Abstract 18336: Accuracy of Antiarrhythmic Drug Dosing Using a Computer-Guided Decision Tree Model

Abstract

Background: Atrial fibrillation (AF) is the most common heart rhythm disorder and often requires inpatient monitoring for antiarrhythmic drug (AAD) initiation, specifically for dofetilide and sotalol. No system currently offers safe at-home AAD initiation, which is needed to expand AAD access for low-risk or underserved patients and reduce mortality. Objective: Assess accuracy of AAD dosing using a computer-guided decision tree algorithm compared to physician dosing decisions. Methods: Patients in an all-comer population were instructed to take a mobile ECG (mECG; KardiaMobile 6L, AliveCor). A proprietary ML algorithm (SafeBeat Rx) interpreted QTc for each mECG and recommended a starting dose for each AAD (dofetilide and sotalol), per drug labeling. For sotalol, an initial dose of 80 mg was recommended if QTc ≤ 450 ms and HR > 60 bpm; if baseline ECG criteria were unmet, 0 mg was recommended. For dofetilide, initial 500 mcg dose was recommended if QTc ≤ 440 ms. For software simulation, creatinine clearance was assumed to be normal (>60 mL/min) for all patients. Recommendations were adjusted based on a higher QTc threshold >550 ms if the patient’s baseline ECG showed wide QRS (e.g. artificially paced, bundle branch block). The software recommendations were compared to the gold standard: dosing selected by 2 physicians manually measuring baseline mECG QTc and selecting a theoretical starting dose accordingly, independent from software-generated results. Results: 95 patients were enrolled (40% healthy, 30% outpatients, 30% inpatients). The algorithm was highly accurate in predicting starting dose; agreement with physicians was 94.68% for sotalol and 95.74% for dofetilide. Median heart rate was 84.8 bpm (SD = 17.5) and median QTc interval was 475.5 ms (SD = 106.4). Physicians chose the manufacturer-recommended full starting dose for 85% of patients for sotalol and 86% for dofetilide, compared with 86% for sotalol and 86% for dofetilide recommended by the software. Conclusion: The decision tree model had 95% agreement with physician dose recommendations, demonstrating that the algorithm can accurately guide patient-specific AAD dosing, which helps guide remote physician-directed initiation of AAD, ultimately expanding drug access for AF patients.