A Novel Breast Cancer Index for Prediction of Distant Recurrence in HR⁺ Early-Stage Breast Cancer with One to Three Positive Nodes

Improvements in patient outcomes with prolonged endocrine therapy are supported by several studies of hormone receptor–positive (HR⁺) breast cancer patients (1–5). However, as extended adjuvant treatment is increasingly adopted, better methods to estimate risks and benefits are important to balance efficacy with potential toxicities and tolerability issues to ensure the best health outcomes for patients that may be receiving anti-estrogen therapy for up to 10 years. Lymph node–positive (LN⁺) disease is associated with approximately twice the risk of recurrence over 10 years compared with lymph node–negative (LN⁻) disease (6, 7), and thus the standard of care includes more intensive treatment regimens such as adjuvant chemotherapy and consideration of extended endocrine therapy for these patients. In findings from large randomized studies in LN⁺ patients, adjuvant chemotherapy and extended endocrine therapy have been shown to reduce the 10-year absolute risk of recurrence by approximately 9% and 3%, respectively (2, 8). However, a substantial subset (>60%) of LN⁺ patients remain disease free for 10 or more years when treated with 5 years of endocrine therapy only with or without adjuvant chemotherapy. Therefore, accurate assessment of individualized risk is an important goal, particularly with regard to LN⁺ patients, because nodal status is weighed heavily relative to other clinicopathologic parameters for guiding more aggressive treatment.

Use of molecular assays to assess prognosis and predict response to therapy in addition to traditional clinicopathologic characteristics are of great interest across many cancer types. The Breast Cancer Index (BCI) is a gene-expression signature developed from the combination of two biomarkers: the HOXB13:IL17BR expression ratio (H/I) and the Molecular Grade Index (MGI; ref. 9). H/I is a biomarker that is associated with response to endocrine therapy in ER⁺ breast cancer, and MGI consists of the average expression of five cell cycle–associated genes and provides quantitative and objective molecular assessment of tumor grade and proliferative status. In previous studies, BCI was demonstrated to significantly stratify patients with breast cancer with LN⁻ disease based on the risk of early (0–5 years), late (5–10 years), and overall (cumulative 0–10 years) distant recurrence (9, 10). A number of genomic biomarkers have demonstrated strong prognostic ability in LN⁻ patients; however, the prognostic value of gene expression alone is reportedly attenuated in LN⁺ patients (11–13). Notably, recent studies within LN⁺ cohorts have demonstrated that integration of clinicopathologic factors with molecular scores improved and strengthened prognostic performance compared with molecular scores alone (14, 15). The objective of this study was to evaluate an optimized model that integrated BCI gene expression with clinicopathologic factors (tumor size and grade) for the prognostic assessment of risk of distant recurrence specifically in patients with ER⁺ breast cancer with one to three positive lymph nodes (pN1).

A distinct recurrence risk model (BCIN⁺), as well as its prespecified cut-point, for patients with pN1 disease was previously developed using Cox proportional hazards regression integrating BCI, tumor size, and grade using ER⁺ pN1 patients as described previously (16). The performance of this BCIN⁺ model was assessed in a well-annotated retrospective case series from a single academic institution [Massachusetts General Hospital (MGH), Boston, MA] that was constructed on the basis of prespecified inclusion criteria and intended to validate the prognostic performance of BCIN⁺. The criteria included women with HR⁺, invasive pN1 breast cancer diagnosed between 1993 and 2007 who received adjuvant endocrine therapy with or without chemotherapy, had at least 5 years of follow-up, and had available FFPE tumor blocks. Corresponding clinical and pathologic information collected included age; tumor size; tumor grade; tumor histology; number of positive lymph nodes; ER, PR, and HER2 status; surgery type; chemotherapy treatment; endocrine therapy regimen and duration of endocrine therapy; and time to any local, regional, or distant disease recurrences. All women were followed up until distant recurrence or last documented follow-up visit. A hematoxylin and eosin (H&E) slide for each case was centrally reviewed to confirm tumor grade according to the Nottingham system. ER, PR, and HER2 statuses were centrally determined via IHC through pathology report review. ER and PR positivity was determined using a cut-off level of 10% nuclear staining of tumor cells (clinical practice standard at the time), whereas HER2 positivity was defined as IHC 3⁺ and/or FISH HER2/NEU gene amplified. The methods for lymph node assessment (sentinel lymph node biopsy alone vs. axillary lymph node dissection) were not collected. The investigation of the collected tumor samples was approved by the Institutional Review Board at MGH. In accordance with the approval, informed consent from patients was not required. The study was conducted in accordance with the U.S. Common Rule.

