Fig. 4

Receiver operator characteristic (ROC); Calibration curve analysis (CCA); Decision Curve Analysis (DCA). In predicting an “Improved” outcome, the ROC curve (A) shows an outstanding performance (AUC = 0.941). The CCA (B) shows a great consistency between the actual predicting performance (Solid blue line) and the ideal predicting performance (Dotted gray line). In the bar-chart (C), the horizontal level of “0” represents the best cut-off point of the model, bars above or below the “0” level respectively represent the two categories classified by the model (Red: Response; Blue: Non-response). The results in this study indicates a high predicting accuracy of 0.911 where only few mis-categorization (The red bars in the blue group) are observed. DCA (D) shows a coverage (Solid red line) of much more net benefit (y-axis) across the majority of the threshold probabilities (x-axis) in the model compared with the “treat-all strategy” (Solid black line) and the “treat-none strategy” (Dotted black line). This finding reveals the promising clinical usefulness of this model