Platform Trials Are Rewriting the Rules of Evidence. Regulators Haven’t Caught Up.

Picture a trial that never ends. New pathogens emerge, old arms close, new interventions rotate in, and the statistical engine hums continuously in the background, allocating patients toward what is working and away from what is not. No press release announcing enrollment completion. No two-year gap while the sponsor waits for an NDA review cycle. Just a living study, producing decisions in near real time. That is the promise of adaptive platform trials, and the infectious disease community has spent recent years proving the concept works at scale. The harder question, the one the Lancet’s recent World Report on platform trials quietly surfaces without quite answering, is whether the regulatory and operational infrastructure surrounding these trials has kept pace with the science inside them.

It has not. And the gap between what platform trials can do biologically and what the system is built to support is where sponsors are getting hurt.

What the Numbers Actually Proved

Start with REMAP-CAP, the Randomized, Embedded, Multifactorial, Adaptive Platform trial, because it is the clearest demonstration of what this design unlocks at scale. Between March 2020 and June 2021, the trial enrolled 4,869 critically ill adult COVID-19 patients across more than 300 sites in 25 countries. By January 2021, the broader REMAP-COVID sub-study had enrolled over 9,000 participants. Those are not Phase III numbers. Those are population surveillance numbers, generated inside a randomized controlled framework, while the pandemic was still actively evolving.

That enrollment velocity matters for a reason most commentators understate. In a conventional parallel-arm trial, you lock your protocol, enumerate your arms, and wait. If a more promising intervention appears mid-study, you watch it from outside the fence. REMAP-CAP did not watch. It rotated arms, updated priors, and reassigned allocation probabilities as evidence accumulated. Immunoglobulin therapy, anticoagulation strategies, corticosteroid dosing, each of these questions got tested inside the same master protocol infrastructure, sharing a control arm, sharing operational overhead, sharing the statistical machinery. The savings in time and sample size are not marginal. A separate conventional RCT for each of those questions would have consumed years that COVID patients did not have.

The RECOVERY trial, run out of the University of Oxford with NHS infrastructure, demonstrated a complementary capability: the platform’s power to generate clean, regulatory-grade efficacy signals at speed. In RECOVERY, the combination of casirivimab and imdevimab (REGEN-COV) significantly reduced 28-day mortality in hospitalized COVID-19 patients who had not mounted their own antibody response, a finding that emerged from a platform already running thousands of patients through other arms. The control arm was shared. The site infrastructure was shared. The result was a clean, interpretable efficacy signal that reached regulators faster than any standalone trial design would have permitted.

So the science works. The operational model works. The enrollment model works. Which raises the uncomfortable question of what, exactly, is not working.

The Regulatory Scaffold Has a Structural Problem

In December 2023, the FDA published a draft guidance titled “Master Protocols for Drug and Biological Product Development,” which addresses the design and analysis of trials conducted under a master protocol, with primary focus on randomized umbrella and platform trials intended to contribute to demonstrations of safety and substantial evidence of effectiveness. That document is useful. It is also, for sponsors actively running adaptive platform trials in 2026, already behind the operational frontier.

The guidance addresses the conceptual architecture of platform trials reasonably well. What it does not resolve is the operational tension that every sponsor running one of these trials eventually collides with: the regulatory review unit is built around discrete submissions, and a platform trial by design never produces a discrete submission moment. Arms open and close on statistical triggers. Interim analyses generate action recommendations before anyone files a document. The master protocol evolves. Amendments accumulate. And the FDA’s review timelines, built around the assumption that a sponsor submits a complete package and waits, do not map cleanly onto a design where the evidence base is perpetually updating.

Consider what happens at the arm-closure stage. When a platform trial drops an arm because it crossed a pre-specified futility boundary, that decision is statistically clean inside the trial. But from a regulatory standpoint, the data from that arm still exists, still requires handling decisions, and still has implications for any future sponsor who wants to revisit that intervention in a different population. The guidance provides no specific framework for how closed-arm data from a platform trial should be archived, accessed, or incorporated into subsequent regulatory submissions. Sponsors are improvising, and improvisations at the evidence layer create downstream problems at the approval layer.

Endpoint selection compounds this. In an infectious disease platform trial designed for pandemic responsiveness, the endpoints often shift with the pathogen’s clinical presentation. Early COVID platform trials used 28-day all-cause mortality as the primary endpoint because that was the appropriate measure for critically ill ICU patients. As the disease shifted to milder presentations in vaccinated populations, that endpoint became less sensitive, less clinically meaningful, and in some populations, nearly impossible to power around. But modifying primary endpoints mid-platform is a regulatory minefield, even when the epidemiological rationale is ironclad. The FDA’s adaptive design review process was not built for endpoints that need to track a moving pathogen.

