Association between standard clinical and pathologic characteristics and the 21-gene recurrence score in breast cancer patients

A population-based study




The 21-gene recurrence score (RS) assay has been reported to accurately predict the risk of disease recurrence and chemotherapy benefit in women with estrogen receptor (ER)-positive, lymph node (LN)-negative breast cancer who are treated with tamoxifen. To the authors' knowledge, the association between the RS and clinicopathologic characteristics has been studied in randomized and case-control trials, but not in the general population.


The authors analyzed the correlation between clinicopathologic breast cancer characteristics and RS among 300 consecutive Israeli patients who were referred to undergo the test between October 2004 and October 2006.


Low, intermediate, and high RS were noted in 109 patients (36%), 134 patients (45%), and 57 patients (19%), respectively. The median age of the patients was 54 years and the median tumor size was 1.6 cm. High tumor grade, low progesterone receptor expression, infiltrating ductal histology, and high HER-2 expression were found to be associated with a high RS, whereas patient age, tumor size, ER expression, and lymph node micrometastasis were found to correlate poorly with the RS. The ability of any of these variables, either alone or in combination, to predict the RS was limited. Similarly, neither commonly used guidelines nor the Adjuvant! Online software were found to be able to predict the RS.


The results of the current study suggest that neither standard clinicopathologic features nor commonly used assessment tools can reliably predict the RS among referred breast cancer patients compared with a clinical trial population. These data also may indicate the need for additional studies regarding the role of the RS among certain subsets of breast cancer patients, including those with noninfiltrating ductal carcinoma histology and the presence of lymph node micrometastasis. Cancer 2008. © 2007 American Cancer Society.

Greater than 50% of women diagnosed with breast cancer have lymph node-negative, hormone receptor-positive disease.1 When treated with tamoxifen, the 5-year recurrence rate among these patients is <20%, and the addition of chemotherapy reduces the recurrence risk by only an additional 2% to 10%.2 The ability to predict which patients will benefit from the addition of chemotherapy is limited and the decision to administer chemotherapy is currently based on clinical and pathologic parameters, including patient age and tumor size, grade, and histology. Several clinical guidelines and Internet-based algorithms use these parameters to predict the risk of recurrence and estimate the relative benefit of chemotherapy. However, these tools tend to recommend chemotherapy for a large proportion of patients.3–6

Recent advancements in the understanding of the molecular basis of breast cancer have led to the development of several gene expression analyses, which aim to predict prognosis as well as response to chemotherapy among breast cancer patients.7–17 It is interesting to note that although using different analyses methods and gene sets, 4 different tests demonstrated significant agreement with regard to outcome prediction.18 The 21-gene recurrence score (RS), also known as Oncotype DX, has been tested in several clinical trials for its ability to predict prognosis and response to therapy among patients with early breast cancer.12 The study is performed by measuring the expression of 16 cancer-related genes and 5 housekeeping genes in paraffin-embedded tumor tissues and using a specific algorithm to analyze the expression data and generate the RS. The RS is currently used only in patients with lymph node-negative, hormone receptor- positive breast cancer and its ability to predict prognosis or response to chemotherapy has to our knowledge been tested in 4 independent studies to date.12, 19–21 In a study of 688 patients with lymph node-negative, estrogen receptor (ER)-positive breast cancer who were treated with tamoxifen in the National Surgical Adjuvant Breast and Bowel Project (NSABP) clinical trial B-14, the RS was found to be a strong and independent prognostic factor, either as a continuous function or when grouped into 3 categories of low (<18), intermediate (18–30), or high (≥31) risk.12 A study of patients enrolled in the NSABP-B20 trial identified the RS not only as a prognostic factor but also as a reliable predictor of chemotherapy benefit.19 The RS also was found to be associated with the risk of breast cancer death in a retrospective, population-based study of 770 breast cancer patients.20 It is interesting to note that a smaller study failed to identify any association between the RS and breast cancer recurrence.21 A recent economic analysis suggests that the use of the RS can save considerable costs.22

To our knowledge to date, the association between the RS and various clinical and pathologic characteristics has not been tested in the general population of patients who are referred to undergo the test. However, these patients may vary considerably from those recruited to the NSABP trials. For example, physicians may recommend the test only for those patients whom they consider to be of particularly low risk for disease recurrence.

In the current study, we analyzed the correlation between standard clinical and pathologic breast cancer characteristics and the RS in a cohort of 300 consecutive Israeli breast cancer patients. In addition, risk stratification of these patients according to the RS was compared with commonly used guidelines.


