Clinical Electrophysiology -> Atrial Fibrillation & Atrial Flutter: -> Pharmacology (Antiarrhythmic drugs and anticoagulants) D-PO02 - Poster Session II (ID 47) Poster

D-PO02-226 - Development Of A Heparin Dosing Algorithm Through Machine Learning To Match Heparin Requirements For Uninterrupted Apixaban During Catheter Ablation Of Atrial Fibrillation (ID 1069)


Background: While therapeutic heparinization during catheter ablation of AF is critical to reduce risk of thromboembolism, dosing requirements for patients on direct oral anticoagulants (DOACs) differ from warfarin (W).
Objective: To create a modified heparin dosing algorithm for patients on apixaban informed by analysis of ACT trends with intraprocedural heparinization.
Methods: Heparin requirements for 256 consecutive patients undergoing AF ablation on uninterrupted DOAC or W were compared by one-way ANOVA and the Tukey post hoc test. Ensemble learning, a method of machine learning, was used to facilitate the development of a heparin dosing algorithm for patients on apixaban.
Results: Compared to rivaroxaban, dabigatran, and W, apixaban patients had the longest time to achieve target ACT (130.5 min vs. 76.5, 70.8, 71.3; p<0.05), more heparin to achieve target ACT (259.8 u/kg vs. 197.2, 165.4, and 184.2; p < 0.05), and higher total intraprocedural heparin administered (352.8 u/kg vs. 313.5, 293.8, 260.6; p < 0.05). Modeling from these data with cross-validation was able to generate an improved dosing algorithm with an area under the curve of receiver operating characteristic of 0.83.
Conclusion: Anticoagulation with uninterrupted DOACs increase intraprocedural heparin requirements to achieve target ACT during catheter ablation of AF. Historical dosing algorithms result in initial under-dosing, delays to reaching target ACT, and less time within therapeutic range. Machine learning can inform new dosing algorithms, such as that developed in this study for apixaban, which can both improve and be tested prospectively over time.