Predicting patient-related outcomes after atrial fibrillation ablation: insights from explainable artificial intelligence and digital health

TL;DR

This study uses machine learning and explainable AI to identify patient groups with different quality of life outcomes after atrial fibrillation ablation, revealing how factors like age and anxiety influence recovery.

Contribution

The novel use of explainable AI and digital health data to identify distinct patient phenotypes with unique post-ablation quality of life trajectories and outcomes.

Findings

Three distinct patient phenotypes were identified with different QoL trajectories after AF ablation.

Anxiety, age, and AF duration were key predictors of post-ablation QoL outcomes.

Patients with AF recurrence had significantly lower QoL gains compared to those without recurrence.

Abstract

Quality of life (QoL) improvement is a primary driver for atrial fibrillation (AF) catheter ablation (CA), yet its determinants remain unclear. We aimed to identify patient phenotypes with distinct post-ablation QoL trajectories, determine their key predictors, and clarify their association with arrhythmia recurrence and reintervention.

We prospectively followed 213 patients (median age 60 years, 31% female) undergoing AF CA at a tertiary hospital for 2.2 years [interquartile range (IQR): 1.6–2.6]. A digital health application collected real-time electronic patient-reported outcomes (PROs), including the AF Effect on QoL (AFEQT) questionnaire. Reference charts were generated from QoL trajectories of recurrence-free patients. Machine learning (ML) identified subgroups with distinct QoL trajectories, and explainable artificial intelligence (AI) highlighted key predictors. Quality of life improved by +26 AFEQT points [95% confidence interval (CI): 18–33] within 3 months post-ablation and remained stable thereafter, despite significant heterogeneity in individual responses. Patients with AF recurrence showed significantly lower QoL gains (P = 0.010). Machine learning identified three phenotypes: a younger cluster with the largest QoL improvements, an emotive cluster with higher recurrence rates and minimal QoL benefits despite additional antiarrhythmic reinterventions, and an older cluster with established cardiovascular risk factors. Anxiety, age, and AF duration emerged as key discriminators.

ML defined three clinically coherent phenotypes, each exhibiting distinct QoL trajectories and ablation outcomes. Explainable AI clarified how individual psychological and biological traits interact to shape these outcomes, highlighting the potential for tailored multidisciplinary care beyond individualized rhythm control strategies.

Graphical AbstractClinical phenotypes and associated outcomes. Electronic patient-reported outcomes measurements from patients undergoing their first AF ablation were prospectively collected through a digital health app. By applying machine learning and explainable artificial intelligence techniques, we identified three distinct clinical profiles based on the interaction of clinical characteristics and outcomes. AF, atrial fibrillation; Dx, diagnosis.

Clinical phenotypes and associated outcomes. Electronic patient-reported outcomes measurements from patients undergoing their first AF ablation were prospectively collected through a digital health app. By applying machine learning and explainable artificial intelligence techniques, we identified three distinct clinical profiles based on the interaction of clinical characteristics and outcomes. AF, atrial fibrillation; Dx, diagnosis.

Figures