key: cord-0975105-a78kpj05
authors: Li, Y.; He, W.
title: Comparative Efficacy and Safety of Current Drugs against COVID-19: a Systematic Review and Net-work Meta Analysis
date: 2020-11-18
journal: nan
DOI: 10.1101/2020.11.16.20232884
sha: 16a8dde1789a5bc636f0160b09325eea19ac4f50
doc_id: 975105
cord_uid: a78kpj05

The rapid spread of coronavirus disease (COVID-19) has greatly disrupted the livelihood of many people around the world. To date, more than 35.16 million COVID-19 cases with 1.037million total deaths have been reported worldwide. Compared with China, where the disease was first reported, cases of COVID-19, the number of confirmed cases for the disease in the rest of the world have been incredibly high. Even though several dugs have been suggested to be used against the disease, the said interventions should be backed by empirical clinical evidence. Therefore, this paper provides a systematic review and a meta-analysis of efficacy and safety of different COVID-19 drugs.

harms is still poorly studied. Although many drugs are available for COVID 19, their use should be guided by integrated evidence to inform the best choice for each patient 3 .

Network meta-analysis provides a comparative efficacy and a broader evidence-based picture of several interventions, the reason it was selected for exploring relative merits of multiple COVID-19 control modalities 4 . This is critical for COVID-19 because the view on best drugs is constantly changing, and in particularly, in the race against managing this pandemic. Because it can provide a complete, extensive and updated viewpoint of evidence, network meta-analysis is an ideal tool to guide practical development and application of drugs against COVID 19. Our study provides a thorough comparison on effectiveness and safety of currently available interventions against COVID 19 , likely not to have been experienced in some clinical settings. This network metaanalysis was performed using Bayesian Markov chain Monte Carlo (MCMC) random effect model that is capable of fitting models of virtually unlimited complexity, and as such have revolutionized the practice of Bayesian data analysis 5 . Specifically, we evaluated the safety and effectiveness of 14 drugs, along with placebos and standard care, according to four clinical outcomes: average recovery time, response rate, mortality and the rate of adverse events. Findings of this research aimed at guiding the rationale of clinical decisions among COVID-19 patients.

Sear ch str ategy and selection cr iter ia

The systematic review and network meta-analysis was performed on numerous reports in the Cochrane Central Register of Controlled Trials, PubMed, Embase and Web of Science. We included relevant studies published since inception the above databases till August 11, 2020, regardless of the language. The search terms were severe acute respiratory syndrome coronavirus 2*" (MeSH) OR 2019nCoV*" OR SARSCoV2*" OR 2019 novel coronavirus*" OR COVID19 virus disease" OR COVID19 virus infection", combined with a list of all included drugs.

Randomized controlled trials (RCTs) were included if they evaluated comparable outcome of drugs with standard care, placebos or safety and effectiveness of a monotherapy in adults (≥18 years old and of both sexes) with COVID-19. The interventional modules must have been in line with NIH COVID-19 Treatment Guidelines 6 . Notable though, the included reports were not limited by dosage and mode of administration. Additionally, included studies must have reported on at least one of the aforementioned outcomes. Finally, quasi-randomised controlled clinical trials (CCTs) and incomplete trials, non-drug treatment such as ventilation or trials on less than 10 participants were all excluded from the meta-analysis. For duplicated publications, we only included the latest and most complete edition. Reviews, case reports, letters and conference abstract were also excluded from our analyses.

Relevant studies were selected independently by two reviewers. In particular, they reviewed the main reports and supplementary materials, extracted the relevant information from the included trials and assessed the risk of bias according to modified Jadad scale 7 . The quality of included studies was rated based on four items: (1). generation of random sequence, (2) . blindlessness of the trails, (3.)concealment and (4). withdraws and dropouts. Any discrepancies were resolved by a third arbitrator. Duplicates were first excluded using EndNote (version 9.3.3). Studies with irrelevant title and abstract were then removed. The remaining papers were read in whole to assess their suitability. Finally, relevant data were extracted independently according to pre-determined criteria.

