Selecting Patients for Oncotype DX Testing Using Standard Clinicopathologic Information

Original Study

Selecting Patients for Oncotype DX Testing Using Standard Clinicopathologic Information Susan J. Robertson,1 Greg R. Pond,2 John Hilton,3,4 Stephanie L. Petkiewicz,1 Yasmin Ayroud,1 Zuzana Kos,1 Denis H. Gravel,1 Carol Stober,3 Lisa Vandermeer,3 Angel Arnaout,3,5 Mark Clemons3,4,6 Abstract In this study, we evaluated whether the use of surrogate prognostic scores, such as the Magee equations, Gage algorithm, and University of Tennessee predictive algorithm, could be used to identify patients unlikely to benefit from additional Oncotype DX (ODX) testing. In this hypothesis generating study, retrospective data was collected from 302 patients with invasive ductal breast cancer with available ODX scores. Use of all formulae, and the Magee 3 equation in particular, is proposed as a screening tool prior to ODX testing. This strategy requires validation in other cohorts, but if effective, could significantly reduce the frequency of ODX testing with resulting benefits to both patients and the health care system. Introduction: Indiscriminate ordering of Oncotype DX (ODX) is expensive and of poor value to patients, physicians, and health care providers. The 3 Magee equations, Gage Algorithm, and University of Tennessee predictive algorithm all use standard clinicopathologic data to provide surrogate ODX scores. In this hypothesis-generating study, we evaluated whether these prognostic scores could be used to identify patients unlikely to benefit from additional ODX testing. Patients and Methods: Retrospective data was collected from 302 patients with invasive ductal breast cancer and available ODX scores. Additional data was available for: Magee equations 1 (212 patients), 2 (299 patients), 3 (212 patients), Gage Algorithm (299 patients), and University of Tennessee predictive algorithm (286 patients). ODX scores were banded according to the TAILORx results. Results: Correlation with ODX scores was between 0.7 and 0.8 (Gage), 0.8 and 0.9 (Magee 2, University of Tennessee predictive algorithm), and > 0.9 (Magee 1 and 3). Magee 3 was the most robust and is proposed as a screening tool: for patients aged
50 years, ODX testing would be not required if the Magee 3 score was < 14 or  20; for those aged > 50 years, ODX would not be required if the Magee 3 score was < 18 or  26. Using these cut-offs, 110 (51.9%) of 212 patients would be deemed as not requiring ODX testing, and 109 (99.1%) of110 patients would be appropriately managed. Conclusions: Use of all formulae, and the Magee 3 equation in particular, are proposed as possible screening tools for ODX testing, resulting in significantly reduced frequency of ODX testing. This requires validation in other populations. Clinical Breast Cancer, Vol. -, No. -, --- ª 2019 Elsevier Inc. All rights reserved. Keywords: Adjuvant, Algorithms, Breast cancer, Pathology, Recurrence score

Introduction Oncotype DX (ODX; Genomic Health, Redwood City, CA) is used to give both prognostic and predictive information in patients with lymph node-negative and N1mic, estrogen receptor-positive 1 Eastern Ontario Regional Laboratory, Department of Pathology and Laboratory Medicine, The Ottawa Hospital and the University of Ottawa, Ottawa, Ontario, Canada 2 Department of Oncology, McMaster University, Hamilton, Ontario, Canada 3 Cancer Research Group, Ottawa Hospital Research Institute, Ottawa, Ontario, Canada 4 Division of Medical Oncology, Department of Medicine, The Ottawa Hospital Cancer Centre and the University of Ottawa, Ottawa, Ontario, Canada 5 Division of Surgical Oncology, Department of Surgery, The Ottawa Hospital and the University of Ottawa, Ottawa, Ontario, Canada

1526-8209/$ - see frontmatter ª 2019 Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.clbc.2019.07.006

(ERþ), human epidermal growth factor receptor 2-negative (HER2) disease.1-3 Traditionally, the ODX recurrence score has been categorized as low (< 18, in 48.8% of patients), intermediate (18-30, in 39.0% of patients), and high risk ( 31, in 12.2% of 6 Clinical Epidemiology Program, The Ottawa Hospital Research Institute, Ottawa, Ontario, Canada

Submitted: May 6, 2019; Revised: Jul 8, 2019; Accepted: Jul 9, 2019 Address for correspondence: Mark Clemons, MD, Division of Medical Oncology, The Ottawa Hospital Cancer Centre, 501 Smyth Rd, Ottawa, Ontario, Canada E-mail contact: [email protected]

