Skip to main content
. 2023 Nov 2;5:1249258. doi: 10.3389/fdgth.2023.1249258

Table 4.

Performance metrics, including confidence intervals for area under the receiver operater characteristics curve (ROC-AUC), for all mortality models using trauma score and injury severity score (TRISS) as baseline, comparing models trained on either danish trauma dataset (DTD), the American trauma quality Improvement program dataset (TQIPD) or a mixed traning dataset consisting of a random forest-selected subset of TQIPD and DTD training data (mixed). Neural network with transfer learning was retrained on DTD. When a resampling method was applied on training data either Synthetic Minority Over-Sampling Technique (SMOTE) or SMOTE with down-sampling removing Tomek links was used (SMOTETomek). When class weighting was applied on a neural network loss function, weights are denoted first by the negative class and secondly for the postive class.

Model Training data source Resampling method Weights ROC-AUC Lower CI ROC-AUC Upper CI ROC-AUC F2 score Precision Recall
TRISS DTD None None 0.853 0.751 0.955 0.581 0.251 0.867
RF DTD None None 0.971 0.922 1.0 0.746 0.654 0.773
SMOTE None 0.971 0.922 1.0 0.752 0.680 0.773
SMOTETomek None 0.966 0.913 1.0 0.783 0.667 0.818
Mixed None None 0.986 0.951 1.0 0.756 0.581 0.818
SMOTE None 0.973 0.926 1.0 0.721 0.696 0.727
SMOTETomek None 0.977 0.933 1.0 0.763 0.600 0.818
TQIPD None None 0.985 0.949 1.0 0.861 0.618 0.955
SMOTE None 0.967 0.915 1.0 0.734 0.382 0.955
SMOTETomek None 0.964 0.909 1.0 0.732 0.514 0.818
AdaBoost DTD None None 0.969 0.918 1.0 0.642 0.667 0.636
SMOTE None 0.968 0.916 1.0 0.746 0.654 0.773
SMOTETomek None 0.963 0.908 1.0 0.826 0.704 0.864
Mixed None None 0.982 0.943 1.0 0.734 0.762 0.727
SMOTE None 0.977 0.933 1.0 0.746 0.654 0.773
SMOTETomek None 0.977 0.932 1.0 0.738 0.529 0.818
TQIPD None None 0.983 0.944 1.0 0.766 0.739 0.773
SMOTE None 0.970 0.921 1.0 0.692 0.429 0.818
SMOTETomek None 0.971 0.922 1.0 0.736 0.463 0.864
XGB DTD None None 0.937 0.866 1.0 0.642 0.667 0.636
SMOTE None 0.960 0.903 1.0 0.697 0.500 0.773
SMOTETomek None 0.952 0.889 1.0 0.759 0.708 0.773
Mixed None None 0.979 0.937 1.0 0.776 0.643 0.818
SMOTE None 0.985 0.950 1.0 0.812 0.655 0.864
SMOTETomek None 0.981 0.940 1.0 0.750 0.562 0.818
TQIPD None None 0.975 0.929 1.0 0.769 0.621 0.818
SMOTE None 0.978 0.934 1.0 0.784 0.457 0.955
SMOTETomek None 0.975 0.928 1.0 0.748 0.487 0.864
EBM DTD None None 0.951 0.887 1.0 0.733 0.607 0.773
SMOTE None 0.933 0.859 1.0 0.690 0.571 0.727
SMOTETomek None 0.952 0.890 1.0 0.750 0.562 0.818
Mixed None None 0.980 0.938 1.0 0.739 0.630 0.773
SMOTE None 0.984 0.946 1.0 0.800 0.541 0.909
SMOTETomek None 0.984 0.946 1.0 0.756 0.581 0.818
TQIPD None None 0.982 0.942 1.0 0.812 0.655 0.864
SMOTE None 0.964 0.910 1.0 0.664 0.425 0.773
SMOTETomek None 0.960 0.902 1.0 0.714 0.422 0.864
Neural network DTD None 0.5439, 6.1970 0.974 0.928 1.0 0.745 0.396 0.955
SMOTE None 0.906 0.821 0.991 0.732 0.514 0.818
SMOTETomek None 0.867 0.769 0.964 0.678 0.533 0.727
Mixed None 0.5503, 5.4705 0.986 0.951 1.0 0.821 0.478 1.0
SMOTE None 0.930 0.855 1.0 0.709 0.462 0.818
SMOTETomek None 0.930 0.855 1.0 0.714 0.422 0.864
TQIPD None None 0.974 0.928 1.0 0.775 0.488 0.909
SMOTE None 0.979 0.936 1.0 0.753 0.379 1.0
SMOTETomek None 0.967 0.914 1.0 0.697 0.500 0.773
Neural network (Transfer learning) Mixed None 0.5503, 5.4705 0.988 0.955 1.0 0.866 0.564 1.000
SMOTE None 0.963 0.908 1.0 0.720 0.432 0.864
SMOTETomek None 0.972 0.923 1.0 0.755 0.412 0.955
TQIPD None None 0.981 0.940 1.0 0.814 0.512 0.955
SMOTE None 0.976 0.931 1.0 0.787 0.513 0.909
SMOTETomek None 0.978 0.935 1.0 0.789 0.467 0.955