Abstract
Introduction: The Prosigna-PAM50 risk of recurrence (ROR) score has been validated in randomized clinical trials to predict 10-year distant recurrence (DR) in hormone receptor-positive breast cancer. Here, we examine the ability of Prosigna for predicting DR at 10 years in a subgroup of postmenopausal breast cancer patients with special histological subtypes.
Methods: Using the population based Danish Breast Cancer Group database, follow-up data were collected on all patients diagnosed from 2000 to 2003 with estrogen receptor (ER)-positive and human epidermal growth factor receptor 2 (HER2) normal breast cancer who by nationwide guidelines were treated with 5 year of endocrine therapy (N = 2558). Among patients with 1 to 3 positive lymph nodes or a tumor size >20 mm, we identified 1570 with invasive ductal carcinoma (IDC) and 89 with special histological subtypes (apocrine, medullary, mucinous, papillary, secretory, tubular, neuroendocrine) who were tested with Prosigna. Fine and Gray models were applied to determine the prognostic value of the Prosigna-PAM50 ROR score for DR special subtypes as compared to IDC.
Results: Median follow-up for DR was 9.2 year and for OS 15.2 year. The 10-year DR rate for the special subtypes was 9.2% (95% CI: 4.0% to 17.2%) as compared to 13.7% (95% CI: 11.9% to 15.7%) for IDC. The 10-year OS was 74.2% (95% CI: 63.7% to 82.0%) for the special subtypes and 75.4% (95% CI: 73.2% to 77.4%) for IDC. Prosigna showed a statistical significant association of the continuous ROR score with risk of DR for both IDC and the special subtypes (IDC: p < .0001; special subtypes: p = .01).
Conclusion: In the present study, we demonstrated that Prosigna-PAM50 continuous ROR score added significant prognostic information for 10-year DR in postmenopausal patients with special subtypes (tumor size >20 mm or 1 to 3 positive lymph nodes) and ER-positive, HER2-normal early breast cancer.
Introduction
Invasive breast cancer encompasses a heterogeneous group of tumors with specific morphologic and phenotypic features [Citation1]. At least 90% of breast cancers present histologic characteristics allowing their assignment to one of 21 histologic subtypes endorsed by the WHO [Citation1]. The vast majority of breast cancers are classified as invasive ductal (IDC) or invasive lobular carcinomas (ILC) and less than 8% are invasive tumors with other specific or special histologic features, for example, invasive apocrine, medullary, mucinous, neuroendocrine, papillary, secretory and tubular carcinomas.
Some of these special subtypes are characterized by good long-term outcome [Citation2,Citation3] but published follow-up data are based on small series of patients due to the rarity of these tumors [Citation1]. A better understanding at the molecular level by the application of genomic assays to a sufficiently large cohort might provide improved treatment guidance for these patients.
Previous reports have linked the ER-negative basal-like and claudin-low genomic profiles with metaplastic carcinoma, carcinomas with medullary features, and the molecular apocrine subtype with increased androgen receptor signaling [Citation4–7]. However, the majority of the histological special subtypes, that is, tubular, mucinous, neuroendocrine or papillary carcinomas are ER-positive and are thought to belong to luminal intrinsic molecular subtypes [Citation4,Citation8]. Several multigene assays have been evaluated for the prognostication of early-stage node-negative disease and have confirmed the clinical utility of these tests [Citation9–17]. The Prosigna-PAM50 ROR score adds significant prognostic information above standard clinico-pathological factors in postmenopausal ER-positive HER2 normal, node-negative as well as node-positive (1 to 3 positive lymph nodes) patients [Citation11,Citation15,Citation18]. However, little is known about the performance of the test when applied on tumors of special subtypes.
In this study, we evaluated the ability of the Prosigna-PAM50 ROR score to predict distant recurrence at 10 years in a subgroup of clinically ER-positive postmenopausal breast cancer patients with special subtypes as compared to IDC following allocation without adjuvant chemotherapy to 5 years of endocrine therapy.
Material and methods
The organization of the DBCG and current study cohort has previously been described in detail [Citation18–20]. The study was approved by the National Danish Ethics Committee (H-D-2007-0034).
