publication date: Jul. 7, 2017
Conversation with The Cancer Letter
Keith Flaherty: We will certainly stop if we enroll all subprotocols
ECOG-ACRIN study chair
Director, Developmental Therapeutics,
Massachusetts General Hospital Cancer Center
Associate professor of medicine, Harvard Medical School
NCI and ECOG-ACRIN officials said the NCI-MATCH trial will keep going, continuing to match patients with treatment arms based primarily on their molecular characteristics.
The institute will no longer pay for genotyping—which it has done to biopsy and genotype nearly 6,000 patients—but it will make use of genomic sequencing that’s being done by commercial labs and at some cancer centers to guide clinical care.
No data on prevalence of genotyping exist.
Keith Flaherty, the ECOG-ACRIN study chair and director of Developmental Therapeutics at the Massachusetts General Hospital Cancer Center, and associate professor of medicine at Harvard Medical School, ventures a guess that about 100,000 tests are performed every year.
“Would it be possible for us to try to feed off of sequencing outside of the trial to be able to fill the particular rare arms? We recognized from the outset that we did not have insight into the alignment between our assay and the genes it sequences, and exactly how these same genes are being analyzed with various other tests, be they academic medical centers or for-profit companies,” Flaherty said.
“We began to communicate with entities of both types to see if they’d be willing, essentially, to share their analytic approach with us, and help us try to understand if it would be possible to align their testing with the set of alterations that we need to have tested, to be able to identify which of their patients could be triaged to our therapeutic subprotocols.”
Flaherty spoke with Paul Goldberg, editor and publisher of The Cancer Letter.
I hear there is a big change in the MATCH trial. Could I ask you to walk me through it?
I guess we start with the major milestone, which is that the trial has reached its screening goal, where we performed centralized screening of patient tumor samples. The goal from the outset was to screen several thousand patients, with the idea that we would then find a fraction of those patients for whom we had drugs being matched to molecular alterations.
We had an approximate goal for the percentage of patients that we were hoping to be able to assign to therapy, but that number was not pre-specified, because we couldn’t have known going into the trial exactly how many patients that might be, but we knew how many patients we could screen.
Just to remind you, the initial number when we very first were designing the trial in collaboration with colleagues at the NCI—3,000 patient screened was the initial number, and then after the interim analysis, that number was increased to 6,000, which is the goal that we reached.
What was the biggest surprise here for you?
I guess number one was the really striking amount of demand in the national research community. Remember that this is a trial that was conducted broadly across the clinical trial network that is connected with the NCI-funded cooperative groups, which you may know now includes four major groups focused on cancers in adults.
Then, affiliated with each of those four major groups are very large academic medical centers, but also a full spectrum of cancer care-providing centers, down to smaller hospitals and even private practices that participate in clinical research as well.
That’s a very large menu of potential sites, and we enlisted nearly 1,100 of them ultimately. It was open to the entire network, but that was the number that basically stepped forward and registered themselves specifically for this trial.
That’s one measure of demand. The other was the amount of enthusiasm at those sites, and the rate at which enrollment happened, which was far faster than has ever been seen in any adult trial in the cooperative groups, or for that matter, in any trial, adult or pediatric. The rate per unit of time was definitely surprise number two.
What about prevalence of these mutations?
This has never been done before that an advanced cancer population who have exhausted the available therapies have been characterized at this scale. The significance of that is in relation to relatively recent analyses of this patient population, particularly genetic analyses of this patient population. When I say genetics, I mean the tumor genetics.
The studies that had been conducted and published in the past several years have been almost completely of surgical specimens, so-called primary tumors, as opposed to patients who have metastatic or advanced disease, who have essentially received and unfortunately are no longer benefiting from available therapy.
In this so-called relapsed, refractory population, there have not been large-scale analyses done. We wondered about the potential shift from the representation of certain mutations or genetic features in this population we were screening versus published datasets.
