Artificial intelligence-enabled electrocardiography (AI-ECG) has emerged as a promising tool to improve risk stratification by leveraging machine learning algorithms to detect subtle ECG abnormalities. This systematic review evaluates the performance and clinical utility of AI-ECG in risk prediction among patients with chronic liver disease (CLD).

A comprehensive literature search was conducted in PubMed, EMBASE, Cochrane Library, and Scopus databases for studies published through November 28th, 2024. Eligible studies assessed AI-enhanced ECG models for risk prediction of cirrhosis, esophageal varices and metabolic dysfunction-associated steatotic liver disease (MASLD) in patients with CLD. Relevant data on were extracted and synthesized.

Four studies, encompassing 133,408 participants were included. The AI-ECG-Cirrhosis (ACE) 12‑lead ECG model was the highest performing model (AUC:0.908, sensitivity: 84.9 %, specificity: 83.2 %), followed by the convolutional neural network (CNN)-based deep learning algorithm for the detection of cirrhosis (AUC: 0.86, 95 %CI: 0.85–0.87, sensitivity: 79.5 %, specificity: 76.1 %). The Detection of Undiagnosed Liver Cirrhosis via ECG (DULCE) model in combination with platelet count for the detection of large esophageal varices (AUC: 0.636) and the ECG in combination with clinical parameters (age, sex, body mass index, diabetes mellitus, alanine aminotransferase) models for the detection of MASLD (AUC: 0.76, 95 %CI: 0.74–0.78), sensitivity: 71.9 %, specificity: 67.1 %), had lower performances. There was a positive correlation of the ACE score and MELD-Na score (Spearman's correlation coefficient r = 0.3267, p < 0.001) for the grading of cirrhosis severity.

AI-enabled ECG models could offer a novel non-invasive approach to early subclinical disease detection and risk stratification in patients with CLD, however, their sensitivities and specificities remain to be improved prior to routine clinical use. Future research should therefore focus on optimizing, refining, prospectively validating and standardizing these models to facilitate clinical integration.