Community Clinical Practice Patterns and Mortality in Patients with Oncotype DX Intermediate Score: Who benefits from Chemotherapy?

Abstract

Background:

Oncotype DX recurrence score (RS) is used as a tool for making decision about chemotherapy in hormone receptor-positive, HER2-negative breast cancer. There is no benefit of chemotherapy in node negative, RS 11–25, age ≥50 patients but the benefit of chemotherapy in the node positive group remains unknown.

Methods:

Based on National Cancer Database between 2010 and 2014, a nationwide retrospective cohort study included 73,185 female breast cancer patients with stage 1–3A and RS 11–30.

Results:

Receipt of chemotherapy was associated with a reduced risk of death among patients with node positive breast cancer (hazard ratio [HR] 0.58, 95% CI 0.45–0.74; p<0.001), after adjusting for other prognostic factors in multivariable Cox model. The 5-year survival gain ranges from 1.3%, 3.3% to 6.7% for the RS 11–17, 18–25, and 26–30 subgroups respectively. Among patients with node negative disease, chemotherapy was associated with a reduced risk of death for patients with RS 25–30 (HR 0.68, 95% CI 0.48–0.96; p=0.03; 5-year survival gain 1.8%), but there was no benefit of chemotherapy for patients with RS 11–17 (HR 0.97, 95% CI 0.61–1.55; p=0.90) and marginally significant benefit for women with RS 18–25 (HR 0.79, 95% CI 0.62–1.00; p=0.05). Similar results were observed using propensity score matching method.

Conclusion:

The benefit of chemotherapy for breast cancer patients with intermediate RS is driven in a non-linear fashion by RS: the higher the RS, the larger the absolute benefit. Findings from this study underscore the utility of real world data to inform joint decision making in practice.

Precis

Using real world data from 73,185 female patients with breast cancer in the US National Cancer Database, we found a statistically significant protective effect of chemotherapy for patients with node positive disease and recurrence score 11–30, and patients with node negative disease and recurrence score 26–30. In patients with node negative disease and recurrence score 11–25, we found no benefit with the use of chemotherapy, which is consistent with the newly reported TAILORx study.

INTRODUCTION

Approximately 70% of newly diagnosed breast cancers are hormone receptor positive,¹ and the 10–15 year recurrence rate in these women following adjuvant chemotherapy is estimated to be 15–35%.^2–4 Based on increasing knowledge of tumor heterogeneity and biological mechanisms of resistance to systemic therapy, there is a need to develop and evaluate tools for personalized treatment of hormone receptor positive breast cancer to attain improved clinical outcomes with good quality of life.

The 21-gene recurrence score (RS) assay (Oncotype DX™) (ODX; Genomic Health, Redwood City, CA) was developed to guide treatment decisions based on prognostication of subsequent outcomes in clinical trials.⁵ ODX has been validated in node negative and positive patients as a prognostic tool using archival specimens from clinical trials with long-term follow up.^5–10 In practice, it is used by oncologists to counsel patients about the benefit of adding chemotherapy given the substantial benefit of hormonal therapy in hormone receptor-positive human epidermal growth factor receptor 2 (HER2)-negative breast cancer.¹¹ The ODX results are expressed as a computed recurrence score ranging from 0–100, with different cutoffs for low, intermediate or high scores based on specified endpoints in prospective clinical trials or as generated by Genomic Health.^{5, 12, 13}

In the TAILORx clinical trial, which was designed to prospectively validate RS in lymph node–negative breast cancer, the RS categories used were RS<11 as low risk, RS 11–25 as intermediate risk, and RS>25 as high risk.¹² Results from TAILORx demonstrated that patients with low RS (<11) who underwent hormonal therapy alone experienced excellent clinical outcomes: the 5-year and 9-year rates of freedom from distant recurrence were 98.8% and 95.0% respectively, while the overall survival rate at 5 years and 9 years were 98.0% and 93.7%, respectively.^{14, 15} Recently, the TAILORx trial reported that the use of hormone therapy was non-inferior to chemotherapy plus hormone therapy in intermediate RS 11–25 node negative disease and about 85% of women could be spared chemotherapy especially those older than 50 years of age or women younger than 50 with a RS lower than 16.¹⁵ Similarly, an initial report from the West German Study Group Plan B (WGS), an ongoing prospective phase III clinical trial that included node positive breast cancer with RS 0–11, showed an excellent 3-year disease-free survival in patients who received hormonal therapy alone.¹⁶ The RxPONDER trial is another ongoing prospective phase III study, which randomized node positive patients with RS of 25 or less to endocrine therapy alone versus chemotherapy followed by endocrine therapy, but outcome from this study is yet to be reported.¹³ ODX has been adopted by oncologists in clinical practice using cutoff points from the company: RS<18 classified as low risk, RS 18–30 as intermediate risk, and RS>30 as high risk.⁵ Therefore, questions remain unanswered to inform clinical decision for patients with node positive disease and node negative RS 26–30. A significant proportion of these patients may receive unnecessary adjuvant chemotherapy while others may be under treated by forgoing chemotherapy.

