Association of mammographic density measures and breast cancer ‘intrinsic’ molecular subtypes

Background

Mammographic breast density or the variation in the composition of breast tissue (fat, stromal, and epithelial tissues) on the mammogram image, is known to be a strong and established BC risk factor [1]. Prior studies suggested that it may also be associated specifically with aggressive tumor characteristics, such as high grade and large tumor size [2, 3]. Others suggested that mammographic density may be associated with molecular tumor subtypes [4-9]. The results from these studies are inconsistent, with some studies suggesting the association between mammographic density and all tumor subtypes have similar magnitudes, while other studies report that these associations are primarily driven by HER2 positive, estrogen receptor (ER)-positive or ER-negative tumors [1, 10-17]. Moreover, PMD was found to be strongly associated with BC of tumors of large size, and positive lymph nodes across all ages (<55 years, 55–64 years, and ≥65 years), and ER-negative status among those age<55 years [16]. A study of Chinese breast cancer cases has shown stronger association with HER2-expressing tumors, but this was limited to the subgroup of cases with normal BMI [12].

Dense area (DA) and non-dense area (NDA) may be associated independently with BC risk [18-21]. In our previous study we showed that DA was associated with BC risk, while NDA was associated with a decreased risk across all ages. Among invasive tumor characteristics, we found significant trends in the effect sizes of associations for DA and NDA with BC with larger tumor size, however, no differences were found by nodal status. In addition, among women age<55 years, DA was strongly associated with ER-positive versus ER-negative tumors, and NDA was strongly associated with decreased risk of ER-negative versus ER-positive tumors [22].

Gene expression profiling has identified molecular signatures that classify invasive BC into distinct subtypes that vary in their clinical behavior, response to treatment and etiology [23-25]. Immunohistochemical (IHC) staining of tumor sections using antibody panels can be used as a surrogate to classify these ‘intrinsic’ molecular subtypes. We evaluated the association of density measures (i.e., PMD, DA and NDA), with ‘intrinsic’ molecular subtypes, and whether the strength of associations varies by age and BMI.

Methods

Results

Our pooled analysis of six cohort or case-control studies included 3492 women with invasive BC and 10,148 without BC. Screening mammography was performed a mean of 4.7±2.7 years prior to diagnosis (for cases). Mean age at mammogram was 56 years among both cases and controls. Of 3492 invasive BC cases, 2217 (63%) were classified as Luminal A, 747 (21%) as Luminal B, 159 (5%) as HER2 expressing, and 369 (11%) as TN. Of the 369 TN, 207 had tumor tissue available and 171 were basal while 36 were unclassified. Mean PMD and DA were higher among cases versus controls, while mean NDA was lower among cases versus controls (Table 1). Among the subtypes, unadjusted mean PMD and DA were the highest in HER2 expressing, while the mean NDA was lowest in HER2 expressing compared to other subtypes (Table 1). Invasive tumor characteristics are presented in Table 2 by ‘intrinsic’ molecular subtypes.

Discussion

Overall, our data suggest that mammographic density measures are associated with BC of all ‘intrinsic’ molecular subtypes. Although mammographic density is a strong established risk factor for BC, there have been inconsistent results across the limited studies regarding its association with ‘intrinsic’ molecular subtypes [8, 11, 13, 14, 17, 40-43]. Our study has the largest sample of breast cancer cases to date to examine the association of density phenotypes with breast cancer subtype using mammograms prior to diagnosis, as well to as examine these associations across BMI and age groups. We found similar associations of density phenotypes with breast cancer across subtypes and no differences by BMI. However, we did find differences by age that were driven by stronger associations of PMD and NDA and the luminal A subtype among younger vs. older women.

Findings from prior studies have been inconsistent. The largest study to date consisted of a case-control study from Sweden with 2632 BC patients and 15,945 controls and found an increased risk for Luminal A, Luminal B, HER2-expression and basal-like with increasing absolute dense area with no evidence of subtype heterogeneity [11]. Another study of 477 BC patients and 588 controls also did not find strong evidence that mammographic density parameters differentially affected specific BC tumor characteristics[17]. Other studies have found stronger density and breast cancer associations for HER2-expressing cancer compared to other subtypes; however, they used different measurements for density including volumetric breast density and the clinical BI-RADS measure. These included a North American study of 457 BC patients (n=59 HER-2 expressing), which found stronger associations of volumetric breast density with HER-2 expressing cancer [40], a case-only study of 1321 invasive female BCs from the Piedmont Cancer Registry that found that triple negative and luminal BH+ (ER+ and/or PR + , HER2+) were negatively and positively associated with higher breast density (by BI-RADS), respectively, compared to luminal A [8], and a case-only study from China with 2001 BC patients (n=180 HER-2 expressing) that also found increased risk of HER-2 expressing cancer among women with extremely dense breasts (by BI-RADS) relative to women with almost entirely fat or scattered fibroglandular dense breasts [12], however, this association was limited to those with normal BMI, <25 kg/m².

