Effect of Noninvasive Respiratory Strategies on Intubation or Mortality Among Patients With Acute Hypoxemic Respiratory Failure and COVID-19: The RECOVERY-RS Randomized Clinical Trial

Question  What is the effect of initial noninvasive respiratory strategies using continuous positive airway pressure (CPAP) or high-flow nasal oxygen (HFNO), compared with an initial strategy of conventional oxygen therapy, on the risk of tracheal intubation or mortality among hospitalized adults with acute hypoxemic respiratory failure due to COVID-19?

Findings  In this randomized clinical trial of 1273 patients, the composite primary outcome of tracheal intubation or mortality within 30 days occurred in 36% of the patients in the CPAP group compared with 44% in the conventional oxygen therapy group, a difference that was statistically significant, and occurred in 44% in the HFNO group compared with 45% in the conventional oxygen therapy group, a difference that was not significantly different.

Meaning  Among patients with acute hypoxemic respiratory failure and COVID-19, an initial strategy of CPAP significantly reduced the risk of tracheal intubation or mortality compared with conventional oxygen therapy, but there was no significant difference between an initial strategy of HFNO compared with conventional oxygen therapy.

Importance  Continuous positive airway pressure (CPAP) and high-flow nasal oxygen (HFNO) have been recommended for acute hypoxemic respiratory failure in patients with COVID-19. Uncertainty exists regarding the effectiveness and safety of these noninvasive respiratory strategies.

Objective  To determine whether either CPAP or HFNO, compared with conventional oxygen therapy, improves clinical outcomes in hospitalized patients with COVID-19–related acute hypoxemic respiratory failure.

Design, Setting, and Participants  A parallel group, adaptive, randomized clinical trial of 1273 hospitalized adults with COVID-19–related acute hypoxemic respiratory failure. The trial was conducted between April 6, 2020, and May 3, 2021, across 48 acute care hospitals in the UK and Jersey. Final follow-up occurred on June 20, 2021.

Interventions  Adult patients were randomized to receive CPAP (n = 380), HFNO (n = 418), or conventional oxygen therapy (n = 475).

Main Outcomes and Measures  The primary outcome was a composite of tracheal intubation or mortality within 30 days.

Results  The trial was stopped prematurely due to declining COVID-19 case numbers in the UK and the end of the funded recruitment period. Of the 1273 randomized patients (mean age, 57.4 [95% CI, 56.7 to 58.1] years; 66% male; 65% White race), primary outcome data were available for 1260. Crossover between interventions occurred in 17.1% of participants (15.3% in the CPAP group, 11.5% in the HFNO group, and 23.6% in the conventional oxygen therapy group). The requirement for tracheal intubation or mortality within 30 days was significantly lower with CPAP (36.3%; 137 of 377 participants) vs conventional oxygen therapy (44.4%; 158 of 356 participants) (absolute difference, −8% [95% CI, −15% to −1%], P = .03), but was not significantly different with HFNO (44.3%; 184 of 415 participants) vs conventional oxygen therapy (45.1%; 166 of 368 participants) (absolute difference, −1% [95% CI, −8% to 6%], P = .83). Adverse events occurred in 34.2% (130/380) of participants in the CPAP group, 20.6% (86/418) in the HFNO group, and 13.9% (66/475) in the conventional oxygen therapy group.

Conclusions and Relevance  Among patients with acute hypoxemic respiratory failure due to COVID-19, an initial strategy of CPAP significantly reduced the risk of tracheal intubation or mortality compared with conventional oxygen therapy, but there was no significant difference between an initial strategy of HFNO compared with conventional oxygen therapy. The study may have been underpowered for the comparison of HFNO vs conventional oxygen therapy, and early study termination and crossover among the groups should be considered when interpreting the findings.

Trial Registration  isrctn.org Identifier: ISRCTN16912075

Acute hypoxemic respiratory failure is a key clinical characteristic of COVID-19 pneumonitis. In a study of 63 792 patients with COVID-19 hospitalized in the UK between March and August 2020, 76% required supplemental oxygen and 9% required tracheal intubation and invasive mechanical ventilation.¹ Early in the pandemic, international experiences highlighted the potential risk that intensive care units (ICUs) might become overwhelmed, and high mortality was observed in patients requiring invasive mechanical ventilation.^2-4 This drove an urgent public health need to identify strategies to reduce the demand for invasive mechanical ventilation.