Using the largest dataset to date, we examined real world data from the National Cancer Database (NCDB) on treatment outcomes for patients diagnosed with breast cancer and ODX recurrence score in the intermediate range defined as RS 11–30. The study takes advantage of ODX results which were recently added as required data items in cancer registry operations nationwide in the U.S. and are suitably complete for valid interpretation.^{17, 18}

METHODS

Data Source and Study Population

The NCDB, a joint project of the Commission on Cancer of the American College of Surgeons and the American Cancer Society, is a nationwide, facility-based, oncology dataset that includes information on approximately 70% of all incident cancers diagnosed in the U.S. collected from more than 1400 hospitals participating in the American College of Surgeons Commission on Cancer.¹⁹ Routinely collected data items include patient demographics, tumor characteristics such as tumor site and histology, and American Joint Committee on Cancer (AJCC) stage.^{20, 21} This study used de-identified data and was deemed exempt from human protection oversight by the institutional review board of the University of Chicago. We identified 1,092,106 patients diagnosed with breast cancer from January 1, 2010 to December 31, 2014 in the NCDB. The study was limited to women 18 years or older diagnosed with stages I to IIIA, first primary, hormone receptor positive (estrogen receptor (ER) or progesterone receptor (PgR)), HER2 negative breast cancer (as per ASCO-CAP guidelines²²) who received all or part of their first course of treatment in the reporting facility. We included patients with recurrence score of 11–30. We further excluded patients with missing data for hormone receptor status, chemotherapy receipt, survival, number of positive lymph nodes, and ODX test results. After these selections, 73,185 patients were included in the final analysis (Supplemental Figure 1).

Study Variables

In the primary analysis, recurrence score was analyzed as a categorical variable at pre-specified cutoffs: 11–17, 18–25, and 26–30. This subdivision is to accommodate both the classification of intermediate RS in the ongoing clinical trials (17–25)^{12, 13, 23} and current clinical intermediate RS classification (18–30). Patients were further stratified by nodal status (positive vs. negative). Recurrence score was also analyzed as a continuous variable using restricted cubic spline (RCS) function with 3 knots at 12, 17, and 25. Spline functions are piecewise polynomials used in curve fitting, and possess smoothness at the places where the polynomial pieces connect (known as knots). A spline represents a less biased and efficient method of modeling the shape of continuous covariates.

Patient demographics, including age at diagnosis and race, were analyzed as categorical variables. Patient comorbidity status was represented by the adoption by Deyo et al of the Charlson Comorbidity Index (CCI),²⁴ which is a cumulative score of 15 health conditions. Tumor characteristics, including tumor size, lymph node involvement, grade, histologic type, and PgR status, were analyzed as categorical variables. The status of first course of chemotherapy and hormone therapy are recorded but neither the specific types of chemotherapy nor the type of hormonal therapy were available in the NCDB. Data on ovarian suppression were no recorded in the NCDB.

Statistical Analysis

Demographic and tumor characteristics were compared between patients receiving chemotherapy and those not receiving chemotherapy using chi-square tests. To evaluate the prognostic value of RS, we modeled the relationship between RS and overall survival using Cox proportional hazard models, stratified by nodal status. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated from Cox models. Linearity of prognostic effect of RS was tested by comparing the model in which RS was fitted with a simple linear function and the model in which RS was fitted with RCS functions.