Patient population
The patients included in this study were all postmenopausal women who by national Danish guidelines from 2000 through 2003 were allocated to 5 years of endocrine treatment as the only adjuvant systemic treatment following a first diagnosis of ER-positive invasive breast cancer. Eligible patients for this DBCG cohort were 50 years or older and met at least one of the following risk criteria: a tumor size > 20 mm (any subtype), ductal histology with malignancy grade 2 or 3; or one to three positive nodes (any subtype). Surgery, radiotherapy and systemic treatment were predetermined and have previously been described as well as the external quality assurance program for ER immunohistochemical staining procedure that was attended by all the Danish pathology departments [Citation19,Citation21]. The prespecified ER cutoff value at the time was ≥10% positive nuclear staining regardless of staining intensity. The histopathological tumor classification was performed by dedicated breast pathologists at the Danish departments of pathology representing real world diagnostics.
For the present subgroup analysis, only patients with carcinomas >20 mm and/or 1–3 positive lymph nodes, and with either rare subtypes (other than ILC) or IDC were included (Supplementary Figure 1).
Central assessment of HER2
Formalin-fixed, paraffin-embedded (FFPE) tumor samples from primary excisional surgery specimens were collected at the Department of Surgical Pathology, Zealand University Hospital, and tissue microarrays (TMA) were constructed with 4 × 1.5 mm tumor tissue cores from, whenever possible, the tumor periphery and then split onto two separate recipient blocks. HER2 status was established centrally on TMAs by standard recommendations [Citation22].
Prosigna analysis
The Prosigna algorithm calculates the tumor-specific molecular subtype (luminal A, luminal B, HER2-enriched or basal like) and calculates a ROR score (1-100 scale) based on a 46 gene subset of the 50 target genes (PAM50) with inclusion of a proliferation score (mean expression of an 18-gene subset of the 50 genes) and tumor size [Citation23]. The Prosigna ROR score can be used as a continuous score or it can categorize ER-positive, HER2-negative postmenopausal breast cancer patients with 0–3 positive lymph nodes into one of three risk groups, or even one of two (low/intermediate versus high) if data are sparse, to determine 10-year risk of recurrence [Citation10,Citation11,Citation24].
RNA extraction and Prosigna testing were performed blinded to any clinical data, following standard operating procedures [Citation11,Citation15,Citation25,Citation26] at the Department of Surgical Pathology, Zealand University Hospital. Once testing of the samples was complete, Prosigna results were transferred to the data manager for preparation of the analysis data set, also blinded to patient data. The analysis data set was then transferred to the DBCG for merging with clinical data and survival analysis according to a prespecified statistical plan.
Endpoints
The primary endpoint was time to distant recurrence (DR) defined as the interval from breast cancer surgery until distant recurrence or death from breast cancer as a first event. Overall survival (OS) was a secondary endpoint. For OS, complete follow-up was achieved until 31 May 2017 by linkage to the Danish Central Population Registry.
Statistical analyses
The statistical analysis was executed by the DBCG statistical office. Categorical characteristics were compared by χ2 or Fisher’s exact test. Follow-up time was quantified in terms of a Kaplan-Meier estimate of potential follow-up. Kaplan–Meier estimates were calculated for OS, and cumulative incidences for DR, handling secondary carcinomas, contralateral breast cancer and death as a first event from causes other than breast cancer as competing risk events. Univariate and multivariate regression analyses were performed for DR and OS. For competing risk analyses, the Fine-Gray subdistribution hazard model was used, while for overall survival the Cox proportional hazard model was employed. Factors included in the multivariable analyses were age (continuous), tumor size (transformed to log(cm)), number of positive lymph nodes (0, 1, 2, 3), histological type and grade (1, 2, 3 and not graded) and for DR also lymphovascular invasion and ER expression (continuous, % positive tumor cells). The assumption of proportional hazards was assessed by Schoenfeld residuals and by including a time-dependent component in the model. The hazard rates of ER and grade were not proportional and were each modeled for early and late periods (<5 years, ≥5 years). Further, the ROR score as a continuous measure (20 point change) and molecular subtype were included in separate models. All p values are two-sided. Statistical analyses were done using SAS v9.4 (SAS Institute, Inc., Cary, NC) and R v3.2.2.
Results
We identified 2558 HER2 normal cases hereof 89 cases of special subtypes and 1570 cases of IDC with tumor size >20 mm or 1 to 3 positive lymph nodes (Supplementary Figure 1). With an estimated median potential follow-up for DR of 9.2 years and for OS of 15.2 years, 191 DR events and 666 deaths were recorded. The median age was 62 years (range, 50 to 89) for IDC and 65 (range, 50 to 87) for the special subtypes (p = .03). Patients in the study cohort were postmenopausal and were without adjuvant chemotherapy assigned to five years of tamoxifen, an aromatase inhibitor (AI) or a sequence of these drugs for early ER-positive breast cancer as described elsewhere [Citation18].
