Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

In the Interim...

In the Interim... artwork

Jumping Hurdles: Interim Analyses for Funding Decisions

Dec 8, 202542 minAudio

In episode 40 of "In the Interim…", Dr. Scott Berry examines the statistical, operational, and behavioral challenges of using interim analyses as triggers for funding in adaptive and seamless Phase II/III clinical trials. The episode presents a typical hypothetical scenario for rare disease drug development, contrasting conventional two-stage development with a seamless design and highlighting efficiency gains in sample size, patient allocation, and trial duration. Scott details the construction of administrative (financial) interim analyses, underscoring their distinction from futility analyses and their role in funding decisions when complete funding is not secured upfront. He addresses FDA operational bias concerns, emphasizing blinding and limiting information sharing to protect trial integrity. Finally, the episode focuses on developing objective interim funding criteria - using Bayesian predictive probability and assurance - and on leveraging illustrative simulation outputs and sample datasets to bridge the “I’ll know it when I see it” divide between scientists and funders. Practical, empirical, and tailored to real funding barriers in clinical research. Key Highlights Statistical structure and efficiency of seamless Phase II/III trial designs Administrative (financial) interim analysis setup as funding decision triggers, distinct from futility analyses FDA operational bias guidance and requirements for trial blinding Predictive probability and assurance as objective interim criteria Sample data and simulation outputs to facilitate stakeholder alignment For more, visit us at https://www.berryconsultants.com/

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Judith: Welcome to Berry's In the Interim podcast, where we explore the cutting edge of innovative clinical trial design for the pharmaceutical and medical industries, and so much more. Let's dive in. Scott Berry: Well, welcome back everybody. To, in the interim, I am your host, Scott Berry. Appreciate you joining me here. Uh, as, as on many of these I have discussed. Uh, my wife Tammy, listens to all of these and her comments to me about,

uh, about the various podcasts. Uh, also asking, you know, where am I looking in all of this? So I, you can see I've, I've this, this sort of setup and I don't. I don't peer into the, the camera here. I'm looking at my large monitor where I have my, my notes about the podcast and my, my other computer down here and the camera's over here. So I, I know it's a little bit odd that I'm always looking in the opposite direction. Maybe this is in part my. Uh, my introvertedness of being a statistician, it's, it's

hard to peer into the camera. There's nobody there, uh, in it. And that, that, that's the old, that's the old joke. Uh, how do you tell an introverted statistician from an extroverted statistician? The extrovert looks at somebody else's shoes. Alright, so I, I think it's fair game for, for me to criticize statisticians and I, I may do that again, uh, because I am one, uh, in this years ago.

We, I was involved. We were involved. Barry consultants were involved. A number of people, really wonderful effort in a project called Adapt It Adapt it was funding from the FDA and the NIH to investigate bringing innovative trial designs into trials that were part of the net. Neurological emergency treatment trials. And we went in and we, we, we took what were gonna be five different trials.

It turned out to be six trials we investigated. And what if we looked to innovate the trial design? What was really. Great about the project is we had, uh, behavioral scientists studying that process of bringing innovative trial designs, uh, to these, the, these particular trials. One of 'em was, uh, which ended up being ice cap cardiac arrest.

Sub people subjects, uh, sorry. Patient suffers cardiac arrest. They're brought into the emergency room. All of these particular scenarios centered on the emergency room. And the question is, does hypothermia create better neurological outcomes? There's a period of time where the, the blood is not flowing in the brain. Lack of oxygen and heat. Makes that worse by putting a patient in a hypothermic state, does that create better outcomes? That was just one of six where there was a stroke trial, there

was a spinal cord injury trial. There were a number of particular trials, so there were behavioral scientists that were studying bringing innovative trial designs to that. We explored adaptive sample size. We explored response, adaptive randomization. We explored futility, interim analyses. Multiple arms modeling, all in the different scenarios. The behavioral scientists in studying this part of their, their goal was what are the barriers to bringing innovative trial design?

