Association of 70-Gene Signature Assay Findings With Physicians’ Treatment Guidance for Patients With Early Breast Cancer Classified as Intermediate Risk by the 21-Gene Assay

Key Points

Question  Are 70-gene signature assay findings associated with physicians’ treatment decisions for patients with intermediate findings on the 21-gene assay?

Findings  In this study of 840 patients with early breast cancer, 282 physicians (33.6%) changed their treatment decision after receiving the 70-gene assay result. Adjuvant chemotherapy was added and removed so that 88% of high-risk patients were recommended to receive chemotherapy and 91% of low-risk patients were recommended no chemotherapy.

Meaning  The 70-gene signature assay provides clinically actionable information, and the findings may be associated with adjuvant therapy decisions.

Importance  Among patients who undergo the 21-gene assay (21-GA), 39% to 67% receive an intermediate risk result and may receive ambiguous treatment guidance. The 70-gene signature assay (70-GS) may be associated with physicians’ treatment decisions in this population with early breast cancer.

Objective  To determine whether 70-GS findings are associated with physicians’ decisions about adjuvant treatment and confidence in their recommendations and to evaluate the dichotomous (high- vs low-risk) and continuous distribution of 70-GS indices among this group of patients with intermediate risk.

Design, Setting, and Participants  The Prospective Study of MammaPrint in Breast Cancer Patients With an Intermediate Recurrence Score (PROMIS trial) was an impact study conducted from May 20, 2012, through December 31, 2015, that enrolled 840 patients with early-stage breast cancer and a 21-gene assay recurrence score of 18 to 30. Patients were treated in 58 US institutions.

Interventions  The 70-GS result was given to physicians before adjuvant treatment.

Main Outcomes and Measures  Change in physician treatment decision before vs after receiving the 70-GS result. With a treatment change of greater than 20%, the odds ratio (OR) was applied.

Results  Among the 840 patients who underwent 70-GS classification (mean age, 59 years; range, 27-93 years), 374 (44.5%) had a low-risk and 466 (55.5%) had a high-risk result. The distribution of 70-GS indices did not correlate with recurrence score within the 21-GA intermediate range, with 70-GS low- and high-risk patients observed at every recurrence score. A significant change in adjuvant treatment was associated with receiving the 70-GS classifications with an OR of 0.64 (95% CI, 0.50-0.82; McNemar test, P < .001) for all patients. Among the low-risk patients, 108 of 374 (28.9%) had chemotherapy removed from their treatment recommendation; among the high-risk patients, 171 of 466 (36.7%) had chemotherapy added. Results of the 70-GS were associated with the physician’s adjuvant treatment recommendation; 409 high-risk patients (87.8%) were recommended to receive adjuvant chemotherapy, and 339 low-risk patients (90.6%) were recommended no chemotherapy. Physicians reported having greater confidence in their treatment recommendation in 660 cases (78.6%) based on 70-GS results.

Conclusions and Relevance  The 70-GS provides clinically actionable information regarding patients classified as intermediate risk by the 21-GA and was associated with a change in treatment decision in 282 of these patients (33.6%). Chemotherapy was added or withheld by the treating physician based on the results of the 70-GS test. Physicians reported more confidence with their treatment recommendation after receiving 70-GS results.

The 70-gene signature assay (70-GS) is a US Food and Drug Administration–cleared molecular diagnostic assay that provides a binary classification for patients with early-stage breast cancer into low or high risk for breast cancer recurrence. The 70-GS was developed by identifying the genes most correlated with disease outcome in systemically untreated patients. The signature probes were selected in a data-driven and unbiased fashion from approximately 25 000 genes representing the entire human genome.¹ The 70-GS was evaluated in a randomized, prospective, phase 3 clinical trial that validated the clinical utility of the assay in patients with early breast cancer, regardless of estrogen receptor (ER) or human epidermal growth factor receptor 2 (HER2) status. The Microarray in Node-Negative and 1 to 3 Positive Lymph Node Disease May Avoid Chemotherapy (MINDACT) trial provided evidence that patients with a 70-GS low risk could safely forego chemotherapy (CT) without detrimentally affecting outcome.² The 70-GS identified 46% of clinically high-risk patients as genomically low risk, with 5-year distant metastasis–free survival of 94.7%, similar to 95.0% among patients classified as clinically low risk. Among patients with ER-positive, lymph node (LN)–negative, HER2-negative disease who were at clinically and 70-GS low risk, 98.4% had a distant metastasis–free interval at 5 years without CT.

