Fig. 3

The AUC of the curve was 0.730 (0.692–0.768) and 0.712 (0.651–0.772) for the (A) training and (B) test groups, respectively. Calibration plot demonstrating good concordance between the predicted probability and actual probability in the (C) training and (D) test sets. DCA results illustrating that the model was clinically useful under the threshold probabilities of 0.2–0.9 and 0.25–0.8 in the training (E) and test (F) group, respectively.