Oncotype DX multigene assay as a prognostic factor in patients treated with tamoxifen

At the 2003 San Antonio meeting, when I presented the initial data on this assay, Dr Osborne raised a question about whether the recurrence score is a prognostic or predictive factor. Frankly, we didn’t really care, as long as it’s a prognostic factor in that specific setting of tamoxifen-treated patients, so that we can identify a cohort of patients who don’t need chemotherapy.

Using the NCCN or St Galen criteria, in the tamoxifen-treated cohort in NSABP-B-14, we would identify about eight percent of patients who don’t need chemotherapy. If we use the Genomic Health assay, we identify 50 percent — a huge increase in the number of patients categorized as low risk and not requiring chemotherapy.

The median 10-year distant failure rate was about 6.8 percent in patients who received tamoxifen with low-risk disease based on the recurrence score, but the individual risk ranged from three percent to 12 percent, which is another strength of this test.

Although the NSABP usually refrains from subset analyses, supplementary information accompanying the New England Journal of Medicine paper (Paik 2004a) details several subset analyses. Questions arose about whether the Oncotype DX assay would work in patients with tumors less than one centimeter, patients older than 60 years old, and other subsets in which the statistical power is much less; however, the overall trends seem to show that the assay works in every subset we evaluated. It always seemed to divide patients into low- or high-risk categories, regardless of histology grade or tumor size.

Oncotype DX assay to predict response to chemotherapy

NSABP-B-20 included women with node-negative, ER-positive disease. It was a three-arm design, and patients were randomly assigned to tamoxifen alone or tamoxifen concurrent with either CMF or methotrexate followed by 5-FU. Our study was a retrospective analysis of that completed trial.

We only had tissue blocks available for approximately 30 percent of the entire study cohort, so it’s a subset; however, the subset and the entire cohort were comparable. We repeated the Oncotype DX assay on the tamoxifen arm to ensure the assay was reproducible, and we demonstrated that it is reproducible, which is encouraging.

Importantly, we evaluated the NSABP-B-20 chemotherapy arms to address whether the assay predicted chemotherapy responsiveness. We went into that study with an a priori hypothesis, based on the data presented at the 2004 ASCO by Dr Luca Gianni’s group in Milan evaluating samples from a neoadjuvant trial they performed with paclitaxel and doxorubicin.

They demonstrated a correlation between the Genomic Health recurrence score and pCR rate (Gianni 2004). The higher recurrence rate correlated strongly with the higher pCR rate. The overall pCR rate was approximately 25 percent in the patients with high-risk disease, and there was no pCR occurred in patients with low-risk disease.

We hypothesized that the benefit from chemotherapy in NSABP-B-20 would be almost negligible in patients with low-risk disease and high in patients with high-risk disease. The results of this study are actually quite striking and unlike anything I’ve ever seen (Paik 2004b). The absolute benefit from chemotherapy is actually negative in the low-risk group and zero in the intermediate-risk group. In high-risk group, the absolute improvement in distant recurrence at 10 years is 28 percent, or a relative risk reduction of 75 percent (5.1).

The data in the low-risk group are, in a sense, not relevant, because the baseline risk after tamoxifen is so low — 6.8 percent — so it’s a moot point of whether they need chemotherapy or not. In the intermediate-risk group the confidence interval overlaps with one, so whether patients with intermediate-risk disease gain any benefit or not remains a question.

Implications of the Oncotype DX assay study results

These data provide an important paradigm shift in the way we think about clinical trial design and patient management. So far, in most clinical trial designs, we presumed that the proportional benefit or incremental gain would be the same degree in patients with low-risk and high-risk disease. All statistical sample size calculations are based on that assumption, but now we have to change that.

It forces us to think about the clinical trial designs in which we preselect patients who are at high risk, because those are the patients who will benefit. We already knew from other studies that ER-positive patients do not benefit much from chemotherapy. In the neoadjuvant trials, the pCR rate is much lower in ER-positive tumors. This study definitely shows that, based on genes related to proliferation or estrogen receptor, we can actually select patients who are the best candidates for chemotherapy trials.

Benefit of chemotherapy in patients with ER-positive versus ER-negative tumors

In the NSABP-B-14 trial of placebo versus tamoxifen, patients had more than 10 fmol/mg of estrogen receptor by ligand binding assay, so these are all ERpositive tumors. We found that based on estrogen receptor messenger RNA quantitation by RT-PCR, we could actually identify patients who don’t gain any benefit from tamoxifen, and they were the patients with low levels of estrogen receptor. It actually correlates well with recurrence score because it’s heavily driven by the estrogen receptor pathway. Patients with a high recurrence score — approximately 25 percent of patients — do not gain any benefit from tamoxifen; however, we certainly need more studies before determining whether we can use the assay to rule out administering tamoxifen to those patients.

Clinical trials for patients with intermediate recurrence scores

Whether patients with intermediate recurrence scores will benefit from chemotherapy remains questionable. The Intergroup is designing a megastudy — the Program for the Assessment of Clinical Cancer Tests (PACCT) trial — with a sample size of 5,000 to 6,000 patients in the intermediate group. Patients will be randomly assigned to hormonal therapy alone versus hormonal therapy plus chemotherapy.

Predictive markers for specific chemotherapeutic agents

The Genomic Health assay does not identify any markers that predict response to specific chemotherapeutic agents. It will be interesting to see whether that can be done. I’ve been working with the NSABP trial in the neoadjuvant setting to determine whether we can use microarray gene expression profiling to predict treatment response. The Genomic Health study of neoadjuvant docetaxel by Luca Gianni’s group showed that proliferation markers are predictive.

Surprisingly, immune-related pathways — histocompatibility genes, the chemokines and immunoglobulin genes — are also somehow predictive. Our neoadjuvant study identified a specific subset of breast cancer that has a high fraction of this so-called immune pathway. I don’t know if it’s expressed by cancer cells or stroma cells, but they seem to have a high pCR rate. It will be interesting to see whether we can use these “blunt tools” of high-surface screening of gene expression or proteomics to sort out markers for response to specific chemotherapies. I suspect that may not be possible with these tools.

The hypothesis-driven studies, like those evaluating topoisomerase II, seem to be generating more interesting data. For example, the Danish group demonstrated that topoisomerase II actually predicts a relative benefit from CEF versus CMF. The MD Anderson study based on microarray analysis has identified a marker, Tau, which might predict response to paclitaxel (Pusztai 2004). We’ll need to determine the reproducibility of those types of markers, but I believe those hypothesis-driven studies will generate more individualized data for each drug.

Oncotype DX data and Ravdin’s Adjuvant! model

Peter Ravdin notes that, in the Adjuvant! Program, the relative benefit of chemotherapy is presumed to be equal for patients at higher and lower risk, but it’s likely that the estimation of chemotherapy benefit in the group with low-risk disease is an overestimation. Conversely, the benefit in the group with higherrisk disease may be underestimated. I believe our studies with Oncotype DX demonstrate this, and Ravdin’s model may need to be modified slightly.

My prediction is that when people see these data, they will want the assay performed because nobody wants to receive chemotherapy when it will not work. I’m sure a lot of competing assays are being developed that will claim to do the same thing. As a clinical trial group, we are interested in supporting all of those studies. In my lab, we are trying to develop competing assays that will be much less expensive and will be based on factors such as histology and estrogen receptors. We must demonstrate — in a clinical study in a stepwise fashion as we did with Genomic Health — that a marker is reliable and reproducible clinically so that patients will have confidence in the result.

Select publications

Dr Paik is the Director of the Division of Pathology at the NSABP Foundation Inc in Pittsburgh, Pennsylvania.