Gene expression analysis of FFPE specimens was performed at Biotheranostics, Inc., blinded to clinical outcome. For each case, three 10-μm tissue sections were cut, and an H&E slide was used to confirm 40% content of invasive cancer before manual macrodissection to enrich tumor content for RNA extraction. Total RNA was reverse transcribed and the resulting cDNA was preamplified using the PreAmp Master Mix Kit per the manufacturer's instructions (Applied Biosystems), followed by TaqMan RT-PCR as previously described (17). H/I, MGI, and BCI were calculated as described previously, blinded to all other variables (9, 17, 18). Precision/reproducibility of the BCI assay was completed under quality system regulation (QSR) as part of the clinical laboratory (CLIA) quality management process and has been reported previously (9). Reproducibility of FFPE sections within blocks was assessed by two serial sections within the same block from 10 patients. The mean within-block SD for the BCIN⁺ score was 0.06 BCI-unit [95% confidence interval (CI), 0.0–0.14].

The primary objective was to evaluate the prognostic performance of the BCIN⁺ model for overall and late (post 5 years) distant recurrence in an independent cohort of patients with pN1 disease. The primary endpoint was time to distant recurrence, defined as the time from diagnosis to the time of first metastasis at distant organs. Death before distant recurrence was considered a censoring event. Locoregional recurrences were not considered either as events or as censoring events. Secondary endpoints included time to any recurrences (locoregional or distant, whichever occurred first). Late distant recurrences were defined as distant recurrences occurring after 5 years from diagnosis and were evaluated within the subset of patients who had remained distant recurrence free for at least 5 years.

It was calculated that a minimum of approximately 226 patients was required to detect an absolute 20% difference in 10-year distant recurrence rate between the two BCIN⁺ risk groups with 80% power at 5% significance level, assuming 20% of patients would be classified as low-risk with 10-year risk of distant recurrence of 5%. The BCIN⁺ model (with scores reported ranging from 0.0 to 10.0), assay cut-points, and all analyses were prespecified and blinded to clinical outcomes. The change in likelihood ratio statistics (ΔLR-χ²) was used to represent the additional prognostic information of gene expression over clinicopathologic factors, and vice versa in a bar graph, with the remaining likelihood ratio statistic representing the overlap in prognostic information between the two. Distant recurrence–free survival for the two prespecified risk groups was evaluated using Kaplan-Meier survival analysis, and the equality of the survival curves was evaluated with the log-rank test. Multivariate Cox proportional hazard models were used to evaluate whether BCIN⁺ as a continuous risk index provided prognostic information independent of other relevant factors (i.e., age, PR status, chemotherapy, duration of endocrine therapy, and type of endocrine therapy) using likelihood ratio tests. Patients with missing data on the covariates were excluded from the multivariate analyses. Interquartile hazard ratios (HRs) for continuous BCIN⁺ corresponding to a change from the 25th to 75th percentile of the distribution, and corresponding 95% CIs, were provided. A receiver operating characteristic (ROC) curve was generated to assess the discriminative power of BCIN⁺ for distant recurrence using 15 years as the censoring time point; patients with <15 years of follow-up and no recurrence were excluded. Concordance index (c-index) was also estimated to assess the predictive ability of BCIN⁺ for distant recurrence–free survival. A two-sided P value of less than 0.05 was considered statistically significant. All analyses were performed using R statistical package (version 3.1.3, http://www.r-project.org).