Clinical Breast Cancer Month 2019

-1

Oncotype DX Testing Using Standard Clinicopathologic Information patients).4 Although decision-making is clear in the low-risk (ie, those patients who do not benefit from adding adjuvant chemotherapy to endocrine therapy) and high-risk (ie, those patients most likely to benefit from chemotherapy) groups, it is the intermediate group where the greatest issue for patient management lies.3,5,6 The results of the TAILORx (Trial Assigning IndividuaLized Options for Treatment [Rx]) trial have somewhat defined management for the intermediate-risk group.7 In this trial, patients in the intermediate-risk group (recurrence score bands adjusted to 11-25 in the trial design) were randomized to endocrine therapy with or without chemotherapy. The results showed that women > 50 years of age were unlikely to derive benefit from adding adjuvant chemotherapy to endocrine therapy.3 Although for those aged
50 years with low or “low-intermediate” scores (score 11-19), there is likely little benefit from chemotherapy, patients with “high-intermediate” scores (score 20-25) may benefit from chemotherapy. Although ODX reduces the overall use of chemotherapy, its value to patients and the health care system will vary depending on the background use of chemotherapy.3,6,8-10 For example, in settings where there is more widespread use of chemotherapy, then the use of the Magee equations, Gage algorithm, and University of Tennessee predictive algorithm as screening tools for ODX use will have a greater impact than settings where chemotherapy use is more judicious. As a result, there is a need to add greater value for each ODX assessment. A number of groups have developed alternative means of assessing prognosis and the magnitude of chemotherapy benefit using readily available clinicopathologic information.11-18 Although all models have their own limitations, they all result in reduced use of chemotherapy.19 The Magee equations,1-3,20-24 Gage et al algorithm,25 and University of Tennessee predictive algorithm26 are being used to triage patients for those who are likely and those who are unlikely to benefit from additional ODX testing (Table 1). Potentially, the use of these models as surrogates for ODX testing may offer a more fiscally responsible strategy for selecting which patients should go on to have the more expensive multigene assays performed.1 In the current study, we explored the utility of the 3 Magee equations and the Gage and Tennessee scores in predicting which patients may or may not benefit from additional ODX testing in view of the of the TAILORx results. It is hypothesized that these models, which utilize readily available clinicopathologic information, could provide a reasonable triage system to select those cases in which there is unlikely to be additional benefit from further ODX testing.

Patients and Methods Patient Population

2

-

All cases of breast cancer where ODX scores are available are catalogued in the Department of Pathology at The Ottawa Hospital, a central testing laboratory serving 4 different regional hospitals in Eastern Ontario. Eligibility criteria included: histologically confirmed primary invasive ductal breast cancer, for which an ODX score was available, no prior chemotherapy, ERþ, HER2, lymph node-negative, or N1mi or isolated tumor cells. Cases that had ODX available that were excluded were: invasive lobular cancers, neoadjuvant treatment, macro-metastatic disease in lymph nodes, rare variants (eg, solid papillary and encapsulated papillary carcinoma), and male patients. In the event that a patient had more than 1 tumor present, clinicopathologic data was collected from the

Clinical Breast Cancer Month 2019

Table 1 Characteristics of Algorithms Used in Current Study Algorithm Name Magee 1

Magee 2

Magee 3

Gage et al (2015) University of Tennessee predictive algorithm

Required Clinicopathology Data -

Tumor size Grade ER, PR HER2 Ki-67 Tumor size Grade ER, PR HER2 ER, PR HER2 Ki-67 ER PR Grade Patient age Tumor size Grade LVI ER PR

Information From Algorithm

References

Surrogate ODX score

20,23

Surrogate ODX score

20,21,23

Surrogate ODX score

20,22,23

25

26

Abbreviations: ER ¼ estrogen receptor; HER2 ¼ human epidermal growth factor receptor 2; LVI ¼ lymphovascular invasion; ODX ¼ Oncotype DX; PR ¼ progesterone receptor.

tumor selected by the treating physician as the one most likely to affect decision making. Use of pathology reports used in the current study was approved by the Ottawa Health Science Network Research Ethics Board.

Standard Pathology Examination Tumor size, Scarff-Bloom-Richardson (SBR) grade,27 and lymphovascular invasion (LVI) status were extracted from the original pathology reports without central review. The representative blocks for submission to Genomic Health were, however, selected at the main Eastern Ontario Regional Laboratory (EORLA) to ensure adequate invasive carcinoma, relatively free from ductal carcinoma in situ, and with avoidance of biopsy site reactions and/or necrosis. Wherever possible, the same block that was used for biomarker and Ki-67 testing was sent for ODX assessment. ER and Progesterone Receptor (PR) Status. American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) guidelines for fixation parameters were met,28 and all cases were tested at EORLA and assessed using the Allred Score and percentage positive cells and average intensity. This reporting was done by 4 experienced breast pathologists (SR, DG, ZK, and YA). ER was tested using the 6F11 antibody (Leica/ncler-6f11) on the BOND platform at 1/150 dilution using HIER with Epitope Retrieval Solution 1 for 20 minutes. PR was tested using the ready-to-use Leica antibody PA0312 on the BOND platform using HIER with Epitope Retrieval Solution 2 for 30 minutes. Surrogate H Scores. A limitation for the use of the Magee equations in broader clinical practice is their requirement for classic H scores.23 In this study, standard pathology practice guidelines were followed; since 2010, ASCO/CAP guidelines require a semi-quantitative score

