Using proliferative markers and Oncotype DX in therapeutic decision-making for breast cancer : the B . C . experience

Background Proliferative scoring of breast tumours can guide treatment recommendations, particularly for estrogen receptor (er)–positive, her2-negative, T1–2, N0 disease. Our objectives were to ■ estimate the proportion of such patients for whom proliferative indices [mitotic count (mc), Ki-67 immunostain, and Oncotype dx (Genomic Health, Redwood City, CA, U.S.A.) recurrence score (rs)] were obtained. ■ compare the indices preferred by oncologists with the indices available to them. ■ correlate Nottingham grade (ng) and its subcomponents with Oncotype dx. ■ assess interobserver variation. Methods All of the er-positive, her2-negative, T1–2, N0 breast cancers diagnosed from 2007 to 2011 (n = 5110) were linked to a dataset of all provincial breast cancers with a rs. A 5% random sample of the 5110 cancers was reviewed to estimate the proportion that had a mc, Ki-67 index, and rs. Correlation coefficients were calculated for the rs with ng subcomponent scores. Interobserver variation in histologic grading between outside and central review pathology reports was assessed using a weighted kappa test. Results During 2007–2011, most cancers were histologically graded and assigned a mc; few had a Ki-67 index or rs. The ng and mc were significantly positively correlated with rs. The level of agreement in histologic scoring between outside and central pathology reports was good or very good. Very few cases with a low mc had a high rs (1.8%). Conclusions Patients with low ng and mc scores are unlikely to have a high rs, and thus are less likely to benefit from chemotherapy. In the context of limited resources, that finding can guide clinicians about when a rs adds the most value.


INTRODUCTION
Proliferation indices can provide valuable information about the prognosis and aggressiveness of breast tumours and can guide treatment recommendations.Patients with estrogen receptor (er)-positive, her2-negative, T1-2, N0 breast cancers represent a group in which the decision to give chemotherapy is often associated with an uncertain or small risk:benefit ratio.Indices of proliferation help clinicians to assess the risk of recurrence.As well, high proliferation rates in patients with early-stage er-positive node-negative breast tumours imply an increased potential benefit from chemotherapy; low proliferation rates suggest a modest recurrence risk, for which the harm of chemotherapy might outweigh the benefit 1 .
Mitotic count (mc) is a count of the number of mitotic figures visible in 10 high-power fields of a light microscope.A score of 0, 1, 2, or 3 is then assigned to the sample according to the cut-off values validated in the Nottingham grading (ng) system 2 .The method is inexpensive and simple, but it lacks the precision and range provided by more modern and automated techniques.The ubiquitous ng for breast cancer is a composite score for tubule formation (tf), nuclear pleomorphism (np), and mc 3 .
The nuclear protein Ki-67 is expressed at its highest levels during mitosis; it is not expressed in G0 phase 4 .
Immunostaining of tumour biopsy tissue with a monoclonal antibody (usually mib1) allows for an assessment of the proportion of neoplastic cells that are mitotically active.That growth fraction is the Ki-67 index, reported as a percentage, which can be correlated with risk categories in breast cancer 3 .Many studies have concluded that the Ki-67 index can provide physicians with information, beyond that available by conventional histopathology, about prognosis and risk of recurrence for early breast cancers.There is evidence that an elevated Ki-67 index is a marker of worse disease-free survival and overall survival [5][6][7] .The Ki-67 index could potentially guide treatment recommendations in future, but more research is needed to link the Ki-67 index with response to a specific therapy and to standardize the scoring system 3 .
The Oncotype dx (Genomic Health, Redwood City, CA, U.S.A.) 21-gene assay generates a recurrence score (rs) that measures the relative expression levels of 16 cancer-related genes and 5 reference genes.The rs quantifies the risk of recurrence so that the benefit of chemotherapy in patients with early-stage node-negative er-positive breast cancers can be assessed.Cases are grouped into 3 categories-low, intermediate, and high risk of recurrence-based on the expression levels of the cancer-related genes.Of the 16 cancer-related genes, 5 (including MKI67) constitute a proliferation group.In er-positive, her2-negative, T1-2, N0 breast cancer, the magnitude of the rs depends mainly on the proliferation group score because of its very high weighting in calculating the final rs and the relatively low estrogen and her2 group scores (Table i).
Several retrospective studies using tumour samples banked by The National Surgical Adjuvant Breast and Bowel Project B14 and B20 trials have found an association between tumour gene expression, recurrence risk, and benefit from chemotherapy.Compared with cancers having a low rs, cancers with a high rs were found, in significantly greater proportion, to be associated with distant recurrence at 10 years 8 .Recurrence risk was also shown to increase as expression levels of genes in the proliferative group increased 8 .Additionally, patients with higher rs cancersand likely a more proliferative tumour-experienced a greater benefit from adjuvant chemotherapy 1 .
In the present study, we examined the use of indices of proliferation-that is, ng and mc, Ki-67 index, and rs-for patients with er-positive, her2-negative, T1-2, N0 breast cancers in British Columbia during 2007-2011.The resulting cohort is one of the largest to be examined with respect to this topic, and it is unique in that it specifically represents a her2-negative population.
Our objectives were to ■ estimate the proportion of patients with er-positive, her2-negative, T1-2, N0 breast cancers for whom the Nottingham mc, Ki-67 index, and Oncotype dx rs proliferative indices were obtained.■ compare the indices preferred by oncologists with the indices available to them.■ correlate ng and its subcomponents, particularly the mc, with rs.■ assess interobserver variation of mc and ng between pathologists.

