Table 5.

Performance metrics, including confidence intervals for area under the receiver operater characteristics curve (ROC-AUC), for all long-term hospitalization models 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.

Model	Training data set	ROC-AUC	Lower CI ROC-AUC	Upper CI ROC-AUC	Precision	Recall	F1 score
Random forest	DTD	0.885	0.849	0.921	0.817	0.958	0.882
	TQIPD	0.860	0.820	0.900	0.840	0.930	0.883
	Mixed	0.884	0.848	0.920	0.858	0.898	0.877
AdaBoost	DTD	0.890	0.855	0.925	0.863	0.935	0.897
	TQIPD	0.885	0.849	0.921	0.834	0.935	0.822
	Mixed	0.884	0.848	0.920	0.858	0.898	0.877
XGBoost	DTD	0.865	0.825	0.904	0.820	0.953	0.882
	TQIPD	0.866	0.826	0.905	0.801	0.935	0.863
	Mixed	0.865	0.826	0.905	0.855	0.930	0.891
Explainable boosting machine	DTD	0.883	0.847	0.919	0.885	0.898	0.891
	TQIPD	0.880	0.843	0.917	0.869	0.893	0.881
	Mixed	0.877	0.839	0.914	0.844	0.907	0.874
Neural network	DTD	0.857	0.817	0.898	0.853	0.916	0.883
	TQIPD	0.888	0.852	0.923	0.810	0.953	0.876
	Mixed	0.875	0.837	0.913	0.831	0.935	0.88
Neural network (Transfer learning)	TQIPD/DTD	0.879	0.842	0.916	0.813	0.949	0.876
	Mixed/DTD	0.874	0.836	0.912	0.830	0.930	0.877