We examined the predictive effect of recurrence score in four steps. First, we compared patients with chemotherapy (plus hormone therapy) and patients without chemotherapy (hormone therapy alone) separately in node positive and negative patients using multivariable Cox models adjusting for age, year of diagnosis, CCI, race/ethnicity, tumor size, histology, grade, PgR, lymphovascular invasion, and recurrence score. We calculated HR as the metric of relative risk. To facilitate clinical interpretation, we estimated marginal cumulative mortality risks and absolute risk differences from the fitted Cox models. The marginal mortality risk can be interpreted as the population average risk if all patients received, or did not receive chemotherapy, while keeping all other prognostic factors unchanged for each patient. Second, we further stratified the patients into pre-specified RS subgroups: 11–17, 18–25, and 26–30 because we hypothesized that the intermediate group is heterogeneous and the relative/absolute benefit of chemotherapy varies by RS subgroups. This is the primary analysis. Multivariable Cox models similar to step 1 were fitted. Third, we fit multivariable Cox models that included an interaction term of continuous RS and chemotherapy to explore the range of RS in which chemotherapy is effective. RS was modeled using aforementioned RCS functions. Last, as a sensitivity analysis, we conducted a propensity score matching. A propensity score was the log odds of receiving chemotherapy estimated from the logistic regression that included RS, number of positive nodes, interaction term of RS and node status, age, year of diagnosis, CCI, race/ethnicity, tumor size, histology, grade, PgR, and lymphovascular invasion. Before and after 1:1 matching, we checked the standardized differences between patients with and without chemotherapy to examine whether matching on the propensity score removed observed differences.²⁵ After the matching, we estimated survival rate using Kaplan-Meier method and calculated HR from Cox models in the 6 pre-specified subgroups (node status by RS category). A two-sided p-value <0.05 was considered statistically significant. Statistical analyses were conducted using STATA 14 (College Station, TX).

RESULTS

Distribution of Recurrence Score and Chemotherapy

Between 2010–2014, 125,556 women who were diagnosed with stage I-IIIA hormone receptor positive breast cancer received ODX as clinically indicated, and 106,439 women had exact recurrence scores. Of them, 73,185 patients (68.8%) had an intermediate ODX RS 11–30, 24,372 patients (22.9%) had RS 0–10, and 8,882 patients (8.3%) had RS 31–100. Figure 1 presents the distribution of RS and receipt of chemotherapy.

Figure 1.

Distribution of OncotypeDx recurrence score (RS) and receipt of chemotherapy

Table 1 shows demographic and clinical characteristics of the patients with an intermediate RS according to the three RS subgroups. 24% of patients were younger than 50 years, 82% of patients had node negative disease, 75% had tumor size equal to or less than 2 cm, and 24% received chemotherapy. Factors strongly associated with higher RS were higher tumor grade and progesterone receptor negativity. Otherwise, most clinical characteristics were fairly distributed across the subgroups. Patients with RS 26–30 (66.6%) were more likely to receive chemotherapy than patients with RS 18–25 (34.6%) and those with RS 11–17 (9.0%). Other strong indicators for receiving chemotherapy include presence of nodal disease, younger age, larger tumor, higher tumor grade, lymphovascular invasion, and clinical high risk defined by MINDACT modification of Adjuvant!Online²⁶ (Supplemental Table 1). There was a decreasing trend of chemotherapy use between 2010 and 2014.

Table 1.

