Dear Editor,

Thanks to Dr. Luo and colleagues for their letter and comments [1]. In the protocol for our systematic review, we decided a priori that all analysis would be conducted using a random effects model, as this analytic plan better accounts for between-study differences [2]. In situations where statistical heterogeneity is minimal, the output from random effects models closely matches that of fixed effects models. However, in response to this letter, we did perform fixed effects analysis for this outcome and the pooled point estimate and 95% confidence intervals do not change.

We focused our TSA analysis on the outcomes most likely to be considered for future RCTs. As escalation of therapy is a composite, including outcomes of variable clinical and patient importance, we intentionally highlighted ‘need for IMV’ and ‘mortality’. For the TSA analysis, we had initially used the relative risk reductions from our pooled analysis to inform the sample size calculation. After peer review of the manuscript, we were asked to choose a threshold consistent with what we believed would be a clinically important difference. As such, we selected a 15% relative risk reduction for this dichotomous outcome.