Lines. Is it the funders? Is it clinicians? Is it regulators? Is it operations? Are it statisticians? Uh, it what, what are the various barriers? Interestingly, the number one barrier that they found worse, statistic. Which is kind of an interesting place that, that interesting role, and I think it was a different time, this was about 2005 to 2010 ish that we

were investigating these trials. It was a different time, but they, they were hurdles that you had to overcome that bringing these innovations, the what ifs in all of that. Um, now. What are, what we, we do this at Berry Consultants. We do a number of what I'll call innovative. And, and now innovative is a self glossing is better. And I I always get concerned saying, oh, we design innovative trials By some sense you mean better trials, smarter trials we're better.

Um, but what I mean by innovative trials is, is things that are rarely or never been done before. Her. Now, you could argue whether they're worse or better, so I wanna leave that aside, but I, but I'll bring that we're doing things that are done relatively rarely or have never been done before within trial designs, I'll call those innovative trial designs. Okay? What turns out to be barriers? Uh, I, I, you know, statisticians in some cases can be barriers.

I don't think there. They're barriers. Uh, I think you can present and create scenarios. What are the way that we make it more, more robust and handle that? What turns out to be a barrier to, to me, I'll speak for myself, is the, the scientists who say, I can't define it. I'll know it when I see it.

Now, what do I mean by that? The scenario is you're, you're trying to look at adaptive designs or you're trying to measure is design A better than design B? You have to be able to understand what's the goal of the trial, what are we trying to do? Are we trying to find the right dose, the right duration? Are we trying to find the right patients? And what is success? If we're trying to compare A to B, we've gotta be able to say, A does a better

job at fi, uh, uh, at being successful than B does, or vice versa, or the amount of resources to get the right answer. But you have to be able to define what is good, what's right what, what trials would you go to Phase three. And when you, when you want to simulate that and you want to compare that and you want to say is a better than tr trial B if, if the, if the scientists on the other side say, I can't define what's better, show me the trial result and I'll

tell you whether I'll go to phase three. That doesn't lend itself to analytical comparison. It doesn't lend itself to building a better design because I don't know what success is. When are we there? What, what, what is good and what is bad? That becomes really the enemy to doing innovative designs. Now, uh, I'll talk as we go through here about ways to combat that. I don't think that's, I, I think you can get out of those individuals good answers.

In that scenario. So I wanna bring up a common scenario where this can happen. And this was, by the way, um, this was a topic suggested to me as somebody said, oh, I'm sure you did a podcast on this. I'll go find it. And I hadn't. But it is, it's a ubiquitous thing for us. This comes up all the time, and that is Here I'll, I'll, I'll, I'll do a setup and then I'll get to the topic is. We're a a, a sponsors looking to design a phase two trial.

The phase two trial has two doses in a placebo, so it's a three arm trial, and they're thinking it's a reasonably rare disease, and they're thinking it is a rare disease. They're thinking of running a trial that is 30, 30 and 30, so 30 patients on placebo, 30 on dose, the low dose, 30 on the high dose. They want to run the phase two trial and then they're gonna run the phase three trial. And this is a scenario where we know the phase 3M.

Point. Uh, there, there are some scenarios where you're running phase two and it's really a wide open space of what the endpoint might be. That the endpoint is, is somewhat dictated by the area measuring clinical benefit. So the, the endpoint is known and then they're gonna run a phase three, which is a one-to-one comparison of the dose that they select to placebo. And they're thinking that trial might be 90 versus 90, that the effect size that they need to have, they, they have a pretty good idea of that trial being 90

versus 90 being powered at their effect. So they're looking at a development pathway that is phase two, which is 30, 30, 30. They, uh, get full follow up on patients. They go through the analysis of that data. They look at that data, they make a decision on going to phase three. They then submit the protocol. They go through all what I'll call the white space between two and three, that they then run a trial of 90 versus 90 in phase three.

So we, we said, well, what about a seamless two three trial where you enroll 30, 30, 30? You select a dose from that, and the trial seamlessly moves into phase three, which is randomization of placebo to the dose that's selected. In phase three, and we can talk about the size of that, but let's, for a minute, just think about the, the separate trials was a 90 patient

trial followed by 180 patients. The seamless II/III trial could be 30, 30, 30. You pick a dose and you enroll 60 against 60 in the second part of the trial. And the patients in the first part and the second part are part of the primary analysis. So the dose that's selected in the seamless trial will have 90 patients against placebo, 90, and you'll have 30 on the unselected dose.