The Data Integrity Pressure Nobody Is Talking About

The operational scale that makes platform trials powerful also creates a data integrity exposure that is genuinely new in kind, not just in degree. REMAP-CAP ran across more than 300 sites in 25 countries simultaneously. RECOVERY enrolled across NHS sites throughout England. When a single master protocol governs data collection at that scale, across that many jurisdictions, with arms opening and closing on rolling timelines, the EDC infrastructure and the data governance model underneath it carry weight that conventional trial systems were not designed to bear.

Response adaptive randomization, the statistical engine that steers enrollment toward better-performing arms, requires near-real-time data cleaning. If site data from a high-enrolling center in one country is systematically delayed or miscoded, the allocation algorithm responds to a distorted signal. It allocates patients based on what the data says, not what is actually happening in the wards. In a two-arm RCT, that kind of data lag creates noise. In a response-adaptive platform, it creates directional error: the algorithm actively steers enrollment away from what might be an effective therapy because the clean data has not arrived yet to reflect its true performance.

Response adaptive randomization is not uncontested methodology. Proschan and Evans argued in Clinical Infectious Diseases in 2020 that adaptive allocation should be resisted in most settings. Korn and Freidlin have shown that time trends make it inefficient and prone to bias. Thall and colleagues documented sample size imbalance running in the wrong direction. Those critiques cut toward this argument rather than against it. If allocation is already vulnerable to drift in the underlying data, the governance of that data while the trial runs is not a secondary operational concern. It is what determines whether the design works at all.

The 2023 master protocols draft guidance is silent on real-time data governance standards for response-adaptive allocation. The adjacent instruments that do exist, ICH E20 on adaptive designs and FDA’s 2019 guidance on adaptive designs for drugs and biologics, govern statistical conduct and trial integrity rather than the quality of the data feeding the allocation algorithm while the trial is running. Sponsors running these trials have built internal protocols to manage it. But internal protocols are not regulatory standards. General data integrity expectations do exist, in ICH E6(R3), in the EMA guideline on computerised systems and electronic data in clinical trials, and in MHRA’s GxP data integrity guidance, and a reviewer can hold a sponsor to those. What none of them supplies is a standard for judging whether real-time governance of an adaptive allocation engine was adequate, which is the question a platform trial submission actually raises. That is not a small gap. That is the kind of gap that becomes a Complete Response Letter after a two-year review cycle.

The irony embedded in all of this is precise. Platform trials emerged partly because conventional trial designs were too slow and too rigid to serve public health in a crisis. REMAP-CAP and RECOVERY proved that the design could produce decisive evidence at pandemic speed. But if the regulatory pathway for converting that evidence into approvals still requires the sponsor to navigate review processes built for static, single-intervention trials, then the speed advantage at the front end of the trial gets consumed by the friction at the back end. Any sponsor filing a fully adaptive platform-derived NDA for a novel infectious disease indication will learn exactly how wide that gap is, and every filing after it gets measured against whatever precedent that first collision sets.

Editor’s note, 28 September 2026: an earlier version of this article stated that no FDA guidance document addresses this failure mode. That overstated the record. The claim has been scoped to the 2023 master protocols draft guidance, and the adjacent instruments that do exist are now named.

References

1. The Lancet, “Platform trials: the study design transforming infectious diseases” (Talha Burki, 2026)
2. FDA Federal Register, “Master Protocols for Drug and Biological Product Development: Draft Guidance for Industry” (December 2023)
3. GCare Research, REMAP-CAP COVID Overview: enrollment figures and site data
4. RECOVERY Trial, Casirivimab and imdevimab (REGEN-COV) 28-day mortality outcome press release
5. Proschan M, Evans S. “Resist the Temptation of Response-Adaptive Randomization.” Clinical Infectious Diseases, 2020.
6. Korn EL, Freidlin B. “Outcome-Adaptive Randomization: Is It Useful?” Journal of Clinical Oncology, 2011; and “Time trends with response-adaptive randomization: The inevitability of inefficiency.” Clinical Trials, 2022.
7. Thall PF, Fox PS, Wathen JK. “Statistical controversies in clinical research: scientific and ethical problems with adaptive randomization in comparative clinical trials.” Annals of Oncology, 2015.