In patients with COVID-19 and increasing oxygen requirements, noninvasive respiratory strategies such as continuous positive airway pressure (CPAP) and high-flow nasal oxygen (HFNO) provide potentially attractive strategies for avoiding invasive mechanical ventilation. In other respiratory diseases, particularly community-acquired pneumonia, both CPAP and HFNO may improve clinical outcomes; however, patients treated with CPAP experience more adverse events.^5,6 In the context of COVID-19, however, there was concern that these strategies might serve only to delay tracheal intubation due to high failure rates, while exacerbating lung injury through generation of large tidal volumes.^7-10

The absence of evidence to support use of CPAP and HFNO in patients with COVID-19 led to significant variability both in international guidelines and clinical practice.^9,11 On this basis, there was a need for a randomized clinical trial to determine whether either CPAP or HFNO, compared with conventional oxygen therapy, reduces the need for tracheal intubation or mortality within 30 days in hospitalized patients with acute hypoxemic respiratory failure due to COVID-19.

The Randomized Evaluation of COVID-19 Therapy–Respiratory Support (RECOVERY-RS) clinical trial was conducted across 48 acute care hospitals in the UK and Jersey. The trial protocol was approved by the London-Brighton and Sussex research ethics committee and the Health Research Authority, sponsored by the University of Warwick, coordinated by the Warwick Clinical Trials Unit, and funded by the National Institute for Health Research. An independent trial steering committee and data and safety monitoring committee provided trial oversight. The RECOVERY-RS trial was conducted in accordance with Good Clinical Practice guidelines, local regulations, and the ethical principles described in the Declaration of Helsinki.¹² In keeping with regional regulations, consent from patients or agreement from their family or another surrogate was obtained orally, with a written record maintained by the researcher. The trial protocol has been published.¹³ The trial protocol and statistical analysis plan appear in Supplement 1.

The trial was a parallel group, open-label, adaptive, 3-group, randomized clinical trial designed to evaluate the clinical effectiveness of CPAP or HFNO, compared with conventional oxygen therapy, in hospitalized patients with acute hypoxemic respiratory failure due to COVID-19 (Figure 1). The multigroup design was essentially conducted as 2 separate trials comparing CPAP and HFNO with a common shared control group of conventional oxygen therapy. A group sequential design allowed early study termination of 1 or both interventions if they were found to be more effective than conventional oxygen therapy, with the final analysis for each comparison adjusted to control the pairwise α value of .05.

Adult hospitalized patients (aged ≥18 years) with known or suspected COVID-19 were eligible if they had acute hypoxemic respiratory failure, defined as an oxygen saturation as measured by pulse oximetry (Spo₂) of 94% or less despite receiving a fraction of inspired oxygen (Fio₂) of at least 0.40, and were deemed suitable for tracheal intubation if treatment escalation was required. Patients with an immediate (<1 hour) need for invasive mechanical ventilation, known pregnancy, or planned withdrawal of treatment were excluded. Based on the judgment of the treating clinician, a contraindication to the intervention precluded randomization to that specific trial group.

Eligible patients were randomized using an internet-based system with allocation concealment. We anticipated that either CPAP or HFNO might be unavailable at the hospital sites on a temporary or permanent basis. As such, the randomization system allowed the hospital site to randomize patients to (1) CPAP, HFNO, or conventional oxygen therapy (on a 1:1:1 basis) or (2) a single intervention (CPAP or HFNO) or conventional oxygen therapy (on a 1:1 basis). These 2 systems were integrated and constantly updated to ensure that the allocation ratio was maintained within the permitted thresholds.

The planned sample size was inflated to account for minor imbalances in the allocation ratio and, if it had been needed, the system allowed randomization weightings to be adjusted. Hospital sites could not randomize only between CPAP and HFNO. Randomization was stratified by hospital site, sex, and age and the allocation was generated by a minimization algorithm, which did not include any random component. Due to the nature of the trial interventions and context, it was not possible to blind patients, treating clinicians, or outcome assessors.