The 21-gene assay (21-GA) was developed by selecting 16 genes associated with recurrence-free survival in patients with ER-positive, HER2-negative findings who received 5 years of tamoxifen therapy. These cancer-related genes (ER, PR, BCL2, SCUBE2, Ki67, STK15, Survivin, cyclin B1, MYBL2, MMP11, CTSL2, GFB7, HER2, GSTM1, CD68, and BAG1) and 5 reference genes were chosen from a panel of 250 candidate genes curated from published literature and genomic databases. The retrospective analysis of tamoxifen-treated patients from the National Surgical Adjuvant Breast and Bowel Project (NSABP) B-14 trial was used to validate the 21-GA. The trial showed significantly better 10-year distant recurrence–free survival among patients classified as low risk by the 21-GA (recurrence score [RS], <18) compared with patients classified as high risk (RS, >30), but the results for the group with intermediate risk (RS, 18-30) were inconclusive.³ Similar to the MINDACT trial, the Trial Assigning Individualized Options for Treatment (TAILORx) plans to report results validating the clinical utility of the 21-GA in patients with ER-positive, HER2-negative, LN-negative breast cancer.⁴ However, unlike the MINDACT trial, the low- and high-risk patients are not being randomized to receive CT. All patients with an RS less than 11 (TAILORx low risk) are forgoing CT, and all patients with an RS of at least 26 (TAILORx high risk) are receiving CT. Only patients with an RS of 11 to 25 (TAILORx intermediate risk) have been randomized to receive CT. Of note, the categorical risk classifications in TAILORx have lower RS cutoffs for low, intermediate, and high risk than the clinically validated ranges from the NSABP B-14 and B-20 trials.

The 21-GA gives an intermediate result (RS clinical range, 18-30; TAILORx trial range, 11-25) for 39% to 67% of patients undergoing testing,^4,5 leaving many physicians and patients without definitive treatment guidance. The frequency of intermediate RS results has been higher in practice than Paik et al³ anticipated when the assay was first developed and later validated.^4,5 The purpose of the Prospective Study of MammaPrint in Breast Cancer Patients With an Intermediate Recurrence Score (PROMIS trial) was to evaluate the clinical utility of the 70-GS in a US clinical practice setting, specifically in a patient population with an intermediate 21-GA result (RS, 18-30).

The PROMIS trial enrolled 876 patients with breast cancer and an intermediate 21-GA result from May 20, 2012, through December 31, 2015. Eighteen patients did not pass the 70-GS quality check. An additional 18 patients were removed from this analysis for the following reasons: 4 were ineligible for unknown reasons, 1 did not complete the informed consent form, 2 withdrew for unknown reasons, 3 started treatment before receiving the test result, 3 had a history of cancer treatment, 3 were duplicated in the enrolled population, and 2 had unresolved case report form queries. The 840 eligible patients had LN-negative or LN-positive (N1) disease, were negative for HER2, and were positive for hormone receptor. Enrollment was contingent on receiving a successful 70-GS (ie, MammaPrint) result. The industry-sponsored trial was approved by institutional review boards in all 58 participating US institutions and registered with clinicaltrials.gov (NCT01617954).⁶ Before enrollment, patients were required to provide written informed consent to participate and for research use of their tumor samples.

Originally, the trial was only open to patients with LN-negative disease; the protocol was amended in March 2013 to allow for the inclusion of patients with LN-positive disease. By this time, the validity of the 70-GS and 21-GA in patients with as many as 3 positive nodes had been validated in multiple studies.^7,8 The inclusion of patients with LN-positive disease in the PROMIS trial also coordinated with the concurrently running MINDACT (70-GS) and Rx for Positive Node, Endocrine Responsive Breast Cancer (RxPONDER; 21-GA) trials,⁹ both of which were enrolling patients with as many as 3 positive nodes. Adjuvant therapy was at the discretion of the treating physician. Physicians provided their recommended adjuvant treatment plan before the 70-GS results were available. After receiving the 70-GS results, the same treating physician provided his or her recommended treatment plan again.