Now what, what does that mean In the seamless trial, you do have to adjust the alpha. At the end, if you're gonna include those patients because you, you quote unquote picked the better dose in phase II what you cherry picked the better dose in that, and then you, you are including that data at the end of phase III straightforward ways to adjust alpha and do that, that, that, what's great about that and what is, is kind of amazing is that you have more power.

Even though it's still 90 patients, than you will have, if you run a separate 90 vs 90 phase III the power of the Seamless II/III trial, which is 90 patients followed by 120 patients, 210 total compared to 270 total, it's higher power for demonstrating success at the end of that phase III So huge benefits in that scenario for resources.

But the other part of it that's just incredible is the lack of white space between them. The time for that trial to read out, the time to then select the dose. Start phase three, enroll 180, uh, another 180 patients. So we, we simulated that. And by the way, you run into this case of how do we pick the dose? Between them, and I don't wanna spend a ton of time on this. Maybe that's a whole separate podcast of picking that dose.

It can be purely on efficacy. It can be a combination of efficacy and safety. It can be efficacy with A-D-S-M-B, weighing in on whether there's a, a safety or tolerability issue on it. But for now, let's think about picking the better dose based on efficacy. And in this particular case, the primary endpoint at the end of phase three is really a combination of both. It's it's the clinical outcome of the patient, so we're gonna pick one that has the better clinical outcome.

That's a case by the way, that you could run into this enemy that I referred to of, I can't tell you how I'll pick the dose. You need to show me the data and then I'll pick it. It, but that's not the topic. When we presented this and we showed that this trial of 210 patients has more power than 270 patients adjusting for Alpha, the incredible part of this is they estimated the time between phase two and phase three will be almost a year.

That, that's just the natural space that will happen if they run a separate two A and two B, that space disappears in a seamless two. A uh, seamless two, three trial, uh, in that setting. We, we asked them how long that will take. We looked at enrollment. We simulated, uh, uh, the trials using the numbers presented. And again, one trial is 270 patients and the average time to readout

of the phase three portion of that was approximately 50 months. It was a range to that, that we had a, a, a confidence interval or a prediction interval for the range of that. And the, the median time for the inferentially seamless trial was 25 months. Just call it 45 months to 25 months, 20 months different. Less resources, higher probability of success, win, win win.

In that scenario, the feedback was, no, we don't want to do the inferentially seamless trial, and it's a case where it's sort of, I have this, wait a minute, what, what? What do I not understand? This is a very common response actually, in a scenario like this. Uh, another aspect of this that's so important is the chain of communication. If the communication is somebody said no, and that went through person A and person

B, and then we hear the answer's, no. It's hard to understand why, and it's hard to ask about it. It's hard to potentially understand. Is there a remedy to that? No. And many times over the years we brought, I brought this up that we're trying to do innovative things, things that are not commonly done. It's not infrequent that somebody says, no operation says no. Regulators say no, somebody says no, and understanding why.

A lot of times there's a reasonably straightforward way to remedy that to then get cheaper, faster, stronger. The reason for this, which is a very common reason, is funding the company is on the smaller side of the pharmaceutical industry. A big pharma may be able to fund the phase two and phase three, and they can set

up those criteria and they can run it. What about a smaller biotech? They currently don't have the money to finish the 210 patients. They can run the 90 patient trial. They have the funding for that, but they need to get funding in order to enroll the rest. So that's a very common response, a very real, uh, thing. I'm not gonna laugh at that in any particular thing, but it, it, it's part of designing these trials, innovative trials is can we combat that?

So it's a very common scenario, by the way. It's the importance of being there. When somebody says no, it says maybe we have a remedy for that. So can we set up funding solutions? And triggers of the phase two that triggers funding. And, and by the way, these funders could be private equity, the funders could be a larger pharmaceutical company, could be a range of individuals that might

be interested in investing in the drug. They would like to invest in a drug that's going to be successful in this scenario. So in the separate trials, what typically happens is the phase two data reads out and they unblind themselves to that data, and they can share that data with a potential funder, and the funder can play with it, and they go back to this. We know it. When we see it, the funder then can decide, will I invest X dollars in this?

