Statistical analysis
Descriptive statistical methods were applied to depict the study population at baseline. Continuous, variables are presented as the median and interquartile range (25th and 75th percentiles). Categorical variables are presented as counts and percentages Differences between groups were compared with Student’s t-test for normally distributed variables and the Mann-Whitney U test for non-normally distributed variables. Categorical variables are summarized as the number and percentage of subjects in each category, and differences were compared with the Pearson chi-square test.
Missing data were not superior to 10%. Pattern of missing values was investigated and multiple imputation process was used to manage them. Briefly, we used fully specified chained equations in the R package.19. Five imputed datasets were created and combined using between/within variance techniques to appropriately propagate uncertainty about the missing data4.
The propensity score was obtained using logistic regression. The variables included in the propensity model were plotted in the supplementary Figure 1. Inverse probability of treatment weighting (IPTW) was obtained for the average treatment effect (ATE). The balance was tested with the standardized mean difference (SMD), which was considered optimal below 0.10. The results were then weighted for IPTW (Figure 1 supplementary).
Early adverse events were analyzed as proportions of the number of patients and described as rates (%). Weighted logistic regression was used as multivariable analysis for early binary outcome. In multivariable models we decided to include, besides different treatment (DHCA+DR vs MHCA+ACP), those operative variables that were not included in the propensity score model (type of arterial cannulation, type of venous cannulation, type of myocardial protection, cardiopulmonary bypass (CPB) duration and nadir temperature reached). ROC curve analysis has been applied to evaluate the correlation between endpoints and variables. All reported p-values were considered statistically significant if below 0.05. R-Studio version 1.1.463 (2009-2018) and SPSS were used for all statistical analyses.