The 70-GS was performed at the centralized Agendia laboratory in Irvine, California, according to standard protocols described previously.¹⁰ This test is based on microarray gene expression analysis of RNA extracted from formalin-fixed, paraffin-embedded breast tumor tissue and uses custom-designed array chips manufactured by Agilent Technologies. The 70-GS provides a numerical index ranging from −1.000 to 1.000 and a dichotomous categorization of low risk or high risk depending on whether the index is greater than 0 (low risk) or 0 or less (high risk). The 70-GS index is calculated from the correlation of a patient’s tumor profile to known low- and high-risk tumor profiles. Therefore, the more extreme the 70-GS index value, the more similar the tumor is to the hallmark low- or high-risk profiles. The statuses of ER, progesterone receptor (PR), HER2, and Ki67 were determined by the local institution using pretreatment samples according to American Society of Clinical Oncology and College of American Pathologists testing guidelines.^11-13

Available data from previous studies were used to calculate the target sample size needed to investigate the hypothesis of a 20% overall treatment change with sufficient power after revealing 70-GS results to the investigator (820 participants for a significance level of .05 and power of 0.90). The overall treatment change was defined as the percentage of patients with a change from their baseline treatment recommendation after receiving the 70-GS result. The frequency of CT recommendations for patients with intermediate 21-GA findings was calculated before and after receiving the 70-GS test results. The odds ratio (OR) for patients recommended not to receive CT after receiving the 70-GS result compared with those recommended to undergo CT before receiving the 70-GS result was determined, and the McNemar test was performed to assess the significance of the 2 proportions (mcnemar.exact function in the exact 2 × 2 library). The χ² test (binary variables), nonparametric Mann-Whitney test (for continuous variables [2 groups]), and nonparametric Kruskall-Wallis tests (continuous variables [>2 groups]) were used for the comparison of population characteristics in different subgroups. Linear regression was used to model the association between the 70-GS index and 21-GA RS. Multivariate logistic regression analysis was performed to determine whether clinicopathologic factors were associated with the change in treatment decision. All statistical tests were 2 tailed, and the statistical significance threshold was P < .05. Most calculations were performed with R software (version 3.2.2; R Project); multivariate analysis was calculated using SPSS software (version 22.0; IBM).

The 840 participants in the PROMIS study included predominantly postmenopausal women (mean age, 59 years; range, 27-93 years), most of whom were white (733 [87.3%]) (eTable 1 in the Supplement). The tumors included 636 T1 (75.7%), 198 T2 (23.6%), 744 LN negative (88.6%), 203 well differentiated (24.2%), 479 moderately differentiated (57.0%), 157 poorly differentiated (18.7%), and 665 invasive ductal carcinoma (79.2%). As expected, most of the women were ER positive (839 [99.9%]; 1 patient had unknown ER immunohistochemical status) and HER2 negative (821 [97.7%]; 19 [2.3%] had unknown or equivocal findings) by locally assessed immunohistochemical analysis.

Per protocol, all patients had an intermediate 21-GA RS from 18 to 30, with a median RS of 22 (interquartile range, 20-25). Patients with LN-positive disease were present at every RS in the intermediate range. Based on the 21-GA result (before 70-GS assessment), 382 patients (45.5%) were recommended to receive adjuvant CT and 458 (54.5%) were recommended not to receive CT as part of their adjuvant treatment regimen (Table 1). The recommendation to receive CT was more frequent among patients with an RS above 26 (68 [66.0%] CT and 35 [34.0%] no CT) and less frequent among those with an RS below 21 (96 [31.9%] CT and 205 [68.1%] no CT). In the middle of the intermediate range (RS, 21-26), which represents 436 patients (51.9% of the trial population), an equal number of patients (218 each) were recommended to receive and to forego CT (Figure 1).