Susan J. Robertson et al

Ki-67 Proliferation Scores. From September 2015 to May 2017, routine Ki-67 testing was performed on all new ERþ, lymph nodenegative invasive breast cancers at the time of excision. After these dates, testing was performed if requested by the treating oncologist.29 Ki-67 testing was done using the MIB-1 antibody (DAKO/ M7240) tested on the BOND platform at 1/75 dilution using HIER with Epitope Retrieval Solution 2 for 20 minutes. Assessment was done as previously described,29,30 using computer image assisted analysis Leica LASV3.8) and counting 1000 cells with the criteria set at any nuclear brown staining. The assessment is a HOT SPOT assessment rather than an average, a method similar to what is done for mitotic counts in standard SBR scoring. This methodology, when used in Magee score calculations, has been previously validated and is correlated with ODX scores.30

score and fluorescence in situ hybridization (HER2:[chromosome enumeration probe 17] CEP17 ratio), Ki-67 index (if performed), tumor size, SBR score, and presence of LVI (if recorded). Spearman rank correlation coefficients were calculated to measure the level of association between each score of interest and the ODX result. C-statistics were used as a measure of discrimination for determining whether a patient would benefit from chemotherapy based on the TAILORx categorizations. A c-statistic of 0.50 is no better than flipping a coin, whereas 1.00 is perfect ability. Univariate c-statistics were calculated for each covariate after accounting for patient age. Predictive scores that gave a c-statistic of 0.90 or higher were evaluated further to evaluate concordance by determining the positive and negative predictive ability based on selected cut points. The final recommendations from the TAILORx trial were considered the gold standard; for patients aged > 50 years with low or intermediate scores (ie, < 26) derived no benefit from adding chemotherapy to endocrine therapy, whereas for those with high scores (ie, > 30), there was a definite need for chemotherapy.3 Although for those aged
50 years with low scores or “low-intermediate” (scores 11-19), there is likely little benefit from chemotherapy, patients with “high-intermediate” (score 20-25) and high (> 25) scores may benefit from chemotherapy.

Equations and Risk Scores

Results

that includes both the percentage positivity and a measure of intensity, such as the Allred score or H score.28 As manual H scores are labor-intensive and time-consuming, several groups have used the Allred score28 to provide a surrogate H score.24 This involves multiplying the average reported intensity (1 ¼ weak; 2 ¼ moderate; 3 ¼ strong) by the midpoint of the reported Allred positive score range (or absolute percentage if that was between 1% and 10%).29

All of the equations used to generate risk scores are presented in Table 1. Magee Equations. Magee equations were calculated by inserting tumor size, grade, and Allred scores as recorded in the clinical records, surrogate H scores as calculated from EORLA reports, and Ki-67 counts (where available) into the online tool.23 Gage Criteria. The Gage model uses 2 rules to classify patients into groups likely to either benefit or not from adjuvant chemotherapy (Table 1). Using grade as recorded in the original pathology report (without review) and ER and PR status from EORLA, each case was categorized as either: LOW (low grade and PRþ [> 1%]) or HIGH (high grade or low ER [< 20%]).25 University of Tennessee. University of Tennessee probability scores were calculated using the online tool, which incorporated patient age, tumor size, grade, and LVI status as recorded in the original report and ER/PR status (positive or negative) as designated by ASCO/CAP cut-offs.26

Objective of the Study The primary objective was to evaluate whether the use of the 3 Magee, Gage, or Tennessee equations could be used to screen for patients unlikely to benefit from additional ODX testing using the risk categories established by the TAILORx trial. The secondary objective was to evaluate how Magee equations, when performed in a service laboratory, correlate with ODX.

Statistical Analysis Clinicopathologic data was collected for ER and PR (staining percentage, intensity score, Allred score), HER2 immunohistochemistry

From January 2010 to 2017, ODX results were available for 299 patients.29-31 The 3 Magee equations, Gage model, and Tennessee nomogram were applied to all cases that had all the necessary information (Table 2).

Magee Equations Ki-67 is required for assessment of Magee 1 and 3 equations and was available for 212 patients, and the Ki-67-independent Magee equation 2 was available for all 299 patients.

Magee 1 and 3 Equations Spearman correlation with ODX and Magee 1 and 3 identified a strong association (ie, rho > |0.60| with a c-statistic > 0.90) (Table 3). Given that the Magee equations had a higher observed correlation coefficient and higher c-statistic compared with the other models, the ability of the Magee 3 (as this had the highest correlation) equations to act as a screening tool for ODX testing was assessed. Results for selected cut-offs are presented in Table 4. For Magee 3 (Table 4A), one could define cut-offs such that patients with low Magee scores (< 14 if < 50 years old and < 18 if  50 years old) would not require ODX testing and would be considered as not benefitting from chemotherapy. Patients with high Magee scores ( 20 if < 50 years and  26 if  50 years) would not require ODX testing, however, as they would be offered adjuvant chemotherapy as they would be considered to be high risk. Based on these cut-offs, of 8 patients < 50 years old who had a Magee 3 score < 14 and 75 patients who were  50 years old with a Magee S3 score < 18, none of them had an ODX test score which would have indicated a benefit for adjuvant chemotherapy. Similarly, there were 11 patients < 50 years old with a Magee score  20, and all 11 would be indicated to benefit from adjuvant chemotherapy, whereas 15 (93.8%) of 16 of patients  50 years and

