Risk of bias for analysis 1.1 All‐cause mortality up to 30 days.
| Study | Bias | |||||||||||
| Randomisation process | Deviations from intended interventions | Missing outcome data | Measurement of the outcome | Selection of the reported results | Overall | |||||||
| Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | |
| Angus 2020 | Low risk of bias | Participants were randomised to each locally available group using balanced assignment. Participants were randomly assigned via a computer software program to each locally available group using proportional assignment (e.g., 1:1 if 2 groups available and 1:1:1 if 3 groups available) Protocol: The RAR will occur centrally as part of the computerised randomisation process. Sites will receive the allocation status and will not be informed of the randomisation proportions. Each region will maintain its own computer‐based randomisation program that is accessed by sites in that region but the RAR proportions will be determined by a SAC and provided monthly to the administrator of each region's randomisation program who will update the RAR proportions Baseline characteristics were similar across groups. |
Some concerns | 15% of the participants in the no hydrocortisone group received hydrocortisone | Low risk of bias | Data were available for this outcome. 238 participants were randomised and 238 were analysed. | Low risk of bias | Mortality is an observer‐reported outcome not involving judgement. | Low risk of bias | Protocol and SAP are available. | Some concerns | Overall judged some concerns due to deviations from the intended intervention |
| Angus 2020 | Low risk of bias | Participants were randomised to each locally available group using balanced assignment. Participants were randomly assigned via a computer software program to each locally available group using proportional assignment (e.g., 1:1 if 2 groups available and 1:1:1 if 3 groups available) Protocol: The RAR will occur centrally as part of the computerised randomisation process. Sites will receive the allocation status and will not be informed of the randomisation proportions. Each region will maintain its own computer‐based randomisation program that is accessed by sites in that region but the RAR proportions will be determined by a SAC and provided monthly to the administrator of each region's randomisation program who will update the RAR proportions Baseline characteristics were similar across groups. |
Some concerns | 15% of the participants in the no hydrocortisone group received hydrocortisone | Low risk of bias | Data were available for this outcome. 238 participants were randomised and 238 were analysed. | Low risk of bias | Mortality is an observer‐reported outcome not involving judgement. | Low risk of bias | Protocol and SAP are available. | Some concerns | Overall judged some concerns due to deviations from the intended intervention |
| Corral‐Gudino 2021 | Some concerns | We judged the domain some concerns due to missing information about the allocation concealment. | Low risk of bias | Open‐label design (participants and clinicians were aware of the assigned treatment). ITT data presented. | Low risk of bias | No substantial data missing. | Low risk of bias | Both groups were measured at the same time and the measurements were similar between groups. | Some concerns | Outcome was registered in Eudra‐CT as in‐hospital mortality but reported as 28‐day‐mortality in the supplemental material. | Some concerns | Overall judged some concerns due to selection of reported results and missing information about the allocation concealment. |
| Dequin 2020 | Low risk of bias | Randomisation was centralised and performed electronically. Allocation sequences were generated in a 1:1 ratio by a computer‐generated random number using a blocking schema. There were no baseline differences between the groups. | Low risk of bias | Protocol: Patients, investigators and care providers will be blinded for the patient‐arm. ITT was used. | Low risk of bias | Data were available for this outcome. | Low risk of bias | The measurements were similar between groups. | Some concerns | The protocol and statistical analysis plan were available. 21‐day mortality was a post hoc outcome. | Some concerns | Overall judged some concerns due to selection of reported results. |
| Horby 2021 | Low risk of bias | Randomisation was performed with the use of a web‐based system with concealment of the trial‐group assignment. No baseline differences between the groups. | Some concerns | 8% of the participants in the control group received dexamethasone. | Low risk of bias | 6425 participants were randomised and 6425 were analysed. | Low risk of bias | Both groups were measured at the same time. The measurements were similar between groups. | Low risk of bias | Protocol and statistical plan available. | Some concerns | Overall judged some concerns due to deviations from the intended intervention in the control group. |
