This is a post examining the new DCVax-L endpoints, likely alpha allocation, and the consideration of historical controls (same as external controls) new FDA Biological Products Guideline.
The post that I’m replying to is one where I’d listed multiple pathways for DCVax-L to be successful and to get to market, and I think those may now be obsolete.
With the change in the endpoints listed on the EU GB site (https://www.clinicaltrialsregister.eu/ctr-search/trial/2011-001977-13/GB
), we can now see the changes that were made to the DCVax-L trial that were set forth by the phone book-sized SAP that took about a year to write. And those endpoint changes establish new pathways forward that make most of the possible pathways I’d presented in this earlier post of mine basically irrelevant now.
So let’s look at these new DCVAx-L endpoints, and what they might mean when we receive our long-awaited top line PR, and how they might predict regulatory approval to a filed BLA (Biological License Application).
As many of us know, the previous PFS primary endpoint was to be measured applying an alpha of .02. This linked PR indicated this to be the case (https://nwbio.com/nw-bio-obtains-approvals-for-enhancements-of-phase-iii-trial-of-dcvax-l-for-gbm-brain-cancer/
). Typically, most trials will use .05 alpha, and so it was my thought that the company had split the alpha allocating .02 to PFS and .03 to OS. It’s very possible the company went with the low .02 because if PFS been proven efficacious on that low .02 alpha, the FDA would have likely granted them AA (accelerated approval), and OS would have been the confirmatory trial to prove the AA was fairly granted.
Under the old protocol, had PFS been stat sig, the .02 would have then carried over to OS and OS would have been measured with the full alpha of .05. This concept of carrying over the alpha comes from the 2013 protocol for the trial which indicated that the endpoints were to be measured within the context of what’s known as a “Closed Testing Procedure”.
From 2013 and 2014 DCVax-L protocols:
Quote: Defining Closed Testing Procedure
The primary, and secondary, and tertiary
endpoints efficacy endpoints will be reviewed. They should be considered
within the context of a closed testing procedure, and in the order of the
stated efficacy hypotheses, specifically the primary endpoint
(progression-free survival) first, followed by the secondary endpoint
(all-cause mortality) followed by the tertiary endpoint.).
This concept means that each endpoint is evaluated in the order they are stated, followed by the next endpoint, and so on. So if the first endpoint in a trial fails (under a closed testing procedure), and that endpoint had all the assigned alpha allocated first to it, then there would be no more alpha left to measure the next in line endpoint. In fact, this is exactly what happened to Dendreon in that 2007 BLA filing. Their primary of PFS failed (for whatever reason - psPD?) while the secondary OS was showing promise and I believe was indicating a few to several months difference in OS between arms. But there was no alpha left to measure that OS endpoint. I believe that is what that now infamous 2007 DNDN AdCom was set to argue… that in spite of the fact that there was no more alpha to measure OS, that DNDN’s Provenge should still be given marketing approval. And in that 2007 instance, the company lost that battle.
Now to be clear, so some people here don’t think I sound super smart with all this use of acronyms and alpha allocations, I wouldn’t know how to apply the alpha to the statistical formula, even if it could save my life (well, maybe then I could figure it out, lol). However… I do understand the concept of how alpha is applied.
So what exactly is alpha? In fancy words, alpha represents the probability of rejecting the null hypothesis (the opposite of the hypothesis) when it is true. Therefore, if a hypothesis looks to be supported by the data, and one used and alpha of .05 to determine the efficacy, that would mean that there is a 5% chance there is no difference between the data, and a 95% chance that the data indicates that there is a difference.
So what does the recently discovered change to the endpoints mean for the DCVax trial?
First, instead of measuring the previous primary endpoint of PFS at an alpha of .02, the trial will measure OS in the first position… but what is the alpha to be applied?
My thinking, and a logical assumption is, that the alpha will equal.05. So what does that extra .03 difference between the possible new .05 and previous .02 mean? If I’m correct and this is the case, it means that there is more power to successfully prove the trial’s hypothesis. The company likely had originally chosen an alpha of .02 to measure the primary endpoint (PFS at the time) for two reasons: first) to save .03 for OS in case PFS failed; and two) because the thought was likely that by using such a low alpha - .02 - and still achieving a stat sig PFS endpoint (which they thought they could do - had pseudo progression caused by DCVax not entered the picture), then the FDA granting accelerated approval (AA) would have been very likely.