Clinical Breast Cancer Month 2019

-3

Oncotype DX Testing Using Standard Clinicopathologic Information Table 2 Patient Characteristics Characteristic

Table 3 Spearman Correlation Coefficients With Oncotype DX

n

Statistic

Results

Age, y

299

Tumor size SBR of 9 Ki-67 H Score ER H Score PR Magee 1 Magee 2 Magee 3 University of Tennessee tail probability University of Tennessee Tail low University of Tennessee reg high University of Tennessee reg low LVI Markers done on Gage high grade or ER < 20 Gage low grade and PRþ Gage bin

299 299 211 299 299 212 299 212 286

Mean (SD) Median (range) Median (range) Median (range) Median (range) Median (range) Median (range) Median (range) Median (range) Median (range) Median (range)

57.0 (9.3) 57 (30-79) 1.8 (0.5-11.7) 7 (3-9) 7.9 (0.01-96.7) 240 (15-285) 240 (0-285) 18.6 (9.7-43.6) 20.2 (8.9-39.4) 18.1 (3.3-42.2) 33.5 (0-99)

286

Median (range)

66 (0-94)

286

Median (range)

8 (0-97)

Magee 1 bin

Magee 1 Tailor bin

No Ki-67 S2 Magee bin

No Ki-67 Magee S2 Tailor bin

-

Clinicopathologic data Patient age

0.002

Tumor size

0.487

0.533

0.01

SBR

0.769

0.788

0.38

LVI

0.627

0.651

0.18

Surrogate H Score ER

0.652

0.681

0.28

Surrogate H Score PR

0.741

0.763

0.53

Magee Score 1

0.901

0.936

0.74

Magee Score 2

0.863

0.883

0.65

Magee Score 3

0.910

0.944

0.74

Magee 1 (bin)

0.793

0.853

0.70

Magee 1 tailor (bin)

0.722

0.795

0.53

No Ki-67 S2 Magee (bin)

0.757

0.809

0.60

Ki-67

0.27

286

Median (range)

91 (1-99)

292 299 299

N (%) N (%) E N (%)

66/292 (22.6) 243/299 (81.3) 82 (27.4%)

299

N (%)

37 (12.3%)

No Ki-67 S2 Magee tail

0.729

0.794

0.52

278

34 (12.2) 177 (63.7) 67 (24.1) 87 (41.6) 110 (52.6) 12 (5.7) 1 (0.5) 178 (87.3) 25 (12.3) 99 (35.6)

Gage

0.714

0.751

0.29

Gage (bin)

0.733

0.777

0.36

University of Tennessee prob (Hi 26)

0.835

0.859

0.52

University of Tennessee tail (Lo 0-10)

0.826

0.849

0.50

University of Tennessee regular cut-off (Hi 31)

0.816

0.862

0.49

278

1 2 3 1 2 3 1 2 3 1

170 (61.2) 9 (3.2) 1 (0.4)

University of Tennessee regular cut-off (Lo 0-17)

0.808

0.854

0.46

266

2 3 1 2 3 Mean (SD) Median (range) 0-17 18-30 31þ

223 (83.8) 42 (15.8) 18.4 (10.4) 17 (0-70) 159 (53.2) 109 (36.5) 31 (10.4)

209

204

ODX score

299

ODX

299

Abbrevations: ER ¼ estrogen receptor; ODX ¼ Oncotype DX; PR ¼ progesterone receptor; S ¼ surrogate; SBR ¼ Scarff-Bloom-Richardson; SD ¼ standard deviation.

4

Variable

Correlation Coefficient Raw Raw D Age-50 (Spearman r) With ODX Score C-statistic C-stat

with a Magee score  26 would similarly be indicated to benefit. Thus, as a potential screening tool, using the cut-offs described above, 110 (51.9%; 95% confidence interval [CI], 44.9%-58.8%) of 212 patients would be deemed as not requiring ODX testing, and 109 (99.1%; 95% CI, 95.0%-99.98%) of these 110 patients would be appropriately managed.

Clinical Breast Cancer Month 2019

Abbreviations: ER ¼ estrogen receptor; LVI ¼ lymphovascular invasion; ODX ¼ Oncotype DX; PR ¼ progesterone receptor; SBR ¼ Scarff-Bloom-Richardson.

Magee 2 Equation The Magee 2 equation does not require Ki-67 and therefore would be more using in general clinical practice. Spearman correlation with ODX and Magee 2 identified a modest association (ie, rho ¼ 0.65 with a c-statistic ¼ 0.883 with age included) (Table 3). Based on the results presented for the Magee 2 equation (Table 4B), for patients aged < 50 years, if patients with a Magee score 2 < 14 or  24 were not sent for further ODX testing, or those with a score < 20 and  32 and age  50 years were not sent for further ODX testing, then 141 (47.2%; 95% CI, 41.4%-53.0%) of 299 patients would not have undergone ODX testing. In this setting, the predictive accuracy would have been 137 (97.2%; 95% CI, 92.9%99.2%) of 141 patients.

