Publication

• Title: Restrictive vs Liberal Physical Restraint Strategies in Critically Ill Patients: The R2D2-ICU Randomized Clinical Trial
• Acronym: R2D2-ICU
• Year: 2026
• Journal published in: JAMA
• Citation: Sonneville R, Couffignal C, Sigaud F, et al; R2D2-ICU Investigator Study Group. Restrictive vs liberal physical restraint strategies in critically ill patients: the R2D2-ICU randomized clinical trial. JAMA. Published online March 17, 2026.

Context & Rationale

• Background

  Physical restraint has long been used in ventilated ICU patients to prevent self-extubation and inadvertent device removal, but actual practice varies markedly between systems and units.
  A recent systematic review estimated that 41.6% of ICU patients are exposed to physical restraint, with very high between-study heterogeneity, underscoring how strongly practice depends on local culture, staffing, and local policy.
  The clinical rationale has always been largely inferential rather than trial-proven. Adverse-event literature links physical restraint with skin injury, subsequent delirium, neurofunctional impairment, and post-traumatic stress symptoms.
  Major ICU guidance on pain, agitation, and delirium had addressed sedation and delirium prevention, but did not establish an evidence-based case for routine prophylactic wrist restraint use. ¹²³
• Research Question/Hypothesis

  The published protocol hypothesised that, in mechanically ventilated ICU patients, a restrictive, low-use physical restraint strategy would increase the number of days alive without delirium or coma over the first 14 days, as compared with a systematic high-use strategy. ⁴
• Why This Matters

  The question is both clinical and ethical. Restraints may reduce immediate treatment interference, but they restrict autonomy, may worsen distress, and are often normalised despite weak evidence.
  A rigorous randomised trial was therefore necessary before routine prophylactic restraint could be defended as standard ICU care.
  Recent ethical analyses have argued that restraint should be a last-resort intervention, used only when the expected benefit clearly outweighs harm; R2D2-ICU directly tested whether routine liberal restraint actually delivers that benefit. ⁵

Design & Methods

• Research Question: In adults receiving invasive mechanical ventilation in ICU, does a restrictive, low-use wrist-strap physical restraint strategy increase the number of days alive without delirium or coma during the first 14 days, as compared with a liberal, high-use strategy?
• Study Type: Institutionally sponsored, multicentre, open-label, parallel-group, randomised, superiority clinical trial conducted in 10 ICUs in France.
• Population:
  • Adults admitted to ICU who had received invasive mechanical ventilation for less than 6 hours and were expected to require ventilation for at least 48 hours.
  • Patients had to be eligible for physical restraint prescription, meaning not already physically restrained at screening and without a pre-existing written prescription for restraint.
  • Key exclusions: delirium before ICU admission, dementia, anticipated alcohol withdrawal, primary neurological admission diagnoses, major visual or auditory impairment, pregnancy or lactation, Simplified Acute Physiology Score II greater than 65, do-not-resuscitate order, opposition to participation, or concurrent interventional delirium-related research.
• Intervention:
  • Restrictive, low-use physical restraint strategy using 2-point wrist straps only when severe agitation occurred, defined as Richmond Agitation-Sedation Scale score of 3 or higher.
  • The allocated strategy continued until day 14 in ICU, readiness for ICU discharge, or death.
  • All patients otherwise received standardised management of analgesia, sedation, delirium screening, ventilator weaning, and mobilisation.
• Comparison:
  • Liberal, high-use strategy with wrist straps applied systematically from day 0 and reassessed daily.
  • Restraints could be discontinued if the patient was awake and not delirious, or extubated and not delirious; restraints could be restarted for recurrent severe agitation.
  • The same co-intervention framework for analgesia, sedation, delirium assessment, weaning, and mobilisation was used in both groups.
• Blinding: Open label. Bedside blinding was not feasible. The primary outcome depended on structured RASS and CAM-ICU assessments, so outcome ascertainment was partly standardised but not fully blinded.
• Statistics: A total of 422 patients were required to detect a 1-day increase in days alive without delirium or coma, assuming a standard deviation of 3 days, with 90% power at a two-sided 5% significance level. The primary analysis included all randomised patients who provided consent for use of their data and analysed them as randomised; prespecified per-protocol, subgroup, and additional weighted per-protocol analyses were also performed.
• Follow-Up Period: Primary endpoint at 14 days; additional mortality, functional, cognitive, and psychological outcomes through day 90.

Key Results

This trial was not stopped early. A total of 405 patients were randomised and 396 were included in the primary analysis after 9 post-randomisation exclusions.