In our study, the risk estimates for PMD and NDA were more strongly associated with HER2-expressing compared to other BC subtypes, however, the heterogeneity across subtypes was not shown to be statistically significant. Also, we found no statistical significant differences in the density and subtype associations by BMI; in fact, estimates between PMD and DA with HER2 expressing subtype were slightly stronger (and for NDA, weaker) among those with BMI ≥25 kg/m², which is in contrast to prior findings seen among Chinese breast cancer cases [12].

We also found a significant interaction between age and PMD and NDA on breast cancer by subtype. This was primarily driven by the stronger associations of PMD and NDA with breast cancer seen among younger women with luminal A tumors. Given that luminal A tumors are the most common subtype, the significant interaction was likely due to the sufficient power to examine interactions in this subgroup. In fact, the stronger risk estimates for PMD and NDA with breast cancer in younger vs. older age groups were similar in magnitude for both luminal A and TN subtypes, but only statistically significant for luminal A. Among luminal A, the ORs per SD PMD were 1.69 and 1.44 for those <45 and 75+ years, respectively, while among TN, these estimates were 1.72 and 1.41, respectively. Thus, we cannot rule out a differential association of PMD and NDA with TN breast cancer by age due to limited power in this smaller subtype. Interestingly, differential associations by age were not seen with DA, which is often suggested as the more biologically relevant density phenotype[44]. The differences by age may provide insight to mechanisms important to the luminal A and TN subtypes, including age-related involution, which may be reflected in the NDA and PMD measures, and confer greater protection for these subtypes at younger ages. In fact, prior studies by our group suggested a correlation between age-related involution with NDA, but not DA [45]; Gierach et al. confirmed these findings, which were stronger among premenopausal women [46].

Our data confirm prior studies of density measures as risk factors for multiple breast cancer subtypes as well as for aggressive and non-aggressive cancers [47-49]. This underscores the importance of the inclusion of density measures in breast cancer risk models for breast cancer overall and for breast cancer subtypes [50]. Three risk models that incorporate breast density have improved discrimination compared to models without breast density [51-53]. A risk model that predicts ER-positive cancer could be clinically useful since women informed they are at high risk of ER-positive breast cancer may be more likely to consider taking primary prevention medication. To be clinically applicable, risk models for HER2-expessing tumors and triple negative tumors would need to be examined in the context of screening strategies aimed at early cancer detection of these tumors. Understanding the appropriate subtype-specific model estimates for risk factors and their combinations, in particular by age, will require larger sample sizes than those to date. Our findings of differential risk of breast density with luminal A cancers, and possibly TN, by age (higher risk at younger ages) will be important to consider, as inclusion of differential estimates associated with breast density and subtypes by age could be included in risk models. Given the particular rarity of HER2+ and TN (basal) subtypes, in particular, studies will require collaborative efforts to estimate accurate risk estimates by age group as well as consider the possibility of differential estimates by race and ethnic groups.

Limitations of the study have been previously described [16, 22], including variation in populations and study design of the six studies, use of clinical pathology rather than a central pathology review for ER, PR and HER2; modifications in diagnostic criteria over time that may influence tumor characteristics and receptor status. Further, our study is primarily representative of Caucasians women and therefore cannot be generalized to other ethnicities. Our power was also limited for age-and density interactions by subtype analyses for the TN and HER-2 expressing subtypes. We also used digitized film mammograms instead of digital mammography or tomosynthesis and did not have information on the clinical BI-RADS density measure on all studies. The strengths of this pooled analysis, however, include a large sample size from six studies with mammograms available for cases years before the cancer, standardized estimates of DA and NDA, detailed data on covariates and tumor characteristics from pathology reports, with screening mammograms assessed for breast density in a systematic fashion. Further, we and others have shown similar associations of the quantitative density measures and BI-RADS with breast cancer risk from film and digital mammograms [1, 54-56]. The more precise quantitative measures also provide increased power for examining differential associations of breast density with subtype by age and BMI.

Conclusions

Our results suggest inclusion of breast density measures is important for all overall breast cancer and subtype-specific risk models, with no evidence for differential association by BMI but potentially greater impact in younger women depending on the breast density measure.

Supplementary Material