Susan J. Robertson et al Table 4 The Positive and Negative Predictive Values for Magee Score 3 (A) and Magee Score 2 (B) Formulae, for a Variety of Cut Points Magee Score 3 (A)

Age, y <50

50

Magee 3 Score

n

<14 <16 <18 <20 <18 <20 <22 <24 <26 <28

8 21 28 34 75 105 125 138 151 152

Recommended for No Chemotherapy Based on ODX Score, N (%) 0 1 2 4 0 2 6 8 15 16

(0) (4.8) (7.1) (11.8) (0) (1.9) (4.8) (5.8) (9.9) (10.5)

Magee 3 Score

n

14 16 18 20 18 20 22 24 26 28

37 24 17 11 92 62 42 29 16 15

Recommended for Chemotherapy Based on ODX Score, N (%) 15 14 13 11 30 28 24 22 15 14

(40.5) (58.3) (76.5) (100.0) (32.6) (45.2) (57.1) (75.9) (93.8) (93.3)

Magee Score 2 (B)

Age, y <50

50

Magee 2 Score

N

<14 <16 <24 <18 <20 <30 <32

10 20 59 77 110 215 219

Recommended for No Chemotherapy Based on ODX Score, N (%) 0 1 10 0 3 31 33

(0) (5.0) (17.0) (0) (2.7) (14.4) (15.1)

Magee 2 Score

N

14 16 24 18 20 30 32

60 50 11 152 119 14 10

Recommended for Chemotherapy Based on ODX Score, N (%) 20 19 10 43 40 12 10

(33.3) (38.0) (90.9) (28.3) (33.6) (85.7) (100.0)

Abbreviation: ODX ¼ Oncotype DX.

Based on these cut-offs, 120 patients would be considered at low risk and deemed to not benefit from chemotherapy, and 117 (97.5%) patients had similar results when verified by ODX. In contrast, 21 patients would be considered at high risk, and 20 (95.2%) of them had similar results observed by ODX. All 10 patients < 50 years old who had a Magee S2 score < 14 and 107 (97%) of 110 patients who were  50 years old with a Magee S2 score < 20 had an ODX test score that would have indicated no benefit for adjuvant chemotherapy. There were 10 (91%) of 11 patients < 50 years old with a Magee 2 score  24 and 10 (100%) of 10 patients  50 years old and with a Magee 2 score  32 who would similarly be indicated to benefit.

University of Tennessee Predictive Algorithm

Gage Algorithm

Clinicopathology

The Gage classification criteria, available on all 302 patients, separated into 37 (12.4%) low-risk, 180 (60.2%) intermediate-risk, and 82 (27.4%) high-risk patients. Of Gage low-risk patients, 30 (81.1%) had an ODX score
17, whereas 0 (0%) had an ODX score  31. Alternatively, for Gage intermediate-risk patients, 108 (60.0%) and 4 (2.2%) had ODX scores
17 and  31 respectively, whereas there were 21 (25.6%) and 27 (32.9%) among the Gage high-risk groups, suggesting that Gage scores fail to discriminate risk in patients that fall in the intermediate- or high-risk groups.

Spearman correlation with ODX was performed using the clinicopathology of the various tumors. Concordance statistics based on the raw predictive score, as well as the raw predictive score adjusted for age, are presented in Table 3. The clinicopathology data did not substantially improve the ability of scores to screen for ODX testing.

University of Tennessee probability scores were possible in 287 patients. The scores were not possible in 15 patients either owing to lack of LVI status (7 patients) or non-compliance with the size limit of 6 to 50 mm for that algorithm (8 patients). Spearman correlation coefficients were calculated to evaluate the association between the probability scores and ODX, with rates of 0.52, 0.50, 0.49, and 0.46 for the 4 probability scores respectively (Hi 26-100, lo 0-10, hi 31, lo 0-17). This level of association would be considered of moderate strength, but in the current analysis was not deemed sufficiently strong to pursue further as a potential screening tool for ODX.

