Machine-learning versus traditional scoring methods for prediction of atrial fibrillation recurrence after catheter ablation: A systematic review and meta-analysis.
Thomas Koh, Stefany Jeong, Edward Dababneh et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis compared machine-learning models with traditional risk scores for predicting atrial fibrillation recurrence after catheter ablation. Eleven head-to-head studies involving 17,150 patients were included, with discrimination assessed using the C-statistic and pooled using random-effects methods. Machine-learning models had higher pooled discrimination than traditional scores: AUC 0.80, 95% CI 0.72-0.86 vs. 0.58, 95% CI 0.53-0.62 (QM 18.96, p < 0.001). Multimodal fusion models performed best in subgroup analysis, while tree-based models did not significantly outperform traditional scores.
Key findings
- Across 11 studies (n = 17,150), pooled AUC was 0.80, 95% CI 0.72-0.86 for machine learning versus 0.58, 95% CI 0.53-0.62 for traditional scores.
- The difference in discrimination favored machine learning (QM 18.96, p < 0.001).
- Subgroup analysis ranked multimodal fusion models highest, followed by neural-network models.
- Tree-based models did not significantly outperform traditional risk scores in the primary analysis.
Why it matters
Estimating recurrence risk after ablation may help clinicians interpret prognosis and support risk stratification. The pooled results suggest machine-learning approaches can discriminate recurrence better than traditional scores, but discrimination alone does not establish accurate individual risk estimates or clinical benefit. The reported gaps in calibration and external validation are important when considering whether these models can be applied beyond the studies reviewed.
What to keep in mind
The review reports methodological limitations, limited calibration reporting, and absence of external validation, which constrain confidence in immediate clinical adoption.