Patients randomized to CPAP or HFNO started treatment as soon as possible. Breaks from treatment were permitted for comfort. Patients randomized to conventional oxygen therapy received oxygen via a standard face mask or low-flow nasal cannula. The patients in the HFNO group received heated humidified HFNO. The patients in the CPAP group received CPAP that did not permit the incorporation of any inspiratory positive airway pressure. Across all groups, local policies and clinical discretion informed decisions regarding device choice and setup, titration (eg, Fio₂, flow, positive end-expiratory pressure), treatment targets (eg, Spo₂), and treatment discontinuation. Tracheal intubation was performed when clinically indicated based on the judgment of the treating clinician. Treatment crossover was defined as a patient who received a nonallocated intervention (CPAP or HFNO) for a period of more than 6 hours unless the intervention was used as a bridge to tracheal intubation or for palliative care.

At enrollment, we collected information on demographics (including investigator-classified sex and race and ethnicity), comorbid state, and physiological levels (including blood pressure, respiratory rate, peripheral oxygen saturation, and blood gas measurements). Collection and reporting of race and ethnicity was based on fixed categories and mandated by the funder due to the disproportionate effect of COVID-19 infection on non-White populations.¹⁴ Participants were followed up throughout their hospital stay to record intervention use, crossover, adverse events, and outcomes. We undertook data linkage with national data sets to support collection of demographic information and outcomes, including death after hospital discharge.

The primary outcome was a composite of tracheal intubation or mortality within 30 days of randomization. Tracheal intubation, as an outcome, reflects the need for invasive mechanical ventilation, which is typically delivered in high-resource ICUs. Secondary outcomes included the individual incidence of tracheal intubation or mortality within 30 days, time to tracheal intubation, duration of invasive mechanical ventilation, time to death, mortality (during ICU or hospital stay), admission to the ICU, length of stay in the ICU, and length of stay in the hospital (included time from emergency department arrival to hospital discharge).

Early data on COVID-19 informed the anticipated event rate in the conventional oxygen therapy group.¹⁵ Assuming a conservative incidence of 15% for the composite outcome of tracheal intubation or mortality (with a 2-sided significance level of .05 and 90% power), a total of 3000 participants (1000 per group across 3 groups) were required. This equated to detecting a reduction of 5% for tracheal intubation or mortality or an odds ratio of 0.625. This minimally important clinical difference for the primary outcome aligns with what was used in the RECOVERY trial.^16,17 The sample size was inflated to 4002 because of the uncertainties in relation to COVID-19 and event rates.

Efficacy monitoring of each pairwise comparison with conventional oxygen therapy was based on an α spending function approach with a 1-sided pairwise type I error rate of 0.025 and a type I error spent at the interim analyses that was proportional to the observed Fisher information. This allowed the trial to be stopped early if 1 or both interventions were more effective than conventional oxygen therapy. Any decision to stop the trial or drop a group due to futility or safety was determined by the data and safety monitoring committee. The sample size calculation assumed the conduct of 11 interim analyses and 1 final analysis.

The primary and secondary analyses were performed for all participants based on their randomized intervention. Outcome data were compared between each intervention group and the conventional oxygen therapy group. Participants in the conventional oxygen therapy group were only included in a comparison with HFNO or CPAP if they had the opportunity to be randomized to that intervention. For the primary outcome, we undertook a post hoc analysis that compared the CPAP and HFNO groups.

Continuous data were summarized using the number of participants (percentage), mean (SD or 95% CI), and median (IQR). Categorical data were summarized with frequency count, percentage, and missing. Odds ratios (95% CIs) were reported for categorical outcomes using logistic regression models. Mean or median differences (95% CIs) were reported for continuous outcomes. Proportional (absolute) differences (95% CIs) were reported for categorical outcomes. For the time to event analysis, hazard ratios (95% CIs) were reported and the proportional hazards assumption was assessed using the score test. In accordance with the statistical analysis plan (Supplement 1), we planned multiple imputation only if there was substantial missingness (≥20%) in relation to the primary outcome.

The primary analysis was unadjusted. For the adjusted secondary analyses, the covariates of age, sex, morbid obesity (defined as a body mass index >35), race and ethnicity, Fio₂, respiratory rate, and treatment phases were used with hospital site included as a random effect.^18,19 Treatment phases were defined as before July 2020, July 2020 to January 2021, and after January 2021 based on the introduction of dexamethasone as standard care in June 2020 and tocilizumab in January 2021.^17,20,21 Due to the unavailability of data from NHS Digital, we could not include social deprivation in the adjusted analyses.