As we move forward for the phase three trial, and then I'm a partner with the biotech. If this ends up being successful and the drug makes money, we're invested in that and we bring in our experts and we're we're able to decipher from the phase two data whether we're gonna be successful in phase three, what the market share looks like, and is this a good investment for us? And that's a very common scenario, and that's what happens in a typical phase two truck.

Now what are we looking to do? We're looking to take that phase two trial and setting up objective criteria that would trigger funding. So you go to a funder and you say, okay, you're interested in funding this. Can we define what you would need to see in phase two to fund phase three? If we can do that. We can do that interim analysis where we select the dose and we

can also let the funder know, did you hit your funding criteria? Sometimes this is done separate from dose selection. By the way, this can be done in a phase three trial, so this isn't unique to Seamless two three trial. That was just the particular recent example that we're working on. Sometimes this happens where the company's ready to run phase three, but they can't fund phase three in its entirety. And a funder says, you know, I'm interested in this drug,

but I need to see more. I need more proof of proof of concept of this before I'm willing to invest the big pot of money for phase three. Sometimes you can start phase three and do an interim analysis. Of whether we hit that proof of concept, proof of funding that the funder says yes. If you could jump this hurdle, we would fund the rest of phase three.

If we can objectively set that up. Agreement can be made that an analysis takes place during the trial. I'm not gonna call it an interim analysis, but it, and I'll come back to that, an analysis takes place. It's conducted by an independent statistical group, an independent CRO, and literally it's a checkbox. Did we hit the funding criteria? Maybe it's two or three check boxes.

Did we jump the following hurdle? And that is signed by the independent group and that is given to the sponsor and we'll come back to who, who knows and who sees what when. But that checkbox could be presented to the funder in agreement. That they've agreed to now fund the rest of the trial based on the deal they have with the biotech. And they know they hit proof of concept in phase two, even though they don't get to see the data.

Now what that, that's the big question of what you get to see during the course of it. So. The data in a phase three trial with one of these analyses or a seamless two, three trial? It doesn't really, neither one is different because as far as regulators in the scientific community are concerned, those patients in the phase two part or the earlier part of phase three, they're phase three patients. It is, it is considered bad form and we'll come back.

I'll come back to that. To unblind the data during the course of a phase three trial. Sponsors are remained blinded to what is the effect size of the treatment. During the course of that, we don't want Jane Doe to be unblinded. We don't want the pharmaceutical company to know that Jane Doe or John Doe was treatment or placebo. It does the data. You know, is there altered, is there some aspect of knowing that unblinded data that changes the trial?

And, and this is this thing called operational bias. So we'll come back to why, why, typically you don't, but let's just assume that you can't. Unblind the data. Go to a conference present. Ooh, look how good our data are in the middle of the phase three trial. Will you help fund the rest? Which is why a lot of times it's a separate trial so that they can do that or, or, or present their data. Unblind the data in the middle of a phase three with one of these analyses, or a two, three.

You can't interrogate the data itself. You don't want to unblind yourself to any individual patient, but you can come back with a limited set of information of is the data progressing well at this point above a particular hurdle? Sometimes if futility analysis are done in trials a lot, this is different than a futility analysis. What do we call this by the way? We tend to call it either an administrative analysis

or a financial analysis. We'll write into the protocol that an administrative or financial analysis will be done. We don't want to call it futility because writing in futility means if you don't hit that, we're gonna stop enrolling patients. We're gonna change the trial. But this analysis doesn't change. The trial at all doesn't change the sample size doesn't change how a patient is treated.

All it does is trigger funding. Potentially. It could be that that funder, it doesn't trigger the funding, but the company gets funding another way and the trial continues on. So this is not something that affects. Patients, the, the, the scientific aspect of it. And so we don't want to call it futility. It's a misusing of futility. If you do that for funding, call it administrative, call it financial, and have that done. So let's come back a little bit to, to why we do that and who knows what, when.

Ha. But if you read the FDA guidance document on adaptive designs where interim analysis are done, they have a great deal of concern of what they call operational bias. Operational bias is, is, it's a hard thing to define largely, it's that by knowing the knowledge of the data, it alters the future of the trial or the interpretation of the data in some way.