Histological subtype
The histopathological classification of the special subtypes is presented in and included mucinous: N = 51 (2.0%); tubular: N = 16 (0.6%); papillary: N = 12 (0.5%); medullary: N = 5 (0.2%); apocrine: N = 3 (0.1%) and neuroendocrine: N = 1 (0.01%). It should be noted that pure papillary carcinomas are rare and that some cases of the invasive variant of solid papillary carcinoma might be included in this subgroup [Citation1]. In addition, both the medullary and apocrine subtype predominantly but not exclusively are ER negative [Citation1,Citation27]. Overall, the proportion of the individual histopathological subtypes in this study are in accordance with WHO classification of tumors of the breast [Citation1]. Patient characteristics for both special subtypes and IDC are shown in ; In general patients with a special subtype had larger tumor size, but fewer positive lymph nodes. The overall 10-year DR rate for the special subtypes was 9.2% (95% CI: 4.0% to 17.2%) as compared to 13.7% (95% CI: 11.9% to 15.7%) for IDC. The 10-year OS was 74.2% (95% CI: 63.7% to 82.0%) for the special subtypes and 75.4% (95% CI: 73.2% to 77.4%) for IDC.
Table 1. Histological special subtypes and distribution of molecular intrinsic subtype including DR events and Deaths (any cause) (N = 89).
Table 2. Patient characteristics. Special subtypes (except for invasive lobular carcinoma) are not graded according to national guidelines [www.dbcg.dk].
Prosigna ROR score
The distribution of ROR score (low/intermediate vs. high) based on previously defined cutoff levels [Citation11] is presented in . Due to the small sample size for special type analyzes, cases within the range of the low/intermediate ROR score were merged. Prosigna showed a statistical significant association of the continuous ROR score with risk of DR not only, as expected, for IDC, but also for the special subtypes (IDC: p < .0001; special subtypes: p = .01) (). Associations were maintained in adjusted estimates (). For the categorical ROR score no events were registered for the special subtypes in the low/intermediate risk group as compared to IDC with 10-year DR of 3.8% (95% CI: 2.3 to 6.0). In the high-risk group, the 10-year DR rate for the special subtypes was 13.1% (95% CI: 5.7% to 23.8%) and 19.9% (95% CI: 16.3% to 21.7% for IDC (). For OS, the prognostic information for both the special histological subtypes and IDC were somewhat lower, and while highly statistical significant for IDC, neither in univariate nor multivariate analysis was a significant association for ROR score found in the small subgroup of patients with special subtypes (). For both univariate and multivariate analyzes, and considering both DR and OS, no statistically significant heterogeneity in estimates of the ROR score could be found.
Figure 1. Continuous relationship between 10-year risk of distant recurrence and the continuous PAM50-Prosigna ROR score by IDC and special subtypes. The figure shows for continuous ROR score (stated on x-axis) the 10-year risk of distant recurrence in percent (stated on y-axis) for each of the two subgroups, modeled by the Fine-Gray subdistribution hazards model.
Figure 2. Cumulative incidence and overall survival according to ROR score low/intermediate versus high (A + C) and molecular subtype (B + D) for IDC and special types (black = ductal, red = special types).
Figure 3. ▪
Table 3. Hazard Ratio (HR) estimates from multivariate analyzes for distant recurrence and overall survival.
PAM50 intrinsic subtypes
Almost half of the patients with rare subtypes (N = 42, 47%) were assigned to Luminal A, while 37 (42%) were assigned to Luminal B. Only a minor proportion of the patients (N = 7, 8%) had a HER2-enriched or a basal like (N = 3, 3%) subtype as shown in . The subtype assignment for IDC was comparable with the special subtypes (p = .052). illustrates the distribution of the special subtypes with relation to the molecular subtypes, number of events and number of deaths (any cause). The tubular carcinomas are all luminal A, whereas both the mucinous and papillary carcinomas are distributed as luminal A and B with a few tumors allocated to the HER2-enriched subtype. As expected, the tumors with medullary features are mainly basal-like and the apocrine subtype mainly HER2 enriched.
