A new direction in personalized medicine: multimodal joint prediction of hepatic encephalopathy risk post-TIPS

Background: Overt hepatic encephalopathy (OHE) is a common complication after transjugular intrahepatic portosystemic shunt (TIPS), adversely affecting quality of life. This study aimed to develop a predictive model integrating manual imaging, radiomics, and clinical data to forecast OHE within 1 year post-TIPS. Methods: -test, least absolute shrinkage and selection operator, and logistic regression. Three models were built using manual CT features (Model M), radiomics (Model R), and clinical data (Model C), respectively. A combined model (Model MRC) integrated all three. Model performance was evaluated via ROC curves, calibration plots, and decision curve analysis. The primary endpoint was OHE occurrence within 1 year post-TIPS. Results: Within 1 year after TIPS, 79 (33.4%) participants in the training group and 34 (33.3%) participants in validation group developed OHE. Three independent models and one combined model were established and evaluated in terms of their performance. The areas under the ROC curve of Model M, Model R, Model C, and Model MRC were 0.858 (95% CI: 0.809-0.907), 0.744 (95% CI: 0.681-0.808), 0.757 (95% CI: 0.692-0.821), and 0.902 (95% CI: 0.863-0.941), respectively. F1 scores were 0.861, 0.765, 0.797, and 0.891, respectively. Model MRC demonstrated superior performance compared to the other three models. Conclusion: Model MRC exhibited a considerable predictive ability for OHE within the first year after TIPS.