METHODS
All patients with er-positive, her2-negative, T1-2, N0, M0 breast cancers, diagnosed and referred to the BC Cancer Agency (bcca) between 2007 and 2011 were identified using data collected by the bcca Breast Cancer Outcomes Unit.The 2007-20011 period was chosen because it coincided with the period of recruitment for two large Oncotype dxbased trials that provided the only public access to Oncotype dx testing in British Columbia at the time 9,10 .Patients who had been treated with neoadjuvant chemotherapy or who had a prior or synchronous invasive or in situ breast cancer diagnosis were excluded.
To estimate the use of ng, the Ki-67 index, and Oncotype dx in this patient population, a 5% random sample of patient charts from the period of interest were retrospectively reviewed.The Ki-67 index was abstracted from pathology reports.Numerous pathology departments perform Ki-67 testing using various assays, and no uniform assay or cut-point had been established.During abstraction, the Ki-67 index was scored as "elevated" or "not elevated" based on the qualitative description in the pathology report.(The most common cut-points for elevation were >10% and >20%.) The patient and disease characteristics collected were age; TNM staging; lymphovascular invasion (lvi) status; strength of er, progesterone receptor, and her2 expression; ng, Ki-67, and Oncotype dx (when performed) scores; and the type of breast surgery, radiation treatment, hormonal therapy, and chemotherapy prescribed.Medical oncologists at the bcca were surveyed about the accessibility and utility of the proliferative markers in this patient population.
To determine the correlations of ng overall, tf, np, and mc with rs risk group in this patient population, all patients with er-positive, her2-negative, T1-2, N0 breast cancer who were referred to the bcca during 2007-2011 and were known to have had an Oncotype dx test performed on clinical study 9,10 were identified (n = 226).Histologic grading was assessed on the specimen used for the Oncotype dx test.Patient and tumour characteristics were collected from pathology reports.Spearman correlation coefficients and 95% confidence intervals (cis) were calculated for rs risk group compared with ng overall, tf, np, and mc.The same correlations were also performed for lvi-negative and lvipositive cancers.
A kappa correlation was used to assess the level of interobserver variation between external pathology reports and internal pathology reviews of ng, where available.Reasons for the reviews included lvi status, margin status, nodal status, and quality control.