Okay. Imagine in the middle of a trial if, if patients find out that the investigational treatment isn't working very well and many of them drop out of the trial, now you've got a huge missing data problem. The interpretation of the trial, you've, you've. Uh, risk. The whole interpretation and the ability to understand is a treatment beneficial. The opposite could happen if there's a really positive signal that patients

drop out and go elsewhere, that future patients enrolled look very different, that the investigator will no longer en enroll this patient or that patient, or something changes. Now this, these are loosely defined and it's, it's, by the way, it's hard to know if you ever have operational bias. Uh, it's hard to defend that you didn't, if somebody says, oh, you knew this, and the it, it, it, it changed your trial. It's hard to prove that it wasn't there. So we do a great deal to, to make sure that there's no concerns about this

mysterious thing called operational bias. So during the course of a phase three trial, we don't want the pharmaceutical company, the sponsor to know the data. We don't want clinical sites to know the data patients, to know the data and all that. It's important within the scientific credibility of the trial. So in this circumstance. What? It's not uncommon that you get a piece of paper that says, was this criteria met?

And I'll come back to what do those criteria look like? What kind of things do we do for that? But is it met? Yes. No. Is criteria one met or criteria two met or no? So three check boxes. So what does that mean? Yes, you can reverse engineer from the checkbox roughly. Are the data above or below this largely that's not considered to create the risk of operational bias. You still want to keep that information relatively confidential.

You don't want to publicly disclose if you can prevent it, that our effect size was above 20%. At the interim, does that cause this potential operational bias? So setting up a relationship with a funder that says you're gonna get this checkbox, the at the sponsor, maybe somebody in the C-suite gets this as well, A CFO. Somebody who's not going to sites and talking to clinicians and investigators

that are enrolling patients, you'd like to create that separation. You'd like the people involved in running the trial to remain as blinded as possible during the course of it. So it might be that the, the CFO, the CEO. Seize that data, but otherwise it is firewalled from the rest of the company. And then the funder can get this information also agreeing not to publicly disclose it, but they agree to write a check if that's there.

Now, what kind of criteria do we tend to use here in, in, in a phase three trial in that, uh, it's not uncommon that we use the predictive probability of trial success. Now, what is that? That is a calculation that's integrating over the uncertainty of the treatment. What's the probability that statistical significance is gonna be met when we enroll out the full trial?

Sometimes this is called assurance based on a particular prior distribution. Uh, in that scenario, sometimes conditional power is used given the effect size we've seen. Now, suppose that's the real effect size. What's the probability of success? The Bayesian predictive probability of success doesn't assume a single effect. It assumes the uncertainty of that effect given the data we have, it's a very natural Bayesian thing to integrate over the uncertainty of the treatment effect.

And calculate the probability that we're gonna meet statistical significance at the end of the trial. What I've described that is a lot of times, and, and you'll, you'll, you'll not be surprised that I'll go to the sports world here. Uh, you can watch an NFL football game, you can watch a basketball game. Uh, you can go to ESPN and it will tell you the probability that one team's gonna win the game at any point in the, in the. The the game. So at halftime, one team's ahead by 14 points.

It might say the probability they win this game is 92%, or if it's a tie game, might be 50/50 In that setting. That's the predictive probability of success. That's exactly what we're doing in a clinical trial, that we're calculating that given what we understand now and the strength of the two teams, the strength of the two treatments, what's the probability that we win the trial? Now, a clinical trial is not a game where the high score wins.

You've got to win by a point spread. You've gotta be statistically significantly better. So if it's, if the two treatments are tied at this point, the chance of winning this trial is gonna be less than 50%. It depends on what sample size it is in that. If you are seeing exactly the effect size you need. To show at the end of the trial, that's where it's about 50 50, that you're gonna win the trial. If the data gets a little bit better, you're gonna win. If it's a little bit worse, you're gonna lose statistical significance.

That would be a 50% probability of success. That's pretty good data by the way, depending on the, the, the disease. If this were an Alzheimer's trial, for example, or a acute stroke trial, and you've got a 50 50 shot of winning the trial. You might want to invest in that trial. That's a really good particular gamble in that scenario, uh, because the, the outcome is such a positive outcome for you. The, the revenue to reap in that scenario is, is positive enough to win the trial.

