Skip to main content

Breast Cancer Prognostic Test Comparison Considers Strengths, Weaknesses of Different Signatures


NEW YORK (GenomeWeb) – A statistical analysis comparing the prognostic power of six signatures found that all of them are good predictors of breast cancer recurrence 10 years after diagnosis and identify node-negative patients at low risk of disease recurrence, but some signatures may be better than others in the node-positive population and in assessing recurrence risk during the late follow-up period.

Researchers led by Ivana Sestak from Queen Mary University of London compared the prognosis signatures — a score based on clinical features of cancers, and five molecular tests, including Genoptix's IHC4, Genomic Health's Oncotype DX, BioTheranostics' Breast Cancer Index (BCI), NanoString's Prosigna, and Myriad Genetics' EndoPredict.

Susan Hilsenbeck, a biostatistician from Baylor College of Medicine, reviewed the data presented at the San Antonio Breast Cancer Symposium last week and said that the study demonstrated the importance of not using the molecular prognostic signatures by themselves, but rather in the context of clinical information on the patient and her tumor. The incorporation of clinical information with molecular signatures becomes particularly important, she observed, when gauging recurrence risk between five and 10 years after diagnosis, and in women whose disease has spread to the lymph nodes.

Despite the availability of multiple breast cancer recurrence tests, there isn't clear evidence of their comparative ability to determine prognosis in the first 10 years after diagnosis and their ability to predict late recurrence — between five to 10 years after diagnosis — in node-negative and node-positive women. The first piece of information can help guide doctors as to whether a patient should receive chemotherapy, and the latter can help guide whether to extend endocrine therapy for a woman.

However, comparisons are challenging, since the six signatures analyze different biomarkers and each have their own ways of reporting results. For example, IHC4 factors in four markers; Oncotype DX gauges the expression on 21 genes; BCI factors in genes in the estrogen signaling pathway and five proliferative genes; Prosigna uses the PAM50 gene signature; and EndoPredict analyzes the expression of eight disease-related genes. Prosigna and EndoPredict factor in different clinical features of the tumor.

Oncotype DX reports high-, intermediate-, and low-risk scores. Prosigna reports high-, intermediate-, and low-risk scores in node-negative patients, and high and low risk in node-positive patients. BCI reports overall distant recurrence risk across three categories, but late distant recurrence as only high and low risk. EndoPredict gives either a high- or low-risk score.

The one commonality for all these tests is that they've either been developed (clinical treatment score and IHC4) or validated (all other tests) using samples from the translational substudy of the Arimidex, Tamoxifen, Alone or in Combination trial (transATAC). Sestak and colleagues — a number of whom have accepted grants and speaking fees from, are employees of, or own shares in companies that developed these tests — compared the prognostic capabilities of the six signatures for around 800 women previously enrolled in transATAC.

Using cox regression models, a statistical method for measuring the prognostic power, researchers found that in node-negative patients "all six signatures provided significant prognostic information for distant recurrence," with Prosigna and BCI the strongest predictors, Sestak said at the meeting.

In node-positive disease, "the prognostic strength was weaker," she said, though the clinical treatment score and EndoPredict came out on top in terms of providing significant distant recurrence risk information. All the molecular tests added only a small amount of prognostic information beyond the clinical treatment score, she observed.

BCI, Oncotype DX, Prosigna, and EndoPredict all placed the majority of node-negative patients into low-risk groups. However, in the node-positive population, only EndoPredict and Prosigna "provided good discrimination between risk groups," Sestak said. "These two signatures identified a good proportion of women who were at low risk of distant recurrence." She added, however, that the risk stratification capabilities of BCI and Oncotype DX were not as clear for node-positive women, and most patients fell into the low-risk group.

In evaluating the prognostic power of these signatures between five and 10 years after diagnosis, the statistical analysis showed IHC4 and Oncotype DX to the be weakest signatures in node-negative patients, while Prosigna performed the best, followed by EndoPredict and BCI.. In the late follow-up period, Sestak said only Prosigna "provided good discrimination between [risk] groups."

In node-positive women, only the clinical treatment signature and EndoPredict provided significant prognostic information. According to Sestak, EndoPredict and Prosigna again performed the best in this subgroup, while the performance of BCI and Oncotype were not as clear.

Based on this information, Sestak concluded that molecular prognostic signatures should not be looked at in isolation and that the incorporation of clinical information would improve their prognostic power. Though the study was intended to compare breast cancer prognostic tests, when asked by a meeting attendee as to which test is best, Sestak declined comment. She noted, however, that it's important to look at node-positive and node-negative women separately, as these signatures clearly perform differently in these subpopulations.