Now let’s take this discussion of alpha further and apply it as it would likely be applied to the newly discovered endpoints for this trial. The primary endpoint is:
The primary objective of this study is to compare overall survival (OS) between patients randomized to DCVax-L and control patients from comparable, contemporaneous trials who received standard of care therapy only, in patients with newly diagnosed glioblastoma. This endpoint will be assessed using 3 different analyses
As stated, it’s logical to think that they’ve applied the full .05 to the primary as I can’t see any reason to just use half of it - like .02 or .03, especially given that they’ve picked an easy, IMO, to achieve endpoint so better to use the whole .05.
It’s also important to note that it’s the DCVax-L treatment arm which will be measured against the control arms of the three historical trials. I stress this because in the past, some stat people here have compared the blended trial data to the treatment arms of other GBM trials to see how DCVax compared - like to like. Some have hypothesized what the treatment arm data for DCVax-L might look like, but these comparisons were hypothetical, and not based on any actual data.
So it’s difficult to see where the DCVax-L treatment arm doesn’t compare extremely well against other comparable GBM trial control arms. To that point, the UCLA Phase 1 trial in newly diagnosed GBM using DCVax had a mOS of 35 months. The current DCVax-L trial’s protocol had assumed a mOS of at least 25.3 (from randomization, so about 3 months from surgery) therefore, that’s about a 28.3 mOS from surgery.
By the way, I’m only using the mOS measurement of survival as it’s an easier metric to discuss comparisons. I’m not suggesting that this trial will use mOS as the final measurement of efficacy, as medians are not what the trial measures.
Moving on, if this trial is still using the same “closed testing procedure” that the trial used in the earlier protocol, then should the primary end point be stat sig, then the full alpha (I’m suggesting .05) carries down to the following end point - the secondary endpoint:
The first secondary objective is to compare overall survival (OS) between patients randomized to placebo who received DCVax-L treatment following disease recurrence and control patients from comparable, contemporaneous clinical trials, in patients with recurrent GBM.
Frankly, the company can project that this endpoint has a strong likelihood to be proven stat sig simply by looking at the blended data. I’m not going to go into great detail on that point at this time because I don’t want to wade into the weeds on that for now.
But I’d like to again point out a few things.
With the former endpoints, had PFS been stat sig using an established alpha of .02, it was my theory that this alpha (under the rules of the “closed testing procedure”) would have then passed on to measure OS. The alpha of .02 that had measured PFS would have been added to the missing .03 alpha to equal .05 to measure OS. The problem, as I saw it, was that if PFS had not been stat sig, then OS would have only had .03 left to measure the secondary endpoint. This concept is somewhat detailed in the post that I’ve responded to (which has been up in the yellow sticky section for some time now).
The beauty of this endpoint change is that this secondary endpoint will now, likely (remember, I am guessing here, but this seems logical, to me anyway), be able to access the full .05 alpha to measure this recurrent GBM OS endpoint. This endpoint is really similar to the earlier secondary endpoint from the original trial, EXCEPT, that the control arm will be historical arms, and not the control arm for this trial. And my bigger point is that this endpoint will likely be measured by the full .05 alpha. I love that part.
Now, should this secondary endpoint be stat sig (which is likely), then the full alpha of .05, theoretically, would move down to the endpoint that follows:
The second secondary endpoint, confirmed progression-free survival (cPFS), is confirmed disease progression (cPD) compared between subjects randomized to DCVax®-L and those randomized to Placebo within Study 020221.
This endpoint is very much like the original primary endpoint, except that now, all of the patients have had their progression adjudicated by a team of radiology experts that have (it’s already done) closely determined, on a blinded basis, when the patient actually progressed, and this endpoint will use that date. That’s why they’re calling it “confirmed progression-free survival”. This endpoint is hoping to remove the pseudo-progression problem, and to look at the real progression events for the trial patients.
Now… this may likely be very clear to some, and may be becoming clear now to others, this particular endpoint would now also be measured, not with that puny .02 alpha, but with the full .05 alpha. And to note, this is the alpha almost every trial uses to measure their endpoints.
And if this third in the line up of endpoints is also stat sig… and to my thinking, it would now have been measured using the additional .03 alpha to help ensure a better outcome - and that was not going to be used previously. So again, if I’m correct, then full .05 alpha would then move on to the next endpoint… and please note that this endpoint represents the original (let us call it the problem primary endpoint) primary endpoint measuring the trial’s actual arms from the previous trial protocol.
The third secondary endpoint, PFS, is progression-free survival compared between subjects randomized to DCVax®-L and those randomized to Placebo within Study 020221.