The DR rate was significantly lower (p < .0001) in luminal A, IDC as compared to luminal B, with an absolute 10-year DR rate of 7.4% (95% CI: 5.5% to 9.5%) for the luminal A as compared to 20.7% (95% CI: 17.2% to 24.5%) for the luminal B subtype (). For the special subtypes, the number of events was limited (N = 7), with no events registered for the Luminal A molecular subtype; for Luminal B the absolute 10-year DR rate was 13.1% (4.0% to 27.9%). For OS, there was a statistical significant (p < .0001) better outcome in patients with Luminal A as compared to Luminal B () for IDC, and a similar trend is seen for patients with special types, although not significant (p = .21). Similarly, results are shown from the adjusted estimates for Luminal B as compared to Luminal A in . No statistical heterogeneity in effect of Luminal B compared to Luminal A according to type (IDC vs. special) was identified, with Pinteraction = .55 in multivariate analysis.
Discussion
The present study showed that the Prosigna-PAM50 continuous ROR score added significant prognostic information in postmenopausal breast cancers patients with special ER-positive and HER2-normal histological subtypes. Furthermore, we found no evidence to support a differential impact on prognosis by ROR-score or molecular subtype as determined by Prosigna among patients with a special histological subtype as compared to IDC.
Some caution should be exercised when interpreting our results. First, the number of cases with special subtypes, even in this large study set, was limited. Given that half of the special subtypes were mucinous one could expect that this subtype would drive the results for the entire group, but a formal analysis was prohibited by the small sample size. However, no statistical heterogeneity for neither luminal subtype (B vs. A) nor ROR score according to type (mucinous vs other special subtypes) was observed (data not shown). The small sample size probably in addition explain why the continuous ROR provided a significant result for DR but not for ROR groups or intrinsic subtypes and neither for OS, although a similar pattern was observed. Strengths of our study include a formal prospective–retrospective design with long and detailed clinical follow up. In addition, the histopathological subclassification of breast tumors was performed by experienced breast pathologists at the Danish departments of pathology dedicated to diagnostic breast pathology.
Several multigene assays can robustly identify early stage node negative patients at sufficiently low risk of 10-year DR that they can be safely spared chemotherapy [Citation9–17,Citation28]. The patients recommended for molecular testing are primarily patients with ductal histology and little is known about the benefit of applying these tests in patients with breast carcinomas of special subtypes. Due to their relative rarity, these subtypes are difficult to study outside of very large patient cohorts, as are available through the DBCG.
In daily practice, the histological special subtypes are mainly classified by light microscopy on standard haematoxylin–eosin (H&E) stained sections according to specific morphological growth patterns [Citation1]. Correct histological diagnosis is dependent on the experience of the pathologist as well as standardized sampling of tumor sections for conclusive diagnosis since some of these tumors have mixed growth pattern. Previous studies have shown only moderate agreement (Cohen's K) between observers with respect to the histopathological classification of breast tumors [Citation29]. Not many immunohistochemical biomarkers are available to support the morphological classification apart from loss of E-cadherin as indication of ILC, the presence of androgen receptor and Gross Cystic Disease Fluid Protein in apocrine carcinomas, the myoepithelial cytokeratins in metaplastic carcinomas and synaptophysin and chromogranin A in neuroendocrine tumors. Correct histological classification of malignant breast tumors is mandatory to guide both optimal surgery and postoperative systemic treatment [Citation1]. Also, the increased application of neoadjuvant treatment for tumor down-staging emphasizes the need for accurate tumor subclassification even further, since low proliferative, highly ER-positive, HER2-negative tumors (luminal A-like tumors) respond poorly to neoadjuvant treatment [Citation30]. Heng et al. [Citation31] demonstrated that integration of gene signatures associated with morphological features might add important information with respect to drug resistance and potential targeted treatment options. In addition, the intrinsic molecular subtypes seem to show different patterns of metastasis with luminal A tumors metastasizing primarily to bone [Citation32,Citation33].
On the other hand, both Bomeisl et al. [Citation34] and Turashvili et al. [Citation35] questioned whether the OncotypeDX recurrence score (RS) was a necessary supplement for breast cancers with favorable histology. In the latter, Turasvili et al. did not identify any tumors with high RS among their cohort of 57 patients with special histological subtypes consisting of mucinous, papillary (encapsulated or solid) and tubular carcinomas [Citation35]. This might be explained by the fact that OncotypeDX RS is based on a supervised signature that was not trained on special types.
