Interleukin-6 Receptor Antagonists in Critically Ill Patients with Covid-19

Trial Design and Oversight

REMAP-CAP is an international, adaptive platform trial designed to determine effective treatment strategies for patients with severe pneumonia in both pandemic and nonpandemic settings. The design of REMAP-CAP and its first results, regarding glucocorticoids in patients with Covid-19, were published previously.^12,13

Patients eligible for the platform are assessed for eligibility to potentially undergo randomization to multiple interventions across multiple domains. A “domain” covers a common therapeutic area (e.g., antiviral therapy) and contains two or more interventions (including control; e.g., “no antiviral”). Patients are randomly assigned to one intervention in each domain for which they are eligible. REMAP-CAP is defined by a master (“core”) protocol with individual appendixes for each domain, regional governance, and adaptations for a declared pandemic (see the protocol, available with the full text of this article at NEJM.org).

The trial was designed and managed by an international trial steering committee whose members were unaware of the trial group assignments and an independent data and safety monitoring board whose members were aware of the trial group assignments. The trial is approved by relevant regional ethics committees and is conducted in accordance with Good Clinical Practice guidelines and the principles of the Declaration of Helsinki. Written or verbal informed consent, in accordance with regional legislation, is obtained from all the patients or their surrogates.

The trial has multiple international funders. Roche Products and Sanofi supported the trial through provision of tocilizumab and sarilumab in the United Kingdom. The funders as well as Roche and Sanofi had no role in designing the trial, analyzing the data, writing the manuscript, or making the decision to submit the manuscript for publication. All the authors vouch for the completeness and accuracy of the data and for the fidelity of the trial to the protocol and statistical analysis plan.

Patients

Critically ill patients, 18 years of age or older, with either clinically suspected or microbiologically confirmed Covid-19 who were admitted to an intensive care unit (ICU) and receiving respiratory or cardiovascular organ support were classified as having a severe disease state and were eligible for enrollment in the Covid-19 Immune Modulation Therapy domain. Respiratory organ support was defined as invasive or noninvasive mechanical ventilation, including through high-flow nasal cannulae if the flow rate was more than 30 liters per minute and the fraction of inspired oxygen was more than 0.4. Cardiovascular organ support was defined as the intravenous infusion of any vasopressor or inotrope. Patients were excluded if there was a presumption that death was imminent with a lack of commitment to full support or if they had previously participated in REMAP-CAP within 90 days. Patients had to be enrolled within 24 hours after starting organ support in the ICU. Additional exclusion criteria, specific to the Immune Modulation Therapy domain, are listed in the Supplementary Appendix, available at NEJM.org.

Randomization

The Immune Modulation Therapy domain included five interventions — two interleukin-6 receptor antagonists, tocilizumab and sarilumab; an interleukin-1 receptor antagonist, anakinra; interferon beta-1a; and control (no immune modulation). Investigators at each site prespecified at least two interventions, one of which had to be control, to which patients would be randomly assigned. Participants were assigned by means of a centralized computer program to each intervention, starting with balanced assignment for tocilizumab, sarilumab, or control, with actual proportions dependent on the number of interventions available at each site.

Tocilizumab, at a dose of 8 mg per kilogram of actual body weight (up to a maximum of 800 mg), was administered as an intravenous infusion over a period of 1 hour; this dose could be repeated 12 to 24 hours later at the discretion of the treating clinician if clinical improvement was judged insufficient. Sarilumab, at a dose of 400 mg, was administered as an intravenous infusion once only. All investigational drugs were dispensed by local pharmacies and were open label.

Procedures

Other aspects of patient care were provided according to the standard of care at each site. In addition to receiving assignments in this domain, participants could be randomly assigned to other interventions within other domains, depending on domains active at the site, patient eligibility, and consent (see the protocol and www.remapcap.org). Randomization to the Corticosteroid domain for Covid-19 closed on June 17, 2020.¹³ Thereafter, glucocorticoids were allowed according to the recommended standard of care. Although clinical staff were aware of the intervention assignment of individual patients, neither they nor the members of the international trial steering committee were provided any information about aggregate patient outcomes.