We used inverse probability weighting as a secondary exploratory analysis. This method corrects for bias that may be introduced into the treatment effect as a result of the crossover. Weights were estimated using propensity scores with the response variable as those participants who did crossover or who did not crossover. These weights were then introduced into the main logistic regression models to diminish the bias introduced by treatment change. Subgroup analyses were performed using logistic regression models, with the primary outcome as the response variable and the subgroup × treatment interaction term included in the model. No adjustment was made for the multiple comparisons. Thus, the type I error control was the same as if CPAP and HFNO had each been compared with conventional oxygen therapy in separate trials.

Cutoff values for the final P value for the primary analysis were calculated to correct for the type I error spent at the interim analyses.²² The final cutoff values depended on the information available at the interim analyses and are reported in the Results section below. No correction for the interim analyses was made to the cutoff values for the secondary outcomes or analyses, with a significance threshold of .05 used. All significance testing was 2-sided. Because of the potential for type I error due to multiple comparisons, the findings for the analyses of the secondary outcomes should be interpreted as exploratory. Analyses were conducted using SAS version 9.4 (SAS Institute Inc) and R version 4.0.3 (R Foundation for Statistical Computing).

Trial recruitment was stopped early. Toward the end of the funded 12-month recruitment period, there was a rapid decline in hospitalized patients with COVID-19 in the UK. During the trial period, recruitment was closely tracked with the overall cases of hospitalized patients with COVID-19 in the UK (eFigure 1 in Supplement 2). The trial management group decided not to seek additional funding and to prioritize the sharing of accumulated data to inform international clinical care.

Prior to stopping trial recruitment, 3 formal interim analyses had been conducted of 36, 160, and 387 participants and the trial continued after each analysis. The results of the interim analyses (other than the decision to continue the trial) were not known to the trial management group, trial steering committee, trial sponsor, or trial funder. The trial management group’s recommendation to stop trial recruitment was agreed on by the trial’s steering committee and funder. The trial sponsor made the decision to stop recruitment and the trial closed recruitment on May 3, 2021.

Between April 6, 2020, and May 3, 2021, there were 1278 patients who were randomized across 48 acute care hospitals in the UK and Jersey. Five patients mistakenly underwent double randomization, leaving 1273 patients included in the trial (380 in the CPAP group, 418 in the HFNO group, and 475 in the conventional oxygen therapy group) (Figure 1). Final follow-up occurred on June 20, 2021. Eight patients withdrew and 5 were lost to follow-up. Primary outcome data were available for 99.0% (1260/1273) of participants.

A total of 733 participants (377 in the CPAP group and 356 in the conventional oxygen therapy group) were included in the comparison of CPAP with conventional oxygen therapy. A total of 783 participants (415 in the HFNO group and 368 in the conventional oxygen therapy group) were included in the comparison of HFNO with conventional oxygen therapy (Figure 1 and eTable 1 in Supplement 2).

Participant characteristics were similar at baseline (Table 1 and eTable 2 in Supplement 2). The mean age was 57.4 years (95% CI, 56.7-58.1 years), 66.3% were male, and 65.3% of White race. The median time from first COVID-19 symptoms to randomization was 9 days (IQR, 7.0-12.0 days). The baseline median Spo₂ was 93% (IQR, 91%-95%) and Fio₂ was 0.60 (IQR, 0.40-0.80).

The allocated intervention was received by 91.6% (348/380) of participants in the CPAP group, 91.9% (384/418) in the HFNO group, and 98.3% (467/475) in the conventional oxygen therapy group (Figure 1). In the CPAP group, initial positive end-expiratory pressure was set at a mean of 8.3 cm H₂O (95% CI, 8.1-8.5 cm H₂O; Table 2). In the HFNO group, initial flow was set at a mean of 52.4 L/min (95% CI, 51.4-53.5 L/min). Of those who required tracheal intubation, their preprocedure clinical conditions appear in Table 2.

Treatment crossover occurred in 17.1% of participants (15.3% [58/380] in the CPAP group, 11.5% [48/418] in the HFNO group, and 23.6% [112/475] in the conventional oxygen therapy group; Figure 1 and eTable 3 in Supplement 2).