The 70-GS reclassified the 840 intermediate patients as low risk in 374 cases (44.5%) and high risk in 466 cases (55.5%) (eTable 1 in the Supplement). These high-risk and low-risk patients were found at every RS in the intermediate range. The scatterplot comparing the 2 continuous variables (Figure 2) shows that the distribution of indices across the dynamic range of the 70-GS index was similar between patients with RS from 18 to 30. In a regression model, a comparison of the 70-GS index and 21-GA RS showed that the 2 test results were independent in the 21-GA intermediate range (adjusted R² = 0.14; P < .001), meaning that patients with a more extreme low-risk 70-GS index (closer to 1.0) did not also have a lower RS score (closer to 18) and vice versa. We found no statistical difference in distributions between 70-GS low-risk and high-risk patients by tumor staging (T1c, 163 [43.6%] vs 216 [46.4%]), nodal staging (N0, 272 [72.7%] vs 326 [70.0%]), age (mean of 58 [range, 27-87] years vs 59 [range, 2-93] years), ER status (374 [100%] vs 465 [99.8%] positive), PR status (318 [85.0%] vs 400 [85.8%] positive), or HER2 status (369 [98.7%] vs 452 [97.0%] negative) (χ² test, P > .05). In the intermediate group, we found a significant difference between 70-GS high- and low-risk patients in histologic grade (G1, 136 [36.4%] vs 67 [14.4%]; G2, 216 [57.8%] vs 263 [56.4%]; and G3, 21 [5.6%] vs 136 [29.2%]; P < .001), tumor type (invasive lobular cancer, 75 [20.1%] vs 47 [10.1%]; P < .001), menopausal status (premenopausal and perimenopausal status, 102 [27.3%] vs 87 [18.7%]; P = .005), and ethnicity (black, 18 [4.8%] vs 48 [10.3%]; P = .032).

The 70-GS classifications were associated with significant changes in CT treatment decisions to add or withhold CT (Table 1) (χ² test, P < .001). The OR for removing CT was significantly higher for 70-GS low-risk patients who were initially recommended CT (OR, 108.00; 95% CI, 18.98-4304.77; McNemar P < .001) and was significantly lower for 70-GS high-risk patients who were initially recommended CT (OR, 0.01; 95% CI, 0.001-0.04; McNemar P < .001). Among all patients, the odds of patients recommended not to receive CT after receiving the 70-GS compared with patients initially recommended to receive CT before receiving the 70-GS test was 0.64 (95% CI, 0.50-0.82; McNemar P < .001). Overall, 282 adjuvant treatment recommendations (33.6%) changed after receipt of the 70-GS result; 108 of 374 patients (28.9%) with a low-risk 70-GS result had CT removed from their treatment recommendation, and 171 of 466 patients (36.7%) with a high-risk 70-GS result had CT added to their treatment recommendation. One patient with a low-risk 70-GS result had CT added, and 2 patients with high-risk 70-GS results had CT removed from their treatment recommendations. In this ER-positive population, 794 of 840 patients (94.5%) were recommended endocrine therapy (HT) before and 803 (95.6%) after the 70-GS (Table 2). Thirty-one patients had a change in HT recommendation. For 9 patients for whom CT was removed, HT was added to the treatment recommendation. Among 11 patients for whom CT was added, 7 had HT removed and 4 had HT added. Among 11 patients with no change in CT recommendation, 4 had HT removed and 7 had HT added.

When analyzing the subgroup of 368 patients for whom the original physician-selected treatment recommendation (based on the intermediate 21-GA) conflicted with that indicated by the 70-GS, 279 (75.8%) of these patients had their treatment regimen changed; 108 of 142 70-GS low-risk patients who originally were recommended to receive CT had CT removed from their treatment recommendation, and 171 of 226 70-GS high-risk patients who originally were not recommended to receive CT had CT added to their treatment recommendation after the 70-GS result (Table 1). Overall, results of the 70-GS were associated with physician adjuvant treatment recommendations; 339 of 374 low-risk patients (90.6%) were recommended no adjuvant CT, and 409 of 466 high-risk patients (87.8%) were recommended adjuvant CT.

The change in treatment decision was the same regardless of LN status; 251 of 744 patients with LN-negative disease (33.7%) and 27 of 84 patients with LN-positive disease (32.1%) had a change to their CT treatment decision (eTable 2 in the Supplement). The decision to change treatment was significant in both groups (χ² test, P < .001 for LN-negative and P = .004 for LN-positive disease). Multivariable analysis showed that the 70-GS was the only independent variable significantly associated with the change in CT decision (hazard ratio, 1.85; 95% CI, 1.31-2.63) (Table 3). Nodal status, tumor stage, grade, histologic type, age, race, and menopausal or PR status were not associated with the change in CT decision.