RESULTS
A survey (Table ii) distributed to the 31 medical oncologists who treat patients with breast cancer at the bcca assessed the preferred and practiced methods of measuring proliferation rates in er-positive, her2-negative, T1-2, N0 disease in British Columbia.The response rate was 32%.Among responders, 90% stated that they consider indicators of proliferation, particularly the Nottingham mc and the Oncotype dx rs, when making chemotherapy recommendations to patients.However, when asked which indices are easily accessible, 87.5% answered the Nottingham mc only.When asked which measure of proliferation would be preferred if there were no barriers to access, the unanimous answer was the rs.All responders stated they would be less likely to overtreat with chemotherapy if they had better access to robust indicators of proliferation.The implication of "more robust indicators" was that respondents wanted tests that have been validated to both consistently and accurately indicate proliferation.
During 2007-2011, 5110 patients with er-positive, her2negative, T1-2, N0 breast cancers were referred to the bcca.Table iii shows, based on the 5% random sample of those patients (n = 256), the case mix and proportion of patients for whom a mc, Ki-67 index, and rs was obtained.A mc was reported for 85% of the sample (n = 217), a Ki-67 index for 9% (n = 24), and a rs for 5% (n = 14).
Of the 5110 identified patients, 226 had both a rs and a ng based on the specimen used for the rs.The median age in that subpopulation was 52 years (range: 23-75 years).Median tumour size was 1.7 cm.All patients underwent a nodal procedure (sentinel lymph node biopsy, axillary dissection, or both), and the median number of nodes removed was 3 (range: 1-27).
Table iii also further describes disease and treatment characteristics in the subpopulations.Despite a "positive" er status, 7 patients received no hormonal therapy.Reasons included refusal of adjuvant treatment (n = 3), weakly positive er expression on immunohistochemistry (Allred score: 3) and negative er reported from Oncotype dx (even though er positivity was reported on immunohistochemistry, n = 2), severe depression (n = 1), and development of metastatic lung cancer (n = 1) shortly after the breast cancer diagnosis.
The 5% random sample cohort (n = 256) and the Oncotype dx cohort (n = 226) showed several significant differences (Table iii).On average, women for whom an Oncotype dx test was performed had more high-risk features-for example, higher T stage and grade, and greater lvi.Patients in the Oncotype dx cohort also more often received chemotherapy and hormonal therapy.Explanations for the differences include specific enrolment criteria for the trials within which the Oncotype dx testing occurred, higher general compliance among patients who choose to take part in trials, and lack of availability of the trials in all bcca centres.
The ng overall and the tf, np, and mc were collected from the pathology reports for the 5% sample of diagnosed patients.Only patients whose overall ng was determined using the specimen also used for the rs were included.When a central pathology review had been conducted (n = 136 patients, 60% of the sample), the histopathologic values assigned during the central review, rather than the values from the external pathology report, were used in our analysis.Moderately significant positive correlations between rs risk group and overall grade (Spearman coefficient: 0.45; 95% ci: 0.34 to 0.55) and between rs and mc (Spearman coefficient: 0.43; 95% ci: 0.32 to 0.53; Table iv) were observed 11 .Weakly significant positive correlations between rs and tf (Spearman coefficient: 0.24; 95% ci: 0.11 to 0.36) and between rs and np (Spearman coefficient: 0.34; 95% ci: 0.22 to 0.45) were observed.
The Ki-67 index was available for only 24 patients, of whom only 20 had an evaluable score, with 8 being below the cut-point, and 12 being above it.A detailed analysis and determination of correlations with mc and rs was therefore not possible.Of the 8 cases with a Ki-67 index below the cut-point, none had a high mc, and none underwent Oncotype dx testing.
Published retrospective reports suggest that chemotherapy has value predominantly in cancers with a high rs 1 , and so a second analysis of correlation that grouped the women having low and intermediate rs scores and compared them with the women having high rs scores was performed.When the ng, tf, np, and mc scores and the rs risk group scores were converted into binary variables, the strength of the correlations increased (Table iv).A strongly significant positive correlation between the mc and rs (Spearman correlation: 0.61; 95% ci: 0.52 to 0.69) was observed.
None of the cancers with a low ng overall or a low tf or np had an Oncotype dx rs that was classified as high risk; 4 cancers with a mc of 1 (1.8%) had a high Oncotype dx rs.
Because the presence of lvi is a risk factor used by some clinicians in determining the risk:benefit ratio for chemotherapy, a secondary analysis considered just the lvi-negative cases.Spearman correlation coefficients of the ng overall and the tf, np, and mc with the Oncotype dx rs were stronger for the lvi-negative cancers (n = 190) than for the lvi-positive (n = 35) cancers (Table v).The lvi-negative cases showed moderately significant positive correlations between rs risk group and ng overall (Spearman coefficient: 0.46; 95% ci: 0.34 to 0.56) and between rs and mc (Spearman coefficient: 0.44; 95% ci: 0.32 to 0.55).When the analysis in lvi-negative cases was repeated using the binary ng and mc scores compared with the rs, moderately significant positive correlations between ng overall and rs (Spearman coefficient: 0.56; 95% ci: 0.45 to 0.65) and between mc and rs (Spearman coefficient: 0.59; 95% ci: 0.49 to 0.68) were observed.
Most of the 226 er-positive, her2-negative, T1-2, N0 breast cancers for which an Oncotype dx was performed (n = 136, 60%) also underwent an internal bcca pathology review based on the specimen used for the Oncotype dx test.Applying a weighted kappa test to assess interobserver variation (Table vi), we observed a very good significant correlation for ng overall between the review and nonreview pathology reports (κ = 0.84, p < 0.001) and tf (κ = 0.84, p < 0.001).The pathology reports also showed a good significant correlation for np (κ = 0.66, p < 0.001) and mc (κ = 0.69, p < 0.001) 12 .Because discordance in mc and ng between the initial and the review pathology was observed in only 10 cases, analyses that excluded the review pathology showed no significant differences in the correlation of mc and ng with rs.

