Practice patterns and outcomes for patients with node-negative hormone receptor-positive breast cancer and intermediate 21-gene Recurrence Scores

The recommendation for chemotherapy in early-stage breast cancer patients has been refined by the 21-gene Recurrence Score. However, uncertainty remains whether patients in the Intermediate Risk category benefit from chemotherapy. We analyzed female patients from the National Cancer Database from 2006 to 2012 who had pT1c-T2N0M0 breast cancer, were ER/PR-positive and HER2-negative, received endocrine therapy, and had a 21-gene Recurrence Score from 11 to 25. We performed univariate and multivariate logistic regression analyses to see what impacted chemotherapy receipt. We compared overall survival using Kaplan–Meier curves and the log-rank test. A multivariable Cox proportional hazards regression model was used to assess what variables impacted overall survival. Of 21,991 patients who met all inclusion and exclusion criteria, 4646 (21.1%) received chemotherapy and 17,345 (78.9%) did not. Chemotherapy was more often received by patients who were younger (adjusted odds ratios (aORs) compared to age < 40 years, 0.48 for 40s, 0.34 for 50s, 0.20 for 60s, 0.10 for 70s, and 0.07 for 80+), had private insurance vs Medicare (aOR = 1.37), were from metro vs urban counties (aOR = 1.15), and were treated in community cancer centers vs academic programs (aOR = 1.26), and those with tumors of higher grade (grade 2 vs 1, aOR = 1.72; grade 3 vs 1, aOR = 3.76), higher tumor stage (pT2 vs pT1c, aOR = 1.62), or presence of lymphovascular invasion (LVI) (aOR = 1.41). At a median follow-up of 46.4 months, there was no significant difference in overall survival between patients who received chemotherapy vs those who did not (5-year estimated overall survival, 97.4% vs 97.8%, p = 0.89). On multivariable analysis, worse overall survival was associated with Black race, treatment at a community program, Medicaid, high-grade tumors, pT2 vs pT1c, higher Charlson–Deyo score, and no radiotherapy. Utilization trends showed that chemotherapy receipt in these patients has been decreasing from 25.8% in 2010 to 18.4% in 2013 (p < 0.001). In these patients where the benefit of chemotherapy remains uncertain, current practices see chemotherapy more likely to be used in patients with younger age, higher pathologic T stage, higher grade tumors, and LVI. No apparent difference was seen in overall survival between those who received chemotherapy and those who did not.


