To date, there is no definite effective treatment for the novel coronavirus (SARS-CoV-2) pandemic. To compare and rank SARS-Cov-2 treatment according to their efficacy and safety. Using the terms Covid-19 or SARS-CoV-2 and treatment, a literature search was performed from MEDLINE, GOOGLE, and CENTRAL databases until July 01, 2020. Randomized clinical trials (RCTs) against SARS-CoV-2 disease were included. The studies excluded were those with nonrandomized design or those with a lack of information on outcomes. To evaluate studies methods, the Cochrane Risk of Bias Tools was used. Efficacy and adverse reaction number were extracted. A frequentist network meta-analysis using random-effect model was conducted. The risk ratio (RR) and 95% CI were calculated for clinical improvement, all-cause mortality, and any adverse event 28-days after randomization. The study protocol is registered with PROSPERO, number CRD42020176977. A total of 14 RCTs, which assessed 11 different treatments and 2,898 participants (range of mean age; 44.7 to 70 years; 1,731 [59.7%] men) were included in the analysis. The overall quality of evidence was rated as high to moderate. 1,658 (57.2%) patients had a clinical improvement, and 5-day of remdesivir was ranked as the better treatment (P-score 0.86). RR compared with standard of care was 1.39 (95% CI 1.00-1.93). 246 (8.5%) patients died within a 28-days after randomization. None difference between treatments in terms of reducing mortality was found. Among the 1,166 (40.2%) reported adverse events (AEs), 467 (40%) were severe. Arbidol (RR, 0.22, [0.07-0.74]), 450 mg of HCQ (0.31, [0.12-0.84]), remdesivir for both 5-day (0.35, [0.16-0.78]) and 10-day (0.36, [0.18-0.72]), and standard of care (0.38, [0.21-0.70]) were associated with low risk of any AEs relative to colchicine. In this study, different treatments were associated with similar effects in reducing deaths, remdesivir for 5-day was associated with more clinical improvement, and colchicine and hydroxychloroquine had more safety concern. Data from ongoing clinical trials are need to drive more precise conclusions on efficacy and safety.
SARS-CoV-2, Network meta-analysis, Treatment
RCT: Randomized Controlled Trials; RR: Risk Ratio; RDVs: Remdesivir for 5-day; RDV: Remdesivir more than 5-day; FPV: Favipiravir; LPVRTV: Lopinavir/Ritonavir; LPVRTVRBV: Lopinavir/Ritonavir and Ribavirin; ARB: Arbidol (Umifenovir); Plasma: convalescent plasma; StdCare: Standard of care; AZT: azithromycin; HCQ low: Low Dose of Hydroxychloroquine (450 mg); HCQ: Hydroxychloroquine; HCQAZT: Association Hydroxychloroquine and Azithromycin
The ongoing pandemic responsible for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) commonly designated Covid-19 is by far the worst and deadliest worldwide infection in the past 20-year. To date, more than 10.7 million cases and 512,331 deaths had been globally reported . The emergency of this situation accelerated the randomized trials of many repurposed drugs which efficacy had been highlighted in vitro or the therapeutic experience from the SARS-CoV-1, and Middle East respiratory syndrome (MERS)-CoV infection .
Of 14 published randomized trials in SARS-CoV-2 virus to date, only three has been shown a clinical benefit for compared to the standard of care. Chen Z, et al.  conclude that 400 mg/d of hydroxychloroquine (HCQ) for 5-day improve pneumonia in 80.6% of patients; Deftereos, et al.  indicate that patients who received colchicine (1.5 mg loading dose followed by 0.5 mg after 1 h and maintenance doses of 0.5 mg twice daily) had significantly improved time to clinical deterioration, and remdesivir was found to be superior to placebo to shortening the time to recovery . Although these previous studies included a small sample, trials involving thousands of patients are ongoing especially RECOVERY (NCT04381936), DISCOVERY (NCT04315948), and SOLIDARITY (NCT04330690).
Furthermore, while the question of HCQ efficacy for SARS-CoV-2 virus has raised many debates, literatures data are conflicting. Data from Million, et al. , Lagier, et al. , and Arshad, et al.  indicated the effective of HCQ to reduce mortality in COVID-19 disease, while those from Singh, et al.  completely showed the opposite sense with 2.17-fold increase in mortality for patients treated by HCQ, and those from four observational comparative studies [10-13] concluded to the ineffective of HCQ.