DISCUSSION
Patients with er-positive, her2-negative, T1-2, N0 breast cancers represent a group for whom decision-making about adjuvant chemotherapy can be difficult.Prognostic estimates and markers of proliferation embedded within the ng and the Oncotype dx rs can help group patients into risk categories, which might enhance confidence about treatment recommendations.The Oncotype dx assay appears to be informative for scores classified as low-and high-risk, as demonstrated by several retrospective studies 1,8 .For example, retrospective analysis of the National Surgical Adjuvant Breast and Bowel Project B20 trial showed significant benefit from adjuvant chemotherapy only in the high-risk group, and little-to-no benefit in the low-risk group 1 .The benefit in the intermediate risk category is less clear.Arguably, the Oncotype dx test is not yet informative for those cases; results of the tailorx study are awaited 9 .
Our findings are consistent with several retrospective studies also investigating the correlations between rs and routine pathology parameters 2,[13][14][15] .In a study of 138 invasive breast cancer cases, Auerbach et al. 13 showed that a combination of negativity for the progesterone receptor and a mc greater than 1 correlated with an intermediate or high Oncotype dx rs.However, those results were based only on a subset of 12 cases.In a test dataset of 104 cases, Allison et al. 14 found that a combination of ng 1, strong positivity for the progesterone receptor (Allred score ≥ 5), and Ki-67 below 10% predicted a low Oncotype dx rs.In a study of 42 cases, Flanagan et al. 15 also demonstrated that the Oncotype dx rs correlated with ng and mc.The study by Zybtek et al. 2 of 137 er-positive invasive breast cancers also found a high correlation with Nottingham mc and Oncotype dx rs.Our study, which encompasses one of the largest cohorts used in an examination of this topic, adds to the literature and focuses on a her2-negative patient population.The data from both the practitioner survey and the 5% retrospective review indicate that, of most such patients in British Columbia whose ng and mc were determined during the period of interest, relatively few underwent Ki-67 or Oncotype dx testing.The survey suggests that oncologists favour Nottingham mc and Oncotype dx rs to inform chemotherapy decisions.The Ki-67 assay is considered less valuable (in part for the reasons outlined in the Introduction).Oncotype dx is the least accessible and most expensive, but also the most desired proliferative index for er-positive, her2-negative, T1-2, N0 patients.
In the present study, we observed moderately significant positive correlations of ng and mc with Oncotype dx rs and weakly positive correlations of tf and np.Mirroring the findings of the National Surgical Adjuvant Breast and Bowel Project B20 trial, we observed that when the raw scores for ng overall, tf, np, mc, and rs are converted to binary variables, the strength of correlations increases.The very good strength of interobserver correlation between the non-  (that is, be reclassified as high-risk).In the current health care climate influenced by limited resources, clinicians can use those data to guide their determination of when an Oncotype dx rs will add the most value.That finding is important given that most clinicians responding to a survey preferred the rs above other measures of proliferation, presumably on the basis that it provided more or better information, which it does not appear to do for cancers with low mc and ng values.If an Oncotype dx test cannot be offered to all such patients, our data show that testing will be most useful and clinically actionable in patients with high histologic scores.Our analysis has several limitations.Relatively few patients in British Columbia underwent Ki-67 testing during the era of interest, limiting our ability to calculate correlations with that index.No provincial policy on centralized testing for Ki-67 or the consistency of its cut-point has been articulated, rendering its interpretation at a population level difficult, and belief about the utility of the Ki-67 index (compared with ng and rs) is less consistent.Also, not all cases underwent a central review of their ng.Use of Oncotype dx testing was relatively low.Our analysis in some ways simplifies the complex decisions made by medical oncologists in individual cases, and as always, there will be exceptions for whom an Oncotype dx rs might have additional value.For example, factors that require additional consideration are tumour size, patient age, and comorbidities.Within a T stage, tumour size (much like er status, grade, and lvi) is a factor that informs the decision about chemotherapy.Specifically when considering chemotherapy, oncologists might be more inclined to want additional guidance from a rs for a large low-grade tumour.Similarly, when the potential benefit of chemotherapy is diminished because of a patient's advanced age or significant comorbidity, the rs might help to justify a disinclination toward chemotherapy in tumours with a high ng or mc score.
Despite the foregoing limitations, our series addressing various proliferative scoring methods is one of the largest and is specifically focused on er-positive, her2-negative, T1-2, N0 disease.