Background
One of the pillars in cancer treatment, chemotherapy has played a significant role in improving the outcomes of breast cancer patients over the years. Traditionally, chemotherapy was offered to premenopausal women with early-stage node-negative breast cancer based primarily on tumor size (e.g., > 1 cm) and receptor status (e.g., triple-negative or HER2-positive) [1][2][3][4].
A paradigm shift in the recommendation for adjuvant chemotherapy was enabled by the advent of the Oncotype DX Recurrence Score (Genomic Health Inc., Redwood City, CA, USA), a 21-gene assay that was validated to predict the rate of distant metastases [5], the risk of breast cancer-related mortality [6], and the benefit of chemotherapy [7] in hormone receptorpositive, node-negative early-stage breast cancer patients treated with endocrine therapy. Use of this assay is now recommended by the American Society of Clinical Oncology [8] and the National Comprehensive Cancer Network guidelines [9]. The original score thresholds were chosen based on the results of the NSABP B-20 trial [5], and generated common guidelines in directing the decision to forego chemotherapy in patients with a Low Risk score (defined originally as 0-17) [10] and to recommend chemotherapy in patients with a High Risk score (originally defined as 31-100). Studies show that approximately one-third of treatment decisions were changed by the use of the Recurrence Score, and that it is a phenomenon with worldwide penetrance [11][12][13][14][15][16]. Overwhelmingly, the changes seen were in reducing the recommendation for chemotherapy in as many as half of patients [14,17].
Subsequently published reports confirmed the safety of omitting chemotherapy in those with a Low Risk score [18]. Initial results from the low-risk arm of the Trial Assigning Individualized Options for Treatment (TAILORx trial) (21-gene Recurrence Score 0-10) recently published with a median follow-up of 69 months show excellent rates of invasive disease-free survival (93. 8%) and overall survival (98.0%) [19]. However, less clarity remains for patients with an Intermediate Risk score, defined originally as a score from 18 to 30. In the analysis of the NSABP B-20 trial, the benefit of chemotherapy was clear in the High Risk group, with 10-year distant recurrence-free rates improved from 60% to 88% as well as superior overall survival; however, the addition of adjuvant chemotherapy failed to result in significantly better distant recurrence-free rates (91% vs 89%) or overall survival in the Intermediate Risk group [7].
The currently ongoing TAILORx trial is a prospective randomized trial studying whether patients with Oncotype DX Intermediate Risk scores benefit from chemotherapy [10,20]. Of note, the investigators shifted the thresholds defining the risk groups to be more conservative, with Intermediate Risk defined as a score of 11-25, to minimize the risk of undertreatment. In the NSABP B-20 trial, a Recurrence Score of 11 was associated with a distant recurrence risk of 10%, a commonly used threshold for recommending adjuvant chemotherapy [10].
Until the results of the TAILORx trial are reported, we rely on retrospective data to guide adjuvant chemotherapy recommendations in women with Oncotype DX scores of 11-25. A recent retrospective review from MD Anderson Cancer Center at a median follow-up of 58 months reported no additional benefit from adjuvant chemotherapy (invasive DFS, RFS, DRFS, and OS) among 894 patients with hormone receptor-positive, HER2-negative, lymph node-negative, early-stage breast cancer with an Oncotype DX Recurrence Score of 11-25 treated at their institution [21]. Another retrospective analysis of a prospectively designed registry showed excellent outcomes for this same population without chemotherapy [22].
We queried the National Cancer Database (NCDB) for practice patterns and survival outcomes in nodenegative T1c-T2 female breast cancer patients with hormone receptor-positive/HER2-negative tumors and Oncotype DX Recurrence Score of 11-25, with a focus on the use of adjuvant chemotherapy.

Data source
The NCDB is a joint project of the American College of Surgeons and the American Cancer Society that draws data from more than 1500 accredited cancer programs accounting for 70% of all newly diagnosed cancer cases in the USA (National Cancer Institute, Bethesda, MD, USA).

Patient selection
Patients were selected to mimic the criteria of the TAI-LORx trial. Women included in our analysis had pathologic stage T1c-T2N0M0 (AJCC 6th edition, 2004+) breast cancer, ER-positive or PR-positive and HER2negative receptor status, and an Oncotype DX score from 11 to 25, diagnosed between 2006 and 2012. Only women who received endocrine therapy were included. Exclusion criteria included patients who received neoadjuvant chemotherapy. In total, 21,991 cases were included in the analysis. These criteria are displayed in a CONSORT diagram in Fig. 1.

Definition of variables
Patients and treatment characteristics included facility type, age, ethnicity, insurance type, median income, education, geographic location, Charlson-Deyo comorbidity score, year of diagnosis, breast cancer laterality, grade, clinical and pathologic tumor size, clinical and pathologic stage, presence of lymphovascular invasion (LVI), type of surgery, surgical margin status, receipt of adjuvant radiotherapy and/or chemotherapy, and nodal irradiation.

