Table 2.

Performance of the prognostic models of the algorithms and the surgeon in the test set (n = 400)

Model/ Surgeon	Balanced Accuracy	Accuracy	Sensitivity	Specificity	AUCa	Brier Score
LRb	0.580	0.90 (CIc 0.86–0.92)	0.235	0.924	0.624	0.041
RFd	0.591	0.81 (CI 0.77–0.85)	0.353	0.830	0.671	0.043
XGBe	0.608	0.84 (CI 0.80–0.88)	0.353	0.864	0.646	0.042
KNNf	0.616	0.70 (CI 0.65–0.74)	0.529	0.702	0.623	0.042
Senior Surgeon	0.527	0.74 (CI 0.69–0.78)	0.294	0.760	-	-

^aAUC: area under the receiver operating characteristic curve; ^bLR: logistic regression; ^cCI: Confidence interval; ^dRF: random forest; ^eXGB: eXtreme gradient boost; ^fKNN: K-nearest neighbors.

95% confidence intervals are provided for accuracy. No confidence interval is reported for balanced accuracy, as it is a composite metric whose uncertainty depends on both sensitivity and specificity, and no standard analytical method is established for its estimation.