In our study, both the mucinous and papillary group of tumors were classified into three of the four intrinsic molecular subtypes, excluding only the basal-like type. Given that the cohort was selected based on hormone receptor positivity, it is perhaps not surprising that few cases, apart from three of the five medullary carcinomas, were basal-like by Prosigna. The DBCG used a 10% cut point for estrogen receptor positivity; other cohorts employing the less stringent 1% or Allred three cut points have been found to include larger numbers of basal cases [Citation36].
In conclusion, when applied to histological special subtypes, the continuous PAM50/Prosigna ROR score in particular remain prognostic for distant recurrence in a large population-based series of hormone receptor positive women treated with adjuvant endocrine therapy.
Anne-Vibeke_Laenkholm_et_al._Supplementary_Figure_1.tif
Download TIFF Image (97.9 KB)Disclosure statement
Anne-Vibeke Laenkholm, Maj-Britt Jensen and Bent Ejlertsen have received research funding from Nanostring Technologies Inc. Sean Ferree and Wesley Buckingham: Employment and stockholders, Nanostring Technologies Inc. Torsten O. Nielsen: Advisory role for Nanostring Technologies Inc. Torsten O. Nielsen also declare stock or other ownership for Bioclassifier. Jens Ole Eriksen declare stock or Other ownership for Novo Nordisk.
Additional information
Funding
References
- WHO Classification of Tumours of the Breast, 4th edition. Washington: WHO; 2012.
- Di Saverio S, Gutierrez J, Avisar E. A retrospective review with long term follow up of 11,400 cases of pure mucinous breast carcinoma. Breast Cancer Res Treat. 2008;11:541–547.
- Oakley GJ III Tubbs RR, Crowe J, et al. HER-2 amplification in tubular carcinoma of the breast. Am J Clin Pathol. 2006;126:55–58.
- Curtis C, Shah SP, Chin SF, et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature. 2012;486:346–352.
- Prat A, Parker JS, Karginova O, et al. Phenotypic and molecular characterization of the claudin-low intrinsic subtype of breast cancer. Breast Cancer Res. 2010;12:R68.
- Lehmann BD, Bauer JA, Chen X, et al. Identification of human triple-negative breast cancer subtypes and preclinical models for selection of targeted therapies. J Clin Invest. 2011;21:2750–2767.
- Reis-Filho JS, Pusztai L. Gene expression profiling in breast cancer: classification, prognostication, and prediction. Lancet. 2011;378:1812–1823.
- Weigelt B, Horlings HM, Kreike B, et al. Refinement of breast cancer classification by molecular characterization of histological special types. J Pathol. 2008;16:141–150.
- Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med. 2004;351:2817–2826.
- Dowsett M, Sestak I, Lopez-Knowles E, et al. Comparison of PAM50 risk of recurrence score with oncotype DX and IHC4 for predicting risk of distant recurrence after endocrine therapy. Jco. 2013;1:2783–2790.
- Gnant M, Filipits M, Greil R, et al. Predicting distant recurrence in receptor-positive breast cancer patients with limited clinicopathological risk: using the PAM50 Risk of Recurrence score in 1478 postmenopausal patients of the ABCSG-8 trial treated with adjuvant endocrine therapy alone. Ann Oncol. 2014;25:339–345.
- Sestak I, Cuzick J, Dowsett M, et al. Prediction of late distant recurrence after 5 years of endocrine treatment: a combined analysis of patients from the Austrian breast and colorectal cancer study group 8 and arimidex, tamoxifen alone or in combination randomized trials using the PAM50 risk of recurrence score. J Clin Oncol. 2015;33:916–922.
- Sgroi DC, Sestak I, Cuzick J, et al. Prediction of late distant recurrence in patients with oestrogen-receptor-positive breast cancer: a prospective comparison of the breast-cancer index (BCI) assay, 21-gene recurrence score, and IHC4 in the TransATAC study population. Lancet Oncol. 2013;4:1067–1076.
- Buus R, Sestak I, Kronenwett R, et al. Comparison of EndoPredict and EPclin with Oncotype DX Recurrence Score for prediction of risk of distant recurrence after endocrine therapy. J Natl Cancer Inst. 2016;108:1–7.
- Gnant M, Sestak I, Filipits M, et al. Identifying clinically relevant prognostic subgroups of postmenopausal women with node-positive hormone receptor-positive early-stage breast cancer treated with endocrine therapy: a combined analysis of ABCSG-8 and ATAC using the PAM50 risk of recurrence score and intrinsic subtype. Ann Oncol. 2015;26:1685–1691.