Patient and tumor characteristics are summarized in Table 1. Of the 402 cases that met the inclusion/exclusion criteria and were included in the study cohort, 39% were <50 years old and 61% were ≥50 years old. The majority of cases (81%) were treated with adjuvant doxorubicin and cyclophosphamide (AC) or AC-T (AC followed by paclitaxel)–based chemotherapy. For adjuvant endocrine therapy, 191 (48%) patients received tamoxifen only, 69 (17%) received an aromatase inhibitor (AI) only, and 142 (35%) received a sequence of tamoxifen and an AI. A total of 276 patients were treated with up to 5 years of endocrine therapy, whereas 126 patients received more than 5 years of endocrine therapy. There was no relationship between tumor size and treatment duration, although a smaller proportion of patients that received more than 5 years of endocrine therapy were grade 3 (16%) compared with patients that received ≤5 years (30%). A majority of patients had ductal histology (86%), and most patients had one (58%) or two (26%) positive nodes. The cohort included 17% grade 1, 56% grade 2, and 26% grade 3, 62% ≤2 cm, and 38% >2 cm, and the proportion of distant recurrences that occurred early (<5 years) versus late (≥5 years) were similar. There were 87 (22% of patients) distant recurrences, 44% of which occurred more than 5 years after diagnosis (i.e., late recurrence). Patient characteristics in patients with early versus late recurrences are shown in Supplementary Table S1. Median follow-up was 12 years, with 78 patients (33% of event-free population) having ≥15 years of follow-up.

BCIN⁺ significantly stratified patients into low-risk and high-risk groups for both overall (P < 0.0001) and late (P = 0.0014) distant recurrence (Fig. 1A and B). The median follow-up was 12 years, and all follow up was censored at 15 years. For overall distant recurrence, BCIN⁺ classified 20% of patients (N = 81) into a low-risk group with a 15-year risk of distant recurrence of 1.3% (95% CI, 0.0%–3.7%), whereas 15-year risk for the BCIN⁺ high-risk group was 29.0% (23.2%–34.4%), giving an HR of 25.93 (95% CI, 3.61–186.22; Fig. 1A). Patients (n = 349) who remained distant recurrence free for at least 5 years were analyzed for late (post 5 year) recurrence. Of these, 23% were classified as BCIN⁺ low risk. Risk of distant recurrence in years 5 to 15 was 1.3% (95% CI, 0.0%–3.7%) versus 16.1% (95% CI, 10.6%–21.3%) in the low- and high-risk groups, respectively (Fig. 1B). Patients in the BCIN⁺ high-risk group had 12.39-fold (95% CI, 1.7–90.35) greater risk for late distant recurrence compared with the patients in the low-risk group. For the prespecified secondary endpoint of all recurrences, including locoregional recurrences, the BCIN⁺ low-risk group had a 15-year risk of 3.6% for both overall and late recurrence (Supplementary Fig. S1A and S1B).

The nonlinearity of BCIN⁺ was assessed by restricted cubic spline, quadratic, and cubic transformation using a likelihood ratio test, which showed that there was no nonlinear effect (P = 0.45, 0.37, and 0.53, respectively). As a continuous variable, predefined BCIN⁺ was a highly significant prognostic factor, with an interquartile HR of 2.14 (95% CI, 1.56–2.94; P < 0.0001) and 1.54 (95% CI, 1.13–2.09; P = 0.006) for overall and late distant recurrence, respectively. Multivariate analysis adjusting for age, PR status, chemotherapy treatment, duration of endocrine treatment (>5-year vs. ≤5-year), type of endocrine treatment (any AI versus tamoxifen) and number of positive lymph nodes showed that BCIN⁺ remained the most significant prognostic factor, with an HR of 1.81 (95% CI, 1.33–2.45; P < 0.0001) and 1.41 (95% CI, 1.06–1.89; P = 0.02) for overall and late distant recurrence, respectively. Duration of endocrine therapy, type of endocrine therapy, and number of positive nodes were also statistically significant in the multivariate analysis with BCIN⁺ included (Supplementary Table S2). The 10-year and 15-year risk of distant recurrence estimated as a linear variable from the Cox model increased monotonically as the BCIN⁺ scores increased for both overall and late distant recurrence (Fig. 2; Supplementary Fig. S2). ROC analyses of continuous BCIN⁺ in all patients showed AUCs of 0.76 for 15-year distant recurrences in both all patients and those treated with ≤5 years of endocrine therapy (Supplementary Fig. S3A and S3B), and AUCs of 0.82 and 0.9, respectively, for 15-year distant recurrences in the HER2-negative subset (Supplementary Fig. S3C and S3D). The c-index was 0.72 for both all patients and those treated with ≤5 years of endocrine therapy and was 0.78 and 0.80 for all HER2-negative patients and those HER2-negative but treated with ≤5 years of endocrine therapy, respectively.

