The first step in the NCI-MATCH trial—deciding how many patients to screen—was a guess.
At first, NCI and ECOG-ACRIN thought genotyping 3,000 patients may be sufficient. Then the sample was upped to 6,000, which also proved to be insufficient to fill the trial’s arms.
What’s the right number?
“It will become clearer as we march forward and get some data,” said Barbara Conley, associate director of the Cancer Diagnosis Program in the NCI Division of Cancer Treatment and Diagnosis.
Conley described the rationale for opening the study for participation by two cancer centers and two commercial vendors—and keeping it going.
“First, they’re screening a lot more patients than we could ever do,” Conley said. “And the second, if we are going to move precision medicine toward real-world application, we can’t confine it to one test.”
Conley spoke with Paul Goldberg, editor and publisher of The Cancer Letter.
Paul Goldberg: I don’t think it’s ever happened before that you basically took an existing ongoing clinical trial and then just changed it over to have it continue in a way that really wasn’t pre-specified. I don’t think I’ve ever really seen anything like it. Have you?
Barbara Conley: No, but this is MATCH, so we have a lot of those moments where we haven’t seen anything like it.
I just looked over the demographics of people who enrolled. I didn’t realize that a lot of enrollment comes from red states.
BC: To me, one of the best things about MATCH is that it’s running through the NCTN, and we can reach into the community and supply these agents. These are all off-label at the very least. Even if they’re approved by the FDA for some reason, those are not the reasons we use them in MATCH.
So, I think this has been reasonably well received, and it has taught everybody a lot about how precision medicine is being perceived, what the match rate will be. Of course, the match rate will always depend on what you have to match to.
But it also teaches us what mutations are out there that might match- and in what frequency it adds to the database. You know [AACR GENIE] is great in that way. They are collecting data from a number of centers, and we are collecting data from NCTN sites.
Basically two-thirds of our data will be from community.
Which is really weird. You did not count on that.
BC: Well, I don’t know that you could say that we didn’t count on that.
I’m asking. It’s a question, sort of. Put a question mark on what I said.
BC: What’s really good though, another really good thing, is that from the community we are getting specimens that we can analyze 94 percent of the time. So whether or not in the future we will actually have to biopsy everybody, we don’t know.
But we know that when we did it, even in the community, 94 percent of the time these specimens were fit for analysis.
So how long will this go? What’s the stopping rule? I’ve been covering clinical trials for maybe 30 years, and I’ve never seen anything that doesn’t have a stopping rule.
BC: It’s a signal-finding trial. And I think everybody is anxious to find out if we have any signals. And we don’t know that yet, because most of our variants are pretty rare. And so, we haven’t collected enough patients who have been eligible to be on study long enough to know what our response rate is on any arm yet.
We’re expecting to have that sometime this year. I hesitate to put a date on it because it does depend on lots of things. Since these arms did not accrue all at once, we have to wait until at least … the idea is we should have at least 31 evaluable patients.
Per arm?
BC: Per arm, right. If you think about the genesis of it, if you have a slam-dunk drug—you have a drug where it has gotten approved by the FDA for some purpose—the response rate is somewhere usually between 50 and 70 percent. Let’s just pick 60, right?
But it’s not that way, at least as far as we know, across all tumors. So we lowered our expectations to say that a positive arm, which would collect any tumor that had that mutation, would have at least a 25 percent response rate. And in order to get a handle on that number, we need 31 evaluable patients per arm.
So there’s a lot of discussion going on right now as to what do we do with the signal and how many signals can you get on a single arm. All of which actually needs to wait until we have the data.
Right. Well, Jim [Doroshow] was talking about potentially having this leading to an approval or two.
BC: Well, one of the examples is the LOXO-101 compound, which is very nice; but also the other recent example, of course, was the approval of pembrolizumab across tumor types.
But that wasn’t from MATCH. I’m not being critical by any means, but I have just never seen a trial change this much.
BC: No. We are there, and we just have to be comfortable with that.
That’s the best explanation I’ve ever heard.
BC: Well, we just have to be comfortable with that. We can’t make it go any faster. It’s a different mindset. This is a different kind of trial. You know, in the background are these other precision medicine trials, some of which seem to show that it’s better to do it that way, others of which have been kind of a bust. And so, when we designed this trial, we designed it carefully to have levels of evidence for every what we call actionable mutation of interest and every treatment.
There’s no treatment in MATCH that hasn’t shown a signal of activity somewhere in some tumor in a patient.
Yeah.
BC: So, that’s different; right? And we thought by designing it very carefully, let’s see if that makes any difference, because it doesn’t make any sense to all of us that precision medicine might not work, but it’s still a work in process, right?
Yeah, but looking at the paper that—the Charles Sawyers paper, the GENIE paper—they are saying about 16,000 was the right number to hit every, to accrue 35 or 31 per arm.