For the comparison of CPAP vs conventional oxygen therapy, the primary composite outcome of tracheal intubation or mortality within 30 days occurred in 36.3% (137/377) of the participants in the CPAP group vs 44.4% (158/356) of the participants in the conventional oxygen therapy group (absolute difference, −8% [95% CI, −15% to −1%], P = .03; Table 3).

For the comparison of HFNO vs conventional oxygen therapy, the primary composite outcome occurred in 44.3% (184/415) of the participants in the HFNO group vs 45.1% (166/368) of the participants in the conventional oxygen therapy group (absolute difference, −1% [95% CI, −8% to 6%], P = .83; Table 4).

The cutoff values for the P values to define statistical significance for the primary comparisons of CPAP and HFNO vs conventional oxygen therapy (corrected for the interim analyses) were equivalent to .044 for 2-sided P values.

The secondary outcomes appear in Tables 3-4 and in eFigures 2-4 in Supplement 2. There was a significant difference in the individual incidence of tracheal intubation within 30 days in the CPAP group (33.4% vs 41.3% in the conventional therapy group; absolute difference, −8% [95% CI, −15% to −1%]). However, a significant difference was not observed for the individual incidence of mortality within 30 days in the CPAP group (16.7% vs 19.2% in the conventional therapy group; absolute difference, −3% [95% CI, −8% to 3%]). Compared with conventional oxygen therapy, neither CPAP nor HFNO significantly reduced mortality in the ICU or in the hospital.

Significantly fewer participants in the CPAP group required admission to the ICU compared with participants in the conventional oxygen therapy group (55.4% vs 62.9%, respectively; absolute difference, −7% [95% CI, −15% to −3%]). Among the participants who required tracheal intubation, there was a statistically significant increase in the median time to tracheal intubation in the CPAP group (2.0 days [IQR, 1.0 to 4.0 days]) compared with the conventional oxygen therapy group (1.0 day [IQR, 0 to 4.0 days]) (median difference, 1.0 day [95% CI, 0.2 to 1.8 days]; Table 3). For all other outcomes and comparisons, there was no statistically significant difference between study groups.

The findings from both the adjusted analyses and the inverse probability weighting analysis were consistent with the primary analysis (eTable 4 in Supplement 2). The majority of tests for interaction in the subgroup analyses were not statistically significant (Figure 2 and Figure 3). However, the comparison of the HFNO group vs the conventional oxygen therapy group for Fio₂ was significant (P = .02; Figure 3). The findings were broadly consistent between the unadjusted and adjusted subgroup analyses (Figures 2-3 and eFigure 5 in Supplement 2).

A post hoc analysis, which compared CPAP and HFNO, included 570 participants who were randomized across all 3 groups. The primary outcome occurred in 34.6% (91/263) of participants in the CPAP group and in 44.3% (136/307) of participants in the HFNO group (absolute difference, −10% [95% CI, −18% to −2%], P = .02; eTable 5 in Supplement 2).

Adverse events occurred most frequently in the CPAP group; 34.2% (130/380) of participants in the CPAP group experienced adverse events, followed by 20.6% (86/418) of participants in the HFNO group, and 13.9% (66/475) of participants in the conventional oxygen therapy group (eTable 6 in Supplement 2). Eight participants experienced serious adverse events (7 in the CPAP group and 1 in the conventional oxygen therapy group). Four of the serious adverse events were classified as probably or possibly linked to the trial intervention and all of these events occurred in the CPAP group (1 patient with surgical emphysema and pneumomediastinum, 2 patients with pneumothorax and pneumomediastinum, and 1 patient with vomiting requiring emergency tracheal intubation).

In this randomized clinical trial of patients with acute hypoxemic respiratory failure due to COVID-19, an initial strategy of CPAP, compared with conventional oxygen therapy, significantly reduced the composite outcome of tracheal intubation or mortality within 30 days. In contrast, there was no significant difference between an initial strategy of HFNO and conventional oxygen therapy, although given the width of the 95% CI, the trial may have been underpowered to detect small but clinically important treatment effects.

This decrease in the incidence of the primary outcome with CPAP was attributable to a significant decrease in the need for tracheal intubation. Neither HFNO nor CPAP reduced mortality compared with conventional oxygen therapy. More adverse events were reported in the CPAP group.

