• © 2008 by American Society of Clinical Oncology

An Ideal Prognostic Test for Estrogen Receptor–Positive Breast Cancer?

The decision to administer chemotherapy to patients diagnosed with early-stage breast cancer ideally should be based on assessment of baseline risk and expected degree of benefit from chemotherapy. Because chemotherapy will not be needed if baseline risk is excellent, most of the time decision is purely based on prognostic factors. Currently two independently validated prognostic tests are widely used for breast cancer. The Adjuvant! algorithm is based on clinical and pathological markers and is freely available.¹ The recurrence score (RS) assay is a multigene based molecular assay which is available as a commercial reference laboratory test (Oncotype DX) covered by insurance.^2,3

In this issue of Journal of Clinical Oncology, Goldstein et al compared these two very different prognostic tests in a retrospective study of patients with breast cancer with hormone receptor–positive disease and zero to three positive nodes who received contemporary chemohormone therapy.⁴ While the study suffers from the fact that the study population is quite different from the originally intended clinical target population for recurrence score (RS) assay, it still provides a valuable clinical insight.

Adjuvant! is a mathematical algorithm that provides an estimate of risk of breast cancer–related death at 10-year follow-up based on the tumor size, the number of involved nodes, and estrogen receptor status.¹ These estimates are based on an analysis of data from the Surveillance, Epidemiology, and End-Results (SEER) registry, which follows approximately 10% of all breast cancer cases in the United States. Estimates of the efficacy of therapy are based mainly on the proportional risk reduction (PRR) obtained from the Oxford overview of randomized clinical trials. Adjuvant! is one of the best independently validated clinical prognostication tools currently in use. Olivotto et al⁵ examined the performance of Adjuvant! in 4,083 patients registered in the British Columbia Breast Cancer Outcomes Unit database between 1989 and 1993. Predicted and observed outcomes were within 2% for most demographic, pathologic, and treatment-defined subgroups. The study by Goldstein et al reaffirms this finding for average outcome for the entire study population when centrally performed grade was used.¹ Although not directly addressed in the article, the prediction of 5-year recurrence by Adjuvant! using locally performed grade may not be as accurate according to Figure 2 in the article.¹

The 21-gene RS assay provides individualized risk estimates based on measurement of the expression levels of 16 cancer-related genes in reference to five invariant genes. RS was originally developed as a context specific prognostic assay to help patients diagnosed with estrogen receptor–positive, axillary node-negative breast cancer treated with tamoxifen. However, due to the inclusion of genes from estrogen receptor, proliferation, and HER2 pathways, it is also expected to be prognostic in general outside the original clinical context. In fact, two of the three cohorts used in the initial gene finding and model building steps were outside the intended context. RS obviously was highly prognostic in those two cohorts.⁶ The prognostic role of RS was validated in an independent context-specific cohort from the tamoxifen treated arm of National Surgical Adjuvant Breast and Bowel Project trial B14 as well as in a case control study in a community oncology setting.⁶

There are clear differences between Adjuvant! and RS. Because the information that the Adjuvant! algorithm uses to estimates the risk are binned into categories (eg, size is binned into 0.1 to 1.0 cm, 1.1 to 2.0 cm, 2.1 to 3 cm, 3.1 to 5.0 cm, and > 5 cm), the output is not continuous and fixed in the number of possibilities. Because real biology most likely is continuous, it is safe to state that Adjuvant! risk categories can provide the best estimates of averaged clinical behavior of defined subsets of breast cancer but not true individualized estimates. Unlike Adjuvant!, RS provides continuous risk estimates. Due to this difference, in Goldstein et al study, the results of the two tests had to be transformed to be compared with each other.⁴ This was achieved by either grouping Adjuvant! output into risk groups proportionate to the raw distribution of RS risk groups or transforming the results of both tests based on rank-ordered risk percentile classification.

Most important information from Goldstein et al study is that RS and Adjuvant! are independent of each other. Not surprisingly the concordance between the two tests is only modest (36% and 38% depending on local or central grade was used for Adjuvant!). So, which test is better? Goldstein et al demonstrated that while they are independent of each other, within each risk categories defined by Adjvant!, RS provided additional prognostic information. For the 43% of those who were in the low Adjuvant! risk group, the risk of recurrence was increased 2.6-fold and 4.0-fold for intermediate and high RS, respectively. For the 30% of patients in the intermediate Adjuvant! risk group, the risk of recurrence was increased 9.6-fold and 5.8-fold for intermediate and high RS, respectively. For the 24% of patients in the high Adjuvant! risk group, only high RS was associated with a significantly increased risk of relapse (2.6-fold). However, it is also clear that Adjuvant! can also provide additional information for risk groups defined by RS. Thus, the best scenario for the patient is with low RS score and low risk according to Adjuvant!. If either of the two shows poor risk, then the risk of relapse appears to be increased.

Because patients in the Goldstein et al study are not from the originally intended clinical context for RS (all treated with chemotherapy in addition to tamoxifen, and some are node positive), it is difficult to judge whether the findings can be extrapolated to node-negative, estrogen receptor–positive patients treated only with tamoxifen. However, unpublished analyses of National Surgical Adjuvant Breast and Bowel Project B-14 cohort originally used for the validation of RS largely agreed with the latter data. Even when the Adjuvant! output was low risk, if RS was intermediate or high score the chance of systemic recurrence was still more than 10%, while if RS was low the chance of recurrence was below 10% even if Adjuvant! output was intermediate or high risk. Therefore, RS could provide additional information over and above Adjuvant!. However, it is also true that Adjuvant! can also provide additional information for risk groups defined by RS.

More importantly, the study by Goldstein et al provided valuable information for the potential utility of RS and Adjuant! in identifying patients who will not require more than chemohormone therapy. In a current environment where many trials for targeted agents are being pursued, it is often easy to forget that many patients have such a good prognosis based on legacy treatment and may not derive meaningful clinical benefit from additional targeted agents. This is especially true now that it became evident that most targeted agents are not without significant adverse effects. Patients with hormone receptor–positive tumors with low RS seem to have excellent prognosis even with positive axillary nodes and therefore these patients, especially if Adjuvant! also indicate low risk, should be spared from further systemic therapy.

This study also demonstrated a need to integrate RS and Adjuvant! into a single prognostic algorithm because they are independent of each other. While the genomic version of Adjuvant! allows integration of RS as a prognostic factor into its algorithm, one obvious hurdle to overcome—as demonstrated by Goldstein et al—is the lack of reproducibility of one of the components of Adjuvant!, tumor histological grade.¹⁰ One potential way to overcome this problem is to replace histological grade with so-called genomic grade described by Sotiriou et al.¹¹ In the latter study, investigators were able to develop gene expression signature describing low and high histological grade and were able to use this to divide intermediate grade tumors into either low- or high-grade tumors. Because gene expression assays are in general more reproducible than subjective grading by pathologists, it may be possible to improve Adjuvant! using genomic grade and achieve more continuous prediction than is currently possible. Then, we may be able to combine Adjuvant! with RS to generate a single integrated marker. The Trial Assigning Individualized Options for Treatment is aimed at answering these issues.¹² In this trial, while initial risk stratification is based on RS (with random assignment of only intermediate-risk patients), comparison with Adjuvant! is planned on completion of the trial. Utilization of the collected tumor blocks will allow further optimization of RS as well as testing for other markers.