The interaction between BCIN⁺ and whether patients received more than 5 years endocrine therapy or not did not show statistical significance (P = 0.25 and 0.90 for overall and late distant recurrence, respectively). In the subset of patients that received ≤5 years of endocrine therapy (n = 276), BCIN⁺ significantly separated the patients into low-risk and high-risk groups for both overall (P < 0.0001) and late (P = 0.0067) distant recurrence (Fig. 1C and D). For overall distant recurrence, BCIN⁺ classified 18% patients into a low-risk group with a 15-year risk of distant recurrence of 2.1% (95% CI, 0.0%–6.0%), whereas 15-year risk for the BCIN⁺ high-risk group was 36.8% (95% CI, 29.3%–43.4%) with an HR of 21.88 (95% CI, 3.04–157.32; Fig. 1C). Of the patients that were distant recurrence free for at least 5 years, 22% were classified as BCIN⁺ low risk. Risk of distant recurrence at year 15 was 2.1% (95% CI, 0.0%–6.0%) versus 19.0% (95% CI, 11.6%–25.8%) in the low- and high-risk groups, respectively (Fig. 1D). Patients in the BCIN⁺ high-risk group had 9.49-fold (95% CI, 1.29–69.7) increased risk for late distant recurrence as the patients in the low-risk group. The secondary endpoint of all recurrences is shown in Supplementary Fig. S1C and S1D.

Several prespecified subset analyses were performed to assess BCIN⁺ in clinically relevant subgroups. First, within the subset of 259 HER2-negative patients, 57 patients (22%) were classified as low risk by BCIN⁺, and this group had no distant recurrences over the 15 years of follow-up (Supplementary Fig. S4). HER2⁻ patients in the BCIN⁺ high-risk group had significantly higher risk of overall (0–15 years) and late (5–15 years) distant recurrence of 29.1% (95% CI, 21.6%–35.9%; P < 0.0001) and 17.0% (95% CI, 9.8%–23.7%; P = 0.0037), respectively. Prognostic performance of BCIN⁺ by the number of positive nodes was also examined (Supplementary Fig. S5): more than half of patients (n = 231; 57%) had 1 positive node, whereas 106 patients (26%) and 65 patients (16%) had two or three positive nodes, respectively. BCIN⁺ performed similarly across groups, with the low-risk group having 1.7%, 0%, and 0% 15-year risk of distant recurrence in patients with one, two, or three positive nodes, respectively. The risk of recurrence in the BCIN⁺ high-risk groups increased with the number of positive nodes (Supplementary Fig. S5).

BCI gene expression alone was significantly prognostic (ΔLR-χ² = 20.12; P < 0.0001; Fig. 3). The addition of tumor size (ΔLR-χ² = 13.29; P = 0.0003) and grade (ΔLR-χ² = 12.72; P = 0.0004) individually or combined (ΔLR-χ²=22.28; P < 0.0001) to BCI significantly improved the prognostic performance (Fig. 3). Conversely, BCI added significant prognostic information to tumor size alone (ΔLR-χ²=17.55; P < 0.0001; Fig. 3). However, addition of BCI to tumor grade did not reach statistical significance in this cohort (ΔLR-χ²=2.38; P = 0.1) due to a large overlap between their prognostic values (ΔLR-χ²=17.74), consistent with the overlap of grade and the MGI component of BCI which evaluates proliferative status.