The study population included all Israeli, newly diagnosed, female breast cancer patients who were referred to the RS assay between October 2004 and October 2006. Two male breast cancer patients who underwent the test were excluded from the current analysis. The test was performed among patients with lymph node-negative, ER-positive disease. HER-2 overexpression and lymph node micrometastases (≤2 mm) were allowed. Clinical and pathologic data were collected at the time of referral. Pathologic examination was conducted at the pathology departments of the referring centers. The RS assay (Oncotype DX) was performed on paraffin-embedded tumor samples at Genomic Health Laboratories, as previously described.12

All pathologic reports were reviewed for the current analysis and data regarding tumor histology, tumor grade, hormone receptor status, and HER-2 expression were recorded. Data also were analyzed using 3 clinical tools aimed at assessing the need for chemotherapy: the National Comprehensive Cancer Network (NCCN) clinical practice guidelines,3 the 2005 St. Gallen recommendations,4 and the Adjuvant! Online website.5 The Adjuvant! Online analysis was calculated for the 10-year survival benefit from hormonal therapy and from cyclophosphamide, methotrexate, and 5-fluorouracil (CMF) combination-like protocols, which are currently recommended for patients with lymph node-negative disease.3

The study was approved by the Ethics Committee of the Chaim Sheba Medical Center.

The study variables were compared between the study groups using Student t tests for continuous variables and the Fisher exact test for categoric variables. The Pearson correlation coefficient (r) was used to determine the correlation between continuous variables. All significance tests were 2-tailed. All calculations and statistical tests were performed using the SAS software package (SAS Institute Inc, Cary, NC).


Between October 2004 and October 2006, 308 Israeli breast cancer patients were referred to undergo the RS assay by 70 different physicians from all major Israeli oncology clinics (16 institutions). The test was cancelled in 8 patients (2.6%) because of insufficient RNA, <5% invasive tumor in the sample, or no ER expression as determined by reverse transcriptase-polymerase chain reaction (RT-PCR). The vast majority of the patients (87%) underwent the test between November 2005 and October 2006. Based on breast cancer incidence data from the Israeli health ministry (available at URL:, this population was estimated to be 25% of all newly diagnosed breast cancer patients with hormone receptor-positive, lymph node-negative disease who were diagnosed at that time.

Based on the previously defined risk categories,12 the patients were divided into 3 risk groups. Low-risk (RS of <18), intermediate-risk (RS of 18–30), and high-risk (RS of ≥31) scores were noted in 109 patients (36%), 134 patients (45%), and 57 patients (19%), respectively. Similar age distribution was noted in all risk groups (Table 1) and no correlation was noted between age and the RS value (Fig. 1A). Although significantly larger tumors were observed in the high-risk group, the absolute difference between the groups was found to be of no clinical significance (1.8 cm in the high-risk group vs 1.6 cm in the remainder of the patients). Similarly, the correlation between tumor size and the RS value was found to be statistically significant (P < .0001) but weak (r = 0.2) (Fig. 1B).

Figure 1.

Correlation (Pearson correlation coefficient [r]) between recurrence score (RS) and (A) patient age and (B) tumor size.

Table 1. Distribution of Clinical and Pathologic Characteristics According to Recurrence Score Risk Category
CharacteristicsLow risk (N = 109)Intermediate risk (N = 134)High risk (N = 57)P*
  • SD indicates standard deviation; IDC, infiltrating ductal carcinoma; ILC, infiltrating lobular carcinoma; NA, not applicable; ER, estrogen receptor; PR, progesterone receptor; mic, micrometastasis.

  • *

    Significant differences between low and intermediate risk were noted only for PR intensity. All calculations were conducted without the unknowns.

  • Grade was evaluated for IDC histology only.

  • Up to 2 millimeters.

Age, y (mean ± SD)55 ± 954 ± 955 ± 10.6
Tumor size, cm (mean ± SD)1.6 ± 0.61.6 ± 0.71.8 ± 0.7.007
Histology (no., %)IDC77 (71)100 (74)47 (82).01
ILC21 (19)13 (10)1 (2)
Other4 (4)1 (1)2 (4)
Unknown7 (6)20 (15)7 (12)
Grade (no., %)111 (10)10 (8)1 (2)<.0001
256 (51)80 (60)23 (40)
312 (11)22 (16)30 (53)
NA30 (28)22 (16)3 (5)
ER intensity (no., %)19 (8)19 (14)5 (9).5
232 (29)39 (29)17 (30)
362 (57)62 (46)26 (45)
Unknown6 (6)14 (11)9 (16)
PR intensity (no., %)012 (11)33 (25)29 (50) 
 124 (22)27 (20)11 (19)<.0001
230 (28)27 (20)6 (11)
335 (32)31 (23)5 (9)
Unknown8 (7)16 (12)6 (11)
HER-2 overexpression (no., %)No97 (89)114 (85)38 (66) 
Yes2 (2)3 (2)9 (16).0001
Unknown10 (9)17 (13)10 (18)
Lymph nodes mic (no., %)No101 (93)119 (89)56 (98) 
Yes8 (7)11 (8)1 (2).2
Unknown0 (0)4 (3)0 (0)

