key: cord-0824521-mf16jhyh
authors: Lee, Todd C.; McDonald, Emily G.; Butler-Laporte, Guillaume; Harrison, Luke B.; Cheng, Matthew P.; Brophy, James M.
title: Remdesivir and Systemic Corticosteroids for the Treatment of COVID-19: A Bayesian Reanalysis
date: 2021-02-01
journal: Int J Infect Dis
DOI: 10.1016/j.ijid.2021.01.065
sha: 375051fbd5f31881993ae8d5636b76ec620b7beb
doc_id: 824521
cord_uid: mf16jhyh

Importance The worldwide death toll from COVID-19 has surpassed 2 million and treatments that decrease mortality are urgently needed. Objective To examine the probability of mortality benefit for remdesivir and contrast with systemic corticosteroids. Design, setting, and participants A probabilistic reanalysis of available clinical trial data for corticosteroids or remdesivir in the treatment of hospitalized patients with COVID-19 using a Bayesian random effects meta-analytic approach. Studies were identified from existing meta-analyses performed by the World Health Organization. Main outcomes and measures For both drugs, we calculated the posterior probabilities of an absolute decrease in mortality versus control patients by subgroups based on oxygen requirements. To contextualize, we quantified the probability of at least a 1, 2, or 5% absolute mortality decrease. Results Among patients requiring ventilation, remdesivir had only a 4% probability of 1% or greater mortality benefit compared to 93% for corticosteroids. For patients requiring supplemental oxygen, the probability of 1% or greater survival benefit was 81% for remdesivir compared to 93% for dexamethasone. Finally, for patients without oxygen requirements, the probabilities of 1% or greater mortality benefit were 29% for remdesivir and 4% for dexamethasone. Conclusions and relevance Using a Bayesian analytic approach, remdesivir had a low probability of achieving a clinically meaningful reduction in mortality, except for patients requiring supplemental oxygen. Corticosteroids were more promising for populations requiring oxygen support, especially ventilation. While awaiting more definitive studies, this probabilistic interpretation of the evidence will help guide treatment decisions for clinicians as well as guideline and policy makers.

this probabilistic interpretation of the evidence will help guide treatment decisions for clinicians as well as guideline and policy makers.

The public health crisis caused by COVID-19 has led to unparalleled international scientific collaborations to find a safe and effective treatment, particularly for hospitalized patients. With close to 2 million fatalities, treatments that can reduce mortality are urgently needed. Large multicenter clinical trials are underway, led by groups such as the U.S. National Institutes of Allergy and Infectious Diseases (NIAID) (the Adaptive COVID-19 Treatment Trials or ACTT), the University of Oxford's Nuffield Department of Population Health (the RECOVERY trials), and the World Health Organization (WHO) and participating countries (the SOLIDARITY trials). While the pace of discovery may feel slow under the stress of the pandemic, the speed of accomplishment of groups such as these has been remarkable.

Two of the most promising treatments to date are systemic corticosteroids and remdesivir. Dexamethasone has been established as lifesaving by reducing mortality in the subgroups of patients requiring oxygen (rate ratio, 0.82; 95% CI 0.72 to 0.94) and mechanical ventilation (rate ratio 0.64; 95% CI 0.51 to 0.81) (The RECOVERY Collaborative Group, 2020). However, an effect was not demonstrated among those who were not receiving oxygen (rate ratio 1.19; 95% CI 0.91 to 1.55 

In contrast, a mortality benefit has been harder to demonstrate with remdesivir. The first clinical trial to be published did not show a mortality benefit (Wang et al., 2020) .

Subsequently, the ATCC-1 trial (Beigel et al., 2020) did not conclusively demonstrate a benefit (hazard ratio, 0.73; 95% CI, 0.52 to 1.03) . A third open label trial (Spinner et al., 2020) involving moderate risk patients had low mortality overall (<2%) and did not provide further insight. Finally, data from WHO SOLIDARITY, the largest remdesivir trial to date with 5,451 patients (WHO Solidarity Trial Consortium, 2020). did not show a statistically significant mortality benefit for remdesivir alone (rate ratio 0.95; 95%CI 0.81-1.11) or in their embedded meta-analysis of all available trials (rate ratio 0.91; 95%CI 0.79-1.05).

