Noninvasive Prediction Models of First Decompensation in Compensated Advanced Chronic Liver Disease: A Meta-Analysis

This review aimed to critically appraise available data to determine the accuracy of noninvasive prediction models (noninvasive tests [NITs]) in identifying patients with compensated advanced chronic liver disease (cACLD) who are at increased risk of a first episode of hepatic decompensation.

This systematic review and meta-analysis was conducted from all published articles until February 2025. Studies were included if they evaluated performance of an NIT, defined as 2 or more individual noninvasive markers that had been combined into a prognostic model. Studies were excluded if analysis was done on a population that included those without cACLD, and studies examining only singular prognostic markers (including baseline liver stiffness in isolation). Summary data were extracted from published reports consisting of the C-statistic and/or area under the receiver-operating characteristic curve and model calibration. This review was registered with PROSPERO (CRD42024608001).

Of 6540 screened articles, 30 were included, consisting of 47,647 participants with cACLD. The articles described 39 prognostic models, of which 19 were suitable for meta-analysis. Random effects meta-analysis found models specifically developed and validated in cACLD provide the best prediction, including the SAVE score (summary C-statistic = 0.87; 95% confidence interval, 0.82-0.93) and ABC score (summary C-statistic = 0.85; 95% confidence interval, 0.80-0.89). There was limited validation and calibration of all models. Sensitivity analysis of cohorts using an invasive diagnosis for cACLD provided the least heterogeneous outcomes, with all assessed models having an I² <50%.

cACLD-specific NITs offer the best option in predicting decompensation, and future studies should focus on robust validation, calibration, and external validation with standardized endpoints to ensure that models are reliable for guiding clinical practice in applicable populations.