Now that predictive probability of success sometimes gets confused for power. Power is not something that changes based on the data that you observed. Power is something that if the treatment effect is Delta, here's the chance we win the trial. The trial's gonna be 80 or 90% power. Every trial is 80 or 90% power. You just figure out what Delta makes it powered. So there are times where they say, oh, to invest, we want it to be 90%.

That's a really, really high predictive probability. Think of that sports game. So you, this is, this is exactly the scenario. You're sitting there. You could make a gamble on team A or team B, and at halftime, a team has a 90% chance to win. If, if you're investing a and you get a double A at the end, of course you're gonna invest in that, uh, uh, in that scenario Now. The, the people that the group that you're investing in understands

the odds and the the payout of that is going to be adjusted by it. At the time you make that. Now this is different. You get to make that investment on it. Now if you say, sure, I'll invest it 90%, it's pretty unlikely. Well, I, it may be unlikely that the company can hit that early on. It, but maybe they can hit 50% or 30%, maybe that's where I would jump in. You can go back to thinking about, suppose I had a separate phase two trial. When would I invest?

What effect size would I need to see before I'd be willing to invest in phase three? That can all be incorporated within this administrative financial analysis. Is, and again, the enemy of that is, I can't tell you when I would invest. You have to show me the data now. Many scenarios. This is kind of a weird concept, this predictive probability of success in the trial to the people who are so used to building a trial where

they're used to talking about power and type one error and all of that. This is all of a sudden a little bit different. They may be familiar with conditional power and they may be familiar with futility analyses, uh, but this predictive probability of success is a little bit of a. Different sort of beast in the trial. So sometimes they're concerned that funders might not understand that. The interesting thing is funders understand that immediately. That's the world of funding. What do I invest?

What do I make, and what's the likelihood that I make it? And that can evaluate that investment. Incredibly straightforward. So this is a powerful thing to funders. Now, my why might a, a, a funder, uh, agree to this objective criteria in it? Well, let's go back to the beginning of this, is if they say No, run a separate phase two and show me the data, and then I might invest.

They're investing something that's 45 months to the readout of potential success and it's a bigger investment. 270 patients will, will is the investment and it's 45 months as opposed to 25 months. In a smaller investment, we get it sooner. It's a win, win, win. If we can agree to this potential funding, it's a win for. The pharmaceutical company.

It's a win for the drug. It's a win for investors. It's a win for patients. It's a win for everybody in this agreement if that can be agreed upon. Now it comes back to this. I know it when I see it. Can we define when we would invest in that? It might even be a function of, if we hit this higher hurdle, we'll invest, and here's the outcome. If we hit the lower hurdle, maybe they get a higher fraction of the outcome because it's a bigger risk. I, I don't live in that world, but you can imagine fairly straightforwardly

this being a part of this agreement. This is a very common situation. Many groups are in this, even big pharma needs to get big pharma. To agree to it and maybe large pharma doesn't want to say we're gonna invest in a 1500 patient trial. We haven't seen enough individually to, uh, at, at the corporate level to say, we're gonna invest in 1500. Run a phase two trial, well turn it into a phase three if you're ready for that.

Or a seamless two, three where you have a funding go, no, go internally in the company. That's gotta be much easier to do. And again, it's a win-win, win. For the pharmaceutical company in this scenario, if you can define it and create it, it's a very straightforward thing to do in the trial. Important part in the innovation part of it. A little bit of a warning here is some of the terminology in this that you'll hear, uh, companies talk about a TPP target product profile.

They're going into it where they know to be a competitive in the market. They need to see a 20% improvement, a 10% improvement. What, whatever that 50% improvement, whatever that is, that's what we're targeting for that product. I have to tell you, lots of people talk about that and they all interpret it very differently. Sometimes it's the delta. I want to power my trial to win on that, and so if the effect is 30% difference,

I want the trial to be 90% powered for that 30% difference, but you could win. If you see 20, about two thirds of that effect, you can win the trial. Some people call the TPP, that 20, that when the trial's over, we need to be that value or above to be competitive in the market. Those are two very, very different things, and then they're two very, very different things for funding. So if 30% is what you're powering the trial for.