Patient characteristics by OncotypeDX recurrence score category

	Oncotype DX RS						Chi-square*
	11–17		18–25		26–30
	No.	%	No.	%	No.	%
Age at diagnosis							176.5
18–39	1,154	3.0	1,039	3.7	300	4.2
40–44	2,940	7.7	2,041	7.3	502	7.1
45–49	5,355	14.1	3,451	12.3	754	10.6
50–54	5,980	15.7	4,183	14.9	1,011	14.2
55–59	5,821	15.3	4,657	16.6	1,123	15.8
60–64	6,324	16.6	4,698	16.8	1,321	18.6
65–69	5,532	14.5	4,192	14.9	1,079	15.2
70–74	3,156	8.3	2,391	8.5	628	8.8
75–79	1,370	3.6	1,059	3.8	292	4.1
80+	406	1.1	330	1.2	96	1.4
mean (SD)	57.5 (10.4)		57.8 (10.5)		58.2 (10.7)
Race/ethnicity							48.6
NH White	32,478	85.4	23,742	84.7	5,882	82.8
NH Black	2,607	6.9	2,109	7.5	628	8.8
Hispanic	1,580	4.2	1,170	4.2	290	4.1
Asian PI, Native American	1,373	3.6	1,020	3.6	306	4.3
Year of diagnosis							114
2010	5,095	13.4	4,362	15.6	1,131	15.9
2011	6,826	17.9	5,235	18.7	1,346	18.9
2012	7,897	20.8	5,382	19.2	1,364	19.2
2013	8,728	22.9	6,098	21.7	1,637	23.0
2014	9,492	25.0	6,964	24.8	1,628	22.9
Charlson comorbidity index							14.4
0	32,830	86.3	24,229	86.4	6,042	85.0
1	4,415	11.6	3,287	11.7	896	12.6
2+	793	2.1	525	1.9	168	2.4
Number of positive lymph node							27.5
0	30,946	81.4	23,146	82.5	5,930	83.5
1–3	6,882	18.1	4,752	16.9	1,135	16.0
4–9	210	0.6	143	0.5	41	0.6
Tumor size							141.8
≤ 1.0 cm	9,488	24.9	6,624	23.6	1,489	21.0
1.1–2.0 cm	19,450	51.1	14,205	50.7	3,522	49.6
2.1–5.0 cm	8,520	22.4	6,833	24.4	2,000	28.1
>5.0 cm	580	1.5	379	1.4	95	1.3
Progesterone receptor							2540.7
Negative	1,573	4.1	3,181	11.3	1,456	20.5
Positive	36,465	95.9	24,860	88.7	5,650	79.5
Histology							226.1
Ductal	28,414	74.7	21,305	76.0	5,872	82.6
Lobular	5,499	14.5	3,907	13.9	628	8.8
Ductal & Lobular	2,920	7.7	2,059	7.3	432	6.1
Other	1,205	3.2	770	2.7	174	2.4
Tumor grade							4195.1
1	12,324	32.4	6,770	24.1	821	11.6
2	20,737	54.5	15,523	55.4	3,552	50.0
3	3,162	8.3	4,447	15.9	2,424	34.1
Unknown	1,815	4.8	1,301	4.6	309	4.3
Lymph vascular invasion							250.4
No	29,374	77.2	21,124	75.3	5,004	70.4
Yes	4,583	12.0	3,883	13.8	1,338	18.8
Unknown	4,081	10.7	3,034	10.8	764	10.8
Clinical risk group†							991.1
Low risk	25,004	67.8	16,887	62.1	3,346	48.5
High risk	11,851	32.2	10,290	37.9	3,551	51.5
Chemotherapy							13341.1
No	34,615	91.0	18,341	65.4	2,371	33.4
Yes	3,423	9.0	9,700	34.6	4,735	66.6
Surgery							49.2
Lumpectomy	25,452	66.9	19,484	69.5	4,856	68.4
Mastectomy	12,575	33.1	8,552	30.5	2,245	31.6
Radiotherapy							32.9
No	11,976	31.5	8,245	29.4	2,184	30.8
Yes	26,018	68.5	19,762	70.6	4,910	69.2

Abbreviation: RS, recurrence score, SD, standard deviation; PI, pacific islanders; NH, non-Hispanic

^*Chi-square statistics were presented to indicate strength of statistical significance. All p-value <0.001 except for Charlson comorbidity index (p=0.006).

^†MINDACT modification of Adjuvant!Online assessment of clinical risk

Prognostic Value of Recurrence Score on Overall Survival

Over a median follow-up of 41 months, 1450 patients died. The recurrence score in the intermediate range was highly prognostic for both node positive and negative breast cancers if examining RS as a continuous variable (Figure 2). There was a monotonic increasing trend between RS and mortality risk, but the trend was not linear (test for linearity, p=0.03). Mortality risk in patient with node negative tumor increased by 1% per 1 unit increment of RS in the category RS 11–17 (p=0.61) and 6% per 1 unit increment of RS in the category RS 18–30 (p<0.001). In node positive patients, mortality risk increased 2% per 1 unit increment of RS in the category RS 11–17 (p=0.39) and 9% per 1 unit increment of RS in the category RS 18–30 (p<0.001). Nodal status was also a strong prognostic factor regardless of RS (Figure 2). Furthermore, we found that the prognostic value of RS was stronger in women 50 years or younger than in women older than 50 (Supplemental Figure 2). Other prognostic factors for overall mortality included age, race/ethnicity, number of positive lymph node, tumor size, histological grade, lymph vascular invasion, Charlson comorbidity index, as well as clinical risk group defined by the MINDACT trial (Supplemental Table 2).

