Artificial Intelligence for Precision Antiarrhythmic Drug Therapy in Atrial Fibrillation: From Recurrence Prediction to Comparative Treatment Selection

Rhythm control therapy has an important role in atrial fibrillation (AF) management, and antiarrhythmic drugs (AADs) remain essential for pharmacological cardioversion, maintenance of sinus rhythm, reduction in AF burden, and treatment before or after catheter ablation. However, their efficacy varies substantially among patients, while proarrhythmia, organ toxicity, drug interactions, and treatment discontinuation frequently limit their use. Current drug selection therefore relies mainly on safety-based exclusion based on structural heart disease, ventricular function, coronary disease, renal or hepatic function, and baseline conduction and repolarization characteristics, rather than on individualized prediction of comparative therapeutic benefit. This narrative review examines the potential role of artificial intelligence (AI), machine learning (ML), computational electrophysiology, and cardiac digital twins across the AAD treatment pathway. Particular attention is given to patient selection, comparative drug choice, prediction of cardioversion success and sinus-rhythm maintenance, dose optimization, proarrhythmia assessment, extracardiac toxicity, and longitudinal safety surveillance. AI can potentially integrate clinical, electrocardiographic (ECG), imaging, wearable, genomic, and pharmacological data to estimate patient-specific efficacy and toxicity. ML models have already demonstrated the feasibility of predicting drug-induced QT prolongation from electronic health records and detecting ECG signatures associated with drug-induced arrhythmic risk. Moreover, patient-specific AF digital twins have been used to simulate electrophysiological responses to amiodarone and identify patients with different subsequent rhythm outcomes. Nevertheless, most available applications remain retrospective, single-center, non-comparative, or proof-of-concept, and few directly support selection among alternative AADs. Most are prognostic, estimate outcomes under observed care, or predict drug-specific toxicity; models that estimate outcomes under alternative AADs remain the essential missing element. AI-supported antiarrhythmic therapy represents a promising transition from population-based prescribing toward individualized estimation of efficacy, toxicity, and monitoring requirements. Its clinical adoption will require multicenter external validation, causal treatment-effect modeling, prospective workflow evaluation, randomized impact trials, transparent uncertainty reporting, and continued clinician oversight.