Outcome Measures

The primary outcome was the number of respiratory and cardiovascular organ support–free days up to day 21. The definitions of respiratory and cardiovascular organ support were the same as in the inclusion criteria. In this composite ordinal outcome, all deaths within the hospital are assigned the worst outcome (–1). Among survivors, days free of respiratory and cardiovascular organ support are calculated up to day 21, such that a higher number represents faster recovery. In a previous Food and Drug Administration–approved trial, 1.5 days was considered to be a minimal clinically important difference.¹⁴ Secondary outcomes were all prespecified, and details are provided in the Supplementary Appendix.

Statistical Analysis

REMAP-CAP uses a Bayesian design with no maximum sample size. Regular, interim analyses are performed and randomization continues, potentially with response-adaptive randomization with preferential assignment to the interventions that appear most favorable, until predefined statistical criteria are met.

The primary analysis was generated from a Bayesian cumulative logistic model, which calculated posterior probability distributions of organ support–free days to day 21 (primary outcome) on the basis of evidence accumulated in the trial and the prior probability distribution (the assumed previous knowledge). Prior distributions for individual treatment effects were neutral.

The primary model was adjusted for location (site, nested within country), age (categorized into six groups), sex, and time period (2-week calendar epochs) to account for rapid changes in clinical care and outcomes over time during the pandemic. The model contained treatment effects for each intervention within each domain and prespecified treatment-by-treatment interactions across domains. The treatment effects for tocilizumab and sarilumab were “nested” in the model with a hierarchical prior distribution sharing the same neutral prior distribution in an overall “interleukin-6 receptor antagonist effect,” but distinct intervention-specific effects were estimated. When consistent effects are observed for tocilizumab and sarilumab, the posterior distribution for each intervention effect is shrunk toward the overall estimate of the interleukin-6 receptor antagonist effect.¹⁵

The primary analysis was conducted by the statistical analysis committee in all the patients with severe disease randomly assigned to any domain up to November 19, 2020 (and with complete follow-up). The inclusion of additional patients who were enrolled outside the Immune Modulation Therapy domain allows incorporation of all information, which provides robust estimation of the coefficients of all covariates, according to the principle of the REMAP-CAP design.^12,13 Not all patients were eligible for all domains or for all interventions (dependent on active domains and interventions at the site, eligibility criteria, and patient or surrogate consent). Therefore, the model included covariate terms reflecting each patient’s domain eligibility, such that the estimate of the effectiveness of an intervention, relative to any other intervention within that domain, was generated from patients who might have been eligible to be randomly assigned to those interventions within the domain.

The cumulative log odds for the primary outcome were modeled such that a value greater than 0 reflects an increase in the cumulative log odds for the outcome of organ support–free days, implying benefit. Missing outcomes were not imputed. The model was fit with the use of a Markov chain Monte Carlo algorithm that drew iteratively (10,000 draws) from the joint posterior distribution, which allowed calculation of posterior odds ratios with their 95% credible intervals and the probability that each intervention (including control) was optimal (i.e., better than all other treatments tested) in the domain, that an intervention was superior to control (efficacy), that two noncontrol interventions were equivalent, or that an intervention was futile as compared with control. An odds ratio greater than 1 represents improved survival, more organ support–free days, or both. The predefined statistical criteria for determining trial conclusions and for triggering the disclosure of results were as follows: a posterior probability of greater than 99% that an intervention was more effective than all other interventions; an inferiority conclusion if the posterior probability that an intervention was more effective than all other interventions was less than 0.25%; intervention efficacy if the posterior probability that the odds ratio was greater than 1.0 as compared with control was greater than 99%; intervention futility if the posterior probability that the odds ratio was greater than 1.2 as compared with control was less than 5%; and equivalence if the probability that the odds ratio was between 0.83 and 1.2 for two noncontrol interventions was greater than 90%.

Analysis of the primary outcome was then repeated in a second model that used only data from patients enrolled in domains that had stopped and were unblinded at the time of analysis with no adjustment for assignment in other ongoing domains. The secondary outcomes were also analyzed in this second model. One subgroup analysis, based on terciles of serum C-reactive protein levels at enrollment, was prespecified. Further details of all analyses are provided in the Supplementary Appendix. Prespecified analyses are listed in Section 15 of the statistical analysis plan, available with the protocol. Data management and summaries were created with the use of R software, version 3.6.0; the primary analysis was computed with R software, version 4.0.0, with the use of the rstan package, version 2.21.1. Additional data management and analyses were performed with SQL Server 2016; SPSS software, version 26; and Stata software, version 14.2.

