Machine Learning Application for Bleeding Risk Prediction in Patients with Atrial Fibrillation Treated with Oral Anticoagulation

Background: Atrial fibrillation (AF) is a prevalent cardiac arrhythmia associated with a significantly increased risk of systemic thromboembolism and stroke. Anticoagulation therapy, particularly with direct oral anticoagulants, has become the standard for stroke prevention but comes at the cost of an increased bleeding risk. With the introduction of effective alternatives to anticoagulation, such as percutaneous left atrial appendage occlusion, bleeding risk stratification has become essential to guide therapeutic decision-making. Conventional statistical methods have been used for bleeding risk stratification scores, such as HEMORR2HAGES, HAS-BLED, and ATRIA. However, these methods may inadequately address the multifactorial nature of bleeding risk in diverse patient populations, and their overall performance has been suboptimal. Summary and Key Messages: Recent advancements in machine learning (ML) offer promising opportunities to enhance bleeding risk prediction and optimize anticoagulation therapy. This review explores ML applications in AF patients receiving anticoagulation therapy, focusing on the development and validation of ML-based bleeding risk scores. These models have demonstrated improved predictive performance compared to traditional tools, leveraging complex datasets to identify nuanced patterns and interactions. Furthermore, ML-driven tools in warfarin management, including dose prediction, optimization of time in the therapeutic range, and the identification of drug-drug interactions, show significant potential to enhance patient safety and treatment efficacy.