Machine learning predicts portal vein thrombosis after splenectomy in patients with portal hypertension: Comparative analysis of three practical models

Table 3 Performance of models for portal vein thrombosis risk prediction in different cohorts

	AUC (95%CI)	Sensitivity (95%CI)	Specificity (95%CI)	PPV (95%CI)	NPV (95%CI)	Accuracy (%)	F1	Kappa	Brier
Overall cohort
GLM	0.812 (0.773-0.851)	0.710 (0.642-0.772)	0.799 (0.747-0.844)	0.714 (0.650-0.775)	0.796 (0.740-0.841)	76.2	0.712	0.509	0.169
LSM	0.822 (0.784-0.861)	0.640 (0.569-0.706)	0.869 (0.824-0.906)	0.776 (0.709-0.824)	0.774 (0.718-0.832)	77.4	0.701	0.522	0.163
RFM	0.814 (0.775-0.854)	0.665 (0.595-0.730)	0.852 (0.805-0.891)	0.760 (0.695-0.812)	0.782 (0.727-0.837)	77.4	0.709	0.526	0.168
Training cohort
GLM	0.833 (0.787-0.879)	0.674 (0.588-0.752)	0.877 (0.824-0.919)	0.784 (0.705-0.842)	0.802 (0.736-0.865)	79.6	0.725	0.564	0.157
LSM	0.839 (0.794-0.884)	0.733 (0.650-0.806)	0.828 (0.768-0.877)	0.739 (0.662-0.810)	0.824 (0.759-0.874)	79.0	0.736	0.562	0.153
RFM	0.838 (0.793-0.883)	0.756 (0.674-0.825)	0.808 (0.747-0.860)	0.723 (0.647-0.800)	0.832 (0.769-0.879)	78.7	0.739	0.559	0.154
Validation cohort
GLM	0.769 (0.691-0.846)	0.646 (0.518-0.761)	0.825 (0.724-0.901)	0.750 (0.625-0.839)	0.742 (0.628-0.847)	74.5	0.694	0.477	0.190
LSM	0.826 (0.755-0.897)	0.708 (0.582-0.814)	0.863 (0.767-0.929)	0.807 (0.687-0.883)	0.784 (0.676-0.884)	79.3	0.754	0.577	0.163
RFM	0.777 (0.700-0.853)	0.631 (0.502-0.747)	0.875 (0.782-0.938)	0.804 (0.678-0.876)	0.745 (0.632-0.864)	76.6	0.707	0.516	0.187

AUC: Area under the receiver operating characteristics curve; CI: Confidence interval; PPV: Positive predictive value; NPV: Negative predictive value; GLM: Generalized linear model; LSM: Least absolute shrinkage and selection operator model; RFM: Random forest model.