| Jamaati 2021 | Some concerns | The selected patients were allocated to either the dexamethasone group or the control group by block randomisation. Ten blocks were generated by the Online Randomiser website. Pulmonary disease was highly significantly more prevalent in the control group. Additionally, baseline level of respiratory support was not reported so that we can only assume the most likely circumstance, i.e. that no support beyond oxygen insufflation had been given until randomisation. |
Low risk of bias | No deviations mentioned. ITT data reported. | Low risk of bias | No missing data. | Low risk of bias | The measurements were similar between groups. | Some concerns | Trial registered after recruitment stop. No analysis plan published beyond registry entry. | Some concerns | Potential bias through probably problematic block randomisation with significant baseline imbalance for pulmonary disease, very delayed registration, and potentially not pre‐specified stop for futility raise some concerns. The issues were not discussed. Potential mild bias towards experimental. |
| Jeronimo 2020 | Low risk of bias | An independent statistician prepared an electronically generated randomisation list with 14 blocks of 30 participants per block, generated via R software version 3.6.1 (blockrand package). The list was accessible only to non‐blinded pharmacists in the study. Participants were randomised by the study pharmacist to their designated treatment regimen at the time of inclusion and were subsequently identified throughout the study only by their allocated study number. There were no major differences in baseline characteristics between the intervention and placebo groups |
Some concerns | Intervention group: 14 excluded before starting treatment, 1 excluded after starting treatment Control group: 5 excluded before starting treatment, 3 excluded after starting treatment |
Low risk of bias | 416 participants randomised and 416 participants analysed. | Low risk of bias | The measurements were similar between groups. | Low risk of bias | Protocol and statistical plan available. Data analysed and presented according to a pre‐specified plan. | Some concerns | Overall judged some concerns due to protocol deviations. |
| Munch 2021a | Low risk of bias | No issues with randomisation process. | Low risk of bias | Withdrawals with consecutive open‐label steroids may have affected the outcome, but withdrawals and non‐compliance was rather balanced after all. | Low risk of bias | No missing data. | Low risk of bias | No issues with measurement. | Low risk of bias | No relevant issues with the reported result. | Low risk of bias | Withdrawals because of the experimental context lead to some concerns (but it is less than 10% of all patients). |
| Tang 2021 | Low risk of bias | Randomisation was stratified by the statistician of the leading site, who produced computer‐generated block randomisation lists with a block size of 4 patients. | Low risk of bias | Single‐blind design (participants), ITT data presented. No deviations reported. | Low risk of bias | No missing data. | Low risk of bias | The data collection and end point judgement were blinded, and the statisticians were also blinded during the statistical analysis. The measurements were similar between groups | Low risk of bias | 30‐day mortality as pre‐specified in the trial registration was not explicitly reported, but could reliably derive from a synopsis of full text, figures, and supplemental material. | Low risk of bias | All domains were rated low risk of bias. |
| Tomazini 2020 | Low risk of bias | Randomisation was performed through an online web‐based system using computer‐generated random numbers and blocks of 2 and 4, unknown to the investigators, and was stratified by centre. The group treatment was disclosed to the investigator only after all information regarding patient enrolment was recorded in the online system. Baseline characteristics were well‐balanced between groups. |
Some concerns | 25 deviations from protocol in the intervention arm (16.55%); 1 patient received a corticosteroid other than dexamethasone. In the control arm, 52 patients received corticosteroids, of which 14 were protocol deviations (9.4%). | Low risk of bias | No missing data. | Low risk of bias | Individuals who assessed the outcomes were not blinded for the assigned treatment. Both groups were measured at the same time. The measurements were similar between groups. | Low risk of bias | The protocol and statistical analysis plan were available. Outcomes reported as prespecified. | Some concerns | Overall judged some concerns because of protocol deviations |