Having said that, while there was a shift in relation to previously published data in primary tumor surgical specimens, it was not a massive shift. It was just different than the estimates we had. Not so much of a difference for the particular common genetic features. Those tended to track reasonably closely, but for some of the less common to rare aberrations, they turned out to be in some cases even less common and even more rare than previous estimates would have suggested.
How has the standard of care changed in the time you’ve been doing this study?
You’re most probably thinking of molecular targeted therapy that’s based on a genetic understanding of cancer. I would say in that regard there have certainly been a few more drug approvals over the three-year interval in which we were first planning, and then finally executing the study in August of 2015.
There have been a few more approvals of specific targeted therapies in specific cancer populations who have certain genetic features, but that joined a larger number. It wasn’t as though there was a big watershed event of new drugs being approved.
Numbers continue to trend higher over time, so that’s, I guess, one answer. The other is, as you’re well aware, infiltration of immune therapies, particularly PD-1 or PD-L1-blocking antibodies, in a rising number of cancer subpopulations, with approvals racking up in exactly this time frame of the few years in which we were designing and executing the trial.
Do either of those changes significantly change the field in terms of the percentage of patients who still, unfortunately, exhaust their available treatment options? I’d say generally no, but I think that the broad population they will impact is notable, but not large, in terms of new treatments that became available during this time frame.
I was more asking about the prevalence of sequencing.
In terms of in clinical practice?
That’s basically not an easy question to answer in a very comprehensive way. What I would tell you is that there’s a handful of major academic medical centers, including my own, that have developed the ability to do so-called next-generation sequencing as a part of routine clinical care.
It continues to be a challenge to negotiate with payers for payment of such testing. Beyond that handful of medical centers, who tend to lose a good deal of capital doing that type of analysis to be able to identify treatment options, there are commercial laboratories of which I’m sure you’re aware. Foundation Medicine and Caris Life Sciences are the two largest in terms of the amount of testing that they do per year.
You have to refer to their annual reports in terms of how many tests they run, but I would give you a ballpark estimate to say about 100,000 tests per year, and that, of course, is distributed across the entire U.S., and actually outside the U.S. population, as well.
If you take that number combined with a handful of centers that have in-house capacity, it’s still a reasonably small minority of patients who are undergoing this type of testing in routine clinical practice. We don’t entirely know, if you look across the entire landscape of oncology in the U.S., why this is the case.
I’d suggest that payment from payers is one potential major impediment, but it also could be that doctors and patients aren’t motivated to do it because they don’t see a likely change in the treatment options that they might face.
That’s still 100,000 tests per year. What was it back when you started the trial? What’s your guess? It could have been a fraction.
I would say probably half that. I’m ballparking it, but if you go back to trial conception, I would say it would be about half that number.
Still, it’s quite a pool.
Yeah, sure. I think if you were to ask the question what’s happening with that population of patients who are identified…where do they go subsequently in terms of treatment, and I don’t just mean changing practices, but what treatment do they receive? Do they receive therapies that are related to those findings? If so, do they receive them outside of a clinical trial context or within a clinical trial?
We don’t have really adequate capture of what happens downstream of this testing. In fact, that’s what payers will oftentimes say in public discussions on this topic, is that they also don’t know, essentially, what the impact of testing is in terms of navigating patient care.
That’s, to a degree, a motivator behind the ASCO TAPUR effort, not to create a completely comprehensive database, but to at least try to offer the opportunity for patients who are undergoing so-called standard of care sequencing the opportunity to receive treatment in a registry type of mode where their treatment, and then to a degree their outcomes, are being captured.
How was the decision to extend and expand the trial made?
We knew relatively early in the study, as we were capturing the types of cancer patients who were being screened, and then the early results in terms of genetic analysis.
We had a much more defined rate of genetic alterations in our population than we could have estimated from the previously published data on primary tumors.
It allowed us to basically recalculate, okay, if you look across the arms of the study, ultimately swelling it to 36 subprotocols, we range from ones that are sufficiently common that we would easily achieve the pre-specified goal of 35 patients enrolled to each arm within even 3,000 patients, but we certainly also had cases where there were arms where the rarity of the alteration was going to mean that we weren’t possibly going to fill or even come close to filling with 6,000 patients screened.