• The primary endpoint was clearly neutral: low-use restraint did not increase days alive without delirium or coma.
• Safety events were similarly uncommon in both groups, with self-extubation 9.2% versus 8.5% and unplanned device removal 1.0% versus 0.5%.
• No subgroup showed convincing heterogeneity. For the primary outcome, the adjusted mean difference was 0.47 days (95% CI −2.00 to 2.94) in patients with COVID-19 and 1.36 days (95% CI −0.28 to 3.00) in those with Simplified Acute Physiology Score II 45 or greater. Day-90 cognitive, functional, and post-traumatic stress outcomes were also similar, but those assessments were incomplete.

Internal Validity

• Randomization and Allocation: Allocation concealment was sound before enrolment. Randomisation was central, computer generated, used variable block sizes, and was stratified by site, age, and coma status.
• Drop out or exclusions: Of 405 randomised patients, 9 were excluded after randomisation, leaving 396 in the primary analysis. In the low-use group, 5 had no data collected; in the high-use group, 2 had no data collected and 2 withdrew consent. For the primary endpoint, 56 of 396 patients (14.1%) had at least 1 missing assessment. At day 90, only 180 of 396 patients completed formal follow-up assessments: 94/196 in the low-use group and 86/200 in the high-use group.
• Performance/Detection Bias: The trial was open label, and bedside nurses knew allocation after randomisation. That matters because the primary endpoint depended on RASS and CAM-ICU assessments rather than purely hard outcomes. Although trained study coordinators collected trial outcomes and standardised protocols were used, the editorial correctly notes that the primary outcome was not blinded. ⁶
• Protocol Adherence: Adherence was a major challenge. In the low-use group, 66/196 patients (33.7%) were exposed to restraint for at least 4 consecutive hours without severe agitation. In the high-use group, 57/200 (28.5%) had restraints removed outside predefined criteria. Despite this, treatment separation remained substantial.
• Baseline Characteristics: Groups were broadly comparable. Median age was 65 versus 66 years, male sex 65.3% versus 58.5%, medical admissions 95.9% versus 91.0%, SOFA score 7 versus 7, and comatose status at randomisation 88.2% versus 87.0% in the low-use and high-use groups, respectively.
• Heterogeneity: Clinical heterogeneity was moderate. Most patients were medical ICU admissions and respiratory failure dominated the admission diagnosis. However, the trial deliberately excluded several high-risk agitation syndromes, including alcohol withdrawal and primary neurological disease, which narrowed the phenotype.
• Timing: Timing of intervention delivery was early and pragmatic: patients had been invasively ventilated for less than 6 hours at screening. Median time from ICU admission to randomisation was 8 hours in the low-use group and 7 hours in the high-use group. However, because nearly 90% were already deeply sedated or comatose, the intervention may have been initiated before many patients entered the behavioural state in which restraint is most clinically debated.
• Dose: The behavioural “dose” separation was real. Over ICU stay up to day 14, mean time spent in wrist restraints was 1.9±4.2 hours in the low-use group versus 15.8±8.2 hours in the high-use group; median time was 0 hours (IQR 0 to 1.4) versus 16.8 hours (IQR 10.3 to 23.6).
• Separation of the Variable of Interest: Separation was strongest in the first several days: day 0 mean restraint exposure was 1.5 versus 17.8 hours, day 1 was 2.0 versus 19.6 hours, day 2 was 1.6 versus 18.3 hours, and day 6 was 2.7 versus 15.3 hours in the low-use and high-use groups, respectively.
• Key Delivery Aspects: Co-interventions were protocolised. Both groups received standardised analgesia, sedation, delirium screening, spontaneous breathing trial practice, and mobilisation pathways, which reduces the chance that the result simply reflects broader differences in ICU care.
• Crossover: There was no classic crossover between named treatments, but there was substantial bidirectional nonadherence to the assigned restraint strategy, which has a similar bias-towards-null effect in practice.
• Adjunctive therapy use: There was no evidence that lower physical restraint led to compensatory escalation of chemical restraint. Cumulative doses at 14 days were similar: sufentanil 0.09 versus 0.10 mg, propofol 1392 versus 1251 mg, midazolam 35.50 versus 30.18 mg, antipsychotics 1.72 versus 2.25 mg, and dexmedetomidine 0.44 versus 0.71 μg/kg.
• Outcome Assessment: The primary outcome was clinically relevant but not purely objective. Delirium assessment was attempted at 2462 of 3120 protocolised CAM-ICU time points (78.9%). Missing primary-outcome data were handled with a deterministic rule-based method and limited last-observation-carried-forward at the final scheduled assessment only.
• Statistical Rigor: The trial followed a prepublished statistical plan and appropriately adjusted the primary model for stratification covariates. However, the achieved analysed sample of 396 was below the planned 422, secondary outcomes were not adjusted for multiplicity, and nonadherence created unstable per-protocol estimates. Notably, the weighted per-protocol estimate for the primary outcome was 1.07 days (95% CI −0.12 to 2.19; P=0.055), whereas the standard per-protocol estimate was −0.81 days (95% CI −2.19 to 0.57), illustrating how contamination complicated causal inference.

