A new economic analysis comparing molecular diagnostics for breast cancer recurrence risk suggests that Agendia's MammaPrint is more cost-effective than Genomic Health's Oncotype DX.
The study, which marks the first head-to-head cost comparison of the two gene-expression-based tests, also underscores the difficulties associated with conducting this type of analysis and, according to at least one personalized medicine expert, highlights the need for academic researchers and payors to collaborate on such work in the future in order to better inform coverage decisions.
In the study, researchers led by Amalia Issa of the University of the Sciences in Philadelphia developed a 10-year Markov model to compare the costs incurred and quality-adjusted life years gained when oncologists used either Oncotype DX or MammaPrint to guide treatment decisions for a hypothetical cohort of 1,000 women with early-stage, estrogen receptor-positive breast cancer that hasn't spread to the lymph nodes.
The decision-analytic model, published in February in the journal Cancer, sought to gauge the cost-effectiveness of these diagnostics based on whether treatment decisions based on the test results resulted in the hypothetical patient cohort experiencing no recurrence, recurrence, or death. The data showed that the patients who received the Oncotype DX test to guide treatment spent an average of $27,882 and gained an average of 7.364 QALYs. Comparatively, patients who were tested by MammaPrint spent $21,598 and gained 7.461 QALYs.
Based on these results, Issa and her colleagues wrote in the Cancer paper that their model suggests that "MammaPrint is a more cost-effective gene-expression profiling test compared with Oncotype DX."
The study, which was conducted "from a third-party payer's perspective," assumed a threshold willingness to pay for breast cancer treatment of $50,000.
Although there have been a number of studies evaluating the cost-effectiveness of the two tests individually, the authors note that this is the first head-to-head comparison of them. And, since Oncotype DX is currently the market leader, they noted that their findings "have implications for health policy, particularly health insurance reimbursement decisions."
Speaking to PGx Reporter, Issa pointed out that as with all cost-effectiveness modeling, the study authors had to make certain assumptions in order to conduct their analysis. For example, while MammaPrint assigns patients into two categories — patients at low risk of recurrence who do not receive chemotherapy and patients at high risk of recurrence who do — Oncotype DX places patients into three categories — low, intermediate, and high risk — with no clear guidance for how to treat patients in the intermediate group. Noting that the "available data for inputs into [their economic]
model were limited," the researchers placed patients who received intermediate scores by Oncotype DX into the high risk category.
Noting that it isn't yet known whether patients who fall into the intermediate risk category by Oncotype DX will benefit from chemotherapy, Issa said that she and her colleagues placed these patients together with high risk patients in the study in order to reflect real-world clinical practice.
Patients who fall into this "gray zone of intermediate risk … typically are offered adjuvant chemotherapy," said Issa, who chairs the health policy and public health department and directs the Program in Personalized Medicine & Targeted Therapeutics at the University of the Sciences.
"We are talking about a life-threatening illness and if patients are told, 'We're not really sure whether your risk of recurrence is going to be high or low,' they are likely going to be risk averse and probably going to opt for chemotherapy," Issa added. "So, our analysis was based on that clinical reality and from the published literature where people seem to be doing that."
The National Cancer Institute-led TAILORx trial, slated for completion in 2015, is expected to yield clinical utility data showing whether patients who fall into this intermediate risk category live longer without experiencing disease recurrence when they are treated with adjuvant chemotherapy as opposed to with just hormonal therapy.
The TAILORx study "will give us higher-level evidence on whether the intermediate risk score is meaningful and what to do with it," Issa said. "In the absence of that evidence, our study [provides] decision makers some information as to how they should think about these kinds of issues in terms of gene expression profiling."
Issa and her colleagues conducted the study on their own and have no affiliation with Genomic Health or Agendia. Their work was funded by a grant from the Institute for Health Technology Studies, a non-profit that underwrites research on the value of medical devices and diagnostics. InHealth was founded in 2003 with donations from the medical technology industry and is currently funded with grants and voluntary donations from individuals.
For their economic modeling, Issa and colleagues gathered previously published validation and cost-effectiveness data for both tests, as well as data from studies in which the tests were compared to Adjuvant!Online, which assesses patients' recurrence risk based on age, tumor size, nodal involvement, and histologic grade.
Based on this literature search, the researchers considered 668 patients as being tested by Oncotype DX who were first classified as either high or low risk by Adjuvant!Online and then reclassified as either being high risk or low risk by the multi-gene expression assay. Of the 354 patients classified as low risk by Adjuvant!Online, 216 patients, or 61 percent, were found to be low risk by Oncotype DX, while 314 patients, or 47 percent, were found to be at high risk. There were 314 patients that fell into the high risk category by Adjuvant!Online, of which 122 patients, or nearly 39 percent, were reclassified as low risk by Oncotype DX, and 192 patients, or 61 percent, were reclassified as high risk by Genomic Health's test.