As there is no recommended treatment or vaccine to contain the disease to date, identifying the most effective treatment is an urgent medical need. To our knowledge, no network meta-analysis has been conducted to summarize publish and unpublished data on promising treatments against Covid-19 infection. In this study, we reported a preliminary result of a network meta-analysis (NMA) of randomized trials to compare and rank the efficacy and safety of tested treatments in patients with SARS-CoV-2 virus.
We searched through MEDLINE, GOOGLE, and Cochrane library (CENTRAL) for randomized controlled trials (RCTs) that investigated the efficacy and safety of treatments against the SARS-CoV-2 virus. The search was restricted to randomized trials conducted in human, and published in any language before July 01, 2020. Trials in which participant were non-randomly allocated to receive SARS-CoV-2 virus treatment were excluded. Using the search terms listed in the Supplementary (eMethod), AD and MT identified all relevant studies, then independently reviewed their full texts, and in case of disagreement, differences were resolved through the arbitration of another author (MCB). Extracted data included: First author name and year of publication, country, RCTs design, study follow-up, age (mean), proportion of men participants, treatment and dosing information, sample size, study sponsorship, proportion or number of participants with clinical improvement, all-cause mortality, and adverse events. The study protocol number is CRD42020176977 (PROSPERO).
We considered any pharmacological medication which was tested to evaluate their efficacy and safety in patients infected by the SARS-CoV-2 virus. Globally, 11 different treatments were compared and ranked (Table 1). For randomized trials, patients were defined as receiving intervention or control if they were randomly allocated to receive either treatment. Almost, all patients received supportive care according to the standard of care for the trial site.
The primary outcome was clinical improvement within a 28-day after randomization. Clinical improvement was defined as patient discharge or a reduction of 2 points on a 6-point disease severity scale which was defined as follow: 6-point, death; 5 points, hospitalization plus extracorporeal membrane oxygenation (ECMO) or invasive mechanical ventilation; 4 points, hospitalization plus noninvasive ventilation or high-flow supplemental oxygen; 3 points, hospitalization plus supplemental oxygen (not high-flow or noninvasive ventilation); 2 points, hospitalization plus supplemental oxygen; 1 point, hospital discharge. Secondary outcomes were all-cause mortality and any Adverse Events (AEs) during treatment course. Because the variability of the endpoint assessment for efficacy and safety outcomes, we considered the lasted evaluation.
Original clinical trials were described using study characteristic summary table and forest plot. The Cochrane risk of bias tools  and Revman version 5.4 were used to assess the risk of bias and to generate its figure respectively. We opted for a frequentist approach to compare efficacy and safety between tested treatments using a random-effects network meta-analysis (NMA) for binary endpoint. Summary estimates were reported as risk ratio (RR) with their reported 95% confidence intervals. For clinical improvement, RRs > 1 correspond to beneficial treatment effects of the first treatment compared to the second, while for the secondary outcomes, it was the reverse. To display the relative efficacy and safety outcomes of all available pairwise comparisons between treatments, a league tables were used. To choose the preferred regimen, the P-score which ranging from 0 (worse treatment) to 1 (best treatment) was computed for each treatment, then treatment with a higher P-score was selected as the better than the competing each treatment. Heterogeneity and inconsistency were quantified using the global Q test proposed by Rucker . The Q statistic is the sum of statistic for heterogeneity, which represent the proportion of total variation in study estimates (within-designs), and a statistic for inconsistency (between-designs), which represents the variability of treatment effect between direct and indirect comparisons at the meta-analytic level . To visualize and identify the nodes of single-design inconsistency, we used a network heat plot. Consistency between direct and indirect comparisons was checked using the so-called node-splitting. Because the small number of included trials that reported all-cause mortality at 28 days and safety, we performed two sensitivity analyses by adding nonrandomized comparative studies in meta-analysis to compare and rank mortality and any adverse event between pharmacological drugs. In these observational comparative studies, patients who were exposed to treatment were those receiving intervention or control at study baseline or received it during the follow-up period before the assessment of efficacy and safety outcomes. None subgroup analysis was performed. All analyses were performed using R package 'netmet' ; P-values < 0.05 was considered significant for the difference between treatments.
