Clinical and Echocardiographic Factors Associated With AI-Estimated Atrial Fibrillation Likelihood During Sinus Rhythm in Patients With Documented Paroxysmal Atrial Fibrillation

Abstract

Background

Artificial intelligence-enabled electrocardiography (AI-ECG) has emerged as a promising tool for identifying patients with atrial fibrillation (AF) using sinus-rhythm ECGs. However, some patients with paroxysmal AF (PAF) may be assigned by AI-ECG to lower AF likelihood categories. Clinical and echocardiographic characteristics associated with AF likelihood assignment by AI-ECG remain unclear.

Methods

This single-center prospective study enrolled adults with documented PAF admitted for catheter ablation who were in sinus rhythm on the admission ECG. The four AF likelihood categories output by AI-ECG were dichotomized into higher and lower groups for analysis. Logistic regression was used to assess factors associated with assignment by AI-ECG to the higher AF likelihood group.

Results

Among 104 patients, 36 were categorized into the lower and 68 into the higher AF likelihood group. Hypertension was associated with lower odds of higher-group assignment (odds ratio [OR], 0.25; 95% confidence interval [CI], 0.10–0.66). In the echocardiographic model, higher left ventricular mass index (LVMI) was associated with lower (OR, 0.72 per 10 g/m² increase; 95% CI, 0.57–0.90) and mild or greater tricuspid regurgitation (TR) with higher (OR, 2.99; 95% CI, 1.14–7.87) odds of higher-group assignment.

Conclusions

A lower AF likelihood assignment by AI-ECG was not uncommon among patients with documented PAF. Hypertension and higher LVMI were inversely associated with assignment to the higher AF likelihood group, whereas mild or greater TR showed a possible positive association. These findings suggest that AI-ECG-based AF likelihood assignment varies among patients with documented PAF during sinus rhythm.