The Untrigger

Updated below: 8/15/16.

I have written a lot about the so-called clinical trial math of Provectus' ongoing pivotal Phase 3 study entitled PV-10 Intralesional Injection vs Systemic Chemotherapy or Oncolytic Viral Therapy for Treatment of Locally Advanced Cutaneous Melanoma (official title). The company's CTO, board of directors member and a co-founder Dr. Eric Wachter, PhD discussed some of this math on Provectus' August 10th 2Q16 business update conference call.

Eric said the number (not the percentage) of disease progression events required to trigger the interim and full study analyses were 81 and 162, respectively. I discussed this as item 5. Clinical trial math under Quick: 2Q16 (August 9, 2016) on the blog's Current News page.

It was helpful and insightful, but disappointing, to hear Eric say the trial does not have a scheduled assessment (a prescribed time trigger):

"So, when we began designing the study several years ago, we looked at the issue of how to schedule an interim assessment. And while that was a rather uncommon idea, we looked at that very carefully. We even had discussions with FDA on that topic. And they were adamant that they were strongly opposed to any sort of effort to schedule an interim assessment. Our hypothesis at the time was that, for example, because the interim assessment and the final assessment are based on a number of progressions, they're modeled using standard statistical models. If there was a longer progression-free survival in the PV-10 arm than expected, there would not be the required number of progressions in that arm to add up to the required number of events."

As such, as I hear and read it, one should consider that there only is one trigger (a prescribed event trigger), stated on the trial's CT.gov page as "[a]n interim assessment of efficacy and safety will be performed by the IRC when 50% of the events required for the primary endpoint have occurred." As a number, rather than a percentage, this figure is 81.

Eric said on the call that "one of our crown jewels is our study protocols because they represent everything that we've invested in the company up to the point that that protocol is issued." He certainly is reluctant to show his jewels if he doesn't have to (although it is possible he may fondle them from time to time). He "teased" two sections of the Phase 3 trial protocol. First, section 4.9 (with my underlined emphasis):

"An interim assessment of efficacy and safety will be performed when the IRC--that's Independent Root [sp] Committee--will be performed by the IRC when 50 percent of the events required for the primary endpoint, that is 81 disease progressions as defined in the study protocol, have occurred. And so, that explains to you that there's a set number. It's designed at the beginning of the study, and that's the point when that number of progressions have occurred."

Second, section 4.10 (with my underlined emphasis):

"It doesn't matter if they're all in the PV-10 arm, if they're all in the comparator arm, which isn't possibly because there aren't enough patients in the comparator arm. We have to have enough progressions to hit that metric. Now, we had hoped, as I said, to have some sort of a mechanism in the study that would allow that to be triggered early if, for example, we're going along and we're not seeing progressions as we expected. And so, while we worked on that with the agency, as I said, they're adamant that that was not appropriate for a pivotal study. So, what we managed to compromise on was that in Section 4.10, which is captioned "Study Duration", that subjects will be monitored until survival--for survival until death, loss to follow up, withdrawal of consent, or study termination by the sponsor."

Rather than hash out or re-hash my work/writing about prescribed event and time triggers, the number of patients needed to be enrolled to hit 81 events, etc., I'd like to state (speculate) the following -- [I believe] there is a trigger Eric does not want to explicitly speak about or affirm, which could/would be an imbalance in the number of events between the treatment and control arms, which could/would cause a statistical and ethical dilemma for the study's IRC. What would an imbalance look like or comprise? In the extreme, or potentially actually, there would be no events in the PV-10 treatment arm, and events of whatever frequency or magnitude in the chemotherapy/oncolytic virus control arm. Takeaway: The pivotal trial does not have to reach 81 events for the interim analysis to be triggered.

In his own special way I believe Eric said this on the call (with my underlined emphasis):

"And so, we are able to monitor at the terminal end of the study, and if the clinical trial data monitoring committee determines that it's in the best interest to end the study, it could end before we had all of the progression events occurring. I doubt that'll happen. It will be a very unusual circumstance, but it would probably be positive for the drug because it would suggest that a large number of patients weren't progressing because of the majority of the patients in a two-to-one randomized study are from the PV-10 arm, and that probably would imply that the PV-10 patients weren't progressing. 

But, this is all speculation."

The statistical dilemma is having a treatment arm with no or very few events, and a control arm with many, many more (like 0 and 15, or 0 and 20). The ethical dilemma is a treatment arm that works and works very, very well (i.e., no disease progression), and a control arm that demonstrates what everyone already predicted (i.e., everybody progresses).

The question then would be, "how many events in total and across the two arms would be required for the IRC to recognize an imbalance, and thus trigger the interim analysis?" Twenty events in the control arm (and none in the treatment arm)? Thirty? A two-to-one study randomization then would suggest enrollment and treatment of 60 patients. Or 90.

Math, math, math.

Updated (8/15/16): My projection for enrollment in the pivotal melanoma Phase 3 trial is below (having adjusted only for 5 months of no enrollment as a result of the major amendment related to Amgen's intralesional (IL) drug Imlygic, and also differentiating enrollment rates between medical and surgical oncology sites). The study may generate enough events to highlight the imbalance between the two arms in 1Q16, where the idea is that most-to-all of the control arm patients generate events, and few-to-none of the treatment arm ones do.

Click to enlarge.

What proportion of events in each arm ("proportion"), and what number of events in each arm ("sample size") gets you to a sufficiently low p-value?

If 75-100% of the control arm generate events, and 0-25% of the treatment do too, who you got?! Using this calculator, which is a simple analysis, and may even be simplistic, may require upwards of 30 patients; however, this may be too low.

E.g., Sample 1: Proportion of Control Arm Events (Number of events generated in the control arm ÷ Number of control arm patients); Sample 2: Proportion of Treatment Arm Events (Number of events generated in the treatment arm ÷ Number of treatment arm patients)

Click to enlarge

Click to enlarge