Figure 2.

Prediction of overall survival by OncotypeDX recurrence score (RS). Solid and dashed lines show hazard ratios and 95% confidence intervals from Cox regression model that modeled RS using restricted cubic spline function with node-, RS=11 as the referent category (red line for node+, blue line for node-). Scatter circles (node-) or diamonds (node+) show hazard ratios of categorical RS relative to node-, RS=11 subgroup.

Predictive Effect of Recurrence Score on Chemotherapy Use

We next examined the effect of chemotherapy on mortality using multivariable Cox models within node positive or negative patients and within the six pre-specified strata defined by node status and RS subgroups (Table 2). After adjusting for multiple covariates, we found that receipt of chemotherapy was statistically significantly associated with better survival in patients with node positive breast cancer as a whole (HR 0.58, 95% CI 0.45–0.74; p<0.001), and this protective effect was significant in all the three RS subgroups though the relative reduction was slightly larger for the RS 26–30 subgroup. In patients with node negative disease, chemotherapy was only significantly associated with survival for the RS 26–30 subgroup (HR 0.68, 95% CI 0.48–0.96; p=0.03). There was no benefit of chemotherapy in patients with node negative RS 11–17 or RS 18–25, which is consistent with recent data from the TAILORx trial. In patients with node negative RS 11–25, there was no significant interaction between chemotherapy and age (≤50 vs. >50) for overall survival (p=0.59) (Supplemental Table 3). To better interpret the clinical benefit of chemotherapy, we calculated the multivariable-adjusted absolute risk difference in patients who received chemotherapy compared to patients who did not (Figure 3). The 5-year survival gain from chemotherapy varied substantially across subgroups: from no benefit for patients with node negative RS 11–17 (0.1%) to large benefit for patients with node positive RS 26–30 (6.7%).

Table 2.

Effectiveness of chemotherapy on mortality risk according to OncotypeDX recurrence score (RS): multivariable Cox models

	# of patient / # of death	5-year absolute risk*		Adjusted hazard ratio (95% CI)**	P
		No chemo	Chemo
In node- patients
All RS 11–30	60022/1074	3.2%	2.7%	0.85 (0.71–1.01)	0.066
RS 11–17	30946/474	2.6%	2.5%	0.97 (0.61–1.55)	0.90
RS 18–25	23146/432	3.6%	2.9%	0.79 (0.62–1.00)	0.052
RS 26–30	5930/172	6.0%	4.2%	0.68 (0.48–0.96)	0.029
In node+ patients
All RS 11–30	13163/376	5.7%	3.4%	0.58 (0.45–0.74)	<0.001
RS 11–17	7092/159	3.5%	2.3%	0.63 (0.40–0.99)	0.044
RS 18–25	4895/152	7.3%	4.0%	0.53 (0.37–0.76)	0.001
RS 26–30	1176/65	14.5%	7.8%	0.50 (0.28–0.89)	0.018

^*Marginal mortality risks were estimated from multivariable Cox models, if all patients receive a specific treatment modality, chemotherapy or no chemotherapy, while keeping the distribution of all other covariates the same.

^**Hazard ratio (95% confidence intervals) comparing chemotherapy versus no chemotherapy in multivariable Cox models adjusting for OncotypeDX recurrence score (continuous), age (categorical), year of diagnosis (categorical), Carlson comorbidity index (categorical), race/ethnicity, tumor size (categorical), histology, grade (categorical), progesterone receptor, and lymph vascular invasion; In the models for node positive group, number of positive lymph node (categorical) was also adjusted for.

Figure 3.