Noninfiltrating ductal carcinoma (non-IDC) histologies, comprised mainly of infiltrating lobular carcinoma (ILC) histology, were found to be significantly less prevalent among the high-risk patients and were noted in only 6% of these patients, compared with 23% among patients in the low-risk category. Different distribution patterns were noted for the ER and progesterone receptor (PR) status. Although similar ER intensities were noted in all the study groups, negative PR staining was found to be significantly correlated with increased risk and was observed in 11%, 25%, and 51% of the low-risk, intermediate-risk, and high-risk groups, respectively (P < .0001).

We analyzed the ability of standard features to predict RS categorization, and noted a non-IDC histology in only 3 of the high-risk RS patients (6%) compared with 25 (23%) and 14 (11%), respectively, of the low-risk and intermediate-risk RS patients. Moreover, 95 of the 101 (94%) PR-positive and grade 1 or 2 samples were found to have a low-risk or intermediate-risk RS (based on the modified Bloom-Richardson grading criteria).

The RS assay is not indicated for patients with lymph node-positive disease and the test is usually not performed in patients with HER-2-positive tumors. However, 14 of the patients had HER-2 overexpression and 20 had lymph node micrometastasis. Although HER-2 was found to be significantly more common in the high-risk group compared with the low-risk and intermediate-risk groups (16% vs 2%; P = .0001), the prevalence of lymph node micrometastasis was found to be similar in all risk categories.

Three commonly used decision tools aimed at either predicting risk or estimating the need for chemotherapy are the NCCN and St. Gallen guidelines and the Adjuvant! Online software.4–6 We compared the recommendations or risk stratifications derived from these tools with the RS risk stratification (Table 2). No correlation was found between the NCCN guidelines and the RS risk categorization. According to the NCCN guidelines, chemotherapy should be either considered or given to all the high-risk RS patients, but also to >90% of the patients in the low-risk and intermediate-risk groups. According to the 2005 St. Gallen guidelines, all patients with hormone receptor-positive, lymph node-negative disease are classified as either low risk (no chemotherapy needed) or intermediate risk (chemotherapy should be considered). The vast majority of the high-risk RS patients (98%) were classified as being at average risk by the St. Gallen guidelines. However, although the stratification according to the St. Gallen guidelines significantly differed between the RS groups (P < .001), 66% of the low-risk and 82% of the intermediate-risk RS patients were classified as being at average risk. Comparison of the RS risk stratification with the benefit of hormonal therapy as predicted by the Adjuvant! Online program demonstrated a significant correlation between the 2 (Table 2) (Fig. 2A). However, the correlation was weak (r = 0.32) and the ability of the Adjuvant! Online software to predict the RS risk categorization appeared to be limited (Table 2). No significant correlation was identified between the RS risk stratification and the chemotherapy benefit predicted by Adjuvant! Online (Table 3) (Fig. 2B).

Figure 2.

Correlation (Pearson correlation coefficient [r]) between recurrence score (RS) and 10-year recurrence-free benefit as estimated by Adjuvant! Online for (A) hormonal therapy and (B) chemotherapy.

Table 2. Comparison of 3 Clinical Predictive Tools and the Recurrence Score Risk Category
Assessment ToolLow risk (No., %)Intermediate risk (No., %)High risk (No., %)P
  • NCCN indicates National Comprehensive Cancer Network.

  • *

    The NCCN categories are: adjuvant chemotherapy not needed, should be considered, and should be administered.

  • The St. Gallen categories are low risk (chemotherapy not needed) or intermediate risk (chemotherapy should be considered).

  • Adjuvant! Online was calculated for 10-year recurrence-free benefit of therapy; calculations were conducted for first-generation chemotherapy (the cyclophosphamide, methotrexate, and 5-fluorouracil (CMF) combination and equivalents).