Statistical analysis
Descriptive statistics (including mean, standard deviation, median, range, frequency, and percentage) were calculated to characterize the study cohort. Demographic, prognostic, and facility characteristics (all categorical variables) were compared between patients who received chemotherapy vs those who did not receive chemotherapy by the chi-square test. Univariate and multivariable logistic regression analyses were employed to determine factors independently associated with receipt of chemotherapy. Factors of interest included facility type, age, race/ethnicity, primary payer, urban/rural location, Charlson-Deyo comorbidity score, grade, AJCC pathologic T stage, LVI, type of surgery, surgical margin status, and radiotherapy status. To evaluate the impact of chemotherapy on all-cause mortality, we first plotted Kaplan-Meier curves and evaluated differences in overall survival between categories of interest with the log-rank test. We then constructed a multivariable Cox proportional hazards regression model for all-cause mortality, adjusting for the same demographic, prognostic, and facility characteristics already listed as these variables were all of interest a priori. Hazard ratios (HRs) and 95% confidence intervals (CIs) were reported. The proportional hazards assumption was examined using Schoenfeld residuals and it was not violated. We also separately analyzed those patients who were diagnosed between 2006 and 2010 to examine a subset with more meaningful length of follow-up. All statistical tests were two-sided, with statistical significance

Results
There were 21,991 patients from the NCDB who met our inclusion and exclusion criteria. From this group, 4646 (21.1%) received chemotherapy and 17,345 (78.9%) did not. Table 1 presents all variables analyzed and compares them based on chemotherapy use via univariate analysis.
As stated earlier, younger women received chemotherapy more often in this study when age groups were divided into decades. We also performed univariate analysis on chemotherapy use among women aged younger than 35, 35-50, and > 50 years. In these age groups, 53.2%, 31.3%, and 16.8% received chemotherapy, respectively (p < 0.001). These data are presented in Additional file 1: Table S1. Age thresholds of 35 and 50 years were selected to correspond to clinically relevant ages of young breast cancer patients and postmenopausal patients. Factors that did not significantly correlate with the receipt of chemotherapy on multivariable analysis included the Charlson-Deyo comorbidity score and margin status.
Kaplan-Meier overall survival curves were generated. At a median follow-up of 32.1 months, there was no significant difference in overall survival between patients who received chemotherapy vs those who did not (p = 0. 37, Fig. 2). At 5 years, the estimated overall survival was 97.4% (95% CI 95.3-98.5%) in the group that received chemotherapy vs 97.6% (95% CI 96.9-98.2%) in the group that did not receive chemotherapy (adjusted HR = 0.83, 95% CI 0.55-1.25). Of the patients included in this study, 4753 were diagnosed between 2006 and 2010. Kaplan-Meier overall survival curves were also generated for this subgroup (Additional file 2: Figure S1).
With a longer median follow-up of 46.4 months, there was also no significance difference in overall survival in this subset between patients who received chemotherapy vs those who did not (p = 0.89).
On multivariable analysis, the following variables were associated with worse overall survival (Table 3): treatment at a community program vs an academic program (adjusted HR = 2.44, p = 0.01), Black vs White race (adjusted HR = 1.99, p = 0.02), Medicaid vs private insurance (adjusted HR = 2.21, p = 0.05), poorly differentiated vs well differentiated tumor (adjusted HR = 2. 00, p = 0.02), pathologic T2 vs T1c (adjusted HR = 2.17, p < 0.001), Charlson-Deyo score of 1 (adjusted HR = 3. 00, p < 0.001) and score of 2 (adjusted HR = 6.74, p < 0. 001) vs score of 0, and no radiotherapy vs radiotherapy (adjusted HR = 2.67, p = 0.02). Conversely, the margin status, LVI, and type of surgery had no significant associations with survival outcome. Figure 3 and Additional file 3: Table S2 show chemotherapy utilization trends stratified by year. The absolute number of patients available from 2006 and 2007 is small (n = 5 and n = 23, respectively), and only starting in 2010 were a significant number of cases captured by the database. A decrease in chemotherapy utilization in this patient population was reported from 25.8% in 2010 to 18.4% in 2013 (p for trend < 0.001). Figure 4 and Additional file 4: Table S3 show chemotherapy utilization trends stratified by Recurrence Score, as well as the absolute incidence of patients with each score. In the specific population of breast cancer patients in this study, there appears to be a higher incidence of Recurrence Scores in the lower half of the range than the upper half. The rate of chemotherapy utilization increased steadily with increasing Recurrence Score, after a sharp increase in rate seen at a score above 17.