The efficacy and safety of interventional drugs were assessed based on four outcomes: the average recovery time (average duration of hospital stay or days to reach clinical rehabilitation standard) and response rate (based on the proportion of patients discharged or reaching clinical rehabilitation standard during the test) for efficacy outcomes. On the other hand, safety outcomes included mortality (the proportion of patients who died during the test) and the rate of adverse events (the proportion of patients exhibiting adverse effect(s) associated with the drug).

Network meta-analysis was performed using a Bayesian Markov chain Monte Carlo (MCMC) random effect model. Key parameters for dichotomous outcomes included odds ratios (RR) and standardized mean difference (SMD) both at 95% CI for continuous outcomes 8 . Node splitting model tested the consistency in closed-loop interventions with both direct and indirect evidence with P ＜ 0.05 considered statistically significant. If direct and indirect evidence were statistical indifferent, then the consistency model was used in the network meta-analysis; otherwise the source and significance of inconsistency was assessed 5 . Additionally, potential scale reduction factors (PSRFs) were calculated to evaluate the convergence of the model. A PSRF value close to 1 is indicative of good convergence, thus the results are high reliability 9 . PSRF ≤ 1.02 was acceptable.

Finally, comparative effectivity and safety of drugs was analyzed according to ranking probability under the Bayesian network model.

Bayesian network model analysis was performed using ADDIS software 10 , whereas the diagrams were plotted using Stata V. 14. 

In general, we identified 1137 articles, and after careful review, we extracted 128 of them for in depth review (Figure 1 ). We constituted the final analyses via including 28 trial studies, 14 drugs together with placebos or standard care and 17078 patients. Of the trial studies, 21 (75%) were standard care-or placebo-controlled trials, whereas the remaining 7 (25%) performed drug-drug comparisons. In addition, 4 (14%), 16 (57%) and 8 (28%) of the trials comprised of patients from North and South America, Asia and Europe, respectively. Characteristics of each report are listed in Table 2 . Meanwhile, the risks of publication bias were assessed based on the modified Jadad scale.

Therefore, studies with a score of 0-3 were considered high degree of bias to publication, converse to those with a score of more than 4 ( Figure 2 ). Bayesian evidence networ k and conver gence evaluation

The network of relevant comparisons for the four outcomes is shown in Figure 3 . The relationship between all interventional methods was plotted bassed on the direct comparison data. Briefly, each cycle represents an interventional drug. The width of the connection lines is proportional to the number of trials comparing every pair of treatments, whereas the size of each circle is proportional to the number of randomly assigned participants (i.e, sample size). The direct or indirect evidence among different interventional methods provided the basic conditions for the network meta-analysis.

Specifically, except for Arbidol, Chloroquine, Febuxostat and Favipiravir, the rest of the drugs were evaluated against at least one standard care-or placebo.

Convergence PSRF of the four outcomes was acceptable, implying minor fluctuations in Markov chain and stable iteration trajectory. As such, the convergence of the model was within the expected distribution in the iterative process and, as a result, can be used in analyzing the data 11 . Specifically, our model contained 4 chains, featuring 20,000 tuning iterations, 50,000 simulation iterations, 10 000 thinning interval inference samples, with a variance of 2.5. The number of iterations and annealing was sufficient; therefore, no additional extension was needed.

Node splitting analysis was performed using ADDIS software. There was no significant difference (P > 0.05) in average recovery time (20 trials, comprising 14169 patients) and mortality (16 trials, comprising 16117 patients) across studies. This points to coherence between direct and indirect comparison. However, there was a significant difference in response rate (24 trials, comprising 15120 patients) between two pairs of intervention drugs (P< 0.05). In particular local inconsistency was found between chloroquine and hydroxychloroquine as well as between lopinavir-ritonavir and chloroquine. Meanwhile, inconsistencies were also observed between interferon β and lopinavirritonavir, and between lopinavir-ritonavir and standard care with regard to rate of adverse events (21 trials, comprising 4198 patients). These significant differences suggest of potential uncertainty All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted November 18, 2020. ;  of predicted effect regarding the above drugs, and as such, their clinical applications against COVID 19 need further evaluation 12 .

Effects and r ank pr obability of the inter ventions Aver age r ecover y days Average duration of hospital stays or days up clinical rehabilitation standard was reported in 20 trials incorporating 14169 participants. Treatment nodes included in the network meta-analysis were arbidol, azvudine, chloroquine, colchicine, favipiravir, hydroxychloroquine, interferon β, ruxolitinib, lopinavir-ritonavir, methylprednisolone and remdesivir, evaluated alongside placebos and standard care. There was no significant difference in average recovery days between interventions (Table 2) .

