From the JAMA Network, this is JAMA Clinical Reviews, interviews and ideas about innovations in medicine, science, and clinical practice. Hello and welcome to JAMA Clinical Reviews. Today we'll be discussing the topic of platform trials and specifically addressing how a clinician should interpret and use the results of new clinical evidence produced through a platform trial. I'm your host, Dr. Roger Lewis, statistical editor for JAMA and co-editor of the JAMA Guide to Statistics and Methods series. I'm here with Jay Park and Ed Mills two well-known experts in this area. Let's start by having you introduce yourselves. Hi, my name's Jay Park. I'm an assistant professor at McMaster part-time. And uh, I currently act as a director of Platform Trials initiative within the Canadian Critical Care Trials Group. And my name is Ed Mills. I'm a professor at McMaster University in the Department of Health Research Methods, evidence and Impact. And I run a platform trial at the moment in Covid. Great. Thank you very much and thank you for joining us today. So I'd like to start with a general question about platform trials specifically for our clinicians. What is a platform trial and why is it important? Sure. Platform trials referred to a type of randomized clinical trials. That allow for multiple interventions to be evaluated simultaneously, and they're designed with the flexibilities of allowing new interventions to be added into the platform after the trial starts. This is a key difference that differentiates between a platform versus more conventional clinical trials that the readers, uh, may be aware of. In conventional clinical trials, you have to pre specify how many interventions and what types of interventions are, um, included in your protocol by design, uh, you can't add a new intervention into the existing clinical trial. Whereas in platform trials, they're designed to allow for new interventions to be added throughout the conduct of the trial. So would it be correct to say that a platform trial is intended to continue longer than the time it takes to evaluate any individual therapy? Yes. Platform trials are generally designed to be perpetual, meaning they're either designed to be last for a very long time or never end with different therapies and different centered care to be added into the clinical trial. Great. So one of the things that I've often heard is the term master protocol. Help me understand the difference between a master protocol and a platform trial, or how those two terms relate to each other. Yes. So a master protocol generally refers to a set of documents that outlines sort of the standardized operating procedures, and they're designed to evaluate multiple interventions. So, the key difference between a master protocol and a platform trial is the master protocol is the protocol documents that guides how platform trials are conducted. Platform trials referred to clinical trial themselves or can be referred to as a type of design as I had mentioned. . Where multiple interventions can leave and enter the platform at different times. So you're both co-authors on our recently published users guide on the use and interpretation of results of a platform trial. Can you explain for our listeners why there was a need for writing a user's guide on this topic? So there was a need for guidance for clinicians on how to interpret platform trials because the Covid Pandemic has created an opportunity for more platform trials than has ever existed before. Prior to the Covid Pandemic platform, trials were relatively rare and had been used predominantly in fields such as oncology. So most clinicians would not be exposed to it. Then Covid came along and big trials like the recovery trial, the solidarity trial, the REMAP cap, the Together trial, the principle trial. Most of the trials giving useful information in Covid have ended up being platform trials. And what has been clear is that decision makers, whether it's clinicians or guideline panels, have difficulty interpreting what are the strengths and weaknesses of, uh, different platform designs. Dr. Mills, you made the point that most of the major therapeutic advances in Covid have come from platform trials. That's really a remarkable statement. What do you think has created this change in the way we generate evidence for covid, say, compared to other questions in critical care or oncology previously? I think one of the reasons that platform trials were adopted quickly in Covid was the experience of the number of the investigators who many had been in contact with each other beforehand. There was clearly a need to respond quickly and to try out a variety of interventions that often were repurposed drugs. So you're seeing that the majority of these platform trials, at least for the first year and a half of the, uh, covid pandemic, we're using repurposed drugs on average with the few exceptions, as the pandemic draws on and there's a longer period of time and the other interventions are becoming available, uh, we're seeing, you know, new pharmaceutical interventions being put into these platform trials. But I think that one of the key successes, or one of the reasons that platform trials have had the profile that they have is that they had a variety of different interventions available that needed to be tested rapidly. If we used the conventional approach of testing them one by one, we would only have a small fraction of those interventions evaluated by now. So it bought us a lot of time and saved a lot of time and resources. So, it sounds like one of the key settings in which a platform trial approach is most useful is when there's multiple available therapies ready for testing that need to be tested quickly. Can you think of other characteristics or other settings in which a platform trial would be particularly