Prediction models for liver decompensation in compensated advanced chronic liver disease: A systematic review.

[BACKGROUND AND AIMS] Identifying individuals with compensated advanced chronic liver disease (cACLD) at risk of decompensation allows for personalized therapy. However, predicting decompensation is challenging, and multiple models have been developed. This study systematically appraises the performance and clinical applications of published multivariable models predicting first decompensation in patients with cACLD or compensated cirrhosis.

[APPROACH AND RESULTS] We searched MEDLINE for liver decompensation prediction models from inception to December 2023. The research was registered with PROSPERO (CRD42023488395). Model risk of bias and applicability were assessed using the Prediction study Risk of Bias Assessment Tool (PROBAST), with results summarized via narrative synthesis. Reporting followed Preferred Reporting Items for Systematic Reviews and Meta-Analysis and Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies guidelines. Sixteen studies (retrospective and prospective) were included. Seven focused on a single etiology. No study specifically predicted outcomes in persons with alcohol-associated liver disease. Outcome definitions varied, with some models predicting HCC together with decompensation. In total, 27 predictors were included in the models. The most frequent predictors were albumin, platelets, age, liver stiffness, bilirubin, international normalized ratio, and the presence of portal hypertension-related findings during upper gastrointestinal endoscopy. All studies reported discrimination measures, but only 10/16 evaluated calibration. External validation was conducted in 9/16 studies. Thirteen studies were rated as having a high overall risk of bias.

[CONCLUSIONS] For clinical utility, a predictive model must accurately describe future risks. Models for predicting decompensation in cACLD are often poorly described, infrequently include patients with ArLD, and lack external validation. These factors are barriers to the clinical utility and uptake of predictive models for first decompensation in patients with cACLD.