BC: Well, we know it’s more than 6,000. You know, those kinds of calculations sort of hinge on a number of things. One is how frequent really are these mutations in the population, and how can you possibly know that?
How can you possibly know that?
BC: Well, eventually, that’s what GENIE is trying to do. You are trying to get a sense of how frequent these things are in the population, among other things. But I think that if we don’t accrue 31 evaluable per arm, and we do accrue some, and some of those still respond, that is still a signal.
We really have no blessed clue?
BC: No, but I think it will become clearer as we march forward and get some data. But we opened it up to outside—well, that was one of the reasons. One is they’re sequencing a lot more patients than we could ever do. And the second is that, if we are going to move precision medicine toward real-world application, we can’t confine it to one test.
Are you planning to make it open to other tests and other labs?
BC: That’s the plan, but we thought that it would be helpful to the community to know how their tests would compare to the MATCH assay for example. And so that’s why we are requesting anybody who gets on the study with outside labs, non-MATCH labs will send a sample, an archived sample. Hopefully, the same one they sent to the outside lab and then we would run the match assay on it; and we’d get some sense.
Now we know we’re not going to be perfect, right, because the world is the way it is. It’s not perfect. But if we get the sense that we’re pretty good, all these labs are pretty good, then that tells us that we maybe, only need a well validated assay. We don’t need a particular assay to assign with.
Is there a next evaluation point? Is there an assessment where you just say, “Hey, we’re done.” Is there an interim analysis? I’m searching for the right word for it. Is there anything protocol pre-specified somewhere out there?
BC: So, we are monitoring on a real time basis what’s going on with accrual and toxicity. But the response analysis is a one point analysis when we have 31 evaluable patients per arm. I mean, we didn’t intend in the protocol to have interim analyses, except for the match rate and the toxicity rate from the biopsies and those kinds of things.
We didn’t really spend a lot of time on formally assessing the strategy as a whole, the entire MATCH protocol, in an interim analysis.
And how many arms do we have to have to say that such a trial is successful? I don’t think anybody will—one is good enough for a lot of people. It would be lovely if we had more than one, in my mind. And I think that we might have more than one, but we’re just waiting to see if we actually have responses now. The downside of having this active through the NCTN is that I don’t know if it’s any worse than the drug company trials because they have to have thousands of sites too, and you have to get the data out of all these sites.
You have 1,100.
BC: Yes.
Is there anything we’ve missed. Anything that we need to add to this?
BC: I think if you go forward and you’re looking at precision medicine and we’re always trying to match the clinical trials to the biology that’s coming out, many people are thinking the next thing might be combinations or maybe getting a handle on what the pathways are and what the potential end runs around those pathways might be.
We’re not anywhere near that right now. But, if you’re thinking about the future of precision medicine, it will always try to get a little more precise; right?
If you were to try to imagine what historians of medicine will say about this time point and this trial let’s say 20 years from now, what will they see, you think?
BC: I think it’s the first foray into broad application of molecularly-guided medicine. These are patients who have been through standard treatment and have progressed after that. Will we be moving that forward? More than likely, given the history of targeted medicine up till now. It’s always moved forward.
Will we be taking more account of the status of the patients, the patient’s immune system, and that interaction with the treatment of the tumor? I do hope so. And studies are in place now that may shed some light on that situation.
Designing and running MATCH has been sort of a career in itself, and we’ve had like 150 people working on it, very hard. It’s not a set it and forget it kind of thing.
It’s a revolution, or am I wrong?
BC: I hope it is.
Well, even if it turns out that it’s wrong, it’s still a revolution.
BC: Yes, even if it’s wrong, we will always learn something from this trial that will help us going forward in precision medicine trials. I’m not sure that we would want to design the same trial again, not only because it was very resource intensive but things will move on. We designed this three, four years ago, maybe five. Anyway, since then things have kind of moved on.
Project GENIE; right? They’ll find prevalences in populations.
BC: They’ll find prevalence and so what does that mean for clinical trials going forward? If we want to treat, not just rare tumors, also rare variants, what does the next trial look like? What do trials look like for that? I think pharma is tackling that right now. We do not have universal molecular testing for the population at this point for cancer patients at any point in their course, except for a few tumors where you should test, right?
Lung cancers, there’s a few things you should know about. That’s one thing. Melanomas, there’s a few things you should know about. But other than that, you know, do we test everybody?
We don’t. Everyone is waiting for the clinical utility of doing that, which means is the patient better off with molecular testing than they were without that? That is the big question for us to answer.
And it’s also an interesting question about who gets tested and who doesn’t. What kind of populations get tested commercially and what kind of populations don’t?
BC: Yes, and a lot of facets on that one. Well, we are monitoring the type of tumor obviously, but we are also monitoring the population, the age, the ethnicity, the race. All of that we’re monitoring. We’re pretty much on track as what we would see in normal clinical trials.
