Comparison of models on the test dataset from a different institution using specific input subsets (subsets A and B) selected for clinical/cost reasons.

The top section indicates the benchmark performance when both training and test-sets include all inputs. In the central section performance on test-data with only subset A of inputs is shown. Models are trained with all inputs and with only subset A of inputs. The lower section of the table repeats these experiments for subset B. Bold font indicates the highest AUC value in the table section. * indicates AUCs that are significantly lower than the highest in that section (p < 0.05).