Abstract
Background
Atypical atrial flutters (AFL) are common in patients with prior history of catheter ablation for atrial fibrillation (AF). However, the occurrence of de-novo atypical AFL’s might suggest an underlying atrial myopathy which has ambiguous markers on EKG.FactorECG is a deep-learning-based algorithm which is based on 21-explainable factors on the EKG.
Purpose
To assess the correlation between electroanatomic and pre-ablation sinus rhythm EKG features using FactorECG and de novo atypical atrial flutters.
Methods
We screened consecutive patients who underwent catheter ablation for de-novo atypical AFL between 01/2020 and 12/2023. We excluded patients with prior catheter ablation, prior cardiac surgery. For comparison, patients without structural heart disease and paroxysmal AF were considered as the "control". We collected data on baseline parameters, tachycardia features and putative mechanism (micro- vs. macro-reentry, if feasible). The data on location and quantity of low voltage areas (LVA) from both the atria were collected from electroanatomic mapping performed either in sinus rhythm or during coronary sinus pacing.
For the purpose of application of the machine-learning algorithm (FactorECG, https://decoder.ecgx.ai), we used baseline 12-lead EKGs in sinus rhythm. LASSO regression analysis was used to analyze the association of individual "factors" of FactorECG algorithm and occurrence of de-novo atypical AFL.
Results
After screening a total of 2,617 patients, we included 74 patients in our study cohort and 51 for the EKG-based analysis (Table). Compared to PAF patients, de-novo atypical AFL patients were older and had a higher CHA2DS2-VASc score. In the subset of 51 patients, there were a total of 61 de-novo atypical AFL’s (mean of 1.2 + 0.4 per patient) with an average TCL of 276 + 63.5 msec. Of the 61 de-novo atypical AFLs, 49 (80.3%) were macro-reentry, 8 micro-reentry (13.1%), and 3 figure-8 (4.9%) and unclear mechanism for 1 (1.7%). The most common location of macro-reentrant atypical AFL was peri-mitral (n=17, 27.9%) and the micro-reentrant AFLs did not have a dominant site in the LA. In comparison to patient with PAF, the patients with de novo atypical AFL had a greater percentage of LVA in all LA wall segments. Upon considering only the important factors (5, 8, 10, 13 and 32) from the FactorECG, the area under the receiver operating characteristic curve (AUROC)/c-statistic using FactorECG was 0.84 [95% CI 0.77-0.92] (Figure). The most important factors were 5, 10 and 32.
Conclusions
In our study, patients with de-novo atypical AFL were older in age and had a significantly greater distribution of LVA in comparison to the patients with PAF. FactorECG can be used for prediction of de-novo atypical AFL in patients based on pre-ablation sinus rhythm EKGs. It is plausible that patients with de-novo atypical AFL might have an underlying atrial myopathy, which can be detected with deep-learning algorithm on 12-lead EKG.
Baseline comparison
ROC curves with FactorECG