Table 2.

Performance of predictive models.

Target	ROC AUC	Average precision	Precision	Recall	Accuracy	F1 Score	Matthews correlation coefficient	Brier score loss	RMSE
Readmitted within 30 days	0.758 [0.755 to 0.762]	0.383 [0.377 to 0.388]	0.632 [0.620 to 0.647]	0.102 [0.098 to 0.106]	0.861 [0.860 to 0.861]	0.176 [0.169 to 0.182]	0.214 [0.208 to 0.220]	0.108 [0.108 to 0.109]	—
Readmitted within 7 days	0.701 [0.696 to 0.707]	0.127 [0.122 to 0.133]	0.586 [0.455 to 0.722]	0.003 [0.002 to 0.004]	0.949 [0.949 to 0.949]	0.006 [0.004 to 0.008]	0.040 [0.030 to 0.051]	0.047 [0.047 to 0.047]	—
Readmitted within 5 days	0.691 [0.684 to 0.698]	0.091 [0.086 to 0.095]	0.456 [0.000 to 1.000]	0.000 [0.000 to 0.001]	0.963 [0.963 to 0.963]	0.001 [0.000 to 0.002]	0.013 [−0.001 to 0.029]	0.035 [0.035 to 0.035]	—
Readmitted within 3 days	0.681 [0.674 to 0.689]	0.057 [0.053 to 0.062]	0.000 [0.000 to 0.000]	0.000 [0.000 to 0.000]	0.978 [0.978 to 0.978]	0.000 [0.000 to 0.000]	0.000 [0.000 to 0.000]	0.021 [0.021 to 0.021]	—
Days to readmissiona	—	—	—	—	—	—	—	—	8.98
Death within 48–72 ha	0.91	—	—	—	—	—	—	0.001	—
Hospital stay >7 days	0.830 [0.827 to 0.833]	0.567 [0.561 to 0.572]	0.653 [0.646 to 0.659]	0.331 [0.325 to 0.337]	0.827 [0.825 to 0.828]	0.439 [0.434 to 0.445]	0.378 [0.371 to 0.384]	0.122 [0.121 to 0.123]	—
Hospital stay >5 days	0.829 [0.827 to 0.832]	0.705 [0.701 to 0.710]	0.690 [0.685 to 0.695]	0.546 [0.541 to 0.552]	0.767 [0.765 to 0.770]	0.609 [0.605 to 0.614]	0.453 [0.447 to 0.459]	0.155 [0.154 to 0.157]	—
Hospital stay >3 days	0.824 [0.822 to 0.827]	0.861 [0.859 to 0.864]	0.760 [0.758 to 0.762]	0.842 [0.839 to 0.845]	0.752 [0.749 to 0.754]	0.799 [0.797 to 0.801]	0.480 [0.475 to 0.485]	0.166 [0.165 to 0.167]	—
Length of stay (days)a	—	—	—	—	—	—	—	—	3.94

^aPerformance on these predictive tasks was poor to the extent that rigorous cross-validation was not performed.