Discussion
Our findings indicate that among 21,991 hormone receptor-positive, HER2-negative, early-stage nodenegative breast cancer patients who received endocrine therapy and had a 21-gene Recurrence Score of 11-25, chemotherapy was more often given to patients (on multivariate analysis) of younger age, with private insurance, not treated at academic programs, from metro counties, and whose tumors had higher pathologic T stage, higher grade, or LVI.
It is not surprising that younger age, pathologic T stage, LVI, and higher grade correlated with increased chemotherapy use as all are recognized as signs of more aggressive cancers that confer a higher risk of recurrence. Therefore, these risk factors could have justified the choice for chemotherapy regardless of the Oncotype DX Recurrence Score.
Margin status was not correlated with chemotherapy use; this factor is generally considered to be more indicative of risk for local recurrence. Accordingly, in our patient population, positive margins were significantly correlated with adjuvant RT (79.8% receiving RT vs 69.9%). Two socioeconomic factors associated with chemotherapy use were type of insurance and type of facility. Patients with private insurance were more likely to receive chemotherapy than those with Medicare, but there was no difference compared to patients with Medicaid. Since we have evidence that age was significantly correlated with chemotherapy use, the most obvious explanation was the age distribution of patients with private insurance vs Medicare. Indeed, when we repeated multivariate analysis only with patients over the age of 65 years, there was no longer any statistical difference in chemotherapy use between any insurance groups.
Chemotherapy was given less often at academic programs compared to nonacademic programs. It is possible that doctors in the community may be more conservative, opting for more aggressive treatment. Alternatively, their patients may have presented with more advanced disease warranting a greater use of chemotherapy. In our study population, there was no difference in pathologic T stage between academic and nonacademic programs, but there may be a difference in other risk factors not captured in the database like multicentric disease. Furthermore, if nonacademic programs saw patients with worse disease, one might expect a greater rate of mastectomies, but this was not the case (data not shown).
In the multivariable analysis of overall survival in these patients, it was alarming to see significantly worse  survival in patients treated in the community, those with Medicaid, and those of Black race. These are potentially indicative of imbalances in access to care in our healthcare system. Our findings also indicate that chemotherapy use has been steadily decreasing in this patient population from 2010 to 2013, perhaps signifying a shift in thinking concerning the need for chemotherapy in this Intermediate Risk group. Additionally, our findings show a clear relationship between Oncotype DX Recurrence Score and chemotherapy use, with a sharp rise in chemotherapy use starting at a score of 18, the original threshold used to distinguish low-risk and intermediate-risk patients, and a linear increase in chemotherapy use from 18 to 25.
Our analysis demonstrates comparable outcomes in patients who received chemotherapy vs those who did not, at a median follow-up of 32 months. This lack of difference persisted when we only analyzed patients   Most importantly, well-known limitations pertain to all retrospective database studies. These types of studies are intrinsically limited in their ability to reveal causal relationships, especially regarding survival endpoints. Selection biases assigning patients to specific treatments are often impossible to identify, and these biases generate confounding variables when defining the causal effect of an intervention. Despite our efforts to take most variables into consideration, other doctors' or patients' related preferences may have exerted an effect not captured in the database. For example, physicians might have used their clinical judgment to preferentially give chemotherapy to patients at higher risk for recurrence, distant metastasis, and/or breast cancer-related mortality using risk factors not recorded in the NCDB. Consistent with this hypothesis, in this study patients who underwent mastectomy were more likely to also receive chemotherapy. It is possible that the treating physicians perceived these patients as higher risk, and consequently recommended more aggressive therapies that included mastectomy, nodal RT, and chemotherapy. One cannot exclude that the addition of chemotherapy in these cases may have mitigated the negative prognostic impact of these risks and in fact reduced the likelihood of recurrence among these recipients, resulting in comparable outcomes.