While failing to reach statistical significance, the point estimate and 95% confidence interval of the pooled remdesivir results includes the potential for an important mortality benefit. Therefore, it could be premature to abandon remdesivir based on statistical significance alone. We sought to reanalyze the remdesivir results using Bayesian methods (Spiegelhalter et al., 1999) to estimate the posterior probability that remdesivir could lead not only to any mortality reduction, but also to a clinically meaningful reduction in mortality compared to usual care. We then contextualized those probabilities against the same analysis performed for systemic corticosteroids including dexamethasone. The purpose of doing so was to help clinicians contextualize the highquality evidence and better practice sensible medicine through Bayesian thinking (Seymour et al., 2020) .

We used Bayesian methods to estimate the absolute mortality reduction of remdesivir and systemic corticosteroids based on the data available from the systematic reviews and meta-analyses performed by the World Health Organization in September (The WHO Rapid Evidence Appraisal for COVID-19 Therapies (REACT) Working Group, 2020) and

December (WHO Solidarity Trial Consortium, 2020) . A PubMed search of the MEDLINE database was conducted on January 10, 2021 and confirmed there were no additional randomized controlled trials available; however, there were 2 corticosteroid trials with additional patient data available after the WHO analysis and we used that data instead (Jeronimo et al., 2020; Tomazini et al., 2020) . The primary outcome was overall reduction in mortality as compared with control patients, and we pre-specified 3 non-overlapping subgroups which matched those pre-specified for the largest trials RECOVERY and SOLIDARITY: patients who needed mechanical ventilation; patients who needed oxygen supplementation without mechanical ventilation; and patients who did not need oxygen supplementation.

Bayesian meta-analysis provides several advantages over frequentist approaches including a more rigorous assessment of overall uncertainty, especially the between study heterogeneity, more reliable analyses of smaller sample sizes, and the ability to provide direct probability statements conditional on the current and prior data.

Two authors (TCL and JMB) extracted the trial results available from each of the four controlled trials for remdesivir (Table 1 ). However, we made two noteworthy decisions as J o u r n a l P r e -p r o o f some of the outcomes were not reported with enough granularity. For the Wang et. al trial (Wang et al., 2020) , the inclusion criteria required the use of oxygen; however, 3 patients in the placebo group were not receiving oxygen at the time of the first dose. Further, there was one mechanically ventilated patient in the placebo group. We included this study in the oxygen without ventilation group as this represented most patients. For the Spinner et. al. trial (Spinner et al., 2020) , oxygen requirement was an exclusion criterion; however, 14% and 19% of remdesivir and control patients respectively had developed an oxygen requirement between screening and the first dose, but the results did not separate mortality by day 1 oxygen need. Because most patients did not receive oxygen and because of the overall low mortality in both arms, we included this study with the "no oxygen" studies.

For corticosteroids, we (TCL and JMB) extracted the results for patients from RECOVERY (The RECOVERY Collaborative Group, 2020), METCOVID (Jeronimo et al., 2020) , and CODEX (Tomazini et al., 2020) and we extracted the remainder of the data from the WHO meta-analysis of corticosteroids (The WHO Rapid Evidence Appraisal for COVID-19 Therapies (REACT) Working Group, 2020). For REMAP-CAP (The Writing Committee for the REMAP-CAP Investigators, 2020) we used the data from the WHO meta-analysis because there were 70 patients included in their final paper who were enrolled at centres where care without steroids was not available and data excluding these subjects was not available with sufficient granularity.

To estimate the final (posterior) probability of differences in outcomes between the remdesivir and control arms as well as the corticosteroids and control arms respectively, the objective data (binomial likelihood) for each study must be combined with previous beliefs according to Bayes theorem (Spiegelhalter et al., 1999) . The estimates of interest were absolute risk differences which are easier for clinicians to conceptualize than hazard or risk ratios and have more meaning for public health decisions.

