40, 95% CI 1.01–1.96) in the heterogeneously dense breast, compared to luminal A. Our study described a similar increase in risk for luminal BH + in fibroglandular breast (versus fat) compared to luminal A, but no significant association emerged for HER2+ . Our findings suggest that dense breast tissue supports a hormonal microenvironment which could promote hormone receptor-positive tumours development (and maybe HER2− enriched subtypes). Thus, being luminal subtypes of the most widespread cancers, the positive and established association between BD and BC could be driven by these subtypes, while this association may be weaker, or null, for triple-negative tumours.

The risk of triple negative appeared also influenced by a high education. Few studies have addressed the association between the level of education and BC risk (an Italian case–control reported that an elevated level of education accounted for 20.3% of BC cases [35], and fewer have focused on the relation among subtypes. A population-based study of 476 Atlanta (USA) women with BC showed, in a case-case analysis, no association between education (college degree versus other) and triple negative [36]. Education is a general indicator of lifestyle habits such as diet, physical exercise and hormone preparation use, which might explain this increase in risk.

Despite the vast amount of literature addressing the relationship between reproductive factors and subtypes-related risk, no clear patterns have already been defined [37]. In a large case-only analysis from the Breast Cancer Association Consortium (BCAC) Studies [23], which considered together luminal A and luminal BH- as reference, no significant differences emerged for the age at menarche (≤ 12 versus ≥ 15) in triple negative (OR 1.08, 95% CI 0.92–1.28) as well as no significant differences emerged for parity (nulliparity versus parity) in luminal BH + (OR 1.00, 95% CI 0.85–1.18) and in HER2+ (OR 0.98, 95% CI 0.81–1.20). Moreover, a more recent case-only study using BCAC data [38] did not observe, comparing to luminal A (grade was used to categorize this subtype instead of Ki67 positivity), associations between parity (yes versus no) and both luminal BH + (OR 1.04, 95% CI 0.88–1.24) and HER2+ (OR 1.04, 95% CI 0.83–1.29). Further, smaller case-only studies, compared to luminal A and BH-, did not find any significant differences in the association between triple negative and age at menarche (< 13 versus ≥ 13) [16, 20], as well as between luminal BH + and number of births (nulliparity, versus 1–2, versus 3 or more) [16, 18, 20]. In contrast with our findings, two case-only studies showed that women with early age at menarche were more likely to be found in triple negative. In particular, a Chinese study [22] including 8067 BCs reported an increase in risk (OR 1.33, 95% CI 1.02–1.74) for age at menarche ≤ 13 (versus ≥ 16) compared to luminal A and a population-based study of 476 BCs [36] reported an increase in risk (OR 1.55, 95% CI 1.08–2.23) for age at menarche < 12 (versus ≥ 12) compared to luminal A and BH-. Similarly to our findings, a case-only study [39], conducted on 1041 women with invasive BCs, suggested, comparing to luminal A and BH-, that triple negative was more likely in women with shorter menstruation duration (OR 1.27, 95%CI 0.88–1.82 for age at menarche ≥ 13), and tumours expressing HER2+ (both luminal BH + and HER2+) were less likely in parous women (OR 0.87, 95% CI 0.61–1.24 comparing at least 3 births with 2 births). Our results indicate that hormonal risk factors have a greater impact on luminal type BC than on triple negative. This is confirmed by our findings about BD, which could suggest that hormonal factors have a minor role on triple-negative development. Indeed, the decrease in risk of luminal BH + (and likely HER2+) in parous women suggests that hormonal factors linked to pregnancy could induce a somewhat protective effect against HER2-positive BCs [38].

Limitations of this study included BD assessment that was relied on BI-RADS classification, a visual and subjective method; held by a single reader, no assessment of intra o inter observer variability was performed. Moreover, prior studies addressing this topic are heterogeneous, due to differences in study populations, BD and receptor status assessment, subtype classification criteria as well as adjustment for covariates. These disparities limited comparison with other studies and could partly explain lack of concordance across studies. Further, our data sources did not include details on alcohol intake and age at first birth; thus, adjustment for these factors was not possible.

Strengths included its case-only design, which reduces selection and information bias. Information bias was also minimized since data were collected by doctors and hormone receptor status was assessed in the same hospital laboratory. Further, considering that women plausibly did not differentially recall according to their BC subtype, recall bias unlikely impact our findings. Moreover, other strengths include the spectrum of BC‐related factors considered (reproductive, sociodemographic, anthropometric and lifestyle risk factors, radiologic index (BD), tumour characteristics) and the accurate criteria adopted to define molecular subtypes. Notably, tumour classification included Ki67 status.

Conclusion

This study provided further quantification of BC heterogeneity. This could redefine the role of some risk factors among subtypes, reflecting, in an etiological perspective, potential mechanisms of carcinogenesis.