advantageous? Sure. Well, I think one of the great advantages to a platform trial is the idea of it being infrastructure built and done in a perpetual way. So, there's no doubt that at some point we will have built all this infrastructure for Covid and hopefully Covid will not be the emergency that it is. And we will be looking at that time to pivot and reapply the infrastructure that's been built into, uh, other conditions. But I think that the strength of platform designs is the infrastructure building. You don't have to have five or six different interventions going on at a trial at any time, so long as you might be prepared to have that occur if the need was there. Great. Thank you. So if one of the goals of a platform trial is to enable the efficient evaluation of multiple experimental therapies, let's talk a little bit about what you need to know ahead of time to get started. So for example, do you need to know ahead of time what the list of potential therapies are that you might be interested in testing? Yeah, so I guess that's the beauty of platform trials. Um, so initially the conduct of the platform trial starts as like a regular conventional clinical trials that you have the list of interventions that you want to test, and as new scientific discoveries gets made and either, um, sort of the standard hearing changes, uh, you can add new intervention into the existing platform throughout. So if you don't necessarily know each of the treatments that you're gonna be evaluating in the clinical trial ahead of time, how do you handle the ethics committee or institutional review board evaluation of those trials? It's difficult for me to see how an IRB could approve a trial without knowing what all the treatments are that would be included. So what some of the investigators of, uh, ongoing platform trials have done is that they submit their master protocol for an ethics review. And as new therapies come in, they still go through an ethics review, but it may be an ethics amendment rather than the brand new review. So you could speed up, uh, the speed in which the, uh, you can start the clinical evaluation for that new intervention that needs to be answered. So, Dr. Park, one of the things that you mentioned in your article is this approach of response, adaptive randomization. Could you describe for our listeners what is response adaptive randomization, and what are its advantages or potential disadvantages, both from the point of view of the individual subject being enrolled in the trial and for us as we try to draw rigorous conclusions from the platform trial. Sure. Uh, great question. Uh, response of adaptive randomization is a type of adaptive trial design in which the allocation ratios are adapted over time based on, uh, accumulated interim results in favor of the interventions arms that are performing better. How this occurs is that a platform trial may start with equal randomization. And as we observe more data throughout the trial, the allocation ratios are preferential increase to the interventions arms that are performing better. So for clinicians and patients who may be wary of, uh, you know, participating in a randomized clinical trial, this actually can relieve some of their concerns. Because they will have a higher probability of randomly being assigned to intervention arms that are performing better. So I was taught many years ago that the most efficient randomization ratio in a trial is one-to-one randomization between two treatment groups. This seems as if response adaptive randomization takes us away from that. So do we lose efficiency by doing this? Yeah, so in a two arm setting, you're gonna have the highest fiscal power when it's, as you said, an equal randomization. Response adaptive randomization may not be efficient in a two arm setting, but in a platform trial setting where you are evaluating multiple interventions at a given time, it actually could improve the overall probability of detecting the best treatment. Given that you maintain an adequate allocation ratio to the control group. Now, one of your motivations in writing this article was to help readers interpret the results of platform trials. So I'd like to come back to that goal and ask you what are the things that readers should look for in assessing the validity or the rigor of a platform trial? So one of the questions is were pre-specified plans in place for interim and statistical analysis that are used in the trials? Were the pre-specified plans applied equally to all interventions? This is important because one of the aspects of platform trials is that we advocate for interim evaluation, so looking at the data in a preplanned early way. But what we want to do is increase the efficiency of the trial, the likelihood of detecting an effect, or deciding not to pursue an intervention if one is not convinced by the interim analysis. Where this can become problematic is if there's a greater emphasis placed on the findings of an interim analysis than, for example, the overall trial findings. And we've seen that in some covid trials so far. So there clearly needs to be in place. Some rules for the data safety monitoring committee to look at the data and how they're going to communicate that and, uh, make judicious decisions at the end of it. The second question to consider is, did the investigator include non concurrently randomized participants into the Cisco analysis or limit the analysis to the concurrent randomized patients? So a concurrent control group refers to participant who were randomized to the control arm during the same time period in which the parts were randomized to a particular intervention group. So this is different from a non-concurrent control participants who may have been randomized to the control group before the intervention groups started enrolling patients