Sanne de Bruin

28 Chapter 2 the intensive care medicine. This survey was initiated by the Cardiovascular Dynamics Section and endorsed by the European Society of Intensive Care Medicine (ESICM). In addition, multiple national intensive care societies distributed the survey to itsmembers by newsletters and/or promoted it on their website (see Additional file 1 for contribu- tors). Study design An online platformwas used to set up the questionnaire (SurveyMonkey; Portland, OR, USA). After designing, the survey was tested by an international panel of intensivists to optimize the validity and accuracy of the questionnaire. The survey included 40 ques- tions, divided into four sections, respondent demographics, transfusion practice regard- ing red blood cells (15-17 questions), platelets (5 questions) and plasma transfusions (8 questions, see Additional file 1 for static version). Multiple clinically relevant subpopula- tions (also non-bleeding) were addressed in each section. For red cell transfusion, first the preferred haemoglobin (Hb) threshold for the general ICU population was asked, followed by the preferred Hb level for each subpopulations. For platelet transfusion, a distinction was made between transfusion prophylactically and prior different invasive procedures. For plasma transfusions, a distinction was made between prophylactic transfusion without a planned procedure and prior to an invasive procedure in general. Statistical analysis Only completed surveys were analysed. Since some questions were not applicable for all doctors, respondents were allowed to leave specific questions about subgroups/ specific interventions open. This missing data was not imputed. Descriptive statistics were used to characterize the respondent demographics. Normal distributed and non-normal distributed data was reported as mean (standard deviation) or as median (first quartile-third quartile) respectively. Categorical data was presented as percentage. Participants were able to fill in Hb thresholds in g/dL, g/L, or mmol/L, and all answers were converted to g/dL for analysis. Transfusion thresholds were not normally distributed, therefore the Kruskall-Wallis test was used to test whether the transfusion thresholds differed significantly subpopula- tions or between interventions. As post hoc test, the Dunn test with Bonferroni correc- tion was used. In addition, transfusion thresholds were analysed using the Wilcoxon sum rank test or Kruskall-Wallis test to test the dependence of two grouping variables or more than 2 grouping variables respectively. Chi-square test with Yates’s correction for continuity was used for categorical variables. For comparing different world regions,