Discussion Although the results of the TAILORx trial showed that overall, patients in the intermediate group obtained no additional benefit from adding chemotherapy to their endocrine therapy, subsequent

Clinical Breast Cancer Month 2019

-5

Oncotype DX Testing Using Standard Clinicopathologic Information

6

-

subgroup analysis made the situation less straightforward.3 The subgroup analyses suggested that for those aged
50 years with low or “low-intermediate” scores (scores 11-19), there is likely little benefit from chemotherapy, whereas patients with “high-intermediate” scores (score 20-25) may benefit from chemotherapy. Many of the limitations of this study have been cited,32 with the most significant being that the study does not evaluate which patients do not need the ODX test performed. The current study is the first we are aware of that has assessed the ability of the Magee, Gage, and University of Tennessee models to be used to triage patients to identify those unlikely to benefit from addition ODX testing in the setting of the TAILORx results. Although accepting the many limitations of both these surrogate models and the current study (outlined below), we hoped that this study would provide some additional information for investigators to use as a means of practically screening which patients would benefit from addition ODX testing, and more importantly, which patients will not. This is particularly important in the setting of a service-based laboratory like our own. Although all models showed some correlation with the ODX results, the Magee 1 and 3 scores appeared to be optimal, with a c-statistic > 0.90 and Spearman r > 0.70. The Magee equations were therefore further evaluated to see how they could be used as a potential screening tool for ODX. Our results indicated that the Magee 3 equation could significantly reduce the use of ODX testing; however, this equation requires Ki-67 analysis, something that is not routinely available in many service-based centers. As a potential screening tool in this population, use of the Magee 3 score could avoid ODX testing in over one-half the patients, while maintaining appropriate management in over 99% of patients. Clearly these findings require validation in other populations and discussion of which cut-off both patients and physicians would find acceptable. However, our results do suggest that there may well be a role for the use of these surrogates to add value to requests for multigene assays.33 The Magee 2 equation, which does not require Ki-67, was not as highly correlated with ODX scores as the Magee 3 equation, could also significantly reduce the frequency of additional ODX requests. The advantage of Magee 2 over the Magee 3 equations would include a range of other factors such as wider availability in general clinical practice, time required to calculate Ki-67, and the variability in Ki-67 assessment between institutions. Understandably there are limitations to this study that can be broadly divided into limitations of the surrogate models, limitations of how these models are used in practice, and Ki-67 use in particular, limitations in ODX testing, and finally, limitations of the current study itself. All these surrogate formulae used ER and PR status along with different combinations of grade, proliferation indices (Ki-67, mitotic rate), HER2 status, and tumor size. The limitations of the models are all well-recognized.24 An important limitation of the Magee formulae is their requirement for H scores for ER and PR expression, as these are labor-intensive if done manually, or require specialized software if done by automated image analysis.29 This limitation has led several groups to use Allred scores to produce a more practical approximation for standard H scores.24,29,30,34 The issues with Ki-67 reliability and significant differences in interpretation between various laboratories is well-recognized and the focus of significant international attention.35 In our center, we have used standardized protocols, image

Clinical Breast Cancer Month 2019

assistance, and involvement of a limited number of assessors to improve reliability. At our center, the average error range of 5% has been seen between 2 pathologists (SR, DG) with less variability with low scores and more with high ones. The variability between the 4 technologists when tested on the same slides was around 4%. However, results may be different in other centers with different levels of variability. A further limitation of the current study could be the lack of central pathology review particularly with respect to tumor grade, which is well-recognized to be associated with significant inter-observer variability. The limitations of ODX testing are well-recognized, and although ODX has been credited with being at least in part responsible for changing adjuvant chemotherapy recommendations,4 its use does lead to higher odds of chemotherapy use in patients with small node-negative cancers, lower odds of use of chemotherapy in node-positive or large node-negative disease, delays in treatment decisions, and an increase in the proportion of test results in the intermediate risk group3,6,8,9 for whom the results of TAILORx are particularly important. Over the same time period that multi-gene assays have been available, there is increasing evidence that the benefits of chemotherapy in this patient population are often modest.5,8,36 These changes in clinical practice have important implications for the optimal use of multi-gene assays33 and are likely reflected through the considerable variability in use of these tests between physicians and between cancer centers in Ontario, Canada (Susan Robertson, personal communication). This equipoise has considerable impact on both patient care and financial impact on the health care system, as indiscriminate ordering of ODX has previously been shown to be a low-value intervention.10 Finally, there are the limitations of the current study. The results coming from 15 pathologists in 4 service-based pathology laboratories with surrogate H scores could be considered a limitation. We also chose to not evaluate ODX use in invasive lobular carcinoma and other rare tumor histologies as ODX use in these subtypes is questionable.37,38 The patients were all referred to medical oncologists at a single, tertiary care center, and the generalizability of results are questionable because they may have different referral patterns (for ODX testing). Further studies are required to explore the external validity of these surrogate measures in other datasets, with larger sample sizes, to also allow further development of the proposed cut-offs (eg, Surveillence, Epidemiology, and End Results and TAILORx cohorts). For example, as the Magee 1 and 3 scores are dependent on Ki-67, there will remain the significant challenge of how to identify those patients that will benefit from a Magee 2 score (with no need for a Ki-67) and those for whom a readily repeatable and valid Ki-67 score is available.

