Does the Choice of Perioperative Inotropes Has Influence on the Morbidity and Mortality of Infants Operated for Complex Heart Defects?

One propensity score will be performed on two groups: Dobu and Milri. After 15 multiple imputations of the datasets using the mice R package, the CBPS R package will be used to perform the propensity score, estimating an Average Treatment Effect (ATE), using covariate balancing and requesting an exact match, which has been showed to be superior to traditional logistic regression approaches and boosted classification and regression trees . An absolute standardized difference less than 10-15% will be considered to support the assumption of balance between the groups because it is not affected by the sample size, unlike P-values, and it may be used to compare the relative balance of variables measured in different units. The mean and standard deviation obtained after matching for continuous variables, and the percentage for categorical variables will be presented. After the propensity score, the investigators will use the survey R package to perform logistic regressions for binary outcome variables and linear regressions for continuous outcomes, which will include the treatment group effect, the weight resulting from the matching and variables present in the propensity score in order to obtain a doubly-robust estimator which will correct the last remaining possible imbalance between the covariates and produces an unbiased treatment effect. The survey R package includes the Huber-White corrected standard errors, which maintains the standard errors unbiased even under heterogeneity of the residuals. Last, the advantage of a doubly-robust estimator is that it needs only one of the two models (propensity score and logistic regression after the propensity score) to be correctly specified. The so-called 'within approach', that is averaging the treatment effects on the 15 logistic and linear regressions has been showed to produce less biased estimates than the 'Across approach' - that is, performing linear or logistic regressions on averaged imputed matched datasets - especially when missing data is related to the treatment group. The R software (R Core Team, 2019), version 3.6.2. will be used to produce the results.