Conclusion on Internal Validity: Internal validity is moderate. Randomisation and treatment separation were good enough to test a policy-level strategy, but open-label assessment, underachievement of the target sample size, substantial nonadherence, and incomplete long-term follow-up reduce confidence that smaller true effects would have been detected reliably.

External Validity

• Population Representativeness: The study population reflects a typical general adult medical ICU population receiving invasive mechanical ventilation in France, with respiratory failure as the commonest admission indication and nurse-to-patient ratios between 1:2.5 and 1:3.
• Centres and Practice Environment: Standard practice at participating centres was relatively restraint intensive: physical restraint at initiation of mechanical ventilation was reported as 75% to 100% across centres. Sedation protocols were present in all 10 centres, but routine delirium-scale use existed in only 5 of 10 centres and routine mobilisation protocols in 7 of 10.
• Important Exclusions: Results do not directly apply to patients with neurological admissions, alcohol withdrawal, delirium on admission, dementia, profound sensory impairment, pregnancy, or very high illness severity.
• Applicability: Applicability is strongest in ICUs where restraints are used prophylactically in ventilated patients and where staffing ratios approximate those in the trial. It is weaker in systems with much lower restraint use, richer bedside staffing, or a practice of using restraints mainly in awake hyperactive delirium.
• State at Randomisation: Because 87% to 88% of participants were comatose at randomisation, the findings are less transferable to jurisdictions where restraint decisions are made primarily in more awake, more behaviourally active patients. ⁶

Conclusion on External Validity: External validity is moderate. The trial is highly relevant to mechanically ventilated adults in general ICUs with similar staffing and restraint practices, but much less generalisable to the highest-risk agitation populations or to units that already use restraint sparingly.

Strengths & Limitations

• Strengths: First multicentre randomised trial directly comparing low-use and high-use physical restraint strategies in mechanically ventilated ICU patients.
  Clear intervention contrast with substantial actual restraint-time separation.
  Protocolised co-interventions for sedation, delirium assessment, weaning, and mobilisation.
  Clinically meaningful primary endpoint focused on brain dysfunction rather than restraint rates alone.
  Extended 90-day follow-up including cognition, function, and psychological outcomes.
• Limitations: Open-label design with unblinded primary outcome assessment.
  Analysed sample size fell short of the planned 422 patients.
  Nonadherence was substantial in both arms.
  Most patients were deeply sedated or comatose at randomisation, potentially diluting any delirium-specific effect of the intervention.
  Key high-risk groups were excluded.
  Long-term functional, cognitive, and PTSD follow-up was incomplete.
  Patient, family, and staff experience, plus bedside workload, were not formally measured.

Interpretation & Why It Matters

• Routine prophylactic restraint

  R2D2-ICU does not support the routine application of wrist restraints to all mechanically ventilated adults in ICU as a way to improve delirium- or coma-related outcomes.
• Safety trade-off

  Reducing restraint use did not produce an obvious penalty in self-extubation, device removal, pressure injuries, ICU mortality, or day-90 mortality. That is the trial’s most clinically useful signal.
• Implication for practice

  In ICUs similar to those studied, clinicians can reasonably move away from universal prophylactic wrist restraint and instead reserve restraint for clearly justified, short-duration situations of severe agitation or immediate safety risk. ⁶⁷
• How it changes assumptions

  The trial refutes a longstanding but weakly evidenced assumption that more restraint necessarily yields safer or more delirium-free ICU care. It also sharpens the distinction between a strategy that is ethically intuitive and one that is demonstrably effective. ⁶