The binary outcome data from each trial was transformed to logarithmic odds ratios (and their associated standard errors), which were then analyzed assuming a normal-normal hierarchical model. Under this model, individual trial outcomes and standard errors are modeled via normal distributions, using their means and their standard errors as sufficient statistics. The second hierarchy level treats the between trial heterogeneity as an additive normal variance model. This provides a random effects model where we want to estimate two parameters, the overall effect, μ, the risk difference, and the positive heterogeneity τ between trials. We have used vague proper informative priors; μ centered at 0 (standard deviation = 4), which corresponds to no effect, and heterogeneity τ assumed to be a halfnormal prior (to assure positive values) with scale 0.03. Sensitivity analyses were performed using different prior distributions (e.g. varying μ and/or using a half-Cauchy distribution for τ) to confirm the estimates were stable. This was operationalized with the bayesmeta package (Röver, 2020) in the R environment (R Core Team, 2019). For comparison a random effects meta-analysis for risk ratio is presented in the supplement.

We then generated figures of posterior density versus absolute difference in mortality between treatment and control patients. From these, we calculated via simulations the posterior probability that there was any mortality benefit and whether the mortality benefit 

In total, we analyzed data from 4 remdesivir trials including 7322 patients (Table 1) and from 8 corticosteroid trials which included 7557 patients (Table 2) . Figures 1 and 2 (a-c) show the posterior density as a function of risk difference for mortality for remdesivir and corticosteroids versus control patients respectively for the three subgroups. Table 3 contains the probabilities that remdesivir or corticosteroids reduce mortality at all and by at least 1, 2 and 5%. Remdesivir had a low probability of clinically meaningful mortality benefit in subgroups other than patients requiring oxygen without ventilation where the probabilities of mortality benefit overall and at least 1, 2, and 5% were 92%, 81%, 61% and 10% respectively. Conversely, corticosteroids (predominantly dexamethasone) showed a high probability of benefit (≥93% exceeding 1 in 100) in all subgroups except patients not requiring oxygen where the probability of any benefit was only 7%.

We evaluated remdesivir clinical trial results by performing a Bayesian meta-analysis to provide estimates of the probability for a clinically meaningful effect and found that remdesivir was unlikely to benefit the critically ill requiring ventilation, with a 93% chance of no effect or increased mortality. By comparison, corticosteroids demonstrated strong evidence of benefit in patients requiring advanced respiratory support or supplemental oxygen. We did find a potential benefit of remdesivir for patients requiring oxygen; J o u r n a l P r e -p r o o f however, using our analytic approach, the probability of a small meaningful effect on mortality (above 1 in 100) was only 81%. Finally, patients not requiring oxygen were unlikely to benefit from either therapy, with remdesivir demonstrating only a 29% chance of mortality benefit of 1% or more and corticosteroids only a 4% chance.

In line with our findings, the National Institutes of Health (NIH) COVID-19 guidelines (as of December 3, 2020 (COVID-19 Treatment Guidelines Panel, 2020)) recommend dexamethasone without remdesivir for patients requiring mechanical ventilation or extracorporeal membrane oxygenation. The NIH panel recommends dexamethasone with remdesivir or dexamethasone alone for patients with high flow or noninvasive ventilation requirements and for those who are hospitalized and require oxygen without advanced support, remdesivir monotherapy with a lesser recommendation for combination therapy with dexamethasone or dexamethasone monotherapy. Our analysis suggests that the probability that dexamethasone and remdesivir reduce mortality by more than 1 in 100 in this population is 93% and 81% respectively. The estimates for dexamethasone are limited by lack of a large replication trial. However, given dexamethasone is inexpensive, has a well-established safety record, and is generally well tolerated it seems reasonable to proceed with the treatment of hypoxic patients without such a confirmatory trial. Whether there is added benefit from giving remdesivir in combination with steroid treatment is unknown. In SOLIDARITY, there was no evidence of effect modification of remdesivir for the approximately 50% of patients who also received steroids (WHO Solidarity Trial Consortium, 2020). The role of remdesivir in this population would be a good target for a rapid and focused randomized controlled trial and J o u r n a l P r e -p r o o f indeed stratifying by the intensity of oxygen requirements provide further clarity. Finally, among patients who do not require supplemental oxygen, the NIH guidelines recommend against dexamethasone and give a contextual recommendation for remdesivir. We show that in this subgroup, neither dexamethasone nor remdesivir are likely to benefit, if one accepts that a 1 in 100 mortality reduction is a reasonable threshold for clinically significant impact.

