Additional outcomes
Incidence of elevation in liver enzymes, infusion-related reaction, neutropenia, and bacteremia.
  1. Search methods for identification of studies

    Information sources:

The following databases have been checked from the first of July 2020 and continued through August to identify relevant studies: ClinicalTrial.gov, ProQuest, PubMed, Embase, Cochrane, Google Scholar, Science direct, Chinese Clinical Trial Registry (ChiCTR), and medRxiv.
The following search terms were used: ”actemra”, ”tocilizumab”, ”SARS” and ”COVID.” In addition, a deep manual search through checking references in bibliography of original articles and relevant reviews was performed.

Study selection:

Studies were selected based on PRISMA flow diagram (Fig. 1 ). Two authors (HKE and MAE) independently selected potentially eligible studies from screened ones based on the title and/or abstract. The full text has been presented in order to review the articles screened for inclusion criteria. The reasons for excluded studies were discussed and any disagreements were resolved through discussion. A third author (MS) revised the process of the study selection and eligibility assessment.

Data collection and analysis:

Data was extracted using a predefined extraction form. The form included:
Study characteristics: title, authors, year of publication, country, and journal name.
Study design: type of the study design, hospital name, time of interventions, time of follow-up, and other settings
Population: age, sex, baseline clinical factors, disease severity definition, inflammation status, inclusion and exclusion criteria, and the disease onset
Intervention: Tocilizumab dose, route of administration, and duration.
Comparator: Any other treatment or care that was given to the patients
Outcomes: All outcomes either efficacy or safety-related were extracted extensively including numbers or data presented with graphs that were transformed with a specific software; Get Data Graph Digitizer 2.26

Assessment of risk of bias in included studies:

The methodological quality of selected cohort studies were assessed based on the basis of ” risk of bias in non-randomized studies of interventions” (ROBINS-I) (Sterne et al., 2016); a tool provides a more comprehensive framework for identifying potential sources of bias (Losilla, Oliveras, Marin-Garcia, & Vives, 2018). The following points were scored as low, moderate, serious, critical or no information (where ’low’ indicated that the study was less risk to bias and thus best quality), and were reported in a ”Risk of bias figure”: bias due to confounding, bias in selection of participants into the study, bias in classification of interventions, bias due to deviations from intended interventions, bias due to missing data, bias in measurement of outcomes, and bias in selection of the reported result.
A consensus criteria in the present meta-analysis for bias judgment included that a study was judged at low risk of bias if all key domains were judged at low risk of bias, a study was judged at high risk of bias if two or more key domains were judged at high risk of bias, otherwise the study was judged at unclear risk of bias.
Studies of low quality were not excluded, instead they were involved in data synthesis after performing sensitivity analysis.

Synthesis of the quantitative results

Measures of treatment effect: risk ratios (RR) with 95%confidence intervals (CI) was used to analyze dichotomous data. None of our included studies reported continuous data.
Data synthesis: The overall treatment effect was estimated by the pooled RR with 95% CI by RevMan version 5.4 using a fixed-effect model (Mantel-Haenszel). A random effects model was used in cases of significant heterogeneity.
Assessment of heterogeneity: Chi-square test of heterogeneity and the I2 statistic of heterogeneity were used to assess effects heterogeneity. For Chi-square test, the data study findings were considered to be heterogeneous if P -value was ≤ 0.05. When a significant heterogeneity occurred, the differences were explained as they related to types of participants and study design.
Sensitivity analysis: Sensitivity analysis was conducted for only those cohort studies assessed as having a low overall risk of bias based on our consensus criteria in key domains.
Publication bias: The publication bias was assessed by using funnel plots when there were more than 10 studies reporting the same effect measure of an outcome.