The standard of care for patients with HR⁺ LN⁺ breast cancer includes more intensive approaches such as adjuvant chemotherapy and, more recently, strong consideration of extended endocrine therapy, based on higher overall risk of recurrence, with an approximately 25% risk at 5 years and approximately 40% risk at 10 years compared with approximately 12% and approximately 20%, respectively, in LN⁻ patients (6, 7). Notably, a majority of early-stage LN⁺ patients do not benefit from extended endocrine therapy. For example, in the MA.17 randomized trial, the absolute benefit of extended endocrine therapy in LN⁺ patients (distant disease–free survival) was approximately 3% with a median follow-up of 30 months (19). Similarly, in the ATLAS randomized trial, the absolute benefit of extended endocrine therapy in LN⁺ patients was approximately 3% (2). Recently disclosed primary data from the MA.17R and NSABP B-42 randomized trials—which contained 51% and 43% LN⁺ patients, respectively, and were the first studies to report safety and efficacy of AIs for up to 10 years—were consistent with earlier extended endocrine trials, demonstrating an absolute benefit of approximately 3% to 4% in terms of disease-free survival and approximately 1% to 2% in terms of distant recurrence–free survival (5, 20). These studies also established that longer durations of endocrine therapy were associated with significant increases in risk of some serious toxicities, including endometrial cancer with tamoxifen and new osteoporosis and fractures with AIs (2, 5, 21).

Results from these extended endocrine trials raise the unmet critical need to weigh risk versus benefit at the individual patient level in determining duration of endocrine therapy. Traditional methods for assessing risk of distant recurrence have used clinical and pathologic factors such as age, tumor size, tumor grade, and extent of nodal involvement; however, these measures can be limited as risk estimates are based on categorical averages. For example, in a recent Early Breast Cancer Trialists' Collaborative Group (EBCTCG) meta-analysis, pN1 patients with small tumors (T1) had a risk of late distant recurrence of 8%, 15%, and 23% between years 5 to 10, 5 to 15, and 5 to 20, respectively (22). Thus, identification of patients with a limited risk of late distant recurrence can be challenging based on clinicopathologic factors.

The current study demonstrated that BCIN⁺-derived risk groups showed distinct rates of distant recurrence based on more than 10 years of follow-up. In particular, the approximately 20% of patients who were classified as BCIN⁺ low risk had limited 15-year rates of distant recurrence of <2%. Notably, these patients have a similar risk of distant recurrence as node-negative patients identified as low risk by BCI gene expression alone (9, 10). Conversely, BCIN⁺ high risk defined patients at significantly higher risk of distant recurrence (29.0%) during 15 years of follow-up, despite aggressive treatment that included adjuvant chemotherapy as well as adjuvant endocrine therapy. These results are of clinical relevance and provide a basis on which to identify patients with one to three positive nodes that may not need extended durations of endocrine therapy. Moreover, BCIN⁺ identified pN1 patients at low risk of recurrence regardless of whether they had one, two, or three positive nodes (Supplementary Fig. S5). Identification of low-risk pN1 disease was also demonstrated in patients treated with endocrine monotherapy, albeit within the training cohort, indicating that the BCIN⁺ model may also have potential utility for decision making regarding adjuvant chemotherapy. Although chemotherapy is recommended for almost all women with four or more positive lymph nodes, further studies in pN1 patients treated with endocrine therapy alone are planned to assess the value of BCIN⁺ to help determine the need for adjuvant chemotherapy at the time of diagnosis.

Characterization of several genomic biomarkers has revealed that molecular signatures have the ability to provide a substantial amount of prognostic information when assessing node-negative patients, but that prognostic capability is substantially enhanced by the addition of clinicopathologic factors to molecular approaches in node-positive patients (11–13). Consistent with this, in previous work, BCI gene expression alone has been shown to provide greater prognostic value than clinicopathologic factors in LN⁻ patients (9, 23). Findings from the current study show that the integration of tumor size and grade statistically enhanced the prognostic performance of BCIN⁺ compared with BCI gene expression alone (Fig. 3). These findings are consistent with studies of other molecular biomarkers in patients with LN⁺ disease. The PAM50 risk of recurrence (ROR) score (including tumor size) and the EndoPredict EPClin score (including tumor size and number of positive lymph nodes) both incorporate clinical factors and molecular scores for risk assessment (11, 12). Similar to BCI in the study presented here, in a comprehensive comparative analysis across biomarkers, molecular information from PAM50 and EndoPredict alone captured substantial prognostic information in LN⁻ patients, whereas in patients with LN⁺ disease, prognostic value was incremental when added to clinicopathologic factors (11–13). As a result, integration of molecular information with clinicopathologic factors led to significantly greater overall prognostic performance in LN⁺ patients. Results of these studies support the notion that the most important factors underlying disease recurrence evolve as disease progresses from LN⁻ to LN⁺ disease, with anatomic and pathologic factors potentially playing a larger role than the underlying biology, and that optimal patient risk stratification of LN⁺ patients may require the involvement of both molecular and clinicopathologic factors.

