(A) ROC curve of the pre-treatment 6-exo-miRNA EMP recurrence prediction model constructed using cox regression with backward elimination for feature selection using the training cohort (AUC = 0.81). (B) Distribution of risk scores according to recurrence status (P < 0.01, Mann Whitney test). (C) The new combination model, EMP and CA19-9, outperformed the prediction accuracy of other variables in the training cohort (AUC = 0.84). (D) ROC curves of the 6-exo-miRNA EMP derived from the training cohort for recurrence prediction in an independent validation cohort (AUC = 0.78). (E) Distribution of risk scores according to recurrence status (P < 0.01, Mann Whitney test). (F) ROC curves showing the recurrence predictive performance of the combination model, EMP and CA19-9, compared to conventional clinical factors in the validation cohort (AUC = 0.82).