VP, MedTech Epidemiology and Real-World Data Sciences Office of the Chief Medical Officer, Johnson & Johnson Philadelphia, United States
Background: Radiofrequency ablation (RFA) using the CARTO-3D mapping system is a common approach for pulmonary vein isolation to treat atrial fibrillation (AF). CARTONET is a cloud-based system that stores and analyzes ablation procedural data using artificial intelligence/machine learning. Linkage between CARTO procedural data and patients’ electronic health records (EHR) provides an opportunity to examine ablation-related parameters that predict AF recurrence.
Objectives: To build a machine learning model to predict AF recurrence after ablation using the CARTO-EHR linkage data.
Methods: Among AF patients undergoing RFA, procedural data generated during cardiac ablation process were downloaded from CARTONET and linked to deidentified Mercy Health EHR data. Data were divided into train data (70%) for model development and test data (30%) for model validation. Automate machine learning (AutoML) with ensemble of multiple models (e.g. random forest, boosted tree, neural network) was used to predict 1 year AF recurrence, defined as a composite of repeat ablation, electrical cardioversion, and AF hospitalization. Patients' demographic and clinical characteristics were used in AutoML base model. AutoML CARTONET model added CARTONET procedural variables, such as number of ablation lesions, contact force, power, RFA duration, in addition to patients’ demographic and clinical characteristics.
Results: Among 306 patients, 67 (21.9%) patients experienced 1-year AF recurrence. The area under a receiver operating characteristic curve increased from 0.663 for AutoML base model to 0.783 for AutoML CARTONET model in test data. AutoML CARTONET model had a sensitivity of 0.75 and specificity of 0.70. Nine of 10 important predictive features for AF recurrence were CARTO procedural data. Patients with lower contact force in right inferior site and having incomplete pulmonary vein isolations had high risk of AF recurrence. Results confirmed that patients with persistent AF and long ablation duration were more likely to have AF recurrence.
Conclusions: Machine learning models based on CARTO-EHR linkage data better predicted 1-year AF recurrence than model with only demographic and clinical variables. Linkage between CARTONET procedural data and patient EHR data provides a valuable means to optimize the ablation strategies to improve success rate of RFA.