Explaining Biomarker Response to Anticoagulant Therapy in Atrial Fibrillation: A Study of Warfarin and Rivaroxaban with Machine Learning Models

• Adriano Veloso UFMG / Kunumi
• Gianlucca Zuin UFMG / Kunumi
• Luan Sena UFMG

Resumo

Atrial fibrillation (AF) is a common arrhythmia that originates in the heart’s upper chambers and can lead to serious complications like strokes and systemic embolism due to atrial thrombi. To mitigate these risks, anticoagulants such as warfarin and rivaroxaban are frequently prescribed. This study examines how known biomarkers behave under treatment with either warfarin or rivaroxaban. We developed high-performance models (AUROC > 0.94) to distinguish patient subpopulations based on their treatment. Additionally, we built models (AUROC > 0.97) to differentiate individuals with AF from a control group without the condition. Using synthetic data generation for training augmentation and explainable machine learning techniques, we analyzed biomarker behavior, uncovering distinct patterns based on whether patients received warfarin or rivaroxaban. Our approach provides valuable insights into the critical factors influencing biomarker variations across different treatments, enhancing our understanding of their roles in AF management.