The initial search through all database identified 1,007 citations, of which 469 were screened by title and abstract after removing duplicates. Of the 27 full-text citations reviewed, 14 RCTs [3-5,16-26] that met the inclusion criteria were finally included in the quantitative network meta-analysis (Figure 1). These 14 RCTs (two phase 2 and five blinded) included together 2,898 patients infected by the SARS-CoV-2 virus with mean age between 44.7 and 70 years, and 1,731 [59.7%] were men, and followed from 6 to 28 days. 1,656 (57.1%) patients had comorbidity with the most common were hypertension (1,029; 35.5%) and diabetes (627; 21.6%).
Figure 1: PRISMA Flowchart of studies selected for meta-analysis of RCT SARS-CoV-2 treatments.
RCT: Randomized clinical trial. View Figure 1
The methodological quality of included RCTs is shown in Figure 2. Overall, the risk of bias was low in two RCTs, moderate in three RCTs, and high in the rest (Supplementary Figure 1 and Supplementary Table 1). A higher risk of attrition bias (incomplete outcome data), and performance bias (blinding participants and personnel) occurred in five and five of 14 RCTs respectively.
Figure 2: Summary of risk bias assessment for RCTs SARS-CoV-2 treatments comparisons.
Number in parenthesis are references. View Figure 2
Table 1: Characteristics of included randomized trials investigating the efficacy and safety of SARS-Cov-2 virus. View Table 1
Data for primary efficacy outcome (clinical improvement) were performed in 12 of the 14 RCTs yielding nine treatments and 14 comparisons [3-5,16-21,22,26]. Of the 2,898 participants, 1,658 (57.2%) had clinical improvement 28 days after randomization. Figure 3 shows the network for clinical improvement captured by the SARS-CoV-2 virus treatment, and the corresponding pairwise comparisons are summarized in Supplementary Table 2. 100 mg of Remdesivir once daily on 5 days was ranked with a higher probability to achieve clinical improvement at 28 days (P-score 0.86). Except between remdesivir and standard of care, no significance difference between treatments was found from the pairwise comparisons (Supplementary Table 2). Risk ratio (RR) for 100 mg of remdesivir once daily on 5 days compared with Standard care was 1.39 (95% CI 1.00-1.93). Likewise, no significant differences between direct and indirect treatment estimates comparisons or evidence of publication bias according to the comparison-adjusted funnel plot were found (Supplementary Figure 2).
Figure 3: Network graph of eligible SARS-CoV-2 treatments comparisons for clinical improvement.
Line width is proportional to the number of trials comparing every pair of treatment. The size of the circle is proportional to the number of participants assigned to receive the treatment; Remdesivir (5-day): Remdesivir for 5-day; remdesivir (10-day): remdesivir for 10-day; Plasma: convalescent plasma. View Figure 3
Data for all-cause mortality were reported in seven trials [4,5,20-24] yielding six treatments and six comparisons. A total of 246 (8.5%) patients died within a 28-days post-randomization, and colchicine (1.5 mg loading dose followed by 0.5 mg after 1 h and maintenance doses of 0.5 mg twice daily) was ranked as the best option with a probability of 83% (P-score 0.83) to be associated with a lower risk of death. No significant difference was observed between treatments (Figure 4 and Supplementary Table 3), RR for colchicine compared to the Standard care was 0.91 (0.45-1.83).
Figure 4: Network meta-analysis comparing single treatment with standard of care (A, C, and D) or colchicine (B) of SARS-CoV-2 outcomes.
