Table 4. Receiver operating characteristic analysis of prediction model for predicting the PHLF grade in total cohort.

Performance index	Grade ≥ A	Grade ≥ B	Grade ≥ C
ROC	0.723 (0.697–0.750)	0.792 (0.735–0.850)	0.821 (0.720–0.922)
Cut-off*	172.63	188.57	209.02
Specificity	0.741 (0.706–0.777)	0.702 (0.679–0.726)	0.848 (0.830–0.867)
Sensitivity	0.604 (0.572–0.638)	0.778 (0.667–0.873)	0.714 (0.524–0.905)
Accuracy	0.661 (0.637–0.686)	0.705 (0.683–0.728)	0.847 (0.829–0.865)
Negative predictive value	0.573 (0.550–0.597)	0.986 (0.979–0.992)	0.995 (0.992–0.998)
Positive predictive value	0.765 (0.738–0.792)	0.107 (0.092–0.122)	0.066 (0.047–0.083)

*, the optimal cut-off value of total points of prediction model were selected based on the Youden index.