After receiving the 70-GS result, physicians were queried in every case as to how the result influenced their confidence level regarding the chosen treatment plan. Physicians reported greater confidence in their treatment recommendations in 660 cases (78.6%), reduced confidence in 49 cases (5.8%), and no effect on confidence in 131 cases (15.6%) after receiving the 70-GS result. Although all of the PROMIS study sites had previously used the 21-GA, one-third of the participating sites were not 70-GS users before the trial and two-thirds were not regular or exclusive users of the 70-GS.

Molecular diagnostic tests supplement a physician’s toolbox by giving information on a disease beyond that provided by traditional clinicopathologic features. Genomic tests, such as the 70-GS (ie, MammaPrint), the 21-GA (ie, Oncotype DX [Genomic Health]), Mapquant Dx (IPSOGEN SA), the 76-gene signature (ie, Rotterdam signature [Veridex]), EPclin (ie, EndoPredict [Myriad Genetics]), and the 50-gene assay (ie, PAM50 or Prosigna [NanoString]), use gene expression profiling for prognostic purposes.¹⁴ Some of these assays then combine the gene expression profile with clinical factors in their algorithms to give a single risk appraisal. These tests are most useful when clinical assessments leave questions unanswered about a patient’s risk for disease recurrence and response to a proposed treatment.

Gage et al¹⁵ have argued that clinicopathologic factors may drive the 21-GA’s overall performance, citing the NSABP B-20 trial validation. They suggest that patients deemed to be at low and high risk by the 21-GA could have been identified from their extreme clinicopathologic characteristics alone and validated simple models using tumor grade, ER, PR, Ki67, and HER2 status that accurately estimate the 21-GA low- and high-risk classifications.^15,16 Furthermore, they argue that testing only a subset of patients with intermediary clinicopathologic features is cost-efficient, albeit in this group the 21-GA has shown the weakest predictive power. A recent impact study of 74 334 patients with breast cancer and a 21-GA result found that the positive predictive value for CT administration in patients with intermediate risk was only 40.1%.¹⁷ In the present trial, the initial decision to administer CT based on the 21-GA result was highly correlated with a high Ki67 expression of at least 14% (Yates χ² test, P = .01) (eTable 3 in the Supplement), indicating that physicians reverted to traditional clinicopathologic features to guide treatment decisions in patients with intermediate risk.

The PROMIS trial showed that physicians have greater confidence in their adjuvant treatment recommendation after receiving results of the 70-GS when the 21-GA yielded an intermediate RS, indicating a limitation of the latter assay in guiding adjuvant treatment recommendations. The MINDACT trial showed that patients classified at low risk by the 70-GS index who belonged to a subset with ER-positive, HER2-negative, and LN-negative disease (744 [88.6%] of PROMIS patients) had a 5-year distant metastasis–free interval of 97.8% with no adjuvant CT.² This risk is sufficiently low to forgo adjuvant CT owing to the negligible absolute benefit in this population. Patients classified as low risk by the 70-GS from the Microarray Prognostics in Breast Cancer (RASTER) trial had a distant recurrence–free interval of 97.0% at 5 years (15% received CT).¹⁸ In addition, in the prospective, randomized Stockholm Tamoxifen (STO-3) trial, 70-GS low-risk patients who received only 2 or 5 years of tamoxifen had a 10-year distant metastasis–free survival of 93% and a 20-year breast cancer–specific survival of 90%.^19,20 Alternately, in 70-GS high-risk patients, adjuvant CT could be considered to optimize their disease-specific outcomes. In the MINDACT study, 70-GS high-risk patients from the same subgroup had a 94.6% distant metastasis–free interval at 5 years with adjuvant CT, whereas untreated 70-GS high-risk patients in the earlier prognostic validation studies had a probability of distant recurrence of 71%.^21,22 Endocrine therapy alone was unlikely to account for the total risk reduction observed in this high-risk group.