And to drive home the point I’m trying to make here, if I’m correct, then instead of measuring the old primary with the originally intended .02 alpha, the old primary endpoint of PFS will now be measured with the full .05 alpha. See how that works? Now I really love that part.
And finally, if this new endpoint (which was the former primary endpoint) is stat sig, and it would now also be measured using the full .05 alpha (and not the .03 I believe would have been used). In other words, the original OS endpoint will have now have more alpha to measure it now. then the full .05 alpha moves on to the next endpoint… which is exactly the former secondary endpoint of OS - comparing the actual arms in the trial. More alpha means it’ll be much easier to reach a stat sig reading.
The fourth secondary objective, OS, is overall survival compared between subjects randomized to DCVax®-L and those randomized to Placebo within Study 020221.
Then there’s the final endpoint, the 5th secondary endpoint, which, under my theory, would be also be measured with the full .05 alpha if the 4th secondary endpoint of OS just discussed is stat sig.
The fifth secondary objective is tumor response compared between subjects randomized to DCVax®-L and those randomized to Placebo within Study 020221.
To wrap up on the topic of these new endpoints, alpha, and closed testing procedures, I’d think most of us realize that these new endpoints are much more achievable, which takes away a great deal of risk from the trial. The former endpoints were a much greater hurtle to achieve a stat sig reading, meaning that on a fail, the company may have had a bigger mountain to climb to convince the regulatory agencies to approve DCVax-L.
That mountain has been trimmed down, IMO, more to like a little mole hill now. Of course, this is all again, my own opinion.
I haven’t seen the latest protocol… I doubt anyone outside has… and the only reason I had the others is that they were… conveniently (to the naysayers)… publicly leaked. So I don’t know that the alpha now being used to measure the primary is .05.
Now I’d like to make a quick examination at the recent December 2019
guidance that points to why these changes to the endpoints are likely to be acceptable to the FDA.
The guidance is entitled, “Demonstrating Substantial Evidence of Effectiveness for Human Drug and Biological Products Guidance for Industry.” https://www.fda.gov/media/133660/download
I’d like to do this because I know that there have been some naysayers, especially on Twitter, that seem to think that using a historical comparison in this trial will never work.
Let’s look first look at the FDA’s recent guidance pertaining to the concept of using “historical comparisons”
in a trial that one hopes to use in order to obtain a marketing approval for a treatment.
The new guidance first addresses the risk versus benefit used in making an approval determination. beginning at line 86:
The finding of substantial evidence of effectiveness is necessary but not sufficient for FDA approval. The approval decision also requires a determination that the drug is safe for the intended use. As all drugs have adverse effects, evaluating whether a drug is “safe” involves weighing whether the benefits of the drug outweigh its risks under the conditions of use defined in labeling. Uncertainties regarding benefits and risks are considered when making an approval determination; a drug with greater risks may require a greater magnitude and certainty of benefit to support approval.
We can pretty much count on DCVax-L passing this “safety” criteria with flying colors. What that means, IMO, is that when the FDA looks at the evidence of efficacy, risk to safety will not weigh against any efficacy because there basically is no risk to safety.
Next, let’s look at how the FDA’s guidance approaches the concept of using a historical control as a basis for determining the effectiveness of a treatment.
In the guidance, the FDA points out that historical controls (also called “external controls”) have been included in the earliest descriptions of adequate and well-controlled trials (AWC). beginning at line 179:
Although randomized double-blinded, concurrently controlled superiority trials are usually regarded as the most rigorous design, as discussed further below, five types of controls are described in section 314.126:10 placebo concurrent control, dose-comparison concurrent control, no treatment concurrent control, active treatment concurrent control, and historical control (a type of external control).11 Of note, when the first version of the rule was published in 1970, historical controls and active treatment controls were included.12 Thus, from its earliest description of adequate and well-controlled trials, FDA included trial designs (as discussed below) that may be more difficult to interpret, which reflected FDA’s recognition that different trial designs (including choice of control) may be appropriate in different disease settings.
And here’s the footnote providing more extensive detail as to what is a historical or external control arm:
11 The regulation uses the term “historical control,” which is a subset of “external control.” FDA also accepts other types of external controls. An externally controlled trial compares a group of subjects receiving the test treatment with a group of patients external to the trial, rather than to an internal control group consisting of patients from the same trial population assigned to a different treatment. The external control can be a group of patients, treated or untreated, at an earlier time (historical control) or a group, treated or untreated, during the same time period but in another setting. An important subset of externally controlled trials are “baseline controlled trials,” where there is not a specific external control group but assurance, based on experience, that no change could occur (e.g., tumors are known not to shrink spontaneously or patients not given general anesthetic remain awake). See International Conference on Harmonisation E10 guidance on Choice of Control Group and Related Issues in Clinical Trials (ICH E10). This guidance uses the term “external control,” except when referring to section 314.126.
