A synthetic minority oversampling technique-based early warning model of postoperative biliary leakage after resection for hepatocellular carcinoma / 中华肝胆外科杂志

ABSTRACT

Objective:To analyze the influencing factors of postoperative bile leakage in laparoscopic liver lobectomy for hepatocellular carcinoma (HCC), and to create and validate an early warning model of postoperative bile leakage based on the synthetic minority oversampling technique (SMOTE).Methods:Clinical data of 120 patients with HCC undergoing laparoscopic lobectomy in Xiaolan People's Hospital of Zhongshan City from January 2016 to January 2022 were retrospectively analyzed, including 72 males and 48 females, aged (58.6±6.7) years old. The patients were divided into two groups according to the occurrence of bile leakage within 30 days after surgery: bile leakage group ( n=32) and non-bile leakage group ( n=88). Clinical data such as lesion size, remnant liver volume, intraoperative blood loss, and serum levels of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) were collected. The positive sample size in the original dataset was expanded according to the SMOTE algorithm, and the SMOTE risk warning model (P 2) was established based on the new dataset. The predictive efficacy of the model was accessed using the receiver operating characteristic (ROC) curve and the area under the curve (AUC). Results:The incidence of postoperative bile leakage was 26.67%(32/120) in the patients. Lesion size, preoperative cholangitis, remnant liver volume, intraoperative blood loss, serum level of ALT and AST differs between the groups (all P<0.05). The sample size of the bile leakage group was expanded to 96 cases by the SMOTE algorithm, and then the sample size ratio of the two groups would be close to 1. Subsequent re-fitting of the expanded data based on the SMOTE algorithm showed that a lesion size of ≥5 cm, preoperative cholangitis, increased intraoperative hemorrhage, elevated ALT and AST were independent risk factors for postoperative bile leakage in patients with HCC (all P<0.05), while a larger remnant liver volume was a protective factor for postoperative bile leakage ( P<0.05). An early warning model P 2 was established based on the above factors. The Hosmer-Lemeshow test showed that the model fitting was good ( P=0.842, coefficient of determination R2=0.647). The sensitivity and specificity of the model for predicting postopera-tive bile leakage was 93.75% and 82.95%, respectively, with an AUC of 0.955 (95% CI: 0.901-0.985). Conclusion:Lesion size, preoperative cholangitis, remnant liver volume, intraoperative blood loss, serum levels of ALT and AST were associated with postoperative bile leakage after surgery for HCC. The early warning model of postoperative bile leakage based on the SMOTE algorithm has a high predictive efficacy.