Recurrence and survival data were not collected as part of this study because the primary objective was to evaluate the change in treatment decision based on the information provided by the 70-GS. The inclusion criteria for the PROMIS trial limited eligible patients to those with ER-positive, HER2-negative disease and the published intermediate-range RS of 18 to 30, not the TAILORx intermediate RS of 11 to 25. Of interest, the decision to change treatment based on the 70-GS was significant (McNemar P ≤ .001) regardless of whether the patient would have been considered at high or intermediate risk by the RS cutoff of at least 26 (eTable 4 in the Supplement).

The purpose of a prognostic test is to provide physicians and patients with definitive and actionable information in combination with other factors for accurate risk assessment. The PROMIS trial observed that treatment recommendations based on the 70-GS or 21-GA were often discordant and that 75.8% of these patients changed their treatment regimen. Overall, the additional information from the 70-GS results was associated with a change in the physician’s adjuvant therapy recommendation for 33.6% of all patients, and physicians followed the adjuvant therapy recommendation based on the 70-GS result for 748 patients (89.0%). The high proportion of cases in which physicians changed their treatment regimens by adding and removing CT, along with a self-reported increase in confidence, substantiates the influence of the 70-GS test when used for adjuvant decision making. Furthermore, obtaining additional molecular testing in cases with ambiguity may improve confidence in the adjuvant treatment plan. Accurately identifying patients who can safely forego CT prevents patients from undergoing a treatment that may minimally benefit them, sparing them from acute and long-term adverse effects, and would reduce the cost of health care for patients and payers.^23-25 Judicious use of genomic testing aids physicians in tailoring the episode of care to appropriate patient populations, both maximizing the value of and minimizing the risk for treatments. Although the recommendation to de-escalate therapy has proven to be one of the most difficult challenges in breast oncology, this study reveals that the 70-GS may provide physicians with clinically actionable information and improve their confidence in de-escalating therapy.

Corresponding Author: Hatem Soliman, MD, Moffitt Cancer Center, 12902 Magnolia Dr, Tampa, FL 33612 (hatem.soliman@moffitt.org).

Accepted for Publication: August 10, 2017.

Correction: This article was corrected on November 22, 2017, to add information about Open Access status to the acknowledgments.

Published Online: October 26, 2017. doi:10.1001/jamaoncol.2017.3470

Open Access: This article is published under the JN-OA license and is free to read on the day of publication.

Author Contributions: Drs Audeh and Soliman had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Whitworth, Untch, Blumencranz, Soliman.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Audeh, Mavromatis, Untch, Treece, Blumencranz, Soliman.

Critical revision of the manuscript for important intellectual content: Tsai, Lo, Audeh, Qamar, Budway, Levine, Whitworth, Zon, Oldham, Untch, Blumencranz, Soliman.

Statistical analysis: Mavromatis, Treece, Blumencranz.

Administrative, technical, or material support: Levine, Untch, Blumencranz, Soliman.

Study supervision: Lo, Audeh, Qamar, Oldham, Untch, Soliman.

Conflict of Interest Disclosures: Drs Audeh, Treece, and Blumencranz and Ms Untch report being employed by Agendia, Inc. Dr Whitworth reports serving on the Agendia speakers bureau and receiving research funding from Agendia at Intact Medical. Dr Mavromatis reports serving on the Celgene speakers bureau and advisory board. Dr Zon reports serving as a medical protective advisor, for which her research site receives funding. Dr. Soliman reports serving on the advisory board for Celgene, AstraZeneca, Eli Lilly, Pfizer, and Novartis. No other disclosures were reported.

Funding/Support: Each site of this study was supported by Agendia, Inc, for the conduct and management of this industry-sponsored study.

Role of the Funder/Sponsor: The sponsor collaborated with the principal investigator and other advising physicians to design the study. Conduct, collection, and management of the study was overseen by the sponsor. Analysis and interpretation of data and preparation, review, approval of the manuscript, and decision to submit were mutually agreed upon among the principal investigator, participating physicians, and the sponsor.

Additional Contributions: Christine Finn, BS, Agendia, Inc, provided clinical trial support. Christa Dreezen, MS, Agendia, NV, provided statistical support in SPSS for multivariate analysis. Both contributors performed this work as part of their employment. We thank all the women who participated in this study and the physicians, research nurses, study coordinators, and investigators who supported the study.