However, rank probability test showed favipiravir (rank probability: 0.46) is most likely the best intervention against COVID 19, followed by arbidol (rank probability: 0.18) and azvudine (rank probability: 0.15) ( Figure 4 and Table 3 ).

Of the 28 included reports, 24 trials encompassing 15120 participants reported on the number of patients discharged or who reached clinical rehabilitation standard during the test. The treatment nodes involved in the network meta-analysis were azvudine, chloroquine, dexamethasone, colchicine, favipiravir, febuxostat, hydroxychloroquine, interferon β, ruxolitinib, lopinavir-ritonavir, remdesivir, placebo and standard care. As shown in Table 2 , there was no convincing evidence that any of the interventions significantly enhanced the response rate of patients with COVID 19.

Nonetheless, rank probability test showed that the drugs, colchicine (rank probability: 0.42), chloroquine (rank probability: 0.28) and favipiravir (rank probability: 0.11), are most likely to be as the best interventions against the disease. Meanwhile, the treatment effects of febuxostat (0.04) dexamethasone (0.03) and ruxolitinib (0.03) were comparable to that of placebos (0.03) ( Figure 4 and Table 3 ). All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted November 18, 2020. ; https://doi.org/10.1101/2020.11.16.20232884 doi: medRxiv preprint

Sixteen trials featuring 16117 participants reported on the number of patients who died of COVID 19 during the trials. Treatment nodes used in the analysis are as follows: colchicine, hydroxychloroquine, interferon β, ruxolitinib, ribavirin, lopinavir-ritonavir, remdesivir, placebo and standard care. Ruxolitinib was closely associated with the reduced mortality-rates [(2.24E+13, 20.03-7.78E+45 for placbo and 2.00E+13, 17.9-6.01E+45 for standard care (Table 2) ]. Moreover, rank probability test showed ruxolitinib (rank probability: 0.99) is most likely the safest intervention against COVID 19 ( Figure 4 and Table 3 ). Other interventions did not inhibited significant difference in mortality.

Meanwhile 21 trials contained 4198 participants demonstrated the adverse effect is related with COVID 19 treatment with drugs such as arbidol, azvudine, chloroquine, colchicine, favipiravir, hydroxychloroquine, interferon β, ruxolitinib, lopinavir-ritonavir, methylprednisolone, remdesivir, placebo and standard care. As shown in Table 2 , moderate certainty evidence showed that compared with standard care, hydroxychloroquine is significantly linked to the adverse events (OR=6.87, 1.52-41.44). Nonetheless, other drugs did not significantly affect the mortality (P > 0.05). Rank probability test showed that azvudine (rank probability: 0.97) may be an intervention associated with least adverse effect. Even so, randomized-controlled trials in patients with COVID-19 has so far provided little definitive evidence about adverse effects for most interventions (Figure 4 and Table 3 ) All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted November 18, 2020. ; https://doi.org/10.1101/2020.11.16.20232884 doi: medRxiv preprint Discussion COVID-19 pandemic is currently the most catastrophic public health emergency in the world, with the disease agent displaying one of the fastest transmission rates ever recorded. Throughout human histroty, severe acute respiratory syndrome strain 2 (SARS Cov 2) virus has been the most difficult infectious agent to control or manage, shown also to cause the most diverse range of complications.

Finding effective and safe treatment is not only the key but also the major challenge in controlling the pendenmic. There have been numerous investment by research communiy across the world, exploring the pathogenic mechanism of COVID-19. Continuous reporting on the disease has elucidated the molecular mechanism and potential infection pathway of COVID-19, offering the chance for the rationalized use of clinical treatment. However, there is still concern over the effectiveness and safety of various interventions. Meanwhile, short-term treatment benefits are averagely minimal, and the balance between long-term benefits and harms is still poorly understood.

Therefore, although clinicians have a wide pool of drugs to select from, however, the best choice for each patient should be backed by empirical evidence.