We looked at the full spectrum of them and asked the question. This is in discussion with the NCI, of course, who are the funders of the study. Basically, what might be a reasonable sample size to grow to try to complete accrual of at least 35 patients per arm?
The discussion began early in the trial. We were reckoning with these uncommon genetic features that there is this foundation of testing being done at both major academic centers and commercial laboratories in a broadly distributed way.
Would it be possible for us to try to feed off of that sequencing to be able to fill the particular rare arms? Recognizing that we, at the outset, did not have insight into the alignment between our assay, the genes that were being sequenced, and exactly how they were being analyzed with various other tests, be they academic medical centers or for-profit companies.
We began to communicate with entities of both types to see if they’d be willing, essentially, to share their analytic approach with us, and help us try to understand would it be possible to align their testing with the set of alterations that we need to have tested, to identify which of their patients can be triaged to our therapeutic subprotocols.
That’s the communication we began while we were still centrally screening, hoping to be able to tee up what we’ve launched as of quite recently. Just before the end of the 6,000-patient screening, in fact, is when we activated this so-called rare variant initiative.
For those rarest of alterations, we set a line for the subprotocols where we would basically begin accepting results of outside tests. That includes Foundation Medicine, Caris, and then two academic medical centers to start.
What’s the new stopping rule? Is it set for each cohort? How is it set?
We actually didn’t declare a stopping rule. Remember that the first motivating principle was that we took therapies into the trial that were already showing efficacy in patients, at least in one tumor type. Could have been more than one tumor type. In other words, futility was already not a concern of having totally inactive therapies in a biomarker-defined population based on a priori evidence that was brought to us by investigators in companies who had been testing these drugs.
The issue in setting up a statistical design for each subprotocol in the 35 patients is that it was nearly impossible from the beginning to predict what types of cancers we would accrue to those arms, because the sponsors who were bringing these drugs forward had not generally run broad pan-cancer studies.
They couldn’t quote to us what they thought the representation was across the cancer population. Recognizing that we might get three patients with one cancer type, two with another, and a bunch of ones, all the way up to 35 patients in the most widely distributed example, we couldn’t really justify a stopping rule where, if we went 0 for 1, 0 for 1, 0 for 1, 0 for 1, in a set of different cancer types, that wouldn’t rule out the possibility that the cancer type that showed up with three patients might actually be one that’s still potentially responsive to therapy.
This is one of the motivating principles behind the trial, that there might be heterogeneity of effect, as has been well described for BRCA mutations across certain cancer types.
We didn’t have a priori rules to stop the subprotocols because we didn’t have efficacy, they said, because there was a priori evidence of efficacy on some number of patients. Then we were very keen to try to understand this cross-cancer effect, either homogeneity or heterogeneity of effect.
When do you stop it? Do we know that?
We’ll certainly stop if we get the patients needed for all of our subprotocols, now relying on the outside assay approach to find those rarest of variants. Once we complete all 36 of the subprotocols that have been launched or are in the process, then that would be the formal end of the study.
Why these four labs? Your institution’s lab, for example, is not on the list. Is that going to be expanded?
Possibly. Not yet decided, though. There’s been a broad agreement to try to expand if we can feasibly do so. It’s no small feat to look at every analyte. Let’s say every gene and every exon within a gene to understand basically the parameters of testing that are being done, and then be able to have demonstration of the methods and the output from each of these laboratories.
Literally, we need to ensure that there is complete coverage of the required analytes that we built into our assay in each of these laboratories. This is essentially a manual process whereby we effectively compare the databases of the centralized NCI-MATCH assay to these outside laboratories.
If we’re satisfied that complete coverage is the case, which we will learn in the demonstration project, then we will be in a position to take on additional laboratories. This information is not publicly available, even from non-profit academic medical centers, and certainly not for the for-profit labs, so laboratory expansion will require individual negotiations with each new entity.