A) Clinical improvement; B) Any adverse event; C) Any severe adverse event; D) All-cause mortality; Treatment are ordered in the rank of their chance of being the best option. Treatment estimates are provided as risk ratios (RR) with 95% CIs. RRs > 1 indicates a beneficial treatment effects compared to standard care (clinical improvement), while RRs > 1 is favor for control (any AEs or serious AEs and mortality). RDVs: Remdesivir for 5-day; RDV: Remdesivir more than 5-day; FPV: Favipiravir; LPVRTV: Lopinavir/ritonavir; LPVRTVRBV: Lopinavir/ritonavir and ribavirin; ARB: Arbidol (umafenovir). Plasma: Convalescent plasma; StdCare: Standard of care; AZT: Azithromycin; HCQlow: Low dose of hydroxychloroquine (450 mg); HCQ: Hydroxychloroquine; HCQAZT: Association hydroxychloroquine and azithromycin. View Figure 4
For the safety outcome, the network meta-analysis was performed in all 14 RCTs, yielding 11 treatments and 16 comparisons. A total of 1,166 (40.2%) adverse events were reported at the treatment end, either 28-day after randomization. Arbidol (200 mg daily twice three times for 14 days) was ranked as the best option with a probability of 86% (P-score 0.86) to be associated with a lower risk of any AEs. Compared to colchicine, we found that arbidol, low dose of HCQ (450 mg), remdesivir for both 5 and 10-day, association lopinavir/ritonavir, and standard of care were significantly associated with low risk of any AEs (Figure 2). The corresponding risk reductions were 78% (0.22, 0.07-0.74) for arbidol, 69% (0.31, 0.12-0.84) for low dose of HCQ, 65% (0.35, 0.16-0.78) for 5-day of remdesivir, 64% (0.36, 0.18-0.72) for 10-day of remdesivir, 62% (0.38, 0.21-0.70) for standard of care, and 53% (0.47, 0.23-1.00) for lopinavir/ritonavir. In addition, we found that a low dose of HCQ reduced the risk of any AEs by 50% (0.50, 0.27-0.90) when compared to high dose of HCQ (Supplementary Table 4).
Among the 1,166 reported adverse events (AEs), 467 (40%) were severe. The most common severe adverse events were acute respiratory failure or acute respiratory distress syndrome (ARDS) reported in 226 patients (117 in remdesivir, 97 in standard of care, and 12 in lopinavir/ritonavir), followed by the secondary infection in 17 cases (13 in standard of care and four in remdesivir), septic shock in 25 patients (14 in remdesivir, three in lopinavir/ritonavir, and eight in standard of care), and pneumothorax in 12 patients (seven in remdesivir and five in standard of care). For any severe AEs, network meta-analysis was performed in six RCTs involving six comparisons of six different treatments. A combination of lopinavir/ritonavir and ribavirin was associated with a risk reduction for any severe AEs with a probability 90% (P-score 0.90). Compared to standard of care, remdesivir for both 5 and 10 days and lopinavir/ritonavir reduced the risk of any severe AEs by 53% (0.47, 0.32-0.69), 23% (0.77, 0.63-0.94), and 40% (0.60, 0.37-0.98) respectively (Figure 2). Moreover, we found that the short exposition of remdesivir (5 days) reduced the risk of any severe AEs by 39% (0.61, 0.44-0.85) compared to the long exposition (10 days) (Supplementary Table 5).
In sensitivity analysis, after adding the 11 nonrandomized comparative studies of treatments against SARS-CoV-2 virus, colchicine (P-score 0.83) and arbidol (P-score 0.79) remained the best options to reduce all-cause mortality and any AEs 28 days after randomization respectively (Supplementary Table 6 and Supplementary Table 7). For mortality outcome, sensitivity analysis involved 36 comparisons of 10 different treatments in 17 studies including 17,251 patients in whom 2,669 (15.5%) died. We not found any difference between treatments in terms of risk reduction of death (Supplementary Table 2). For any AEs, sensitivity analysis involved 30 comparisons of 13 different treatments in 18 studies including 8,637 patients who reported 2,219 (25.7%) AEs. Compared to HCQ, azithromycin, remdesivir for 10-day, and standard of care were associated with 47% to 55% relative risk reductions of any AEs (Supplementary Table 7). The specific relative reductions were as follow: For azithromycin, 55% reduction (0.45, 0.24-0.84); for 10-day remdesivir, 49% reduction (0.51, 0.26-0.99); and for standard of care, 47% reduction (0.53, 0.36-0.80). When compared to colchicine, azithromycin, and standard of care were associated with 62% to 67% relative risk reductions. The specific relative reductions were as follow: for azithromycin, 67% reduction (0.33, 0.11-0.95) and for standard of care, 62% reduction (0.38, 0.16-0.91). Furthermore, in direct comparisons, we found that azithromycin was associated with 58% to 50% relative risk reduction when compared to azithromycin plus HCQ (0.42, 0.22-0.82) and HCQ alone (0.50, 0.26-0.97). Likewise, standard of care was associated with 54% to 48% relative risk reduction when compared to azithromycin plus HCQ (0.46, 0.24-0.89) and HCQ alone (0.52, 0.34-0.80).