Enrollment and Randomization

The first patient with Covid-19 was enrolled in REMAP-CAP on March 9, 2020, and the first patient underwent randomization in the Immune Modulation Therapy domain on April 19 as tocilizumab became available. Sarilumab only became available on June 20. On the basis of an interim analysis as of October 28, the independent data and safety monitoring board reported that tocilizumab had met the statistical criteria for efficacy (posterior probability, 99.75%; odds ratio, 1.87; 95% credible interval, 1.20 to 2.76). According to the protocol, further assignment to control closed on November 19, with randomization continuing between different active immune modulation interventions. At that time, 2046 patients who had severe disease had undergone randomization in at least one REMAP-CAP domain and 895 had undergone randomization in the Immune Modulation Therapy domain (366 were assigned to tocilizumab, 48 to sarilumab, 412 to control, and 69 to other interventions within the domain) at 113 sites across six countries (Figure 1). Thirty patients subsequently withdrew consent, and 11 patients had missing primary outcomes. After a subsequent interim analysis, the data and safety monitoring board reported that sarilumab had also met the statistical criteria for efficacy, so these results are also reported.

Patients

Baseline characteristics were balanced across intervention groups and typical of a critically ill population with Covid-19 (Table 1 and Table S1 in the Supplementary Appendix). All but 3 patients were receiving respiratory support at the time of randomization, including oxygen through high-flow nasal cannulae (29%), noninvasive ventilation (42%), and invasive mechanical ventilation (29%). The majority of patients (707) were enrolled after June 17 and the announcement of the dexamethasone result from the RECOVERY trial^16,17; of these patients, 93% (610 of 654) were treated with glucocorticoids at enrollment or within the following 48 hours (Table S2). Of the 158 patients recruited before June 17, a total of 107 underwent randomization in the Corticosteroid domain within REMAP-CAP, with 41 assigned to a 7-day course of hydrocortisone, 39 to shock-dependent hydrocortisone, and 27 to no hydrocortisone.¹³ Remdesivir use was recorded in 265 of 807 patients (33%).

In the tocilizumab group, 92% of the patients received at least one dose, and 29% received a second dose at the discretion of the treating clinician. In the sarilumab group, 90% of the patients received the assigned drug. In the control group, 2% of the patients were given one of the immune-modulating drugs outside the trial protocol.

Primary Outcome

The median number of organ support–free days was 10 (interquartile range, −1 to 16) in the tocilizumab group, 11 (interquartile range, 0 to 16) in the sarilumab group, and 0 (interquartile range, −1 to 15) in the control group (Table 2 and Figure 2). The median adjusted odds ratios (primary model) were 1.64 (95% credible interval, 1.25 to 2.14) for tocilizumab and 1.76 (95% credible interval, 1.17 to 2.91) for sarilumab as compared with control, yielding posterior probabilities of superiority of more than 99.9% and of 99.5%, respectively. The in-hospital mortality in the pooled interleukin-6 receptor antagonist groups was 27% (108 of 395 patients), as compared with 36% (142 of 397 patients) in the control group. The median adjusted odds ratios for in-hospital survival were 1.64 (95% credible interval, 1.14 to 2.35) for tocilizumab and 2.01 (95% credible interval, 1.18 to 4.71) for sarilumab as compared with control, yielding posterior probabilities of superiority of 99.6% and 99.5%, respectively. The results of the sensitivity analyses were consistent with those of the primary analysis (Table 2, Tables S3 and S4, and the analysis reports by the statistical analysis committee and the international trial steering committee in the Supplementary Appendix). The estimates of the treatment effect in patients treated with either tocilizumab or sarilumab and glucocorticoids in combination were greater than the estimates for any intervention on its own, and the estimated interaction between interleukin-6 receptor antagonists and glucocorticoids was additive and slightly in the direction of synergistic, but with substantial variability in the estimate (Tables S5 and S6).

Secondary Outcomes

The secondary outcomes are listed in Figure 3 and Table S7. Tocilizumab and sarilumab were effective across all secondary outcomes, including 90-day survival, time to ICU and hospital discharge, and improvement in the World Health Organization ordinal scale at day 14.¹⁸ Similar effects were seen across subgroups (Tables S8 and S9).

Nine serious adverse events were reported in the tocilizumab group, including one secondary bacterial infection, five bleeding events, two cardiac events, and one deterioration in vision. Eleven serious adverse events were reported in the control group, including four bleeding events and seven thromboses. No serious adverse events were reported in the sarilumab group.