I believe the FDA was the one who guided NWBio and Dr. Linda Liau to use ECAs for their DCVax-L clinical trial data analysis. This was a critical part of the DCVax-L revised SAP and the revised endpoints.
Slides 18 to 23: Current Challenges of GBM Immunotherapy Trials External Control Arms (ECA) Selection of External Control Arm (ECA) Large RCTs for Newly Diagnosed GBM (2010 to 2019) Comparisons of Treatment with External Pooled Controls Patient Characteristics from Large RCTs for Newly Diagnosed GBM
So, the current challenges for glioblastoma immunotherapy trials are that, as I mentioned, there are challenges with randomized clinical trials. I mentioned, kind of, the biases that do occur with these trials, but also if you’re looking at it from the patient’s standpoint, the patients don’t want to be randomized. They don’t actually want to be in the placebo arm in a big randomized control trial. So it’s hard to recruit patients into randomized trials. Glioblastomas are relatively rare, and even among cancers, and even within glioblastomas, they are very heterogeneous, as I mentioned. There are small numbers of patients, so to do these trials, there is insufficient statistical power, heterogeneity will (garbled), and also there is heterogeneity of the tumor, as well as heterogeneity of the host. There’s lack of (garbled) correlates, or other biomarkers of response, we don’t have any good consensus in the field as to what the biomarkers of response are… and then, there’s rapidly progressing knowledge in the field, even as clinical trials go on. So these large randomized controlled trials take years and years and years, sometimes decades. In that period of time, there are new biomarkers, new things that are discovered, that are, you know, pertinent to the patients. And then, you know, we have to go back and reinterpret these trials. But then, because of regulatory issues, you know, related to FDA approval, you can’t just go back and reinterpret old trials. You have to do a wholeness trial. And, you know, one good example of that is the IDH1 mutation. As you know, for those that are in the field, IDH1 is a mutation that has now redefined the pathology textbooks as to what is a glioblastoma and what’s not. 1p19q deletion as we know it has redefined what’s an oligodendroglioma and what’s not. But those things were not routinely tested five years ago. So when people were talking about response to treatment, you know, these markers that were not tested previously, I guess, would not have been an inclusion criteria. So these trials were really testing on a very diverse population.
So I’ll talk a little bit about potential ways to think about clinical trials for this particular type of disease, but also for other rare diseases. And then maybe we can open it up for comments and discussion.
So one thought that has been, I think, gaining some popularity in the cancer field, is the use of external control arms. And I must say, as a clinical trial purist, initially I was not fond of this concept. And as I mentioned, there is a selection bias when you’re not doing a randomized controlled trial. But this is the concept of external controls. Basically, rather than, you know, randomizing patients, you basically take external data of a large population of external controls. And now, with electronic heath records, and other large data bases, there is that, I guess, probability or that option to do propensity weighting (note: a technique used in controlling for selection bias in non-experimental studies), or dynamic borrowing (note: using Bayesian hierarchical models with informative priors is a methodology which enables a data-driven approach to synthesizing historical data, which can be used as a reference for trials with limited or non-existent control data), or different concepts and basically design an external control arm to, as closely as possible, match the treatment arm. And the reason to do this is also, you know, as I mentioned, patients don’t want to be control patients. And a lot of our trials, sometimes when patients know they’re on control, or even when they’re blinded, the patients want to know for sure that they’re getting something or they tend to drop out of control trials, and, you know, jump around to other kind of early phase trials. And so we’re not getting the data that we want for these large randomized controlled trials. And then there is literature regarding what constitutes a good quality external control arm, and there’s different grades of quality. And what’s deemed as kind of the best data source are, you know, patients from large controlled, randomized controlled, trials, because it’s a rich data base, and the patients are, the data collection is, quite rigorous in those trials.
So when you go and look at, let’s say look at glioblastoma, over the last ten years, there have been many large randomized controlled trials for various treatments that have been published in high-quality journals. And the control arms has, for essentially the last ten years, been the standard of care (SOC), which is the same for all these trials. It’s radiation and Temozolomide (TMZ). So all these patients have the same treatment, and if you add these all up, you can see here, it’s almost fourteen hundred patients. And if you take the control curves of all these trials, they actually, pretty much, overlap.
So we really could, essentially, say there is a large external control arm for any new treatments that we’re trying to test. But the problem is, if we’re going to go to the FDA and get FDA approval, we’d have to do another one of these for three to four hundred randomized control trials, with hundreds more people on this control arm, rather than necessarily borrowing from these prior trials that have been done within the last decade. ‘Cause all these patients are newly diagnosed patients, and it’s pretty standard what the treatment is. The control curves are quite overlapping. So you kind of have to think, “Well, what about the other way around?” Right? So if you’re, let’s say, some of these quote “failed randomized trials”, you want to be sure that they truly did fail. And so what you could do is you could actually take the treatment arm of those trials, and compare it to this kind of large external control. And, for instance, EGFR P3 trial, the Rindopepimut and Temozolomide trial, if you compare the addition of this experimental treatment to standard treatment, which is just the control, and all these lines essentially they’re all Temozolomide and radiation, you can see that trial, it’s actually this blue line in here, but it blends right into all these control arms. So that, in itself, would have been… I guess, you could have, even if you did a single arm trial, and compared it to the external controls, it still would have been a failed trial. There’ve been other trials, for instance dose-dense Temozolmide, which is the kind of yellow line here, and you can see that that kind of falls into that control arm (curve) as well.
So one way to test new treatments is to, like I said, perhaps do a single arm trial and see how it compares to that control arm. And that would speed up patient accrual, because you don’t actually need to put patients on a control arm. You can compare it to the historical controls. One problem with that is that a lot of these trials are owned by the companies or the groups that fit the trials. So it would be good to get that data so that it’s out in the public domain so that we could all use that data as the comparators for how we design our treatments and our trials going forward.
But from the published data, though, you can see what the percentages are of all the kind of known factors. But then, if there was a way to model new predictive factors, that would obviously be of benefit as well.
Note: Similar comment from LL here: UCSF Resident Research Day Lecture Thursday, June 03
Resident Research Day Guest Speaker: Dr. Linda Liau Professor and W. Eugene Stern Chair, Neurosurgery Department Director, UCLA Brain Tumor SPORE David Geffen School of Medicine at UCLA "Immunotherapy in Brain Tumors: New Concepts"