Global heterogeneity was low for clinical improvement (Cochran's Q 9.01; p = 0.11; τ2 = 0.010; I2 = 44.5% [0%-78%]). For adverse event and mortality after including nonrandomized studies, global heterogeneity was significant (47.0; p < 0.0001; τ2 = 0.141; I2 = 76.6% [59.2%-86.6%] and 117.82; p < 0.0001; τ2 = 0.233; I2 = 86.4% [79.7%-90.9%] respectively), mainly due to significant between-design heterogeneity (AEs) and between-design as well as within-design heterogeneity (mortality). These finding were supported by the heat plot displayed in the Supplementary (Supplementary Figure 3, Supplementary Figure 4, Supplementary Figure 5 and Supplementary Figure 6).
In this study, we conducted the first network meta-analysis, based on 14 RCTs including 2,898 patients randomly assigned to 11 different treatments against the SARS-CoV-2 virus. Pooled results suggest that a 5-day course of remdesivir was superior to standard of care in terms of clinical improvement and to reduce adverse event, but with a comparable effectiveness to other pharmacological drugs. Moreover, our findings suggest that arbidol, low dose of HCQ (450 mg), favipiravir, lopinavir/ritonavir, and standard of care were superior to colchicine in risk reduction of any AEs.
None difference between treatments in all-cause mortality reduction was found. Relative risk reduction of all-cause mortality within a 28-day for standard of care compared to HCQ was 0.80, 0.54-1.19 which supported the evidence indicating the absence of the efficacy of HCQ to reduce mortality in COVID-19 disease. These findings were conflicting with those reported previously [6,7]. While, Million, et al.  shows that patients who received HCQ had a risk reduction of death of 68% (0.32, 0.19-0.52) compared to those treated without HCQ, Singh, et al.  completely indicates the opposite sense with 2.17-fold increase in mortality for patients treated by HCQ. Moreover, we found that HCQ was associated with more adverse events than azithromycin, remdesivir for 10-day, and standard of care (Supplementary Table 6).
Although, our study provides the most current evidence to date on the comparative efficacy and safety of available treatments against the SARS-CoV-2 virus, these findings should be interpreted with caution. We are aware that all pharmacological drugs classified as the best options for clinical improvement, all-cause mortality or safety concerns have only tested once. More data is needed to replicate these results. However, after the addition of nonrandomized comparatives studies, the results remained stable for mortality and safety, suggesting a robustness of the data.
To date several trials registered in ClinicalTrials.gov databases are ongoing, and results are expected in the coming months. Recently, the RECOVERY trial investigators had communicated on possible superiority of dexamethasone to manage mortality in hospitalized SAR-CoV-2 patients with severe condition . According to their results, 6 mg of dexamethasone once per day for ten days would reduce deaths by 35% (0.65, 0.48-0.88) in ventilated patients and by 20% (0.80, 0.67-0.96) and by 20% (0.80, 0.67-0.96) in other patients receiving oxygen only. However, the efficacy of dexamethasone against lower respiratory tract infection is not revolutionary, and has been proved in several clinical studies . When all ongoing trials are published, an update of this work will be necessary to draw definitive conclusions about the efficacy and safety of the treatments tested against the SAR-CoV-2 virus.
Despite our efforts to minimize publication bias by including unpublished studies like those posted in preprint database, this preliminary study had some limitations. First, the small number of RCTs included in the network meta-analysis negating the possibility of performing subgroup analyzes according to studies characteristics (design, follow-up, sample size, endpoint assessment, or risk of bias). Second, the different endpoint used for the assessment of efficacy and safety outcomes which may influence the results.
In this study, remdesivir for 5-day was reported to be more effective than standard of care to achieve clinical improvement, different treatments were associated with a similar risk reduction of death, and colchicine and HCQ had more safety concern compared to arbidol, favipiravir, low dose of HCQ (450 mg), remdesivir for both 5 and 10 days, and standard of care. However, data from ongoing clinical trials are need to drive more precise conclusions on efficacy and safety.
The data are available upon request (email@example.com).
We declare no competing interests in relation to this work.
AD conceived the study design, analyzed the data, and drafted the manuscript, MCB supervision of data collection, and critical revision of the manuscript, AD and MT collected data, MHD analyzed the data, BDD, and MT interpreted and substantially revised the manuscript. All authors approved the final version of the manuscript.