Machine learning-based prediction of postoperative pancreatic fistula after laparoscopic pancreaticoduodenectomy

BMC Surgery

Table 5 Predicted Results of Five Models for CR-POPF in Training Data

Model	Methods	Threshold	AUC	Spe	Sen	NPV	PPV
1	RF	0.600	1.000	1.000	1.000	1.000	1.000
	SVM-L	0.838	0.699	0.534	0.842	0.948	0.250
	SVM-P	0.851	0.947	0.922	0.947	0.990	0.692
	Logistic	0.210	0.810	0.816	0.737	0.944	0.424
2	RF	0.529	1.000	1.000	1.000	1.000	1.000
	SVM-L	0.839	0.686	0.771	0.550	0.900	0.314
	SVM-P	0.844	0.974	0.962	0.950	0.990	0.826
	Logistic	0.154	0.793	0.714	0.800	0.949	0.348
3	RF	0.598	1.000	1.000	1.000	1.000	1.000
	SVM-L	0.850	0.715	0.570	0.800	0.934	0.271
	SVM-P	0.832	0.929	0.950	0.850	0.969	0.773
	Logistic	0.171	0.782	0.700	0.800	0.946	0.348
4	RF	0.566	1.000	1.000	1.000	1.000	1.000
	SVM-L	0.868	0.613	0.822	0.471	0.902	0.308
	SVM-P	0.855	0.925	0.950	0.882	0.980	0.750
	Logistic	0.241	0.653	0.921	0.412	0.903	0.467
5	RF	0.567	1.000	1.000	1.000	1.000	1.000
	SVM-L	0.821	0.590	0.800	0.450	0.884	0.300
	SVM-P	0.831	0.904	0.952	0.800	0.962	0.762
	Logistic	0.129	0.760	0.600	0.850	0.955	0.288

Threshold, values above and equal to threshold was classed into the case group; AUC, area under the curve; Spe, specificity; Sen, Sensitivity; NPV: negative predict value; PPV: positive predict value. RF, random forest; SVM-L, support vector machine with linear kernel; SVM-P, support vector machine with polynomial kernel, Logistic, logistic regression model

ISSN: 1471-2482