Evaluation of the incremental contributions of gene expression versus tumor size and grade in this cohort showed a large overlap between BCI and tumor grade (ΔLR-χ²=17.74; P < 0.0001; Fig. 3). This overlap is likely due to the proliferation-based MGI component within BCI, as MGI was developed as molecular surrogate for tumor grade and is composed of five cell cycle–regulation genes (17). The implication of this finding is that tumor proliferative status is the primary determinant of prognosis in this model, however, when analyzed to evaluate what clinicopathologic factors add to BCI gene expression, BCI was a highly significant prognostic factor in the pN1 population, with size and grade adding significant prognostic value in addition to the molecular signature (P = 0.0003 and 0.0004, respectively). Given the variability of clinical and pathologic assessment in routine clinical practice, incorporation of molecular factors may help to standardize and refine risk assessment. This may be particularly relevant for molecular assessments of proliferative status, where marked variability has been documented in pathologic assessment (24–31). Further studies in other independent cohorts are required to adequately characterize the contribution of tumor grade to the prognostic performance of BCIN⁺.

The study had several key strengths and limitations. Validation of the BCIN⁺ model was completed in a clinically relevant cohort, including pre- and post-menopausal patients and patients treated with tamoxifen and/or AIs, most with adjuvant chemotherapy. A key limitation of the study was the use of a retrospective cohort for performance evaluation, though the study was prospectively defined and blinded. In addition, information on duration of endocrine therapy was collected in a categorical manner (≤5 years vs. >5 years), and therefore, further subset analyses based on specific duration of endocrine therapy could not be performed. Furthermore, the proportion of PR⁻ patients was under-represented in this cohort compared with reported estimates in other published studies (8, 32–34). Additional validation studies in prospective or multicenter cohorts will strengthen the evidence base.

In summary, findings from this study indicate that integration of clinicopathologic and molecular factors in BCIN⁺ has the potential to inform treatment decisions for a substantial proportion of HR⁺ pN1 patients—in particular, the 1 in 5 patients with low risk of late distant recurrence. On the basis of these results, BCIN⁺ may provide information to facilitate selection of LN⁺ patients for extended endocrine treatment. Women categorized as BCIN⁺ low risk appear to have been adequately treated by 5 years of adjuvant therapy and can avoid extended adjuvant endocrine therapy. Efforts to further refine risk assessment through the integration of clinicopathologic and molecular biomarkers should continue in an effort to optimize the risk-versus-benefit assessment and individualization of care for ER⁺ breast cancer patients.

D.C. Sgroi co-owns a patent on gene expression biomarkers for breast cancer with Massachusetts General Hospital and Biotheranostics;. MGH licensed the patent to Biotheranostics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Zhang, B.E. Schroeder, C.A. Schnabel, D.C. Sgroi

Development of methodology: Y. Zhang, B.E. Schroeder, P.-L. Jerevall, C.A. Schnabel, D.C. Sgroi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.A. Schnabel

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Zhang, B.E. Schroeder, C.A. Schnabel, D.C. Sgroi

Writing, review, and/or revision of the manuscript: Y. Zhang, B.E. Schroeder, P.-L. Jerevall, C.A. Schnabel, D.C. Sgroi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Zhang, A. Ly, H. Nolan, C.A. Schnabel

Study supervision: Y. Zhang, C.A. Schnabel

Ranelle Salunga, Jose Ramirez, and Tristan Harris provided technical support for this study. This work was supported in part by grants from the Breast Cancer Research Foundation, the NIH/NCI, and the Avon Foundation for Women.

This work was supported by the U.S. DOD Breast Cancer Award BC097711 (to D.C. Sgroi).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.