NEW YORK (GenomeWeb) – A team led by Johns Hopkins Kimmel Cancer Center researchers has developed a model that uses routine pathologic parameters to identify breast cancer patients for whom molecular-based recurrence testing with Oncotype DX is likely to provide added information.
As the team reported yesterday in the Journal of Clinical Oncology, the model relies on histopathological factors like estrogen receptor, progesterone receptor, Ki-67, HER2, and grade findings to predict whether Genomic Health's Oncotype DX would place a patient at high or low risk of tumor recurrence.
While the team's model was able to place a portion of patients into high- or low-risk categories, it's that gray area of patients who the model couldn't place where molecular testing might offer additional information. This, the team added, could shift the population for which such tests are ordered.
"The concern many of us have is that still today, many years after several of these [molecular] assays have been developed, we don't know, optimally, how to use them in a way that adds to the information that we already have — in this case, with high-quality pathology — and not simply replicates information that is already available," Hopkins' Antonio Wolff told GenomeWeb. "The key issue is, we are not trying at all to use these assays less, we are trying to use them better in a manner that further informs clinical decision-making."
Genomic Health's Oncotype DX test examines the expression of 21 genes in samples from breast cancer patients to assign a recurrence risk score and guide chemotherapy. That score is given on a scale of one to 100, with one being a low risk of recurrence.
The TAILORx trial has reported that Genomic Health's Oncotype DX test results can stratify treatment for women with breast cancer and that most women with early-stage breast cancer who have a low-risk score remain disease-free after five years of hormonal therapy. Similarly, the Plan B trial reported that patients with high clinical risk, but low Oncotype DX recurrence scores who didn't receive chemotherapy had good three-year survival rates.
But Wolff and his colleagues suspected that, in some cases, physicians might already have the information they need to estimate a patient's recurrence risk category without the molecular test. The researchers developed two models, a linear model and a random forest model, and trained them on a set of 1,113 patients with ER-positive, lymph-node negative, stage 1 or stage 2 breast cancer to estimate patients' recurrence score. Here, the researchers defined a low-risk score to be 25 or lower and a high-risk score to be above 25.
For each patient, they had access to histopathological data as well as their OncotypeDx Recurrence Score. Tumor grade, HER2 status, and Ki-67 were positively correlated with the Oncotype DX Recurrence Score, the researchers reported, while ER status and progesterone receptor status were inversely correlated with recurrence. Age and tumor size, meanwhile, were not significantly correlated with the Oncotype DX Recurrence Score and were not included in the models. The histopathological markers, the researchers reported, could determine risk category with greater than 95 percent confidence in more than 55 percent of cases.
After training, the models were locked and tested on an additional cohort of 472 patients, drawn from the same hospitals as those in the training set. For these, the researchers reported that the random forest model outperformed the linear one. The model correctly predicted risk category in 52.5 percent of cases, with an accuracy of 97 percent.
For the other 224 patients, the researchers' tool was unable to predict what their Oncotype DX score would be. It's for this group of patients, whose pathological findings are less clear, that the researcher said Oncotype DX testing could be the most informative.
Wolff and his colleagues further developed their tool into a web-based app, dubbed the Breast Cancer Recurrence Score Estimator.
This app, they added, could shift test ordering patterns. The researchers applied their tool to a set of 939 patients from Johns Hopkins. Of these, 299 had undergone Oncotype Dx testing. The app, though, placed 297 of them in the undetermined risk group that the researchers say might benefit from molecular testing. However, only a portion of the cases that had sought and were recommended for testing through the app overlapped. Only 127 of the 299 cases with Oncotype DX results would have been recommended for testing if the app had been used first, the researchers reported.
"This was fascinating for us because, at the end of the day, we would still do the same amount of testing, but we would change the patient population to include individuals where the test would help us with the most information that was not already available," Wolff said.
They encourage clinicians seeking to use their app to first test it using patient data they've already collected to ensure it works in their particular patient population.
In an email to GenomeWeb, a Genomic Health spokesperson called this study and others like it "interesting," adding that the Breast Cancer Recurrence Score Estimator draws on data that is variable between sites and doesn't gauge outcomes like chemotherapy benefit. She noted that the Oncotype DX test is backed by clinical evidence supporting its clinical utility and cost effectiveness to inform treatment decision-making by patients and physicians.