And at the interim you say, we need to see 30% or above. It's disconnected with the trial. 'cause you can win at 20, halfway through the trial. If you see 30 and it's power, you see 29, that's awesome. You are likely to win the trial again. Your probability of success is probably hovering over 90% at that particular time point. Now, if somebody says, oh, but you're not meeting TPP, that's the part where. The terminology gets in the way of good communication for that.

So one of the ways to combat this, I'll know it when I see it, is to show it to them show them a dataset and say, if we have this data in phase II would you fund us? Because it's hard for, for, uh, lots of people, non-statisticians To think about the quantitative aspects of what is a predictive probability of success? What does this mean? What does that mean? But if I can see the data, if I can see the effect size that's meaningful

to me, clinically meaningful, so we show them a dataset and say, would you fund it if we showed you this data set? Sometimes they say, you know, I also need to see this. Okay, we'll show you that as well. Would you fund it? Say, you know, if we see this data, we would fund it. Okay. How about this dataset? Would you fund this dataset? And we show them example datasets Kaplan-Meier plots of what data might look like. Uh, estimated hazard ratio, difference in proportions, mean changes.

We show them these data and then you show them thousands. You show them scatter plots of the various effect sizes. And then you say, by the way, this point right here that you would fund the predictive probability of success is 82%, or it's 63%, or it's this, and now here's this other data set. And you say, here's the predictive prob. And they get to understand it it's meaningful to them. I know it when I see it through that quantity because it's now familiar to me.

So while that initially can be thought of as this is the enemy, that person is somebody that you can work with in simulating this, showing this example, showing example data sets, tying it to a particular criteria, sometimes it changes the criteria. Let's just do the observed effect at that time in that. Many times it can iterate and a lot of times, by the way, you're making this decision in a seamless two three trial where you don't have complete data on the phase two part.

That's okay. You might be building longitudinal modeling of this. You might over enroll phase two a little bit to get enough completers because those patients count in phase three. They're not wasted. Maybe if you are enrolling one to one to one, and you over enroll the three, do the three arms, two of those arms would continue in phase three. So if you don't continue the trial and you stop, they've been over-enrolled.

But if you continue, only a third of those patients have been over-enrolled. The other ones contribute to the end of the phase three analysis. These are all the things that go into these types of trials and these types of comparisons. In order to tackle what is a ubiquitous problem, you've, you have to be able to fund the trial. All right, so I, I, I hope you've enjoyed the discussion of funding Jumping Hurdles, combating the I'll know it when I see it.

And thank you for joining me. Until next time, we'll be here in the interim.

Open the transcript file

Hosts and Guests

Hosts

  • Scott Berry Host Feed-stated

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Can Borrowing Ever Increase Power?

    Nov 16, 202625 minAudio

    In this episode of "In the Interim…", Dr. Kert Viele, Director of Research at Berry Consultants, takes up a requested topic: published results showing that Bayesian borrowing of historical control data cannot increase...

    Credit the people on this episode

  2. Mailbag: Learning Healthcare, Multiplicities, and the Pace of Innovation

    Nov 9, 202647 minAudio

    In this mailbag episode of "In the Interim…", Dr. Scott Berry answers two listener questions and weighs in on the pace of innovation in clinical trial science. Asked what infrastructure a learning healthcare system...

    Credit the people on this episode

  3. Reflecting on EMA Reflections on Platform Trials

    Nov 2, 202644 minAudio

    In this episode of "In the Interim…", Dr. Scott Berry and Dr. Kert Viele, Director of Research at Berry Consultants, review the European Medicines Agency's draft reflection paper on methodological aspects in...

    Credit the people on this episode

  4. Statistical Methodology Groups in the Pharmaceutical Industry

    Oct 19, 202656 minAudio

    In this episode of "In the Interim…", Scott Berry is joined by Jenny Devenport and Kaspar Rufibach of Merck Group, co-authors of the EFSPI Statistical Methodology Leaders paper "Statistical Methodology Groups in the...

    Credit the people on this episode

  5. A Discussion on Virtual Controls

    Oct 12, 202647 minAudio

    In this episode of "In the Interim…", Dr. Scott Berry shares the framework he presented at a CSRC regulatory science hackathon on digital twins and synthetic controls in cardiovascular clinical trials. He separates...

    Credit the people on this episode