NCCN guidelines (N = 286)*Chemotherapy not needed7 (7)5 (4)0 (0).17
Consider chemotherapy11 (11)21 (16)6 (11)
Administer chemotherapy85 (82)101 (80)50 (89)
St. Gallen guidelines (N = 296)Low risk37 (34)24 (18)1 (2)<.0001
Average risk72 (66)106 (82)56 (98)
Adjuvant! Online hormonal therapy (N = 296)<10-y benefit91 (83)97 (75)22 (39)<.001
>10-y benefit19 (17)33 (25)35 (61) 
Adjuvant! Online chemotherapy (N = 296)<5-y benefit40 (37)52 (40)24 (37).7
>5-y benefit69 (63)78 (60)33 (63) 


In the current study, we report on the correlation between standard clinicopathologic features and the 21-gene RS among 300 consecutive Israeli breast cancer patients who were referred to undergo the assay by their physicians. Because previous studies analyzed the RS in archived tissues,12, 19, 20 the current study findings may better reflect the real-life correlation between clinicopathologic features and the RS.

An important finding noted in the current study cohort is the different risk stratification of our patients compared with previously published studies. Although 36% of our patients were considered to be at low risk, 45% at intermediate risk, and 19% at high risk, the distribution was 54%, 21%, and 25%, respectively, in the chemotherapy validation study.19 This difference most likely stems from the tendency of physicians not to refer patients who, based on standard features, are considered to be at either low or high risk. Because the management of patients with an intermediate-risk RS is currently uncertain, the proper RS-based management of nearly half of the referred Israeli patients awaits the results of additional prospective clinical trials.

As previously reported, several standard pathologic features, including tumor grade and PR staining, appear to correlate with the RS. Surprisingly, no association was noted between the RS and patient age or ER intensity and only a modest association was noted between the RS and tumor size. These findings are different from those of previous reports.12, 19, 20 Although a lack of association between the RS and ER intensity may reflect variability in the staining procedures among different pathology laboratories, the lack of an association between the RS and both patient age and tumor size may reflect true differences between the populations. It remains to be elucidated whether these differences between the referred and the validation patient populations will translate into differences in the predictive ability of the RS or the interpretation of the test results.

To our knowledge, the current study is the first to report a possible association between 3 pathologic characteristics, namely HER-2 expression, histology, and lymph node micrometastases, and the RS. A strong association was noted between HER-2 expression and high RS. Although to our knowledge this finding has not been reported previously, it is not surprising because of the high weight given to HER-2 expression in the RS algorithm.12 Because only 14 patients with HER-2 expression were included in the current study cohort, the association between HER-2 expression and RS needs to be validated further. We noticed an association between non-IDC histology, mostly ILC, and a low RS. Despite having a prognosis that is similar to IDC, ILC tumors tend to have different biologic characteristics, including higher ER and PR expression, and lower rates of proliferation and HER-2 expression.23 Because the RS is based on these features, a lower score is expected to be found in ILC tumors compared with IDC tumors, regardless of prognosis or response to chemotherapy. Thus, our data suggest that the prognostic and predictive roles of the RS should be validated separately in ILC and IDC cases and, possibly, a different algorithm or risk stratification should be used for the categorization of patients with ILC. It is interesting to note that no association was noted between the RS and the presence of lymph node micrometastases. Thus, a high score was noted in only 1 of 20 patients with micrometastases. This finding obviates the need to further study the role of the RS in patients with lymph node micrometastases.

It has been suggested that standard pathologic features may be used to predict the RS. Indeed, in the Israeli population, tumors that were found to stain for PR and had a low or intermediate tumor grade were found to have a 95% chance of not being classified as high risk. However, the analysis is based on only 57 patients who were classified as high risk in this cohort and these results should be interpreted with caution. Moreover, none of the standard features were able to differentiate between low-risk and intermediate-risk categorization. Thus, the use of standard pathologic features as an alternative to the RS cannot be justified according to our data.

Both the NCCN guidelines and the St. Gallen recommendations could not accurately predict the RS. Moreover, the use of the RS could eliminate the use of chemotherapy in a large proportion of those patients who, according to these guidelines, would be assigned to receive such therapy. A significant association was identified between the RS and hormonal therapy benefit, as predicted by Adjuvant! Online. However, despite being validated in a population-based study,6 no association was found between chemotherapy benefit, as predicted by Adjuvant! Online, and the RS.

To our knowledge, the current study is the first to report on RS distribution and its association with standard clinicopathologic features in a cohort of consecutively referred breast cancer patients. The data indicate that none of the standard clinicopathologic features or currently used predictive tools can accurately predict the RS. The observed discrepancies between these commonly used prediction methods emphasize the importance of prospective evaluation and comparison before their acceptance as valid clinical tools. Our data also indicate differences between the referred patients and the population described in the validation studies. The meaning of these differences remains to be elucidated.


We thank Dr. Esther Shabtai from the Tel Aviv Medical Center for her assistance in data analysis. We also thank Dr. Lior Soussan-Gutman from Oncotesteva for her help with data collection