This guidance then addresses using “external controls” - which, as noted above, historical controls are a subset of. at line 221
Externally controlled trials differ in several important ways from the other trial designs identified in 21 CFR 314.126. Most notably, random assignment is not a feature of external control designs. As a result, there may be differences in patient characteristics or concomitant treatments in the trial population compared to the external control population that lead to differences in outcomes that are unrelated to the investigational treatment. In addition, the lack of blinding could introduce bias. For these reasons, external control designs are usually reserved for specific circumstances, such as trials of diseases with high and predictable mortality or progressive morbidity (e.g., certain malignancies or certain rare diseases) and trials in which the effect of the drug is self-evident (e.g., general anesthetics).
In this passage the guidance states that problems with using external controls are that these types of trials usually don’t use “random assignment” (however, we know that the DCVax trial did use random assignment), and lack of blinding (and we know that the DCVax trial was quadruple blinded - quite literally). However, one will note that when these two stated problems (which this trial DOES NOT HAVE), these types of externally controlled designs are reserved for diseases with “high and predictable mortality… in which the effect of the drug is self-evident” (which one will also note, describes the GBM disease very accurately).
To that point, this guidance goes on to state, beginning at line 231
Despite the limitations of externally controlled trials compared with concurrently controlled trials, strong support for effectiveness can emerge from externally controlled trials, especially when
(1) the natural history of a disease is well defined,
(2) the external control population is very similar to that of the treatment group,
(3) concomitant treatments that affect the primary endpoint are not substantially different between the external control population and the trial population, and
(4) the results provide compelling evidence of a change in the established progression of disease.
Such results could include partial or complete response in a disease where spontaneous regression is not observed, or stabilization or improvement in function in a disease where progressive functional decline is well documented to occur over the duration of the treatment period in the trial. Another example of where there is strong evidence of drug effectiveness is reversal of clinical signs and symptoms following a toxic exposure or overdose after administration of a drug antidote. In all such circumstances, a detailed understanding of the full range of possible clinical outcomes, with a well-documented natural history of the disease in the absence of treatment, is essential to interpreting trial results and, therefore, drawing a conclusion about the effectiveness of the drug.
We should note that the guidance is quite literally stating that “strong support for effectiveness can come from these types of externally controlled (using historical controls) trials, especially when:
(1) The history of the disease is well defined - as is absolutely the case with GBM
(2) The external control group is similar to the treatment group - most historical arms will be similar, although some of them will have included pseudo-progressors from chemo/radiation (these are the typical longer livers in GBM) in their trial, whereas the DCVax-L trial did their level best to screen these types of pseudo-progression longer livers out of the trial.
(3) Concomitant treatments used are not substantially different between the treatment group and the external control groups - most GBM trials used the same SOC treatment (radiation and temozolmide; and some crossovers may have used Avastin ) that DCVax used.
(4) The results provide compelling evidence of a change in the established progression of the disease - even the blended data (including using even 30 or so non-crossed over controls) still shows compelling evidence that adding DCVax to the treatment mix has increased the life span by a greater percentage for many of the trial participants. The top 100 were showing a mOS of almost 5 years, for goodness sakes!
And the last part that I highlighted about evidence of drug effectiveness is when they see a reversal of clinical signs and symptoms following a toxic (toxicity is not a concern - but definitely after a dosing of DCVax, this trial may see a some possible clinical outcomes that can only be attributed to the effectiveness off the drug - e.g. pseudo-progression). beginning line 247
For example, compelling results may overcome challenges associated with less rigorous trial designs, such as those with an external control. As discussed above, a small externally controlled trial with an outcome markedly superior to the well-established natural history of a disease may provide a compelling case for drug effectiveness.
And in making the above stated point, the FDA guidance is signaling that compelling results - such as what we expect to evidenced in the DCVax-L results - should be enough to overcome any challenges to having used external controls to measure against the treatment arm in their new endpoints.
So anyone that is emphatically stating otherwise, must not be aware, or does not understand specifically what this new approach to historical or external controls that the new FDA guidance on Biologic Products is now signaling they're willing to accept.