- Dubsky P, Brase JC, Jakesz R, et al. The EndoPredict score provides prognostic information on late distant metastases in ER+/HER2-breast cancer patients. Br J Cancer. 2013;109:2959–2964.
- Cardoso F, van't Veer LJ, Bogaerts J, et al. 70-gene signature as an aid to treatment decisions in early-stage breast cancer. N Engl J Med. 2016;375:717–729.
- Laenkholm AV, Jensen MB, Eriksen JO, et al. The PAM50 Risk of Recurrence Score Predicts 10year distant recurrence in a comprehensive Danish cohort of 2558 postmenopausal women allocated to 5yr of endocrine therapy for hormone receptor positive early breast cancer. J Clin Oncol. 2018;57, In press.
- Ejlertsen B, Jensen MB, Mouridsen HT, et al. Excess mortality in postmenopausal high-risk women who only receive adjuvant endocrine therapy for estrogen receptor positive breast cancer. Acta Oncol. 2014;53:174–185.
- Christiansen P, Ejlertsen B, Jensen MB, et al. Danish Breast Cancer Cooperative Group. Clep. 2016;8:445–449.
- Talman ML, Rasmussen BB, Andersen J, et al. Estrogen Receptor analyses in the Danish Breast Cancer Cooperative Group. History, methods, prognosis and clinical implications. Acta Oncologica. 2008;47:789–794.
- Wolff AC, Hammond ME, Hicks DG, et al. Recommendations for human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists clinical practice guideline update. J Clin Oncol. 2013;31:3997–4013.
- Nielsen T, Wallden B, Schaper C, et al. Analytical validation of the PAM50-based Prosigna Breast Cancer Prognostic Gene Signature Assay and nCounter Analysis System using formalin-fixed paraffin-embedded breast tumor specimens. BMC Cancer. 2014;14:177.
- Wallden B, Storhoff J, Nielsen T, et al. Development and verification of the PAM50-based Prosigna breast cancer gene signature assay. BMC Med Genomics. 2015;8:54.
- Nielsen TO, Parker JS, Leung S, et al. A comparison of PAM50 intrinsic subtyping with immunohistochemistry and clinical prognostic factors in tamoxifen-treated estrogen receptor-positive breast cancer. Clin Cancer Res. 2010;16:5222–5232.
- NanoString Technologies S. Prosigna Package Insert. Seattle (WA): NanoString Technologies; 2013.
- Martinez SR, Beal SH, Canter RJ, et al. Medullary carcinoma of the breast: a population-based perspective. Med Oncol. 2011;28:738–744.
- Sestak I. Identifying biomarkers to select patients with early breast cancer suitable for extended adjuvant endocrine therapy. Breast Care (Basel). 2017;12:146–151.
- Stenkvist B, Bengtsson E, Eriksson O, et al. Histopathological systems of breast cancer classification: reproducibility and clinical significance. J Clin Pathol. 1983;36:392–398.
- Cortazar P, Zhang L, Untch M, et al. Pathological complete response and long-term clinical benefit in breast cancer: the CTNeoBC pooled analysis. Lancet. 2014;384:164–172.
- Heng YJ, Lester SC, Tse GM, et al. The molecular basis of breast cancer pathological phenotypes. J Pathol. 2017;241:375–391.
- Kennecke H, Yerushalmi R, Woods R, et al. Metastatic behavior of breast cancer subtypes. JCO. 2010;28:3271–3277.
- Sihto H, Lundin J, Lundin M, et al. Breast cancer biological subtypes and protein expression predict for the preferential distant metastasis sites: a nationwide cohort study. Breast Cancer Res. 2011;13:R87.
- Bomeisl PE, Thompson CL, Harris LN, et al. Comparison of Oncotype DX Recurrence Score by histologic types of breast carcinoma. Arch Pathol Lab Med. 2015;139:1546–1549.
- Turashvili G, Brogi E, Morrow M, et al. The 21-gene recurrence score in special histologic subtypes of breast cancer with favorable prognosis. Breast Cancer Res Treat. 2017;165:65–76.
- Sheffield BS, Kos Z, Asleh-Aburaya K, et al. Molecular subtype profiling of invasive breast cancers weakly positive for estrogen receptor. Breast Cancer Res Treat. 2016;155:483–490.