Data analysis
As the study was retrospective (the sample size was fixed a priori ) and exploratory, the 10 cases per covariate approach to sample size calculation was used to establish, prospectively, that our sample size of 300 was likely to be sufficient.16 Descriptive statistics included medians and inter-quartile ranges for continuous variables (including time-to-event variables) and counts and percentages for categorical variables. Initial summary statistics of the whole cohort and comparison tests between two groups, those successfully extubated by 14 days after starting dexamethasone and those not successfully extubated in that time, were done using GraphPad Prism 8 (GraphPad Software LLC, San Diego, CA, USA).
As the primary outcome in the present study is a binary variable, binary logistic regression was employed. Univariate and multivariable binary logistic regression models were fit to the data to produce unadjusted and adjusted odds ratios, respectively, along with their corresponding 95% confidence intervals. The multivariable model included all potentially important predictors (independent risk factors and confounders) including clinically important variables and those with a p<0.2 in the univariate analysis. All modelling was conducted using the R statistical package (v4.0.2; The R Project for Statistical Computing, www.r-project.org/) and a p-value less than 0.05 was used to represent statistical significance throughout all inferential analyses.