Performance of artificial intelligence for prognostic prediction with the albumin-bilirubin and platelet-albumin-bilirubin for cirrhotic patients with acute variceal bleeding undergoing early transjugular intrahepatic portosystemic shunt

Purpose The aim of this study was to validate and compare the prognostic performance of the albumin-bilirubin (ALBI) grade, platelet-albumin-bilirubin (PALBI) grade, Child-Pugh (CP) grade, and Model for End-Stage Liver Disease (MELD) score in predicting the 1-year variceal rebleeding probability using artificial intelligence for patients with cirrhosis and variceal bleeding undergoing early transjugular intrahepatic portosystemic shunt (TIPS) procedures. Materials and methods This dual-center retrospective study included two cohorts, with patients enrolled between January 2016 and September 2018 in the training cohort and January 2017 and September 2018 in the validation cohort. In the training cohort, independent risk factors associated with the 1-year variceal rebleeding probability were identified using univariate and multivariate logistic analyses. ALBI-, PALBI-, Child-Pugh-, and MELD-based nomograms and an artificial neural network (ANN) model were established and validated internally in the training cohort and externally in the validation cohort, which included patients with variceal bleeding who were treated with preventive TIPS. Results A total of 259 patients were included. The median follow-up periods were 24.1 and 18.9 months, and the 1-year variceal rebleeding rates were 12.3% (14/114) and 10.3% (15/145) in the training and validation cohorts, respectively. In the training cohort, all four variables were identified as independent risk factors. Four nomograms were then established and showed comparable prognostic performances after internal (C-index: 0.879, 0.829, 0.874, and 0.798) and external (C-index: 0.720, 0.719, 0.718, and 0.703) validation. The ANN demonstrated that these four variables had comparable importance in predicting the 1-year variceal rebleeding probability. Conclusion None of the four variables are optimal in predicting the 1-year variceal rebleeding probability for patients with cirrhosis and variceal bleeding undergoing early TIPS.