A decision tree model to predict liver cirrhosis in hepatocellular carcinoma patients: a retrospective study

Abstract

Background The severity of liver cirrhosis in hepatocellular carcinoma (HCC) patients is essential for determining the scope of surgical resection. It also affects the long-term efficacy of systemic anti-tumor therapy and transcatheter arterial chemoembolization (TACE). Non-invasive tools, including aspartate aminotransferase to platelet ratio index (APRI), fibrosis-4 (FIB-4), and γ-glutamyl transferase to platelet ratio (GPR), are less accurate in predicting cirrhosis in HCC patients. We aimed to build a novel decision tree model to improve diagnostic accuracy of liver cirrhosis. Patients and Methods The Mann-Whitney U test, χ2 test, and multivariate logistic regression analysis were used to identify independent cirrhosis predictors. A decision tree model was developed using machine learning algorithms in a training cohort of 141 HCC patients. Internal validation was conducted in 99 HCC patients. The diagnostic accuracy and calibration of the established model were evaluated using receiver operating characteristic (ROC) and calibration curves, respectively. Results Sex and platelet count were identified as independent cirrhosis predictors. A decision tree model integrating imaging-reported cirrhosis, APRI, FIB-4, and GPR was established. The novel model had an excellent diagnostic performance in the training and validation cohorts, with area under the curve (AUC) values of 0.853 and 0.817, respectively. Calibration curves and the Hosmer-Lemeshow test showed good calibration of the novel model. The decision curve analysis (DCA) indicated that the decision tree model could provide a larger net benefit to predict liver cirrhosis. Conclusion Our developed decision tree model could successfully predict liver cirrhosis in HCC patients, which may be helpful in clinical decision-making.

Keywords

• Hepatocellular carcinoma
• Liver cirrhosis
• Diagnosis
• Decision tree model
• Machine learning