Only prospective studies randomly assigned to chemotherapy vs no chemotherapy can generate evidence that avoids these biases. Comparisons of results from prospective randomized trials with those from retrospective analysis asking the same question can result in widely dissimilar findings. For example, the effect of radiation therapy on the outcome of breast cancer patients from the Early Breast Cancer Trialists' Collaborative Group Meta-Analyses drastically differed from a report based on the SEER registries [23].
Two other limitations of our study include excluding patients who did not receive endocrine therapy, and  dissimilarities between our patient population and those being studied in the TAILORx trial. By removing patients who did not receive endocrine therapy, we may have skewed our results by excluding women with more comorbidities, poorer compliance, and so forth. While our study was designed to predominantly match the criteria of the TAILORx trial, deviations include no exclusion of patients with COPD, chronic liver disease, CVA, CHF or other heart disease, and chronic psychiatric conditions (although we attempted to partly compensate by including the Charlson-Deyo comorbidity score in our analysis); inability to assess life expectancy and therefore not excluding patients with < 10-year life expectancy; and not including high-risk pT1b or pT3 patients. Querying the NCDB for pT3N0M0 patients who also met all other inclusion/exclusion criteria used in this study yielded only 281 patients. It is unlikely that the inclusion or exclusion of this small subset of patients would have had an impact on the findings of this study. Another limitation is the lack of data in the NCDB regarding recurrence, pattern of recurrence, and cause of death. This limits the outcomes we are able to evaluate. Lastly, an important caveat is that the patients in this study may not be fully representative of patients in practice. In the NCDB from 2006 to 2013, 23.2% of hormone receptor-positive, HER2-negative, node-negative breast cancer patients had a documented Oncotype DX score, while 51.9% were listed as "Not Applicable: Information not collected for this case" and 8.1% were "Unknown" (data not shown). Those who were selected to undergo the test may have had higher risk disease. Furthermore, insurance may preferentially cover Oncotype DX testing for ER-positive tumors, while we included ER-negative, PR-positive tumors as well.
Our study documents the practice patterns of chemotherapy use in this select population of breast cancer patients, and highlights the importance of the Oncotype DX Recurrence Score in influencing treatment recommendations. Individualized medicine involves separating patients into specific subsets using scientific knowledge to identify meaningful parameters. The establishment of the Low, Intermediate, and High Risk groups by the Oncotype DX score is an important step in this direction. These types of tests provide more clarity for patients and healthcare providers, reduce overtreatment, and potentially reduce healthcare costs. The application of the Recurrence Score has been modeled to be an economically advantageous clinical tool, possibly saving over $1000 per patient [11,24]. Further partitioning of Low, Intermediate, and High Risk groups of patients will optimize therapy for each individual.
To that end, other genetic panels are also being used including the Mammaprint 70-gene recurrence assay (Agendia, Irvine, CA, USA) and the PAM-50 (also known as Prosigna Breast Cancer Prognostic Gene Signature Assay; NanoString Technologies, Seattle, WA, USA). The Mammaprint assay has shown promise in identifying early-stage breast cancer patients with zero to three positive lymph nodes who may not need chemotherapy [25]. Interestingly, one of the selling points of the Mammaprint assay is that it is binary in classifying patients as Low Risk or High Risk, eliminating the ambiguity of an Intermediate Risk result [26]. Preliminary results also show potential for Mammaprint results to substratify Oncotype DX Intermediate Risk patients [27]. The growing importance of these genetic panels is highlighted by their inclusion in the eighth edition of the AJCC staging guidelines [28].

Conclusions
The additional benefit of adding chemotherapy in earlystage hormone receptor-positive breast cancer patients Fig. 4 Incidence of individual 21-gene Recurrence Scores from 11 to 25, and comparative utilization of chemotherapy based on the score. Most patients in this range are in the lower half (score [11][12][13][14][15][16][17][18]. Chemotherapy use in patients sees a dramatic increase starting at a score of 18, and rises steadily for each point increase