Effectiveness of chemotherapy plus hormonal therapy (C+H) versus hormonal therapy alone on all-cause mortality in patients with estrogen receptor positive, HER2 negative breast cancer, stratified by lymph node status and recurrence score (RS) category. Adjusted hazard ratios (HR) and absolute cumulative risk were calculated from multivariable Cox models.

To explore the effect of chemotherapy as a function of RS, we modeled RS as a continuous variable using a restricted cubic spline function. We found that in node negative patients, the benefit from chemotherapy was seen if RS was 23 or higher, whereas in node positive patients the benefit from chemotherapy was seen if RS was >13 (Figure 4). This analysis suggests that our pre-defined RS subgrouping is appropriate. Stratified by age, we did not find the effect of chemotherapy varied by age (Supplemental Figure 3).

Figure 4.

Effectiveness of chemotherapy on mortality risk according to continuous OncotypeDX RS: restricted cubic spline modelling of OncotypeDx RS in multivariable Cox regressions. The gray area indicates confidence zone of chemotherapy effect. RS, recurrence score; CI, confidence interval

As an alternative approach to remove confounding, we conducted propensity score matching analysis. A total of 13,735 patients who received chemotherapy were matched with 13,735 patients without chemotherapy. As shown in Supplemental Table 4, the two groups were quite balanced in demographic and tumor characteristics after the matching. We reassessed the effectiveness of chemotherapy using the matched cohort, and found that the effect estimates were similar to those observed in the multivariable models which used the full cohort although the confidence intervals were wider using the matched cohort (Supplemental Table 5, Supplemental Figure 4). The sample size in the node positive RS 26–30 subgroup was limited (26 events), so the effect did not reach statistical significance.

DISCUSSION

Over the last decade there has been an upward trend in the use of ODX, with about 50% of patients diagnosed with early stage hormone receptor positive breast cancer receiving ODX testing and more than a third of these patients reported to have intermediate ODX score 18–30.²⁷ Similarly, we found a third of patients had RS 18–30 and another third of patients had RS 11–17. As the new AJCC 8 cancer staging system incorporates prognostic/predictive gene markers like ODX,²⁸ a rapid rise in its clinical utility is expected. As such it becomes imperative that there is a more defined approach to addressing patients with intermediate ODX score. We observed an absolute benefit of chemotherapy that increased incrementally with rising RS, with larger absolute benefits for node-positive patients. These findings become relevant in clinical decision-making. Our study adds new knowledge addressing the question of the benefit of chemotherapy in node negative and node positive patients with intermediate recurrence score 11–30 using a large nationwide dataset.

The prognostic value of ODX was validated by the NSABP B14 trial which retrospectively showed that in node negative women higher RS was associated with an elevated 10-year risk of recurrence, where the whole range of RS (0–100) was analyzed,⁵ and the prognostic value of RS has been reproduced by other studies over the years.^{6, 29} Consistent with previous studies, our study demonstrated the higher the RS, the higher the risk of mortality for patients with intermediate RS. Mortality risk differed for patients with regards to the presence or absence of nodal disease and this finding was consistent with the retrospective analysis of the SWOG-8814 trial and the study of SEER registries, which showed that higher RS was associated with poorer overall survival and breast cancer-specific survival in node positive patients.^{7, 18}

The predictive value of ODX in node negative patients was established by the NSABP B-20 trial, which showed that the benefit of chemotherapy was greatest if RS was >30 with a HR of 0.26, while a non-significant effect was observed in the intermediate risk group RS 18–30 (HR 0.61, 95% CI 0.24–1.59) and no benefit was seen in the low risk group RS<18 (RR 1.31).⁹ The prospective TAILORx trial reported that hormonal therapy was non-inferior to hormonal therapy plus chemotherapy in patients with RS 11–25 (HR for invasive disease-free survival 0.96, 95% CI 0.81–1.06; HR for overall survival 1.01, 95% CI 0.82–1.27).¹⁵ We also found that chemotherapy has no significant protective effect on overall survival in patients with RS 11–25. In our study, the 5-year adjusted mortality risk was 3.0% for hormone therapy group and 2.6% in the hormone therapy plus chemotherapy group, which were slightly higher than reported in the TAILORx trial (2.0% and 1.9%, respectively).¹⁵ TAILORx investigators found an interaction between age and recurrence-score category for invasive disease–free survival, but not for overall survival. Our study used overall survival as the study endpoint and did not find differential effects by age, suggesting that overall survival cannot capture this interactive effect. In the TAILORx trial, all patients with RS 26–30 were assigned to chemoendocrine therapy and the 5-year mortality risk was 4.1%. Our study found that chemotherapy has a significant protective effect in patients with RS 26–30 and the 5-year adjusted mortality risk was 4.2% in the chemoendocrine therapy group.