For MammaPrint, there were 302 patients in total who fell into the high or low risk category by Adjuvant!Online. Of 80 patients deemed as being at low risk of recurrence by Adjuvant!Online, 52 patients, or 65 percent, were reclassified as low risk by MammaPrint, and 28 patients, or 35 percent, were found to be at high risk by the gene expression test. Within the 222-patient high risk Adjuvant!Online group, 59 patients, or 27 percent were deemed low risk by MammaPrint, and 163 patients, or 73 percent, were high risk by Agendia's test.
In the model, the researchers assumed that 90 percent of patients who were found to be at high risk by both Adjuvant!Online and Oncotype DX received chemotherapy, and 90 percent of patients deemed to be at low risk by the two tools did not receive chemotherapy. For cases where the Adjuvant!Online prognosis and the molecular tests results disagreed, the researchers assumed that 50 percent of the population received chemotherapy. The same criteria were applied for MammaPrint.
The relative risk reduction for high and low risk patients due to chemotherapy, as well as the costs associated with chemotherapy-related toxicities, were also taken from the literature. Issa and colleagues applied the list prices of $3,975 for Oncotype DX and $4,200 for MammaPrint.
In order to establish the statistical significance of the study, Issa and her colleagues performed sensitivity analysis by running a Monte Carlo simulation 1,000 times. The researchers then represented this data as an incremental cost-effectiveness scatter plot that graphically compared the costs and the effectiveness of the two tests.
The incremental cost effectiveness plot is divided into four quadrants. Trial points that fall in Quadrant I are the instances when Oncotype DX is more costly and more effective than MammaPrint; Quadrant II is when Oncotype DX is more costly and less effective than MammaPrint, Quadrant III is when Oncotype DX is less costly and less effective than MammaPrint; and Quadrant IV represents scenarios when Oncotype DX is less costly and more effective than MammaPrint.
"What we found when we looked at all the different trial points for our study … [is that] 82 percent of trial points fell into Quadrant II," Issa said. "So, the vast majority of our data points fell into Quadrant II, which suggests that MammaPrint is the dominant strategy."
Furthermore, the sensitivity analysis showed that overall treatment costs are statistically different, with mean costs of $27,882 and a standard error of $1,455 for Oncotype DX; and mean costs of $21,598 and a standard error of $1,246 for MammaPrint. Effectiveness was also statistically different, with a mean effectiveness of 7.36 QALY for Oncotype DX and mean effectiveness of 7.46 QALY for MammaPrint. Standard error was 0.07 in both cases.
"In the base case analysis and in the probabilistic sensitivity analysis of recurrence rates under the two testing strategies, the MammaPrint test dominated the Oncotype DX test," the study authors wrote in the Cancer paper.
The authors believe that the study gives health insurers something to think about when making coverage decisions about these tests.
Most major private payors in the US reimburse for Oncotype DX in the node-negative setting. The company claims that payors representing more than 200 million lives in the US cover the test. Recently, Agendia CEO David Macdonald indicated that the company has a similar number of lives covered for MammaPrint in the US, and the company is working to expand reimbursement for the test by conducting cost-effectiveness studies (PGx Reporter 4/25/2012).
Issa pointed out that her team's cost effectiveness analysis is particularly meaningful in the context of the Patient Protection and Affordable Care Act, under which payors are under pressure to cut costs without sacrificing patient outcomes.
"We felt it was important to do this kind of study to provide empirical data about novel genomic technologies that are being used currently in clinical practice," she said. "As molecular diagnostics become more broadly available," this type of cost-effective analysis is "going to have major implications for healthcare delivery.
"One thing that I hope will come out of this study is that more research and other evaluations are needed prior to the adoption of new genomic diagnostics," Issa added.
The study authors noted several limitations of their analysis, including grouping patients with intermediate Oncotype DX risk scores with high-risk patients; not breaking out cost-effectiveness based on whether the hypothetical cohort experienced local or regional recurrence or a secondary primary or contralateral breast cancer; and modeling only one relapse per patient.
"Thus, there may be an underestimate of relapse costs if certain patients developed more than one relapse (and approximately 30 percent of patients with breast cancer do develop more than one relapse)," the researchers wrote.
The study authors' decision to group Oncotype DX intermediate risk scores with high risk scores may reflect the current clinical reality that in the absence of clearer guidance, a large portion of intermediate risk score patients receive chemotherapy. However, it's possible that this assumption could have impacted the cost-effectiveness conclusions in the Cancer paper for Oncotype DX.
