| Bias | Author's judgement | Support for judgement |
| Confounding |
Serious |
Comment: Some confounders of interest (hypertension, diabetes, immunosuppression, obesity) were not controlled for in the propensity score model (see eTable 3). Authors do not adjust for time-varying confounding. Based on Table 1, some patients received tocilizumab as long as >48 hours after intubation, so they may have spent a substantial amount of time in the control group (i.e., not receiving tocilizumab). |
| Selection of participants into the study |
Serious |
Comment: It is unclear when follow-up began. The baseline values reported seem to have been collected within 12 hours of intubation. However, the decision to administrate tocilizumab could have been take as long as >48 hours after intubation (see Table 1). The duration of intubation could have been taken into account by the treating physician in deciding to administer tocilizumab. Also, some patients were treated with tocilizumab prior to intubation. Baseline data for these participants were taken from the last measurement before tocilizumab administration. The start of the intervention and the start of the follow-up do not seem to coincide for all patients. Table 1 shows that some patients were treated with tocilizumab in <24 hours from the time of intubation. |
| Intervention classification |
Low |
Quote: "This analysis focuses on comparative outcomes of mechanically ventilated patients who received tocilizumab and those who did not." Comment: data were collected from clinic records. Treatment groups were classified using clearly defined criteria (receipt vs non receipt of a single tocilizumab dose). |
| Deviation from intended intervention |
Moderate |
Comment: Use of concomittant medication during hospitalization is presented in Table 1. Corticosteroids, hydroxychloroquine and anticoagulants were slightly more frequently administered to patients receiving tocilizumab. Risk assessed to be low for the outcomes: time to death. |
| Missing data |
Low |
Comment: No participants were excluded due to missing data. Multiple imputation was used for missing data in confounders used in the development of the propensity score model. The analysis is also presented in the complete-case dataset. Results for time to death HR are similar in the two analyses. Risk assessed to be low for the outcomes: mortality, time to death, clinical improvement |
| Measurements of outcomes |
Low |
Comment: Death is an observer-reported outcome but requires no clinical judgement however Clinical improvement is based on a 6 score ordinal scale and requires clinical decision making that could be influenced by knowledge of intervention status however this is unlikely in the context of a pandemic. Risk assessed to be "low" for the outcome: mortality/time to death. Risk assessed to be "moderate" for the outcomes: clinical improvement |
| Selection of the reported results |
Moderate |
Comment: The outcomes and analyses are clearly defined in the Methods section. Planned subgroup/sensitivity analysis results are reported. There is no a-priori registered protocol or statistical analysis plan available. Risk assessed to be moderate for the outcomes: mortality, time to death, clinical improvement. |