The predictive value of ODX in patients with node positive intermediate RS was not clear. The retrospective analysis of breast tissue samples in the SWOG-8814 study, conducted in node positive patients, did not report a statistically significant benefit of chemotherapy in patients with RS 18–30 (HR 0.72, 95% CI 0.31–1.31), but the caveat being that the sample size was small.⁷ In contrast, we found a statistically significant protective effect of chemotherapy for patients with RS 11–30 (HR 0.58, 95% CI 0.45–0.74), and more importantly the absolute benefits varied according to recurrence score. Identification of significant effects in our study could be due to larger sample size (376 deaths), compared to SWOG-8814 study (42 events). Other possible reasons include improvements in supportive care during chemotherapy and differences in chemotherapy regimen. Patients in earlier studies used CMF (cyclophosphamide, methotrexate, 5-FU) regimen while therapies used in the contemporary cohort in our study are likely to be doxorubicin or taxane based regimens based on NCCN guidelines and the literature.³⁰ In advance of the publication of the RxPONDER trial in node positive patients,¹³ our study findings suggest that some node positive patients with intermediate RS derive benefit from chemotherapy or participation in clinical trials testing novel targeted treatments. Future studies could be conducted to investigate whether chemotherapy is beneficial for patients with node positive, RS 0–11 breast cancer.

There are several strengths to this study including large, nationwide sample with sufficiently complete data on ODX for the years of analysis and refined subgroup analysis. The study also has several limitations. First, observational studies are subject to confounding. We have used multivariable Cox models to control for known variables related to the receipt of chemotherapy, and conducted propensity score matching to remove potential confounding. However, unmeasured confounders may still exist. Because we found differential effect of chemotherapy across nodal status and RS subgroups, unmeasured confounders, if they exist, have to play a differential role across subgroups. Furthermore, our study findings in node negative patients are consistent with that of the just published randomized trial (TAILORx). Second, the NCDB collects data from Commission on Cancer (CoC) accredited hospitals, which are larger, more frequently located in urban locations, and have more cancer-related services than non-CoC hospitals.²⁰ Hence, our findings may be limited to CoC-approved hospitals, but clinical trials also tend to occur in CoC-approved hospitals and have similar limitation. Third, the follow-up of the patients was short so our study can only assess the short-term effects of chemotherapy and recurrence score but any benefit from chemotherapy is expected to show up early.^{31, 32} Although a previous study showed that recurrence score has better prediction of short-term survival than long-term survival,⁷ further follow-up of this national cohort is warranted. Fourth, exact chemotherapy regimens were not available in the NCDB, which makes the interpretation of the study findings somewhat difficult, although while doxorubicin or taxane based regimens are likely in this contemporary cohort.³⁰

As most of the benefit of chemotherapy occurs during the first few years after diagnosis,^{31, 32} the optimal adjuvant treatment regimen to reduce the risk of late recurrence in ER positive breast cancer is unknown. Nonetheless, this study demonstrated the benefits of chemotherapy according to RS and lymph node status, but the 5-year survival benefit of 1–7% need to be balanced with the known long term adverse effects of chemotherapy which has been reported to range from 2%-27%, especially in patients with other co-morbidities such as hypertension, diabetes, and obesity.^33–35 In the current cohort, 17,858 patients (24.4%) with intermediate RS actually received chemotherapy. If we consider a 5-year gain of 1% in absolute survival benefit to be clinically acceptable, then 9,191 patients (12.6%) with intermediate RS have been treated appropriately, 8,667 patients (11.8%) may have been over-treated, and 9,902 patients (13.5%) could have been under-treated.

In summary, this study demonstrated that careful analysis of real world data can provide valid results, which are often complementary to randomized clinical trials. This study provides additional data that can be used to inform clinical decision-making and underscores the need for further studies to optimize treatment in ER positive breast cancer.