This pragmatic trial was designed to be deliverable in the context of a pandemic and tested interventions that precluded blinding of either the participant or treating clinician. The decision to perform tracheal intubation, and thereby commence invasive mechanical ventilation, was not standardized.¹¹ It is possible that the lower tracheal intubation rate in the CPAP group may have been driven by a greater willingness among clinicians and patients to delay tracheal intubation, and this may be supported by the finding that the time to tracheal intubation was longer in the CPAP group. However, physiological levels at the time of tracheal intubation were similar across groups, suggesting that, irrespective of treatment strategy, clinicians used a similar threshold to determine the need for tracheal intubation. Furthermore, this effect was not observed with HFNO, which should have been susceptible to the same risk of performance bias.

The decision to not standardize escalation to tracheal intubation was driven by clinical uncertainty regarding the optimal timing and threshold of tracheal intubation in patients with COVID-19.^11,23 Although rapidly building clinical consensus may be achievable in trials recruiting patients from a small number of hospitals, such as in the Helmet Noninvasive Ventilation Versus High-Flow Oxygen Therapy in Acute Hypoxemic Respiratory Failure (HENIVOT) trial,²⁴ the RECOVERY-RS trial management group determined that any attempt to stipulate specific criteria might influence clinical equipoise and patient acceptability, affect trial recruitment, and, more importantly, reduce trial generalizability. Previous large trials of noninvasive respiratory strategies have differed in their approach to protocolization of tracheal intubation, which likely reflects these specific challenges, even in respiratory conditions for which the pathophysiology has been well described.^25-27

A recent systematic review and meta-analysis⁵ of 25 randomized clinical trials (3804 patients) summarized evidence on the clinical effectiveness of noninvasive ventilation (with or without pressure support) and HFNO, compared with conventional oxygen therapy, in patients with acute respiratory failure. Across 14 trials (1275 patients), noninvasive ventilation via a face mask was significantly associated with a lower risk of both mortality and tracheal intubation. In contrast, HFNO was significantly associated with a lower risk of tracheal intubation (5 trials; 1479 patients), but not mortality (3 trials; 1279 patients). The RECOVERY-RS trial found that CPAP significantly reduced tracheal intubation, but not mortality, although the wide 95% CI precludes the drawing of a specific conclusion about the effect on mortality. The current trial further found that HFNO did not significantly reduce the need for tracheal intubation. One explanation for these discordant findings is differences in pathophysiology between COVID pneumonitis and other causes of acute respiratory failure.^5,28 Furthermore, in the RECOVERY-RS trial, some hospitals modified care pathways to deliver CPAP and HFNO outside an ICU, which may have influenced the clinical effectiveness of the interventions.

The current trial builds on the findings of 2 other recently published randomized clinical trials that examined the use of noninvasive respiratory strategies in patients with COVID-19.^24,29 The High-Flow Nasal Cannula in Severe COVID-19 With Acute Hypoxemic Respiratory Failure (HiFLo-Covid) trial²⁹ compared HFNO with conventional oxygen therapy in 220 adults with severe COVID-19 across 3 Colombian hospitals. The trial reported that HFNO both reduced the need for tracheal intubation (hazard ratio, 0.62; 95% CI, 0.39-0.96) and time to clinical recovery. In contrast to the current trial (RECOVERY-RS) and the HiFLo-Covid trial,²⁹ the HENIVOT trial²⁴ directly compared 2 noninvasive respiratory strategies, namely noninvasive ventilation via a helmet (with pressure support) vs HFNO. In 110 patients with COVID-19 recruited across 4 ICUs, there was no significant difference for the primary outcome of days free of respiratory support, although significantly fewer patients in the helmet noninvasive ventilation group required tracheal intubation (odds ratio, 0.41; 95% CI, 0.18-0.89). The protocolized approach to the setup and weaning of trial interventions and the decision to perform tracheal intubation in both the HENIVOT and HiFLo-Covid trials potentially limits their generalizability.

This trial has several limitations. First, the trial did not achieve its planned sample size with the decision to stop recruitment driven by the end of the funded recruitment period, together with declining numbers of patients with COVID-19 in the UK, and an ethical obligation to share accumulated data with the international clinical community. The decision to stop trial recruitment early did not involve the members of the data and safety monitoring committee, which was the only group that had seen the interim analyses, such that the risk of bias arising from stopping the trial early is likely to be minimal. However, the trial may have been underpowered to detect small but clinically important treatment effects for the comparison of HFNO vs conventional oxygen therapy.

Second, there was crossover between the allocated treatment groups, principally from the conventional oxygen therapy group to 1 or both interventions. This is a common challenge in trials of noninvasive respiratory strategies^26,27 and reduces the observed effect size of a clinically effective treatment. Nevertheless, the findings from the inverse probability weighting analysis were consistent with the primary analysis. Third, it was determined that it would be impractical to collect screening data, therefore, it was not possible to describe the reasons and number of patients who were not randomized.

Fourth, the trial’s definition of acute hypoxemic respiratory failure was based on objective criteria of oxygenation and oxygen use. In clinical practice, the decision to commence noninvasive respiratory strategies may be based both on objective criteria, such as these, and subjective criteria, such as respiratory distress.

Fifth, the study population, particularly in terms of racial and ethnic groups, may not be generalizable across all populations. Sixth, there were some minor differences across groups at baseline in relation to comorbid state.

Seventh, the trial was rapidly setup early in the pandemic, prior to the development of a core outcome set for COVID-19 trials.³⁰ Even though the outcome list aligns closely to most of the core outcomes subsequently identified, the trial did not capture information on patient recovery after hospital discharge.

Among patients with acute hypoxemic respiratory failure due to COVID-19, an initial strategy of CPAP significantly reduced the risk of tracheal intubation or mortality compared with conventional oxygen therapy, but there was no significant difference between an initial strategy of HFNO compared with conventional oxygen therapy. The study may have been underpowered for the comparison of HFNO vs conventional oxygen therapy, and early study termination and crossover among the groups should be considered when interpreting the findings.

Corresponding Author: Daniel F. McAuley, MD, Wellcome-Wolfson Institute for Experimental Medicine, Queen’s University Belfast, 97 Lisburn Rd, Belfast BT9 7BL, Northern Ireland (d.f.mcauley@qub.ac.uk).

Accepted for Publication: January 4, 2022.

Published Online: January 24, 2022. doi:10.1001/jama.2022.0028

Author Contributions: Drs Ji and Lall had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Perkins, Connolly, Couper, Baillie, Bradley, Dark, De Soyza, Gorman, Hart, Regan, Simonds, Stallard, Yeung, McAuley.

Acquisition, analysis, or interpretation of data: Perkins, Ji, Connolly, Couper, Lall, Bradley, Dark, Dave, De Soyza, Dennis, Devrell, Fairbairn, Ghani, Green, Hart, Hee, Kimbley, Madathil, McGowan, Messer, Naisbitt, Norman, Parekh, Parkin, Patel, Regan, Ross, Rostron, Saim, Skilton, Stallard, Steiner, Vancheeswaran, Yeung, McAuley.

Drafting of the manuscript: Perkins, Connolly, Couper, Lall, De Soyza, Devrell, Hart, Naisbitt, Parekh, Parkin, Patel, Regan, Stallard, Steiner, McAuley.

Critical revision of the manuscript for important intellectual content: Perkins, Ji, Connolly, Couper, Baillie, Bradley, Dark, Dave, De Soyza, Dennis, Fairbairn, Ghani, Gorman, Green, Hart, Hee, Kimbley, Madathil, McGowan, Messer, Norman, Parekh, Patel, Ross, Rostron, Saim, Simonds, Skilton, Stallard, Steiner, Vancheeswaran, Yeung, McAuley.

Statistical analysis: Ji, Lall, Hee, Stallard.

Obtained funding: Perkins, Dark, De Soyza, Stallard, McAuley.

Administrative, technical, or material support: Perkins, Connolly, Couper, Bradley, Dark, Dave, De Soyza, Devrell, Ghani, Gorman, Green, Hart, McGowan, Norman, Regan, Ross, Rostron, Simonds, Skilton, Steiner, McAuley.

Supervision: Perkins, Connolly, Baillie, Dark, Dave, De Soyza, Ghani, Green, Regan, Rostron, Saim, Steiner, Vancheeswaran, McAuley.

Conflict of Interest Disclosures: Dr Perkins reported being supported by the National Institute for Health Research (NIHR) West Midlands Applied Research Collaboration and serving as co-director of research for the Intensive Care Society until recently (term ended in June 2021). Dr Connolly reported receiving grants from the NIHR; receiving personal fees from Fisher & Paykel Healthcare; and serving as the director of research for the Intensive Care Society. Dr Baillie reported receiving grants from the Wellcome Trust, the Biotechnology and Biological Sciences Research Council, and the Medical Research Council. Dr Dark reported receiving grants from the Manchester NIHR Biomedical Research Centre and being a national specialty cluster lead for the NIHR. Dr Dave reported receiving personal fees from Chiesi. Dr De Soyza reported being a national specialty cluster lead for the NIHR and receiving personal fees from AstraZeneca, Bayer, Chiesi, Gilead, GlaxoSmithKline, Forest Labs, Novartis, Insmed, Teva, Zambon, and Pfizer. Mrs Devrell reported receiving personal fees from the NIHR. Dr Gorman reported receiving grants from the NIHR and Wellcome Trust. Dr Hart reported receiving a UK Research and Innovation grant from the Medical Research Council; receiving unrestricted grants and equipment from Philips-Respironics, Fisher & Paykel Healthcare, and ResMed; receiving institutional funding for his role on the Philips Global medical advisory board; receiving personal fees from Philips-Respironics, Philips, ResMed, and Fisher & Paykel Healthcare; and receiving financial support from Philips for the development of the Myotrace technology that has a patent approved in Europe and in the US. Dr Hee reported receiving grants from the British Heart Foundation and the NIHR West Midlands Research Design Service. Mr Messer reported receiving personal fees from Fisher & Paykel Healthcare. Dr Parekh reported receiving a UK Research and Innovation grant from the Medical Research Council and receiving grants from the NIHR. Dr Steiner reported receiving personal fees from GlaxoSmithKline. Dr McAuley reported receiving personal fees from GlaxoSmithKline, Boehringer Ingelheim, Bayer, Novartis, Sobi, Eli Lilly, Vir Biotechnology, and Faron Pharmaceuticals; receiving grants from the NIHR, Randox, Wellcome Trust, Innovate UK, the Medical Research Council, and the Northern Ireland Health and Social Research and Development Division; holding a patent for an anti-inflammatory treatment issued to Queen’s University Belfast; and serving as co-director of research for the Intensive Care Society until recently (term ended in June 2021) and as program director for the NIHR Efficacy and Mechanism Evaluation program. No other disclosures were reported.

Funding/Support: This study was funded by grant COVID-19-RSC from the National Institute for Health Research.

Role of the Funder/Sponsor: The National Institute for Health Research approved the design of the study and monitored the conduct of the study. They played no role in the collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Group Information: The RECOVERY-RS Collaborators appear in Supplement 3.

Disclaimer: The views expressed are those of the authors and do not necessarily represent those of the National Institute for Health Research or the Department of Health and Social Care.

Data Sharing Statement: See Supplement 4.

Additional Contributions: We are grateful to all the patients and families who supported the trial, together with the physicians, nurses, and allied health professionals across all participating hospitals who supported both trial recruitment and delivery of trial interventions in extremely challenging conditions. We thank the National Institute for Health Research Clinical Research Network and the Northern Ireland Clinical Research Network. We also thank Health Data Research UK, the Office for National Statistics, and the Intensive Care National Audit and Research Centre for support with the data linkage. In addition, we are indebted to the following members of both the trial steering and data and safety monitoring committees: Kathy Rowan, PhD (Intensive Care National Audit and Research Centre, London, England), Duncan Young, DM (Oxford University, Oxford, England), Marion Campbell, PhD (University of Aberdeen, Aberdeen, Scotland), Susie Hennings, MSc (Keele University, Keele, England), John Laffey, MD (National University of Ireland, Galway), Martin Landray, PhD (Oxford University, Oxford, England), Gillian McCarmack (patient and public representative), Gary Overton (patient and public representative), Marion Thompson, PhD (patient and public representative), B. Taylor Thompson, MD (Massachusetts General Hospital and Harvard Medical School, Boston), and Timothy Walsh, MD (University of Edinburgh, Edinburgh, Scotland). Mr Overton and Dr Marion Thompson received personal fees for their contribution to the study as patient and public representatives. All others listed did not receive compensation.