Integrating clinical variables with magnetic resonance imaging and electroanatomic mapping parameters to predict response to first-time and repeat atrial fibrillation ablation procedures

Abstract

Listen

Translate article

Abstract Background The heterogeneity in patient response to atrial fibrillation ablation mandates the development of methods to predict clinical outcomes. Several clinical risk scores exist to prognosticate patients post ablation. Magnetic resonance imaging can quantify anatomical and functional atrial properties, whilst electroanatomic mapping data provides unparalleled atrial substrate characterisation. These data could be leveraged to predict ablation outcomes more accurately. Purpose To incorporate magnetic resonance imaging and electroanatomic mapping parameters with clinical variables to predict response to 1) first-time and 2) repeat atrial fibrillation ablation. Methods A retrospective analysis of 123 consecutive patients undergoing first-time atrial fibrillation ablation including pre-procedural atrial magnetic resonance imaging was performed. Data were split into training (n=68) and test (n=55) cohorts depending on data availability for clinical risk score calculation. The APPLE, DR-FLASH, FLAME, HATCH, HATCH+OSA, CHA2Ds2VASc and CAAP-AF scores were calculated for patients in the training cohort. Univariate logistic regression was used to identify variables associated with arrhythmia recurrence which were then combined into a multivariate logistic regression model. Area under the curve receiver operating characteristic (AUC-ROC) analysis was used to compare the ability of clinical risk scores and the combined model to predict arrhythmia recurrence risk after one procedure. The model's ability to predict repeat ablation response was tested in 38 patients who underwent redo procedures. The ROC optimal cut-off score was used to divide patients into two groups (high and low arrhythmia recurrence risk) for time-to-event analysis. Results In univariate analysis, weight, age, hypertension, left atrial ejection fraction and mean left atrial bipolar voltage were associated with arrhythmia recurrence after one procedure. The multivariate logistic regression model achieved an AUC of 0.7491 (95% confidence interval (CI) 0.6269-0.8712) in the training cohort and 0.7457 (95% CI 0.6137-0.8778) in the test cohort for predicting arrhythmia recurrence at 12 months (Figure 1A). Amongst clinical risk prediction tools, CAAP-AF had the highest predictive value (AUC 0.6529, 95% CI 0.5265-0.7793) (Figure 1B). In patients undergoing repeat ablations, the combined model had an AUC of 0.7585 (95% CI 0.5989-0.9182) for predicting arrhythmia recurrence (Figure 2A). The Kaplan-Meier curves showed a higher risk of arrhythmia recurrence in the ‘high recurrence risk’ group at 12 months (log-rank p=0.029) (Figure 2B). Conclusion A logistic regression model integrating clinical, anatomical, functional, and electrophysiological parameters improved the accuracy of atrial fibrillation ablation outcome prediction compared to existing clinical risk scores. The model may be used to predict response to repeat procedures offering the potential to improve the patient selection process.