into the trial. So, this non concurrent control is, issue rises in a platform trial because a given intervention, let's just say treatment A, treatment X may enter into the platform at a different time [INAUDIBLE]. Which means that there may already be a group of patients who have been randomized to the controlled arm at an earlier time point. So this is what we refer to as a non-concurrent control patients. So Dr. Park, what are the risks of associated with including non-concurrently randomized patients in the control group when evaluating a therapy in a platform trial? So when you include non-concurrent controlled patient, you wanna watch out for potential temporal biases or differences that exist between a concurrently randomized to non-concurrently randomized. As patients were randomized at a different time point, there may be a change in patient characteristics, uh, change in standard care and other issues that may raise to temporal variability between concurrently randomized to non concurring randomized clinical trials. So the response rate may differ between these two types of control group. So you wanna account for these temporal variability in your Cisco uh, analysis. So potential downfall is that if you're inadequately able to account for these temporal differences, then you may bias results and reach erroneous conclusions. Great. And I did notice that in your, uh, user guide, you make a distinction between the use of non concurrently randomized control patients. Who may be subject to these temporal changes or biases that you mentioned and the use of historical data or data, say from other trials. Can you clarify that distinction for our listeners? Yes, so historical data generally could be a clinical trial that has been conducted independently in the past, or it could be a source of data from an observational dataset. The difference between a non concurrent control versus historical data is that the use of master protocol. In a platform trial, even though participants were randomized at different time point, they're overseen by standardized operating procedures and same analysis plan that are outlined in the master protocol. So the key distinction here is that historical data, in addition to temporal biases, there are differences in which, how the data were collected and the Cisco analysis was conducted. Great, thank you. Now let's go back to those four specific questions that you outlined in the user's guide that clinicians should ask themselves when evaluating the risk of bias in a platform trial. So we've addressed the use of pre-specified plans for interim and statistical analysis, and the issue of the concurrency or non-concurrency of control subjects. Dr. Mills, can you address the third question that you outlined in your publication? Sure. So the third question is, did the investigators minimize risk of bias from information flow within and outside the specific platform? I think in many ways it's relatively easy to minimize the risk of bias from information within a clinical trial because evidence that's created and presented to a data safety and monitoring committee should, in general, be blinded from the investigators and certainly from those running and participating in the trial. That part is relatively easy to maintain. What I think is very difficult to maintain is the influence of other clinical trials outside of your platform trial and whether or not that's likely to influence either the decision making of participants to be in your clinical trial or whether or not the standard of care is changing due to other clinical trials occurring, and to a certain extent, this hasn't been resolved about how to deal with changing standard of care, but there are statistical methods to evaluate whether or not the underlying, um, control event rate changed according to differences in standard of care that I think needs to be taken into consideration. So, Dr. Mills, how is that different for a platform trial from any other trial that may be conducted over a substantial period of time? Well, I don't think that there's that much of a difference between a platform trial and a long recruiting clinical trial. Obviously, we're learning a lot during the Covid pandemic about these types of designs. Probably more has been gained during the pandemic in terms of, uh, learning the methods and applying new methods for platform trials than ever existed before. But what has clearly become the case during the pandemic was the underlying base control event rates changing, the variants of interest changing and the, uh, available standard of care changing. So, platform trials have had to respond to this in the best way possible. It's been a challenging circumstance and I'm not sure there's consensus on exactly how to deal with it, but it's certainly something that readers will want take a look at. Great. Thank you. Um, and let's go to the fourth point that you make regarding what readers should look at. So the fourth question for the readers to consider is whether the investigators follow existing reporting guidelines for the reporting of their clinical trials. So the, this is not too much different for conventional clinical trials versus, uh, platform trials. That, you know, there are existing, uh, concert [INAUDIBLE], uh, guidelines that the platform trial investigators should use when importing their clinical trial results. And there's been a recent development of a concert extensions. There are specifics to adaptive clinical trials that the platform trials can use to improve the transparency in their reporting. Now I'd like to go back to the issue of changing practice and how a platform trial can maintain its rigor in the setting of rapidly changing clinical standards. And this is obviously something that we've all seen to a tremendous extent with Covid. So one