"I wonder if the difference between the assays found in this cost-utility study was related to the fact that the 21-signature analysis resulted in more patients being classified as high-risk patients (thus receiving chemotherapy), than the MammaPrint test, when both were compared to clinical-pathological risk assessed by the online program Adjuvant!.Online," said Jonas de Souza, an instructor in the University of Chicago's department of medicine who also conducts clinical and outcomes research for cancer treatments.
"This may have been related to the fact that intermediate risk patients were grouped with the high-risk patient group in the 21-gene analysis," he said.
Overall, De Souza said that Issa and her colleagues produced a "good manuscript that aims to answer an important question." However, he suggested that if the study authors had conducted a second analysis that grouped a portion of the intermediate risk patients with the low-risk group, they may have arrived at a "completely different" cost-effectiveness conclusion.
The Right Data
The limitations in the analysis published in Cancer are not uncommon for cost-effectiveness studies done by academics. For Howard McLeod, director of the University of North Carolina's Institute for Pharmacogenomics and Individualized Therapy, this highlights the fact that researchers rarely have access to the data they need to perform the type of cost-effectiveness analysis that will fuel evidence-based coverage decisions.
For example, Issa and colleagues used list prices for MammaPrint and Oncotype in their model because they were readily available, but the cost value that would yield the most accurate cost-effectiveness estimate is what payors are actually reimbursing for the tests.
"We have something on the order of 2,000 health insurance [payors] in this country. So, they have different reimbursement plans based on what contracts they hold with providers. It would be an incredibly arduous task to be able to get all that data, assuming [insurers] would even provide it to us and analyze it for reimbursement," Issa said, adding that insurers are generally unwilling to publicize their contractual payment agreements with providers. "So our decision to go with the list price was based on practical and pragmatic reasoning as to what is feasible and what data is available."
Data access issues can also limit the level of detail in a cost-effectiveness analysis. For example, the incremental cost effectiveness analysis doesn't specify the clinical scenarios under which one test may be less costly and more effective over another.
"Whenever you do this type of analysis, you do this at a population level. So, I can't specifically say that X percent of patients had these clinicopathological characteristics, versus Y percent had these characteristics," Issa explained. .
However, insurers would likely want more granularity if they wanted to provide coverage for one gene expression profiling test over another based on a head-to-head cost-effectiveness comparison.
By using similar variables to compare the two tests, the analysis by Issa and her colleagues may be meaningful for the "typical breast cancer patient … but they didn't look at a lot of the other characteristics of the cancer patient," observed McLeod. "Partly because of the lack of data and in trying to take a straightforward or simple approach, they didn't really [look at] node-positive, node-negative, HER2 positivity versus no HER2, and all these different issues."
When the two tests are separately considered, a number of studies have found both Oncotype DX and MammaPrint to be cost-effective tools under different clinical scenarios. For example, a study published in the American Journal of Managed Care in December 2010 found MammaPrint "is likely to be a cost-effective strategy to guide adjuvant chemotherapy treatment in younger patients with early-stage breast cancer."
In the case of Oncotype DX, an analysis published in 2010 in Cancer found that treatment decisions guided by the 21-gene RT-PCR assay had "greater efficacy with acceptable cost-effectiveness ratios" compared to when node-negative early-stage breast cancer patients were given just hormonal treatment. Additionally, Oncotype DX-guided treatment decisions had "similar efficacy and lower cost" when patients were treated with chemotherapy and tamoxifen, the study found.
Furthermore, Issa and colleagues used QALYs as a measure of effectiveness in relation to the cost of the tests, but payors may not use QALYs to gauge cost-effectiveness over the lifetime of the patient, choosing to look at actual costs associated with interventions over a much shorter time frame.
"While QALYs are convenient for academic analysis, they are not necessarily endpoints that payors live by," McLeod said. "So, if you get a premenopausal patient who is in their 40s and they theoretically have another 30 years left, just the sheer brute force of time suddenly makes almost anything look like an improvement."
Instead, payors are more likely to consider the "absolute costs around the five-year treatment window, or even in one year if there is going to be more utilization in one group over another," McLeod added.
He noted that there needs to be more collaboration between payors and academic researchers to perform comparative-effectiveness and cost-effectiveness analysis. "When looking at these studies, [we should] really issue a call for more interaction with those groups," he noted. "You really need to be doing this backwards, by first learning what kind of data is needed by the end users, including insurance companies, and then do the studies that generate that data."
Overall, McLeod lauded Issa and her colleagues for the study, but added that "there is more to do to make sure that the end result is not happy readers, but rather happy users of the data."
Issa noted that she is currently involved in conducting another grant-funded study analyzing the practice patterns and utilization of gene expression profiling tests for breast cancer in a large health system.
Although she didn't provide further details this investigation, she noted that it "will analyze the real-world impact of these gene expression profiling tests."