Question 27 from the first paper of 2022 asked the trainees to describe the design and characteristics of platform trials, a research instrument with which they are currently surrounded. Park et al (2020) and Berry et al (2015) are two excellent freely available overviews, and this Nature piece (2019) is also great if you can get a hold of it via your institution.
Platform trial design
Put in the simplest way:
"A “platform trial” is a clinical trial with a single master protocol in which multiple treatments are evaluated simultaneously"
- Todd Graves, ACTA
In other words, this is a study where multiple interventions are being compared to the same control group, whereas in a traditional clinical trial one intervention is investigated at a time.
This is a version of adaptive trial design, where the data - as it is being collected - modifies the course of the trial (eg. the distribution of enrolment of the patients into groups) as the trial goes on. In this fashion, where the treatment is clearly beneficial as compared to placebo, the control group shrinks in comparison to the treatment group as the trial runs its course because more and more people end up being enrolled in the treatment arm (Pallman et al, 2018). This calls for several stages of interim analysis during the trial, as the group sizes are reevaluated.
A natural extension of adaptive trial design is to run multiple concurrent intervention arms, alongside the control group. This multiarm multistage approach can then reevaluate each arm during the process of the study using a "drop the loser" model, where an underperforming arm might be closed because of a clear lack of effect (or evidence of harm).
This is already what might be referred to as a "closed" platform trial. An "open" platform trial is basically these same thing except the protocol permits the addition of other interventions during the trial, i.e. intervention groups can be either dropped or "picked up". Additionally, adjustments to the control group can be carried out, adding more control patients as needed (this would be essential if new arms were being added).
In case anyone is interested in the statistical complexities of adding new arms to trials, Lee et al (2021) is available to patiently talk you through it. It also opens a deep rabbit hole leading to the discussion of basket trials, umbrella trials, and adaptive enrichment design, all topics that are probably somewhat less important for the CICM exam candidate (as they have never appeared in the exam).
Advantages of platform trials
Why, one might ask, would we make life so difficult for our statisticians?
- "Disease-focused" instead of "intervention-focused: the trial investigates the best treatment for a disease from a range of investigated options, rather than looking at an intervention to see whether it is better than a comparison
- Efficient use of study resources: because there are often several candidate therapies for the same disease, the alternative would be to run several concurrent RCTs with separate control groups, each duplicating basically the same design and infrastructure. Combining all the intervention arms together reduces this wasteful redundancy.
- Easier to compare interventions: it is easier to compare the intervention groups in the same platform trial, than it is to compare several independent trials for the same therapies, because of the possible differences in methodology
- Resistant to obsolescence: Novel developments can be incorporated into the study protocol, maintaining currency over the lifespan of a long trial; or obsolete therapies can be dumped from the study so that resources can be reallocated (instead of persevering with an intervention that is losing popularity). This was particularly useful for something like an evolving pandemic, where novel therapies were constantly being proposed.
- Perpetual: the trial never actually needs to finish; new arms can be added and old arms removed under the same master protocol
- Minimises the size of the control group: this has multiple benefits, not the least of which is the ethical consideration that "active" or "beneficial" treatment is being provided to the maximum number of potential beneficiaries. This reduces the overall number of patients required.
- Superiority study: unlike using several independent trials, a platform trial can declare one treatment superior to the others
- Prevents underpowered results: multiple stages of interim analysis allow the investigators to adjust the size of the sample, which means you should not run into a situation where you suddenly discover that you did not enrol enough patients to detect an unexpectedly small treatment effect
- Decreased risk of participation: for the patients enrolled, the trial decreases the risk that they will be allocated to a control group (which is smaller) or to an ineffective treatment (which would be dropped from the study). This means that the benefits of participating are more equitably shared across the study population. It also means that - once you have finished studying an apparently ineffective treatment - you can keep using the already collected results and comparing them to new treatments in other arms added later, without having to expose another set of patients to the ineffective treatment.
Disadvantages of platform trials
- Data complexity: the added care that is required to manage the responsive randomisation, multiple stages of interim analysis, multiple arms, the design of a master protocol with longevity - all of these are extremely labour-instenive from a statistical and administrative perspective, which means some of the money saved by preventing multiple simple trials is instead spent on several professors of statistics and countless data management staff.
- Administrative complexity: each intervention may have its own corporate sponsor, its own additional consent requirements, and follow-up might be different, making the process of running a platform trial more complex.
- It never ends: a "perpetual trial" is a mixed blessing, as it may be harder to secure funding for an indefinite project from industry or government agencies (as the question "how much is this going to cost" becomes more difficult to answer).
- Planning is extensive and takes a while: the need to engage many experts and plan carefully may mean that this design is unsuitable for an intervention that needs to be investigated immediately
- Nonconcurrent control group could bias the results, for example where the standard of care changes over time
Examples of platform trials
Question 27 from the first paper of 2022 asked the trainees to list a couple of examples of platform trials, for 10% of the mark. The 2019 Nature article from the Adaptive Platform Trials Coalition contains an excellent table which covers a large number of these, which is much more than what that answer required. In summary, most of the adaptive platform trials over the later 2010s were mainly oncology and Alzheimers disease trials, with a few notable exceptions in the realm of critical care. Platform trials relevant to ICU practice include: