Trial Design and Oversight
The RECOVERY trial was designed to evaluate the effects of potential treatments in patients hospitalized with Covid-19 at 176 National Health Service organizations in the United Kingdom and was supported by the National Institute for Health Research Clinical Research Network. (Details regarding this trial are provided in the Supplementary Appendix, available with the full text of this article at NEJM.org.) The trial is being coordinated by the Nuffield Department of Population Health at the University of Oxford, the trial sponsor. Although the randomization of patients to receive dexamethasone, hydroxychloroquine, or lopinavir–ritonavir has now been stopped, the trial continues randomization to groups receiving azithromycin, tocilizumab, or convalescent plasma.
Hospitalized patients were eligible for the trial if they had clinically suspected or laboratory-confirmed SARS-CoV-2 infection and no medical history that might, in the opinion of the attending clinician, put patients at substantial risk if they were to participate in the trial. Initially, recruitment was limited to patients who were at least 18 years of age, but the age limit was removed starting on May 9, 2020. Pregnant or breast-feeding women were eligible.
Written informed consent was obtained from all the patients or from a legal representative if they were unable to provide consent. The trial was conducted in accordance with the principles of the Good Clinical Practice guidelines of the International Conference on Harmonisation and was approved by the U.K. Medicines and Healthcare Products Regulatory Agency and the Cambridge East Research Ethics Committee. The protocol with its statistical analysis plan is available at NEJM.org and on the trial website at www.recoverytrial.net.
The initial version of the manuscript was drafted by the first and last authors, developed by the writing committee, and approved by all members of the trial steering committee. The funders had no role in the analysis of the data, in the preparation or approval of the manuscript, or in the decision to submit the manuscript for publication. The first and last members of the writing committee vouch for the completeness and accuracy of the data and for the fidelity of the trial to the protocol and statistical analysis plan.
We collected baseline data using a Web-based case-report form that included demographic data, the level of respiratory support, major coexisting illnesses, suitability of the trial treatment for a particular patient, and treatment availability at the trial site. Randomization was performed with the use of a Web-based system with concealment of the trial-group assignment. Eligible and consenting patients were assigned in a 2:1 ratio to receive either the usual standard of care alone or the usual standard of care plus oral or intravenous dexamethasone (at a dose of 6 mg once daily) for up to 10 days (or until hospital discharge if sooner) or to receive one of the other suitable and available treatments that were being evaluated in the trial.
For some patients, dexamethasone was unavailable at the hospital at the time of enrollment or was considered by the managing physician to be either definitely indicated or definitely contraindicated. These patients were excluded from entry in the randomized comparison between dexamethasone and usual care and hence were not included in this report. The randomly assigned treatment was prescribed by the treating clinician. Patients and local members of the trial staff were aware of the assigned treatments.
A single online follow-up form was to be completed when the patients were discharged or had died or at 28 days after randomization, whichever occurred first. Information was recorded regarding the patients’ adherence to the assigned treatment, receipt of other trial treatments, duration of admission, receipt of respiratory support (with duration and type), receipt of renal support, and vital status (including the cause of death). In addition, we obtained routine health care and registry data, including information on vital status (with date and cause of death), discharge from the hospital, and respiratory and renal support therapy.
The primary outcome was all-cause mortality within 28 days after randomization; further analyses were specified at 6 months. Secondary outcomes were the time until discharge from the hospital and, among patients not receiving invasive mechanical ventilation at the time of randomization, subsequent receipt of invasive mechanical ventilation (including extracorporeal membrane oxygenation) or death. Other prespecified clinical outcomes included cause-specific mortality, receipt of renal hemodialysis or hemofiltration, major cardiac arrhythmia (recorded in a subgroup), and receipt and duration of ventilation.
As stated in the protocol, appropriate sample sizes could not be estimated when the trial was being planned at the start of the Covid-19 pandemic. As the trial progressed, the trial steering committee, whose members were unaware of the results of the trial comparisons, determined that if 28-day mortality was 20%, then the enrollment of at least 2000 patients in the dexamethasone group and 4000 in the usual care group would provide a power of at least 90% at a two-sided P value of 0.01 to detect a clinically relevant proportional reduction of 20% (an absolute difference of 4 percentage points) between the two groups. Consequently, on June 8, 2020, the steering committee closed recruitment to the dexamethasone group, since enrollment had exceeded 2000 patients.
For the primary outcome of 28-day mortality, the hazard ratio from Cox regression was used to estimate the mortality rate ratio. Among the few patients (0.1%) who had not been followed for 28 days by the time of the data cutoff on July 6, 2020, data were censored either on that date or on day 29 if the patient had already been discharged. That is, in the absence of any information to the contrary, these patients were assumed to have survived for 28 days. Kaplan–Meier survival curves were constructed to show cumulative mortality over the 28-day period. Cox regression was used to analyze the secondary outcome of hospital discharge within 28 days, with censoring of data on day 29 for patients who had died during hospitalization. For the prespecified composite secondary outcome of invasive mechanical ventilation or death within 28 days (among patients who were not receiving invasive mechanical ventilation at randomization), the precise date of invasive mechanical ventilation was not available, so a log-binomial regression model was used to estimate the risk ratio.
Through the play of chance in the unstratified randomization, the mean age was 1.1 years older among patients in the dexamethasone group than among those in the usual care group (Table 1). To account for this imbalance in an important prognostic factor, estimates of rate ratios were adjusted for the baseline age in three categories (<70 years, 70 to 79 years, and ≥80 years). This adjustment was not specified in the first version of the statistical analysis plan but was added once the imbalance in age became apparent. Results without age adjustment (corresponding to the first version of the analysis plan) are provided in the Supplementary Appendix.
Prespecified analyses of the primary outcome were performed in five subgroups, as defined by characteristics at randomization: age, sex, level of respiratory support, days since symptom onset, and predicted 28-day mortality risk. (One further prespecified subgroup analysis regarding race will be conducted once the data collection has been completed.) In prespecified subgroups, we estimated rate ratios (or risk ratios in some analyses) and their confidence intervals using regression models that included an interaction term between the treatment assignment and the subgroup of interest. Chi-square tests for linear trend across the subgroup-specific log estimates were then performed in accordance with the prespecified plan.
All P values are two-sided and are shown without adjustment for multiple testing. All analyses were performed according to the intention-to-treat principle. The full database is held by the trial team, which collected the data from trial sites and performed the analyses at the Nuffield Department of Population Health, University of Oxford.