Marjon Borgert

180 Chapter 8 Analysis Continuous variables that were normally distributed were expressed as means with standard deviations (SD) and not normally distributed variables as medians and inter- quartile ranges (IQR). To test independent groups of not normally distributed continuous variables, the Kruskal Wallis test was used or Mann-Whitney U test when appropriate. Categorical variables were expressed as percentages, numerators and denominators and were compared with the Chi-square test or Fisher’s exact test. Analysis of variance (ANOVA or unpaired t-test) was used to test for di erences in means across the study periods. In some cases the denominator does not correspond fully with the overall number of patients. This di erence is due to missing values on some variables. The goal of the logistic regression analysis was to quantify the net e ect of the implementation of the transfusion bundle on the likelihood of appropriate transfusions, controlling for other variables. Exploration of interaction (e ect modi cation) and confounding was considered methodologically relevant. We rst focused on the crude (uncorrected) e ect of the implementation of the bundle (independent variable) on appropriate transfusions (dependent variable). Then, statistical and clinically relevant covariates were added as an interaction term. If the interaction term appeared to be signi cant ( P < 0.05), this would indicate that the relation between the implementation and appropriate transfusions could be di erent for various levels of the covariate. This indicates the need for separate models for the levels of the covariate. As a signi cant interaction was not found, the model was examined for confounding. Confounding was de ned as ≥ 10% change in the coe cient of the central determinant implementation as a consequence of adding a covariate. Statistical signi cance was considered to be at P < 0.05. When appropriate statistical uncertainty was expressed by the 95% con dence levels. Analyses were performed using R (version: 3.1.3; R Foundation for Statistical Computing, Vienna, Austria). Ethics The study was approved by the Medical Ethics Committee of the Academic Medical Center of Amsterdam, the Netherlands and the need for informed consent was waived.