Fig. 3

The construction of risk scores for prognosis and evaluation of OP in training and testing cohorts. A–B The significance order of variables related to DFS and OS using the random forest model in training cohort. “%IncMSE” means “increase in mean squared error (%),” and “IncNodePurity” means “increase in node purity,” both of which showed the significance of variables, choosing “%IncMSE” as the primary index. The randomForest R package was used. C–D ROC of the established models in predicting DFS of the training and testing cohort and AUC were compared. E–F ROC of the established models in predicting OS of the training and testing cohort and AUC were compared. G–H The proportion of relapse and death in high- and low-risk groups of all patients as risk scores increased. I–J KM curves of DFS and OS between BSO and OP groups in low DFS and OS risk patients respectively