Developers of newer tests have jumped on the chance to use these study results to highlight the advantages of their products over the current market leader, Oncotype DX. NanoString issued a statement noting that in the analysis by Sestak and colleagues, Prosigna came out on top in assessing 10-year recurrence risk and five-to-10-year recurrence risk in node-negative women.

Myriad, which recently added EndoPredict to its pipeline of tests through its acquisition of Sividon, took the opportunity to discuss the study's positive findings with regard to its tests — mainly that EndoPredict fared better than Oncotype DX in its ability to assess recurrence risk in five to 10 years post-diagnosis, and that low-risk, node-positive patients had a much lower rate of recurrence when assessed by EndoPredict compared to Oncotype DX.

Johnathan Lancaster, chief medical officer of Myriad Genetic Laboratories, echoed what other competitors have previously said, that oncologists and patients don't like Genomic Health's "intermediate" risk category. Getting an intermediate result "often puts patients into a tail spin," Lancaster told GenomeWeb, noting that EndoPredict provides a more straightforward high- and low-risk categorization.

GenomeWeb has previously covered the relative advantages and disadvantages of a continuous recurrence risk score and a binary result. Those that prefer Oncotype DX's continuous score, say they do so because it allows them to weigh the test score in the context of clinical and other factors. Lancaster noted that EndoPredict also reports a raw score for doctors who want to use that instead of a high- or low-risk categorization.

EndoPredict is the market leader in breast cancer prognostic testing in Europe, which is why Myriad became interested in it. Myriad aims to launch it in the US in the first half of 2017, but is still evaluating whether it will launch EndoPredict as a lab-developed test or a kit. 

Meanwhile, Sestak highlighted some of the weaknesses of the statistical comparison. For example, the clinical treatment score and IHC4 were developed using transATAC samples, so their prognostic capabilities might be overestimated. This raises the bar for the other tests, and "raises potential credence at the end for independently defined signatures," Baylor College of Medicine's Hilsenbeck said at the meeting in her review of the study.

She noted that likelihood ratios are "very useful" in comparing contributions of various biomarker signatures within a dataset, such as transATAC, but "a larger likelihood ratio does not necessarily imply a larger effect size."

Hilsenbeck pointed out that many molecular markers lose prognostic value over time, unlike clinical tumor features, such as node involvement and tumor size. "We should expect that signatures that incorporate clinical data, which tend to continue to have value over time and were developed on datasets with very long-term follow up, are going to be more likely to retain their value," she said.

Genomic Health Chief Scientific Officer Steve Shak told GenomeWeb that Sestak's group is comparing apples and oranges in their study precisely because some tests factor in clinical information while others, like Oncotype DX, don't. "We've done work where we've combined the recurrence score with other factors, and [that data] wasn't even looked at," Shak said.

He added that the researchers' analysis failed to address if the statistical differences between the tests translate to differences that are clinically meaningful. Further, Sestak's group combined all node-positive patients in their analysis and included those with up to nine positive nodes, which would have influenced the study findings.

Oncotype DX is not offered for patients whose disease has spread to four nodes and beyond. This subset of patients should receive chemotherapy and therefore wouldn't need any molecular diagnostic. This issue came up following Sestak's presentation, but she responded that the number of node-positive patients were so few that she didn't think it would have had much impact on the study findings.

Having been on the market since 2004, Genomic Health has amassed a hefty evidence base supporting Oncotype DX, a test that 700,000 patients have received to date. And even though there are conflicting guidelines with regard to the use of the test in node-positive patients, clinicians are increasingly using it in this subpopulation, Mayo Clinic's Prema Peethambaram reported at SABCS.

Data from 15,000 patients who received testing in the National Cancer Database showed, however, that Oncotype DX scores influenced the use of adjuvant chemo in patients with one positive node, but didn't influence chemotherapy use in patients with two or three positive nodes.

Moreover, clinicians tend to use Oncotype DX mostly to gauge risk in the first five years following diagnosis. But a Genomic Health spokesperson highlighted a recent paper in the Journal of Clinical Oncology that reported the experiences of patients who were followed for more than a decade, and that analysis backed the test's ability to gauge recurrence over the long term. "In deciding the clinical utility of any test, we look at the body of evidence, and not any single study," Shak said.

Overall, Hilsenbeck found that the conclusions drawn by Sestak and colleagues were reasonable and make a strong case for accounting for clinical information in addition to molecular signatures. But uncertainties still remain for oncologists, since prospective studies, such as those evaluating the ability of Oncotype DX to predict which patients will benefit from chemotherapy, aren't yet completed.

"We need consensus guidelines on how best to blend signature and clinical data," Hilsenbeck said. "And we need to know more about how high-risk patients would actually benefit from chemo or extended therapy."