Risk prediction models for hepatic encephalopathy post-transjugular intrahepatic portosystemic shunt: A systematic review and critical appraisal.
Stijn de Reus, Wenjing Wang, Wichor M Bramer et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review assessed prediction models for hepatic encephalopathy (HE) after transjugular intrahepatic portosystemic shunt (TIPS), including studies with at least 100 patients. It included 27 studies, all based on retrospective cohorts of patients with cirrhosis; 24 developed models. Discrimination varied, and models incorporating imaging predictors reported higher AUCs/C-indices than clinical-only models (0.73-0.97 vs. 0.61-0.86). However, 96% of models had high risk of bias, and the authors concluded that current models cannot reliably support clinical decision-making.
Key findings
- The review included 27 studies, of which 24 developed prediction models; all used retrospective cohorts of patients with cirrhosis.
- Imaging-inclusive models reported higher discrimination than clinical-only models (0.73-0.97 vs. 0.61-0.86).
- Child-Pugh and MELD scores showed moderate performance, with AUCs/C-indices of 0.60-0.75 across nine studies.
- PROBAST assessment found high risk of bias in 96% of models; inadequate validation was reported in 79%.
Why it matters
HE is a major complication after TIPS, so reliable risk estimates could inform patient selection. This review shows that apparently strong discrimination in some models does not establish dependable clinical utility: methodological weaknesses, limited applicability, and inconsistent outcome definitions constrain confidence in predictions. The findings identify priorities for developing and validating models that can be assessed across broader patient populations.
What to keep in mind
All included cohorts were retrospective, and 25 of 27 were from Asia, limiting the breadth of populations represented. Models also varied in HE severity and prediction time frame.