ABSTRACT
PURPOSE: The aim of this study was to construct a prediction model for posthepatectomy liver failure (PHLF), as defined by the International Study Group of Liver Surgery, and evaluate its accuracy in hepatocellular carcinoma (HCC) patients with cirrhosis or chronic hepatitis. METHODS: A total of 277 consecutive hepatectomies for HCC between 2005 and 2013 were analyzed retrospectively. Multivariate logistic regression analysis was used to develop a predictive model for PHLF. The sensitivity, specificity, and area under the receiver operating characteristic (AUROC) curve were evaluated. The Hosmer-Lemeshow goodness-of-fit test was used to assess the model calibration. The constructed model was internally validated by k-fold cross-validation (k=5). RESULTS: PHLF developed in 12.6% of hepatectomies. Multivariate analysis identified the following variables as predictors of PHLF: elevated preoperative serum bilirubin level, elevated preoperative international normalized ratio, and intraoperative packed red blood cell transfusion. The predictive model allowed discrimination between patients who developed PHLF and those who did not, with a sensitivity of 82.9%, specificity of 72.3%, and AUROC curve of 0.81 (95% CI, 0.74 to 0.89). The Hosmer-Lemeshow test indicated a good fit (P=0.545). The AUROC curve of the developed model was significantly greater than that of the model for end-stage liver disease (MELD) score (P=0.014), suggesting that the former model is better at predicting the PHLF than the latter one. CONCLUSIONS: The developed model could be useful for predicting the occurrence of PHLF in HCC patients with underlying liver disease.