We needed to literally set up the process to do this because it’s never been done, to our knowledge, to compare and contrast these platforms. We have set up the process now and are completing it through the two for-profit labs and the two academic labs.
This now allows us to contemplate the idea of carbon copying the approach and taking on other labs.
Do I understand that the match is going to be harder to reach because you’ve already enrolled some of the more prevalent mutations, so you’re chasing rarer mutations?
That’s right. By definition, the likelihood out of any 100 or 1,000 patients who are tested, the number who are going to have the alterations that we are now looking for is inherently much lower. The testing that’s happening is outside of our purview. This is next-generation sequencing that is happening because doctors and patients have decided that they want to do it, for whatever reason that they want to do it.
Then it’s on the back end of that testing that if they have an alteration that’s relevant for NCI-MATCH, then they could consider the idea of approaching the study and being screened for it.
That’s a fundamentally different paradigm than when we were doing centralized screening, where basically patients were signing up and consenting for the entire process.
You were paying for it, and you’re not now?
But it’s being done, so that it’s feasible.
Right. To the numbers we discussed before, it’s already the case that next-generation sequencing is a part of practice for a portion of the cancer population. We are now aligning the rare variant effort to be able to feed off of that.
Do we know who is sequencing and who is not by profile? Is it academic medical centers that are doing the sequencing for the most part? Who is more likely to get sequenced?
Yeah, it’s a great question. I’m not sure I’ve ever seen data presented that really breaks it down quite that way. My sense is that it’s pretty evenly distributed, which is to say that, as you know, two-thirds to three-quarters of cancer patients are treated in a community context. They’re outside of major academic medical centers. My sense is that that’s about proportionate to the amount of testing that’s done, if you focus on the commercial labs, because the commercial labs are the only available option for them.
Again, still a relatively small number of academic medical centers can have their own molecular pathology groups doing this type of testing in a CLIA-approved environment.
Whose idea was it to expand the trial, or extend the trial?
I would say that just as it is with all major decisions in MATCH, it was mutually our idea with NCI collaborators. I wouldn’t say there was any one person who was responsible for that.
It was just an obvious thing because of what you’re facing.
Frankly, just even going back to the early days of launching the study, there were questions from the very beginning from physicians in various centers, be they academic medical centers or in the community, who were wondering if we would take the results of outside testing.
This was already a concept that was bubbling up in a more democratic way. Of course, in the very early days of the trial, it was a very simple answer that we were going to take ownership of the testing centrally to try to have quality control assured by doing it that way. The desire to be able to feed in patients who were getting testing otherwise, I would say, was there from before the beginning.
What are the incentives for patients to enter this trial? You get the drug, I guess, is that how it works?
You get testing, and you get the drug. Remember, the testing that we reported back to sites when we were doing the centralized testing was basically a report of all the alterations that we were testing. The doctors and patients would have that information and then, effectively, could do with it what they wanted.
If they had a so-called actionable alteration for which we had a treatment arm, then of course we offered that as a next step, meaning the patient could then be consented and consider proceeding with the therapeutic arm.
I would break it down into two steps, that, basically, first we were offering the testing at little or no cost to the site or the patient.
Then for those patients who had alterations, we were offering them a therapy. If you consider that there’s some motivating principle behind up to 100,000+ people getting tested a year, currently absent a trial effort like this, then you could say that we are at least serving that function for all patients who are getting tested at our participating sites. Then, for those who had a treatment assignment potential, they will be offered free access to the drug.
Are you expecting that people will jump on this and use this opportunity?
You’re talking about now the extended variant part?
Yeah, I think it’s likely that it will be quite an attractive option. It’s difficult to know, as I said, exactly what’s happening in the broad cancer community downstream of testing. Here we are now, just following the testing that’s being done in routine clinical practice.
We don’t know exactly how patients are triaged to FDA-approved treatments for their exact cancer type versus off-label use of drugs or clinical trials. It’s difficult to know exactly how much unmet need we’re filling. There are certainly sites, including large academic medical centers, that already have clinical trials targeting exactly the same populations we are studying in NCI-MATCH.