Controversies & Other Evidence

• The most important design controversy is whether the trial tested the right patients. It excluded neurological admissions, anticipated alcohol withdrawal, delirium on admission, and dementia, while 87% to 88% of enrolled patients were comatose at randomisation. That makes the result less informative for the very agitation-prone populations in whom restraints are most often defended. ⁶
• The second controversy is contamination. The low-use arm still included restraint exposure in 36.2% of patients, and protocol-level nonadherence reached 66/196 in the low-use group and 57/200 in the high-use group. This likely diluted any real effect. The instability of the per-protocol estimates reinforces that point: weighted per-protocol analysis trended towards benefit, whereas standard per-protocol analysis trended in the opposite direction.
• A further debate concerns whether the primary outcome was the optimal one for the intervention. Delirium- and coma-free days are patient-centred, but in a population that was predominantly deeply sedated at randomisation, the endpoint may have been driven more by underlying illness and sedation practice than by restraint strategy itself. ⁶⁷
• The pretrial literature was mixed. A 2025 meta-analysis of guide-based interventions reported lower delirium incidence and lower unplanned extubation with restraint-reduction approaches, but those studies were heterogeneous, smaller, and usually not direct patient-level comparisons of two explicit restraint strategies. R2D2-ICU is therefore a more direct test and materially tempers the optimism of those pooled estimates. ⁸
• Related commentary immediately before the trial illustrated live equipoise. One 2026 article argued that the restraint-free ICU is “not quite there yet”, whereas another argued that physical restraints should no longer be needed in ICU routine care. R2D2-ICU lands between those poles: it does not justify universal restraint, but neither does it prove that restraint can never be necessary. ⁹¹⁰
• Broader contemporary evidence continues to frame restraint more as a potential harm than a proven protective therapy. Reviews describe associations with skin injury, delirium, neurofunctional impairment, and post-traumatic stress, while ethical analyses recommend restraint as a last-resort intervention that should be brief, justified, and proportionate. ²⁵
• No later formal ICU guideline incorporating R2D2-ICU is yet available. In that vacuum, the PADIS framework remains the dominant critical care reference, and this trial strengthens a practice model centred on analgesia optimisation, lighter and protocolised sedation, delirium screening, spontaneous awakening and breathing trials, and mobilisation rather than reflex prophylactic restraint. ³⁷

Summary

• R2D2-ICU randomised 405 mechanically ventilated adults in 10 French ICUs to restrictive low-use versus liberal high-use wrist-strap restraint strategies; 396 contributed to the primary analysis.
• Low-use restraint did not improve the primary endpoint of days alive without delirium or coma at 14 days: adjusted mean difference 0.37 days; 95% CI −0.71 to 1.46; P=.51.
• Delirium incidence, self-extubation, device removal, pressure ulcers, ICU mortality, and 90-day mortality were all similar between groups.
• No compensatory increase in sedative, opioid, antipsychotic, or dexmedetomidine exposure was observed with the low-use strategy.
• The trial undermines routine prophylactic restraint in similar ICUs, but its applicability is limited by substantial nonadherence, deep baseline coma, and exclusion of the most agitation-prone populations.

Overall Takeaway

R2D2-ICU is a landmark negative trial in an area long dominated by dogma and ethics-based intuition rather than randomised evidence. It shows that routine liberal wrist restraint for mechanically ventilated ICU patients does not confer a measurable delirium, coma, or safety advantage over a restrictive strategy, and it pushes modern practice towards restraint minimisation, targeted use, and better non-restraint approaches to delirium prevention and patient safety.

Bibliography

• 1.Zare-Kaseb A, Sarmadi S, Sanaie N, Emami Zeydi A. Prevalence and variability in use of physical restraints in intensive care units: a systematic review and meta-analysis. Aust Crit Care. 2025;38(3):101210.
• 2.Berger S, Grzonka P, Amacher SA, Hunziker S, Frei AI, Sutter R. Adverse events related to physical restraint use in intensive care units: a review of the literature. J Intensive Med. 2023;4(3):318-325.
• 3.Devlin JW, Skrobik Y, Gélinas C, et al. Clinical practice guidelines for the prevention and management of pain, agitation/sedation, delirium, immobility, and sleep disruption in adult patients in the ICU. Crit Care Med. 2018;46(9):e825-e873.
• 4.Sonneville R, Couffignal C, Sigaud F, et al. Restrictive use of restraints and delirium duration in the intensive care unit (R2D2-ICU): protocol for a French multicentre parallel-group open-label randomised controlled trial. BMJ Open. 2024;14(4):e083414.
• 5.Acevedo-Nuevo M, Velasco-Sanz T, Del Olmo-Somolinos B, Vía-Clavero G. Ethical and legal considerations and recommendations for action in the physical restraint use in critically ill patients. Enferm Intensiva (Engl Ed). 2025;36(2):100497.
• 6.Haines KJ, Nørregaard Kjær MB, Deane AM. Physical restraint use in critical care: balancing safety and harm. JAMA. Published online March 17, 2026.
• 7.Kieswick S, Gibbison B. Critical care delirium: prevention, identification and management: a narrative review. Anaesthesia. Published online March 11, 2026.
• 8.Ping Y, Yang J, Zheng Y, Feng W, Huang Z, Sha R, Cui N, Tian Y. Guide-based interventions aimed at reducing physical restraints in intensive care unit: a systematic review and meta-analysis of randomized controlled trials. Front Med (Lausanne). 2025;12:1606359.
• 9.Rosà T, Tempo G, Antonelli M. The restraint-free ICU: not quite there yet. Intensive Crit Care Nurs. 2026;92:104261.
• 10.Kooken R, Tilburgs B, van den Boogaard M. Do we still need physical restraints in ICU? No! Intensive Crit Care Nurs. 2026;92:104262.