Artificial Intelligence-Enabled Electrocardiography for Prediction of Sudden Cardiac Death and Malignant Ventricular Arrhythmias: A Scoping Review

(1) Background and Objectives: Current risk stratification strategies for primary prevention of sudden cardiac death (SCD) have limited sensitivity and specificity. Artificial intelligence (AI) applied to electrocardiograms (ECGs) has emerged as a promising tool to predict the risk of future cardiac arrhythmias. This scoping review synthesizes evidence from original studies evaluating AI models trained on ECGs for risk stratification of SCD/malignant ventricular arrhythmias. (2) Materials and Methods: A comprehensive search of MEDLINE, Embase, Web of Science, Scopus and IEEE Xplore was conducted to identify peer-reviewed studies from inception to February 2026. Eligible studies included original investigations in which the model input was an ECG, recorded at baseline or during monitoring, and the outcome was either short-term or long-term SCD/malignant ventricular arrhythmia risk prediction. Extracted variables included study characteristics, ECG data, AI model data, model performance metrics, and the validation strategy. Risk of bias was assessed using PROBAST. (3) Results: Twenty studies met the inclusion criteria. High-risk cardiovascular subgroups (e.g., heart failure cohort, ICD cohort, etc.) or datasets from admitted patients, and conventional machine learning models or deep learning models were used in most studies. AI-ECG algorithms achieved moderate-to-high discriminative performance for identifying patients at an increased risk for imminent SCD/malignant ventricular arrhythmias (nine studies, AUROC ≈ 0.77–0.96) or future SCD/malignant ventricular arrhythmias (eleven studies, AUROC ≈ 0.66–0.94). However, multiple methodological limitations were identified, including limited sample sizes, susceptibility to overfitting, data imbalance-related bias, heterogeneity in dataset and endpoint definitions, inadequate external validation, and incomplete assessment and reporting of model calibration. (4) Conclusions: AI-ECG models demonstrate potential for risk stratification of SCD and malignant ventricular arrhythmias. However, the current evidence base is constrained by several methodological limitations, and further research is required to determine the clinical utility of AI-ECG for predicting SCD.