Our analysis has several limitations. Firstly, the absence of individual patient data limits the ability to stratify for important subgroups including age, ethnicity, medical comorbidities, or duration of illness. Such an analysis, though post-hoc, might better define who would benefit most from remdesivir or which groups would be best represented in confirmatory trials. Secondly, we were required to make some assumptions in the subgroups because granular data were not available. However, the number of patients who may have been misclassified was small and/or mortality was unlikely in both control and treatment groups. Thirdly, in terms of contextualizing the effect size of remdesivir with corticosteroids, data for steroid use outside of severe illness was limited and highly influenced by RECOVERY (The RECOVERY Collaborative Group, 2020). Importantly, this was not a network meta-analysis, but in this interpretation, we made some indirect comparisons between corticosteroids and remdesivir. These treatments may not be directly comparable and our objective in doing so was only to contextualize the effect size of remdesivir compared to the only other currently proven effective therapy. Finally, we did not evaluate the benefit on time to "recovery" or "fitness to discharge," which was reduced in ATCC-1 and is an important consideration given Notwithstanding these limitations, we believe this analysis provides a richer and complementary interpretation of the data to help guide clinicians to make appropriate use of remdesivir and corticosteroids in various subgroups of hospitalized patients.

Based on a Bayesian meta-analysis we contextualized the results of the remdesivir and corticosteroid clinical trials in terms of the probability of a meaningful impact on inpatient mortality. When viewed alongside the data for corticosteroids, particularly dexamethasone, the probability of a meaningful effect with remdesivir was lower. We found that remdesivir is unlikely to reduce mortality in the critically ill and for those who do not require oxygen; however, there remains a chance remdesivir may reduce mortality by at least 1% in those requiring non-invasive oxygenation. At an estimated price of $2340-$3120 USD per 5 days course (O'Day, 2020), investing in a moderate probability of a 1% absolute reduction in mortality requires a substantial global commitment of funds.

In the future, a cost-effectiveness analysis examining the potential for reduced hospital length of stay with 5 days of remdesivir would be a meaningful addition to the discussion.

In addition, the added benefit of remdesivir in hypoxic patients requiring non-invasive supplemental oxygen and treated with dexamethasone would be a good target for a rapid and focused randomized controlled trial. While awaiting such a definitive study, this probabilistic interpretation of the evidence may help guide treatment decisions for clinicians as well as guideline and policy makers.

Drs. Lee, Harrison, and Cheng were co-investigators on CATCO, the Canadian arm of the WHO SOLIDARITY trial. 

Montréal, Canada 4. Division of General Internal Medicine

COVID-19) Treatment Guidelines. National Institutes of Health

Methylprednisolone as Adjunctive Therapy for Patients Hospitalized With COVID-19 (Metcovid): A Randomised, Double-Blind, Phase IIb, Placebo-Controlled Trial

R: A language and environment for statistical computing. R Foundation for Statistical Computing

Bayesian Random-Effects Meta-Analysis Using the bayesmeta R Package

Sensible Medicine-Balancing Intervention and Inaction During the COVID-19 Pandemic

An introduction to bayesian methods in health technology assessment

Effect of Remdesivir vs Standard Care on Clinical Status at 11 Days in Patients With Moderate COVID-19: A Randomized Clinical Trial

Dexamethasone in Hospitalized Patients with Covid-19 -Preliminary Report

Association Between Administration of Systemic Corticosteroids and Mortality Among Critically Ill Patients With COVID-19: A Meta-analysis

The Writing Committee for the REMAP-CAP Investigators. Effect of Hydrocortisone on Mortality and Organ Support in Patients With Severe COVID-19: The REMAP-CAP

Corticosteroid Domain Randomized Clinical Trial

Effect of Dexamethasone on Days Alive and Ventilator-Free in Patients With Moderate or Severe Acute Respiratory Distress Syndrome and COVID-19: The CoDEX Randomized Clinical Trial

Remdesivir in adults with severe COVID-19: a randomised, double-blind, placebo-controlled, multicentre trial

Repurposed Antiviral Drugs for Covid-19 -Interim WHO Solidarity Trial Results

We would like to acknowledge all the patients who volunteered for the included studies and the investigators and research teams who performed and published them.J o u r n a l P r e -p r o o f

*3 patients included in placebo arm were not on oxygen at enrollment and 1 patient was on mechanical ventilation; ** includes 55 and 38 patients respectively who went on oxygen between eligibility and receipt of 1st dose