Statistical analysis
Study population was described according to pH groups. Counts and proportions were reported for categorial variables, and means +/- standard deviations (SD) for continuous variables. Normality for continuous variables was explored both in a graphical way with histogram. Groups were compared using Chi-squared test for qualitative variables (unless expected counts were less than 10, in which case Fisher’s exact test was used) and ANOVA for normal continuous variable (homogeneity of variance was analyzed using Bartlett’s test) or non-parametric Kruskal Wallis test for non-normal variables.
Correlation between lactates rate distribution and blood pH rate was explored with Spearman’s Rho correlation test.
The date of initiation of VA-ECMO cannulation was chosen as the date of origin for survival outcomes. Survival data were described with Kaplan-Meier survival curves. Median survival time of the study population was reported with a 95% confidence interval (95%CI). Log-rank test was used to assess comparison between pH groups. Survival rates at 30 days and one year were estimated for each group. Cox proportional hazard models were used for the assessment of the association between pH groups and overall survival and a backward stepwise selection was performed for variable selection. The final model selected covariates with significance level 0.10. Two covariates considered as potential confounders were forced into the model: sex and ECMO indication. Interaction with initial diagnosis was tested. We checked the collinearity between the covariates in the multivariable analysis using the variance inflation factor (VIF) for diagnostic for multicollinearity. A decision curve analysis was performed to graphically analyze the relationship between pH and the futility of the procedure.
Statistical analyses were performed with R version 4.0.1.