Table 1.

Model performance metrics for balanced test set.

Threshold-independent metrics, (95% CI)			Threshold-dependent metrics, (95% CI)
AUC	PR_AUC	${\overline{\mathbf{MSE}}}_{\mathbf{pat}}$	Acc	Sens	Spec	F1	FPR	NPV	PPV
0.893 (0.862–0.924)	0.903 (0.873–0.933)	0.124 (0.090–0.159)	0.825 (0.786–0.863)	0.853 (0.802–0.904)	0.798 (0.741–0.855)	0.826 (0.776–0.876)	0.202 (0.145–0.259)	0.851 (0.799–0.903)	0.801 (0.745–0.857)

n = 350 admissions/patients.

AUC area under curve, PR_AUC precision-recall AUC, ${\overline{\mathrm{MSE}}}_{\mathrm{pat}}$ mean of the brier score of each patient, Acc accuracy, Sens sensitivity, Spec specificity, F1 F₁-score, FPR false-positive rate, NPV negative predictive value, PPV positive predictive value, CI confidence interval. The threshold for positive/negative class prediction was set to 0.41, leading to a sensitivity of 0.850 on cross-validation folds in the training set.