Accordingly, this network meta-analysis was performed to understand the relative balance between merits and potential shortcomings of multiple COVID-19 interventions. In general, 17078 individuals reported in 28 papers were randomly assigned to one of 14 individual drugs against COVID-19, placebo or standard care. Response of patients to various COVID 19 infections across the world was obtained thorough search of published reports in any language. The key parameters of interest were average recovery days, response and mortality rates and rate of occurrence of adverse events. This provided a solid base to comprehensively evaluate the effectiveness and reliability of various drugs against COVID-19. Our analysis found the efficacy of commonly proposed drugs to be insignificantly indifferent when compared with placebos or standard care.

Moreover, the therapeutic effects of the drugs is always modest. For instance, compared with other drugs, patients on remdesivir, favipiravir and azvudine exhibited higher recovery rate and lower occurance of adverse events, though statistically insignificant. By contrast, compared with other All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted November 18, 2020. ; https://doi.org/10.1101/2020.11.16.20232884 doi: medRxiv preprint drugs, chloroquine and hydroxychloroquine are not effective against COVID 19, and thus, may not be a first choice in the clinic. On the other hand, this meta-analysis demonstrated that compared with other interventions, ruxolitinib is associated with lower mortality rate, but the curative effect and other benefits of this drug in patients with COVID 19 remains unclear. In addition, according to a randomized-controlled trial, hydroxychloroquine did not reduce COVID 19-associated mortalities, and moreover, promoted the occurrence of adverse events. In general, so far there is no sufficient evidence demonstrated that the proposed COVID 19 drugs have improved eficacy and safety compared with placebo or standard care.

Our findings notwithstanding, this report has several limitations. First, the study focused on hospitalized patients, thus it may have suffered a possible selection bias. Therefore, it is not clear how asymptomatic individuals with a positive nucleic acid test result or mild COVID 19 symptoms would respond to the proposed drugs. Second, given we included trials from the whole world, the majority of studies involved in this meta-analysis were basically on Chinese population, thus further universal studies should be reviewed. Meanwhile, the heterogeneity between studies included in this report is critical. Sources of heterogeneity included broad inclusion criteria, differences in clinical trial design, patient cooperation and measurement of outcome. Additionally, differences in individual samples with regard to severity of the disease, intervention time, age and gender composition may have contributed to heterogeneity. Thus given the significant heterogeneity between studies related to COVID-19, our findings should be interpreted with caution, random effect model notwithstanding 13 .

Despite these limitations, this network meta-analysis provides preliminary understanding on current COVID-19 treatment. In this report, we first conducted a comprehensive search in five databases including Cochrane, EmBase, PubMed and Web of Science, incorperating a relative a large sample of COVID 19 patients across the world. Although the reports on several drugs such as colchicine are limited, findings of this study may still provide a preliminary reflection of their safety and efficacy. In addition, selection, inclusion and data extraction were performed by two independent All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted November 18, 2020. ; https://doi.org/10.1101/2020.11.16.20232884 doi: medRxiv preprint researchers, thus minimized the selection bias and strenthened the reliability of the results.

Aditionally, all included studies were tested for publication bias through sensitity analysis. The infection. One possible reason is that immune agents may aggravate the degree of immune imbalance, thus causing death. Our findings are thus instructive for selecting the most effective intervention of managing COVID-19 patients. When implemented, our findings will rationalize clinical decisions among COVID-19 patients.

All authors collectively and equally participated in all aspects of the research, from preliminary literature assesment, data extraction, statistical analysis as well as writing and editing of the manuscript.

All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in 17 Cao 18 Cao 19 Chen 20 Chen 21 Chen 22 Corral-Gudino 23 Davoodi 24 Davoudi-Monfared 25 Deftereos 26 Horby 27 Horby 28 Huang 29 Hung 30 Li 31 Mitjà 32 Ren 33 Skipper 34 Tang 35 Wang 36 Geleris 37 Spinner 38 Kim 39 Cavalcanti 40 Gautret 41 Paccoud Each cycle represented an intervention drug, the width of the connection lines is proportional to the number of trials comparing every pair of treatments, and the size of every circle is proportional to the number of randomly assigned participants. Table 2 . Directly compare the results of 14 drugs, placebo and standard therapies. Compared with the row definition process, the data is the SMD or RR (95% CI) of the column definition process. In the average recovery day, if the SMD is greater than 1, the column definition processing takes precedence. For the remission rate, mortality and the incidence of adverse events, RR less than 0 is beneficial to define the treatment. In order to get the reverse comparison of RR, you need to take the inverse number. Important results are shown in bold. perpetuity.

preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

The copyright holder for this this version posted November 18, 2020. ; https://doi.org/10.1101/2020.11.16.20232884 doi: medRxiv preprint

COVID-19 Map -Johns Hopkins Coronavirus Resource Center

A Rush to Judgment? Rapid Reporting and Dissemination of Results and Its Consequences Regarding the Use of Hydroxychloroquine for COVID-19

Evidence-based medicine in times of crisis

Graphical tools for network meta-analysis in STATA

Bayesian Model Choice Via Markov Chain Monte Carlo Methods

COVID-19) Treatment Guidelines

Scales to Assess the Quality of Randomized Controlled Trials: A Systematic Review

Meta-analysis: Principles and procedures

Ratio of means for analyzing continuous outcomes in meta-analysis performed as well as mean difference methods

Evaluation of scientific evidence using Bayesian networks

Consistency and inconsistency in network meta-analysis: concepts and models for multi-arm studies

Predictive distributions for between-study heterogeneity and simple methods for their application in Bayesian meta-analysis

Drug Evaluation during the Covid-19 Pandemic

Discovering drugs to treat coronavirus disease 2019 (COVID-19)

Pathogenic human coronavirus infections: causes and consequences of cytokine storm and immunopathology

Remdesivir for the Treatment of Covid-19 -Preliminary Report

A Trial of Lopinavir-Ritonavir in Adults Hospitalized with Severe Covid-19

Ruxolitinib in treatment of severe coronavirus disease 2019 (COVID-19): A multicenter, single-blind, randomized controlled trial

Favipiravir versus Arbidol for COVID-19: A Randomized Clinical

A Multicenter, randomized, open-label, controlled trial to evaluate the efficacy and tolerability of hydroxychloroquine and a retrospective study in adult patients with mild to moderate Coronavirus disease

Efficacy and safety of chloroquine or hydroxychloroquine in moderate type of COVID-19: a prospective open-label randomized controlled study

A controlled trial of methylprednisolone in adults hospitalized with COVID-19 pneumonia

Febuxostat therapy in outpatients with suspected COVID-19: A clinical trial

Efficacy and safety of interferon beta-1a in treatment of severe COVID-19: A randomized clinical trial

Effect of Colchicine vs Standard Care on Cardiac and Inflammatory Biomarkers and Clinical Outcomes in Patients Hospitalized With Coronavirus Disease 2019: The GRECCO-19 Randomized Clinical Trial

Dexamethasone in Hospitalized Patients with Covid-19 -Preliminary Report

Effect of Hydroxychloroquine in Hospitalized Patients with COVID-19: Preliminary results from a multi-centre, randomized

Treating COVID-19 with Chloroquine

Triple combination of interferon beta-1b, lopinavirritonavir, and ribavirin in the treatment of patients admitted to hospital with COVID-19: an openlabel, randomised, phase 2 trial

An exploratory randomized controlled study on the efficacy and safety of lopinavir/ritonavir or arbidol treating adult patients hospitalized with mild/moderate COVID-19 (ELACOI)

Hydroxychloroquine for Early Treatment of Adults with Mild Covid-19: A Randomized-Controlled Trial

Open-Label, Controlled Clinical Trial of Azvudine Tablets in the Treatment of Mild and Common COVID-19, a Pilot Study

Hydroxychloroquine in Nonhospitalized Adults With Early COVID-19: A Randomized Trial

Hydroxychloroquine in patients with mainly mild to moderate coronavirus disease 2019: open label, randomised controlled trial

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

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

Lopinavir-ritonavir versus hydroxychloroquine for viral clearance and clinical improvement in patients with mild to moderate coronavirus disease 2019. The Korean journal of internal medicine

Hydroxychloroquine with or without Azithromycin in Mild-to-Moderate Covid-19

Hydroxychloroquine and azithromycin as a treatment of COVID-19: results of an open-label non-randomized clinical trial

This study was supported by the National Natural Science Foundation of China (Nos. 81872823 and