Conclusion This is the first study we are aware of that uses a practical strategy to identify patients unlikely to benefit from additional ODX testing in light of the TAILORx results. Further studies are needed for the clinical adoption of current surrogate markers, but these findings could provide important positive results for patients, physicians, and the health care system as a whole. This will become increasingly important for lymph node-positive disease39 and for late recurrence and extended endocrine therapy use,40 as well as risk of long-term recurrence.41

Susan J. Robertson et al Clinical Practice Points  Although widespread use of ODX has certainly reduced the use

of chemotherapy in patients with early stage breast cancer, indiscriminate ordering of ODX is of poor value to patients, physicians, and health care providers.  Models such as the 3 Magee equations, Gage algorithm, and University of Tennessee predictive algorithm use standard clinicopathologic data to provide surrogate ODX scores. In this study, we evaluated whether these prognostic scores could be used to identify patients unlikely to benefit from additional ODX testing.  The results showed that use of all formulae, and the Magee 3 equation in particular, could be proposed as screening tools prior to ODX testing. Use of any of these surrogates could significantly reduce the frequency of ODX testing.  If these findings are validated in other datasets they could rapidly and easily be integrated into clinical practice as a screening tool to determine which patients are unlikely to benefit from additional ODX testing.

Acknowledgments This trial was supported by internal funding from the Ottawa Hospital Research Institute.

Disclosure The authors have stated that they have no conflicts of interest.

References 1. Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifentreated, node-negative breast cancer. N Engl J Med 2004; 351:2817-26. 2. Paik S, Tang G, Shak S, et al. Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer. J Clin Oncol 2006; 24:3726-34. 3. Sparano JA, Gray RJ, Makower DF, et al. Prospective validation of a 21-gene expression assay in breast cancer. N Engl J Med 2015; 373:2005-14. 4. Carlson JJ, Roth JA. The impact of the Oncotype Dx breast cancer assay in clinical practice: a systematic review and meta-analysis. Breast Cancer Res Treat 2013; 141: 13-22. 5. Kelly CM, Bernard PS, Krishnamurthy S, et al. Agreement in risk prediction between the 21-gene recurrence score assay (Oncotype DX(R)) and the PAM50 breast cancer intrinsic Classifier in early-stage estrogen receptor-positive breast cancer. Oncologist 2012; 17:492-8. 6. Kelly CM, Krishnamurthy S, Bianchini G, et al. Utility of oncotype DX risk estimates in clinically intermediate risk hormone receptor-positive, HER2-normal, grade II, lymph node-negative breast cancers. Cancer 2010; 116:5161-7. 7. Sparano JA, Gray RJ, Makower DF, et al. Adjuvant chemotherapy guided by a 21gene expression assay in breast cancer. N Engl J Med 2018; 379:111-21. 8. Hassett MJ, Silver SM, Hughes ME, et al. Adoption of gene expression profile testing and association with use of chemotherapy among women with breast cancer. J Clin Oncol 2012; 30:2218-26. 9. Sotiriou C, Pusztai L. Gene-expression signatures in breast cancer. N Engl J Med 2009; 360:790-800. 10. Mittmann N, Earle CC, Cheng SY, et al. Population-based study to determine the health system costs of using the 21-gene assay. J Clin Oncol 2018; 36:238-43. 11. Goldhirsch A, Wood WC, Gelber RD, Coates AS, Thurlimann B, Senn HJ. Progress and promise: highlights of the international expert consensus on the primary therapy of early breast cancer 2007. Ann Oncol 2007; 18:1133-44. 12. Ravdin PM. A computer based program to assist in adjuvant therapy decisions for individual breast cancer patients. Bull Cancer 1995; 82(Suppl 5):561s-4s. 13. Ravdin PM. A computer program to assist in making breast cancer adjuvant therapy decisions. Semin Oncol 1996; 23:43-50. 14. Candido Dos Reis FJ, Wishart GC, Dicks EM, et al. An updated PREDICT breast cancer prognostication and treatment benefit prediction model with independent validation. Breast Cancer Res 2017; 19:58. 15. Wishart GC, Rakha E, Green A, et al. Inclusion of KI67 significantly improves performance of the PREDICT prognostication and prediction model for early breast cancer. BMC Cancer 2014; 14:908.