CONCLUSIONS
The Oncotype dx rs very rarely provides information that will change treatment recommendations for cancers with low ng or mc scores.Although oncologists responding to a survey identified rs as their preferred method of assessing proliferation and the value of chemotherapy in er-positive, her2-negative, T1-2, N0 disease, our data suggest that rs is no better than mc and ng when those parameters score low.In a health care system with limited resources, such data should inform clinicians and administrators with respect to how the cost of rs can be balanced against additional actionable information to aid in chemotherapy recommendations.

TABLE I
The 21-gene panel (16 cancer-related genes and 5 reference genes) used to calculate the Oncotype DX Recurrence Score a Genomic Health, Redwood City, CA, U.S.A.Current Oncology, Vol.22, No. 3, June 2015 © 2015 Multimed Inc.

TABLE II
Survey for medical oncologists on the utility of proliferative scoring when determining the potential benefit of chemotherapy

TABLE III
Patient characteristics and treatment factors in two patient groups, 2007-2011 a Genomic Health, Redwood City, CA, U.S.A.

TABLE V
Correlation analysis of Nottingham histology with Oncotype DX a recurrence score risk group, by lymphovascular invasion status

TABLE VI
review and central review pathology reports is reassuring for the scoring consistency of the ng and its subcomponents in the pathology community in British Columbia.It suggests that cancers with a low or intermediate score for tf, np, and mc are unlikely to return a treatment-changing rs