Clustering analysis of left atrial smax patterns: predicting atrial fibrillation termination and associations with key clinical factors

Although the maximal slope of the action potential duration restitution curve (Smax) reflects the tendency for the wave-break, the distribution pattern of Smax among left atrium (LA) area has not been studied. We hypothesized that the distinct distribution pattern of Smax has a significant association with atrial fibrillation (AF) sustainability.

We created digital twin models of the LA in patients who underwent atrial fibrillation catheter ablation (AFCA), incorporating computed tomography and electroanatomical mapping data obtained during the ablation procedure. Virtual AF was induced in these models, and the Smax of all nodes comprising each digital twin was calculated during the induction period. The extra pulmonary vein (PV) area was divided into six regions, and the mean Smax for each region was calculated. K-means clustering analysis was applied to determine if different mean Smax distribution patterns corresponded to variations in AF sustainability. We analyzed whether Smax patterns varied according to age, sex, AF type (paroxysmal vs. persistent), and termination rate during virtual AF induction in digital twins. If AF terminated within 32 seconds of virtual induction, it was classified as virtual termination.

Among 323 patients (mean age 60.6 ± 9.91 years; female proportion 23.8%; persistent AF 69.0%), Smax patterns of the extra PV area were divided using K= 4 based on clustering selection criteria. Each cluster exhibited unique Smax distribution patterns within the LA. Cluster 1 demonstrated the smallest Smax values in each region, while Cluster 4 had the largest. Cluster 2 had larger Smax values in the left lateral isthmus and left atrial appendage regions compared to Cluster 3, but smaller Smax values in other regions.

Patients in Cluster 1 had a significantly higher rate of AF termination than those in Clusters 2, 3, and 4 (47.3% vs. 9.6%, 15.4%, and 12.1%, respectively; p < 0.001). The male ratio in Cluster 1 was lower than that in Clusters 4 (63.1% vs. 84.6%; p = 0.007).

In multivariate logistic regression including age, sex, and clusters as variables, Cluster 1 maintained a significantly higher likelihood of AF termination compared to the other clusters (p < 0.001).

Our findings suggest a strong association between Smax distribution patterns and AF sustainability, including correlations with sex, AF type (paroxysmal vs. persistent), and AF termination rate. This study demonstrates the potential to predict AF termination using cluster analysis and logistic regression models.Figure 1