16. Shachar SS, Muss HB. Internet tools to enhance breast cancer care. NPJ Breast Cancer 2016; 2:16011. 17. Wishart GC, Azzato EM, Greenberg DC, et al. PREDICT: a new UK prognostic model that predicts survival following surgery for invasive breast cancer. Breast Cancer Res 2010; 12:R1. 18. Turner B, Tang P, Hicks D. The value of algorithms predicting the Oncotype DX recurrence score should not be underestimated! Breast Cancer Res Treat 2017; 164: 249-50. 19. Wazir U, Mokbel K, Carmichael A, Mokbel K. Are online prediction tools a valid alternative to genomic profiling in the context of systemic treatment of ER-positive breast cancer? Cell Mol Biol Lett 2017; 22:20. 20. Klein ME, Dabbs DJ, Shuai Y, et al. Prediction of the Oncotype DX recurrence score: use of pathology-generated equations derived by linear regression analysis. Mod Pathol 2013; 26:658-64. 21. Hou Y, Tozbikian G, Zynger DL, Li Z. Using the modified Magee equation to identify patients unlikely to benefit from the 21-gene recurrence score assay (Oncotype DX Assay). Am J Clin Pathol 2017; 147:541-8. 22. Flanagan MB, Dabbs DJ, Brufsky AM, Beriwal S, Bhargava R. Histopathologic variables predict Oncotype DX recurrence score. Mod Pathol 2008; 21: 1255-61. 23. Magee Equations for Estimating Oncotype DX Recurrence Score. Available at: https://path.upmc.edu/onlineTools/mageeequations.html. Accessed: July 3, 2019. 24. Harowicz MR, Robinson TJ, Dinan MA, et al. Algorithms for prediction of the Oncotype DX recurrence score using clinicopathologic data: a review and comparison using an independent dataset. Breast Cancer Res Treat 2017; 162:1-10. 25. Gage MM, Rosman M, Mylander WC, et al. A validated model for identifying patients unlikely to benefit from the 21-gene recurrence score assay. Clin Breast Cancer 2015; 15:467-72. 26. Orucevic A, Bell JL, McNabb AP, Heidel RE. Oncotype DX breast cancer recurrence score can be predicted with a novel nomogram using clinicopathologic data. Breast Cancer Res Treat 2017; 163:51-61. 27. Elston CW, Ellis IO. Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: experience from a large study with long-term follow-up. Histopathology 1991; 19:403-10. 28. Hammond ME, Hayes DF, Dowsett M, et al, American Society of Clinical Oncology; College of American Pathologists. American Society of Clinical Oncology/College of American Pathologists guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer (unabridged version). Arch Pathol Lab Med 2010; 134:e48-72. 29. Robertson S, Petkiewicz S, Arnaout A, Clemons M, Gravel D, Pond G. Abstract P3-07-16: Comparing surrogates of oncotype Dx recurrence scores in invasive ductal carcinoma: how complicated does it have to be? Cancer Res 2016; 76, P307-16. 30. Bradshaw SH, Pidutti D, Gravel DH, Song X, Marginean EC, Robertson SJ. Predicting OncoDx recurrence scores with immunohistochemical markers. Appl Immunohistochem Mol Morphol 2013; 21:490-6. 31. Bradshaw SH, Pidutti D, Gravel DH, Song X, Robertson SJ. Predicting OncoDX recurrence scores with immunohistochemical markers: effect of stromelysin. Appl Immunohistochem Mol Morphol 2015; 23:26-30. 32. 2018 ASCO: TAILORx: most women with early-stage breast cancer can forgo chemotherapy when guided by a diagnostic test. The ASCO Post. Available at: https://www.ascopost.com/News/58904. Updated March 6, 2018. Accessed July 3, 2018. 33. Kim HS, Umbricht CB, Illei PB, et al. Optimizing the use of gene expression profiling in early-stage breast cancer. J Clin Oncol 2016; 34:4390-7. 34. Chang N, Arnaout A, Clemons M, Addison C, Pond G, Robertson D. Are oncotype DX and magee recurrence scores concordant in invasive lobular breast cancer?. Pittsburgh, PA: First International ILC Symposium; 2016. 35. Yerushalmi R, Woods R, Ravdin PM, Hayes MM, Gelmon KA. Ki67 in breast cancer: prognostic and predictive potential. Lancet Oncol 2010; 11:174-83. 36. Davidson JA, Cromwell I, Ellard SL, et al. A prospective clinical utility and pharmacoeconomic study of the impact of the 21-gene Recurrence Score(R) assay in oestrogen receptor positive node negative breast cancer. Eur J Cancer 2013; 49: 2469-75. 37. Marmor S, Hui JYC, Huang JL, et al. Relative effectiveness of adjuvant chemotherapy for invasive lobular compared with invasive ductal carcinoma of the breast. Cancer 2017; 123:3015-21. 38. Kizy S, Huang JL, Marmor S, Tuttle TM, Hui JYC. Impact of the 21-gene recurrence score on outcome in patients with invasive lobular carcinoma of the breast. Breast Cancer Res Treat 2017; 165:757-63. 39. Mamounas EP, Liu Q, Paik S, et al. 21-Gene recurrence score and locoregional recurrence in node-positive/ER-positive breast cancer treated with chemoendocrine therapy. J Natl Cancer Inst 2017; 109. 40. Wolmark N, Mamounas EP, Baehner FL, et al. Prognostic impact of the combination of recurrence score and quantitative estrogen receptor expression (ESR1) on predicting late distant recurrence risk in estrogen receptor-positive breast cancer after 5 years of tamoxifen: results from NRG Oncology/National Surgical Adjuvant Breast and Bowel Project B-28 and B-14. J Clin Oncol 2016; 34:2350-8. 41. Sgroi DC, Sestak I, Cuzick J, et al. Prediction of late distant recurrence in patients with oestrogen-receptor-positive breast cancer: a prospective comparison of the breast-cancer index (BCI) assay, 21-gene recurrence score, and IHC4 in the TransATAC study population. Lancet Oncol 2013; 14:1067-76.

Clinical Breast Cancer Month 2019

-7