Table 3.

Performance metrics for random forest model. Random forest model performance metrics for predicting in-hospital mortality using different sets of features. Confidence intervals derived by bootstrapping the predictions using 1000 samples shown in parenthesis. Accuracy=(TP+TN)/(TP+TN+FP+FN). Precision=TP/(TP+FP). Recall=TP/(TP+FN). Specificity=TN/(TN+FP). F1 score=2/([1/Recall]+[1/Precision]). FN, false negatives; FP, false positives; Preop, preoperative; TN, true negatives; TP, true positives

Model	Accuracy	F1 score	Precision	Recall	Specificity
POSPOM	0.861 (0.851–0.869)	0.047 (0.021–0.078)	0.026 (0.012–0.045)	0.201 (0.097–0.318)	0.872 (0.864–0.881)
Charlson comorbidity	0.895 (0.885–0.904)	0.112 (0.064–0.165)	0.065 (0.037–0.098)	0.390 (0.240–0.538)	0.904 (0.895–0.913)
ASA status	0.897 (0.889–0.906)	0.160 (0.110–0.222)	0.093 (0.061–0.133)	0.587 (0.472–0.709)	0.903 (0.895–0.911)
Preop features	0.985 (0.981–0.988)	0.275 (0.115–0.446)	0.610 (0.333–0.814)	0.179 (0.069–0.315)	0.998 (0.997–0.999)
Preop features+ASA status	0.984 (0.980–0.988)	0.284 (0.119–0.464)	0.590 (0.333–0.810)	0.189 (0.074–0.329)	0.998 (0.997–0.999)
Preop+surrogate-ASA	0.984 (0.980–0.988)	0.280 (0.125–0.452)	0.541 (0.294–0.750)	0.191 (0.078–0.331)	0.997 (0.996–0.998)
Preop+ASA status, w/o lab times	0.982 (0.977–0.986)	0.302 (0.172–0.449)	0.420 (0.245–0.615)	0.239 (0.127–0.379)	0.994 (0.992–0.997)
Preop+surrogate-ASA status, w/o lab times	0.980 (0.976–0.985)	0.258 (0.127–0.412)	0.358 (0.180–0.551)	0.204 (0.094–0.342)	0.994 (0.992–0.996)