Single-Centre Retrospective Training Cohort Using Artificial Intelligence for Prognostic Prediction of Encephalopathy, Mortality, and Liver Dysfunction after Early TIPS Creation.

Abstract

Based on an artificial intelligence approach, this study attempted to establish prognostic models to predict 3-month overt hepatic encephalopathy (OHE) occurrence, 1-year mortality, and liver dysfunction for cirrhotic patients with acute variceal bleeding (AVB) treated with early transjugular intrahepatic portosystemic shunt (TIPS) creation. This retrospective study included patients treated with early TIPS between January 2016 and November 2019. Independent risk factors associated with occurrence of OHE within 3months, 1-year mortality, and liver dysfunction after early TIPS were identified using univariate and multivariate logistic analyses. Artificial neural network (ANN) models and prognostic nomograms based on the independent risk factors were established and validated internally. A total of 207 patients were included, with 33 (15.9%) experienced OHE within 3months after TIPS creation. The albumin-bilirubin grade (P = 0.015), age (≤ 65, > 65years) (P < 0.001), gender (P = 0.002), and alcoholic cirrhosis (P = 0.013) was identified as independent risk factors associated with 3-month OHE. Presence of portal vein thrombosis (P = 0.034) and model for end-stage liver disease score (P = 0.063) were identified as independent risk factors associated with 1-year mortality. The platelet-albumin-bilirubin grade (P = 0.041) and a history of hepatic encephalopathy (P = 0.018) were identified as independent risk factors associated with liver dysfunction after TIPS creation. Three ANN models and three nomograms were then established and validated with high accuracy. The ANN and nomogram models have potential to accurately predict early occurrence of OHE, mortality, and liver dysfunction after early TIPS creation for cirrhotic patients with AVB.