- Existing post-TIPS HE models show variable discrimination but pervasive methodological flaws and high risk of bias, precluding reliable clinical implementation.
- Models incorporating imaging predictors achieved higher discrimination but often used specialised imaging and selected Asian cohorts, limiting generalisability.
- Future models must use accessible variables, represent diverse patients, account for competing risks, and undergo rigorous external validation to reduce overfitting.
JHEP Rep. 2026 Aug 22:102006. doi: 10.1016/j.jhepr.2026.102006. Online ahead of print.
ABSTRACT
BACKGROUND & AIMS: Transjugular Intrahepatic Portosystemic Shunt (TIPS) is an established treatment for portal hypertension (PH) with expanding indications. Hepatic encephalopathy (HE) is the main complication, affecting up to 50% of patients. Accurate HE-risk prediction is necessary for patient selection. Hence, we aimed to provide an overview of prediction models for post-TIPS HE.
METHODS: We conducted a systematic review of studies developing or validating prediction models for post-TIPS HE, which included ≥ 100 patients. The search, conducted up to December 2025, covered databases such as Embase and Medline. Outcomes assessed included AUCs/C-indices and bias risk, evaluated using PROBAST.
RESULTS: Twenty-seven studies were included, with 24 developing prediction models. All used retrospective cohorts of cirrhotic patients (population size: 106-621), with 25 cohorts from Asia (93%). Models varied in HE severity and prediction time frame, with 11 studies (41%) not reporting any. Discrimination was variable across studies. Child-Pugh score (CPS) and Model for End-Stage Liver Disease, evaluated in nine studies, showed moderate performance (AUCs/C-indices: 0.60-0.75). Models that included imaging predictors (42%) reported higher discrimination than clinical models alone (0.73-0.97 vs. 0.61-0.86). Common clinical predictors were age (68%), CPS (40%), and creatinine (28%). Models scored a high bias risk in PROBAST analysis (96%). Methodological concerns included inappropriate patient exclusion (63%), low event-per-variable rate (81%), inadequate validation (79%), and neglecting competing risks (93%). Applicability concerns involved using highly selected populations (44%) and specialized imaging predictors (33%).
CONCLUSIONS: Current prediction models for post-TIPS HE, despite strong discrimination, cannot support clinical decision-making due to methodological and applicability concerns. Future research should prioritize models with accessible parameters, broad patient representation, competing risks, and rigorous validation to address overfitting.
IMPACT AND IMPLICATIONS: Despite numerous studies, there remains no consensus on whether prediction models for post-TIPS HE can be integrated into clinical practice. This paper provides a comprehensive overview of existing models by highlighting their strengths and limitations. In addition, it identifies recurrent predictors across models and proposes a framework for future research aimed at improving the prediction of this challenging complication.
PMID:42632423 | DOI:10.1016/j.jhepr.2026.102006
Share Evidence Blueprint
Save to Google Notes
Search Google Scholar
Save as PDF
⭐ My Revision List
