Factors Associated with NBMA Reversal

Variables were described by frequencies (percentage) or mean (SD) and median (IQR), as appropriate. Univariate associations were tested by Chi-Square test, Chi-Square test for trend in proportions, and Mann-Whitney U test, as appropriate, considering a threshold of 0.05 statistical significance. We built a multivariable logistic regression model to describe factors associated with NBMA reversal, and secondary analysis to detect factors associated with the choice of sugammadex in the subpopulation of patients paralyzed with rocuronium. Before inclusion, missing data were analyzed and plotted for each variable to improve model performance. We used random-forest-based imputation of missing data before including variables in the logistic regression model, weighted by the level of missingness per feature (Figure S1). After imputation, the logistic regression model was built including clinically meaningful variables such as year, surgical specialty, and day-hospital surgery by default, and performing the selection of the other variables through a forward-backward stepwise regression approach using Aikake information criteria (AIC) for variable selection. Repeated k-fold cross validation was employed for internal model validation using 10 folds and 10 random repeats. Model discriminative performance was assessed through ROC analysis after cross-validation. Pooled calibration and precision-recall gain curve, which standardized precision against baseline chances expectations, were calculated. Sensitivity analysis on non-missing data only is reported in Figure S2.
All statistical analyses were performed using R software, version 4.1.1.