Fig. 1

Variable importance for the prediction of the extent of BME in the SIJ (Berlin score) from a boosting model with implicit variable selection. This implies that all main effects and interaction terms with no filled bars have been removed in the optimal model. In mboost (23), categorical base learners are transformed to dummy-coded variables. Therefore, single levels of respective categorical variables may enter the model. Variable importance is measured as the summarized contribution of each base learner (linear effect of each base learner) to the final model. The final model is determined by 10-fold cross validation. The percentage (x-axis) of each variable quantifies the contribution of each variable to the optimal model determined by 10-fold cross-validation