The ICU Trial That Should Embarrass Every Sponsor Hiding Behind “Ethical Complexity”

Picture a ventilated patient in a medical ICU. Sedated, intubated, unable to communicate. The standard response across most critical care units worldwide is to reach for a wrist restraint — because extubation risk is real, because nurses are stretched, because no one wants to be the clinician on duty when a delirious patient pulls a lifeline. Physical restraints in ICUs are pervasive not because they have been proven to improve outcomes, but because the absence of evidence has been allowed to function as a substitute for it. That era just ended.

Published in JAMA in 2025, the RESTRAINT trial — a multicenter randomized clinical trial conducted across French ICUs and led by investigators including Dr. Jean-Baptiste Lascarrou — directly compared a restrictive physical restraint strategy against a liberal one in critically ill patients receiving invasive mechanical ventilation. The primary outcome was the number of days alive and free of invasive mechanical ventilation at 28 days. The results carry implications that extend well beyond the ICU: a well-designed RCT can and should operate in the most ethically fraught, operationally complex patient populations in medicine. The question is why so few sponsors are willing to try.

The conventional wisdom — repeated in IRB pre-submission meetings, echoed in sponsor risk registers, and codified into cautious protocol exclusion criteria — is that critically ill patients represent an insurmountable operational and ethical obstacle to rigorous randomized design. Patients cannot consent. Surrogates are difficult to reach. The clinical environment is unstable by definition. Most sponsors confronting a similar design challenge would either exclude the population entirely, run an observational registry, or use a before-and-after cohort study and call it evidence. The RESTRAINT investigators did none of that.

The Protocol That Held

The trial enrolled 447 patients across 13 ICUs. The restrictive arm used physical restraints only as a last resort, after structured non-pharmacological alternatives had been attempted and failed. The liberal arm permitted restraint use according to standard nursing judgment — which, in practice, meant frequent application. This is not a soft pragmatic trial with vague outcome definitions. The investigators tracked patient-centered endpoints including coma-free and delirium-free days, serious adverse events including accidental extubation, and 90-day mortality. They built the protocol around ICU reality, not against it.

And the ethics framework held. Emergency consent procedures and surrogate consent pathways were embedded from the start, not retrofitted when enrollment stalled. This is the operational lesson most sponsors miss: ethics architecture is not a barrier to trial design — it is part of the design. When the IRB sees a protocol that has already answered the hard procedural questions about consent in incapacitated populations, the review is faster, the approval is more durable, and the deviation rate during conduct goes down.

The trial’s ability to randomize at all in this setting required what FDA’s guidance on human subject protection in emergency research has long called the “exception from informed consent” (EFIC) pathway — a mechanism created precisely for situations where the standard consent model collapses under clinical urgency. The RESTRAINT trial’s European context meant a different regulatory framework applied, but the methodological parallel is direct: FDA’s EFIC regulations, codified in 21 CFR 50.24, exist because the agency recognized decades ago that excluding critically ill populations from rigorous research was producing a body of critical care evidence built almost entirely on inference and institutional habit.

That logic holds — until you look at how rarely sponsors actually use it. A 2019 analysis published in Critical Care Medicine found that fewer than 200 trials had utilized EFIC in the United States since the regulation took effect in 1996. Fewer than 200, across three decades, in a country where approximately 5.7 million patients are admitted to ICUs each year according to the Society of Critical Care Medicine. The research gap is not a regulatory problem. It is a sponsor risk-tolerance problem masquerading as an ethical one.

What Rigorous Design Actually Costs

The counterintuitive truth embedded in the RESTRAINT trial is this: designing for complexity is cheaper than designing around it. Every sponsor who excludes fragile populations to simplify enrollment is not reducing trial risk — they are transferring it forward into post-market surveillance, label restrictions, and the inevitable label expansion trial they will have to run anyway when regulators or payers demand real-world evidence in the populations that actually use the product. The FDA’s guidance on real-world evidence issued in 2023 is explicit that RWE submitted in lieu of prospective data faces a higher evidentiary bar, not a lower one.

The RESTRAINT trial enrolled its 447 patients in a population with a median SOFA score indicating severe organ dysfunction — not a mild-to-moderate population selected to generate a clean signal. The 28-day primary endpoint was analytically demanding: days alive and free of invasive mechanical ventilation is a composite that captures both survival and recovery trajectory simultaneously, a design feature that ICH E9(R1)’s 2019 addendum on estimands would recognize as methodologically sophisticated. The investigators were not simplifying. They were doing the harder thing because the harder thing generates the evidence that actually changes practice.

Consider the contrast with the sepsis RCT landscape. The PROWESS trial of drotrecogin alfa, sponsored by Eli Lilly, enrolled 1,690 patients with severe sepsis and generated an approval in 2001 on the strength of a 6.1% absolute mortality reduction. That drug was later withdrawn in 2011 after the PROWESS-SHOCK trial in 1,697 patients found no survival benefit — a reversal that cost the field years of misallocated clinical resources and exposed the danger of trials designed to support approval rather than to answer the biological question cleanly. The RESTRAINT trial is designed the other way around: the question comes first, the enrollment criteria serve the question, and the outcome is defined to capture what matters to patients and families, not what is easiest to measure.

The Regulatory Signal Sponsors Are Missing

FDA’s 2023 guidance on clinical trial diversity made explicit what the agency had been signaling for years: sponsors who systematically exclude vulnerable or hard-to-enroll populations will face mounting pressure to justify those exclusions in pre-IND and Type B meetings. The guidance stops short of mandating inclusion — but the direction of regulatory travel is unmistakable, and sponsors who are still writing “ICU patients” into blanket exclusion criteria are accumulating a liability they have not yet been asked to pay.

The RESTRAINT trial proves the liability is avoidable. A multicenter design across 13 sites, emergency consent protocols, a 447-patient enrollment in a critically ill population, and a primary endpoint with enough statistical power to detect a clinically meaningful difference — this is not a heroic outlier. It is a reproducible methodology. The operational infrastructure exists. The regulatory pathway exists. What has been missing is the institutional willingness to treat ethical complexity as a design challenge rather than a design veto.

Sponsors building protocols in critical care, perioperative medicine, emergency neurology, or any domain where patients arrive incapacitated should read the RESTRAINT trial not as a clinical result but as an operational proof of concept. The 28-day primary outcome data will inform ICU practice. But the consent architecture, the site training model, the deviation management framework embedded in a protocol that had to account for a patient population that could not speak for itself — that is the transferable asset.

The FDA is watching who builds that infrastructure and who continues to file Type B meeting requests asking permission to exclude the hard patients. The answer to that request is coming, and it will not be the one sponsors are hoping for.

References

1. JAMA — “Restrictive vs Liberal Physical Restraint Strategies in Critically Ill Patients” (2025)
2. FDA — Guidance on Exception from Informed Consent Requirements for Emergency Research (21 CFR 50.24)
3. FDA — Diversity Plans to Improve Enrollment of Participants from Underrepresented Populations in Clinical Studies (2023)
4. FDA — Guidance on Real-World Evidence Program Framework (2023)
5. ICH — E9(R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials (2019)
6. Critical Care Medicine — Analysis of EFIC Trial Utilization in the United States (2019)
7. Society of Critical Care Medicine — ICU Admission Statistics (published data)
8. NEJM — PROWESS Trial of Drotrecogin Alfa in Severe Sepsis, Eli Lilly (2001)
9. NEJM — PROWESS-SHOCK Trial of Drotrecogin Alfa, Eli Lilly (2011)