of the things that has happened over the last couple of years is that the standard of care for the treatment of patients with covid, whether in the outpatient or inpatient or intensive care unit setting, has rapidly changed. And as a general rule, the control therapy included in a clinical trial should represent the best available standard of care and the location in which the patient is being treated. So, with Covid or with similarly rapidly developing therapeutic areas, can the control therapy within a platform trial change? And if so, how do you do that? So the reality is that the standard of care has changed in a variety of conditions over the last few years, and in Covid, for example, the standard of care for hospitalized patients has changed since the early days with the, for example, the discovery via the recovery trial of dexamethasone being a useful intervention in the hospitalized setting that changed what the standard of care is. Obviously, we have an ethical responsibility to provide the best level of care to patients, whether they're in the active arm or in the control arm. This creates a problem for platform trials because you're making changes during the conduct of a trial, and it's very rare that a condition would change as rapidly as has occurred during Covid, and it gives you an option of either stopping your trial and starting again. And then you have the risk that there's going to be another big shift in standard of care or pre-planning for changes in your statistical analysis plan, which takes into consideration the different temporal effects of the time effects and what was going on at those particular times. And of course, you can test for whether or not there's important differences to determine whether or not the time period should be combined or not. It's something that we're continued to learn about, but there's no doubt that there's a responsibility of the trial is to maintain their trial being up to date with whatever the standard of care is in that local community. So, one of the goals of using the platform trial approach as opposed to conducting a series of independent trials is to obtain statistical or logistical efficiencies. So we learn faster about which therapies work and which therapies don't give us the benefits that we'd hoped for. Can you briefly summarize the sources of those efficiencies in the platform trial approach? There are several areas for efficiencies in platform trial. By having a common control group, instead of conducting a series of two arm trials where a duplicate number of control arms is created, you can use a one common control group for Cisco efficiencies and platform trials also commonly use adaptive trial design features such as sequential designs that allow for early stopping. For instance, if the intervention arm is not doing so well and they have a low probability demonstrating eventual uh CISCO? Fiscal?, significance, you can cut with the losses early and you can use designs such as response adaptive randomizations in a multi-arm setting as in the platform trial to screen out multiple interventions efficiently over time. And there are operational efficiencies as well. In a platform trial, instead of having to conduct a series of two arm trial where you have to develop the protocol and close out the trial sites and et cetera, you can maintain a single infrastructure and a single protocol and update them over time. Dr. Mills, Covid has really shown us the power of the platform trial approach to rapidly addressing pressing questions regarding the effectiveness of therapies. But as we move forward and try to expand this approach when, when appropriate into other disease areas, or even expand the use of platform trials in Covid or critical care, what do you think the major challenges are for those conducting research and funders of research? So, I'm of the belief that many, many fields of medicine should start developing the infrastructure for platform trials. Now, I will caution that there's a high bar to entry and even those of us involved with clinical trials for the last several decades, maybe unfamiliar with the number of the developments that occurred within adaptive designs. The Bayesian statistics that are often applied to platform trials and some of the challenges of administration and the infrastructure requirements, as well as the long-term funding that may be required for conducting this type of trial. I think that platform trials are likely to do best in the public sector where there are a variety of interventions that can be evaluated for different conditions. It's particularly appealing for diseases that there is less drug development occurring within. I think industry applying platform trials is going to be a real challenge for us to determine whether or not decisions that are being made are legitimately done for the purpose of evaluating a drug or whether or not they're being done to increase the likelihood of detecting an effect or a beneficial effect. So, there's no doubt that there are challenges to platform trials. There are great advantages to it and in the long run I think it'll save money and save resources and get answers to patients a lot quicker than conventional designs. But there's also a lot of opportunity for manipulation of them, and I think clinicians and uh, the methods community needs to pay a great amount of attention to that. This is Roger Lewis and I'd like to thank our guests today, Dr. Jay Park and Ed Mills. This episode was produced by Jesse McQuarters at the JAMA Network. The audio team here includes Daniel Morrow, Shelly Steffans, Lisa Hardin, Audrey Forman, and Mary Lynn Ferkaluk. Robert Golub is the Executive Deputy Editor for Digital Media. To follow this and other JAMA Network podcasts, please visit us online at jamanetworkaudio.com. Thanks for listening.