What does this research mean for the field?

Existing risk prediction models for hepatic encephalopathy in patients with liver cirrhosis demonstrate moderate discrimination performance but suffer from a high overall risk of bias and a lack of clinical effectiveness research. Novelty: ClaimNovelty.SYNTHESIS. Consensus alignment: ConsensusAlignment.NEUTRAL.

April 8, 2026BMC Gastroenterology0 citationsOpen Access

Risk prediction models for hepatic encephalopathy in patients with liver cirrhosis: a systematic review and meta-analysis

ZXZhile Xiao XLXiaonan Li SDShouquan Dong

Key Points

This research aims to review existing prediction models for hepatic encephalopathy in liver cirrhosis patients.
Systematic review of prediction models from Chinese and English databases
Data extraction by two independent researchers
Quality assessed using the PROBAST tool
Meta-analysis conducted with Review Manager 5.4 and Stata 17.0
Thirty-eight prediction models included from thirty-eight studies
Moderate predictive performance with combined AUC of 0.802
Age, prior HE, low albumin, total bilirubin, Child–Pugh grade, ascites, and co-infection identified as significant predictors
High risk of bias found in thirty-four studies, raising concerns about applicability

Abstract

To conduct a systematic review and meta-analysis of existing prediction models for hepatic encephalopathy (HE) in patients with cirrhosis. Four Chinese and five English databases were searched for studies on prediction models for the risk of HE in cirrhosis from inception to April 12, 2025. Two researchers independently conducted the literature search and data extraction, and the quality of the literature was evaluated using the Prediction Model Risk of Bias Assessment Tool (PROBAST). The meta-analysis was conducted using Review Manager 5.4 and Stata 17.0 software. Thirty-eight best prediction models from thirty-eight studies were ultimately included in this review. Among them, 17 studies predicted HE after transjugular intrahepatic portosystemic shunt (TIPS). The incidence of HE ranged from 7.2% to 50.4%. The most commonly used predictors were age and Child–Pugh grade/score. The reported area under the curve (AUC) or c-statistic values ranged from 0.667 to 0.969. Thirty-four studies were found to have a high risk of bias, and 27 studies raised applicability concerns, primarily due to inappropriate data sources, limitations in the domain of analysis, and homogenous study populations. Four externally validated logistic regression models had a combined AUC of 0.802 (95% CI: 0.785—0.820), indicating moderate predictive performance. In meta-analysis, age (OR = 1.04, 95% CI: 1.03, 1.05), prior HE (OR = 4.42, 95% CI: 2.67, 7.31), low albumin (OR = 1.78, 95% CI: 1.25, 2.56), total bilirubin (OR = 2.22, 95% CI: 1.73, 2.85), Child–Pugh grade/score (OR = 2.41, 95% CI: 1.87, 3.09; OR = 1.65, 95% CI: 1.16, 2.33, respectively), ascites (OR = 1.96, 95% CI: 1.48, 2.60), and co-infection (OR = 2.57, 95% CI: 1.66, 3.98) were significant predictors of HE in cirrhosis (P < 0.01). Prediction models for estimating the risk of incident HE with cirrhosis demonstrate moderate discrimination performance, while with a high overall risk of bias and a lack of clinical effectiveness research. Future research should focus on developing new models through optimised study design and analysis, increased sample sizes, and external validation, and applying them to clinical practice. The protocol for this review was registered on PROSPERO (CRD420251040913).

问 AI

Bookmark

View Full Paper