Day-Night Patterns of Wearable-Derived Electrocardiographic Age and Left Atrial Remodeling in Atrial Fibrillation-Naïve Individuals.

Abstract

Background: Artificial intelligence-derived electrocardiographic (AI-ECG) age is conventionally treated as a static marker. Its diurnal dynamics and structural correlates in individuals without atrial fibrillation (AF) remain unknown.

Objective: To characterize diurnal AI-ECG age patterns, their reproducibility, and associations with left atrial (LA) remodeling.

Methods: We analyzed continuous wearable ECGs from two prospective registries (NCT05119725, NCT05355948). In 499 AF-naïve participants with ≥48-hour recordings, AI-ECG age was estimated every 10 minutes. Unsupervised clustering identified day-night trajectories, summarized by day-night difference (ΔAge) into Restorative (>0.6 years) or Disrupted (≤0.6 years) phenotypes. Reproducibility was evaluated across sequential periods. Age- and sex-adjusted associations of binary and continuous ΔAge with LA volume index (LAVI) were pooled in a cross-sectional echocardiographic substudy (n=122).

Results: Three trajectories emerged in both registries. Restorative prevalence was 47.6%-50.2%. Agreement between two consecutive 48-hour periods was 75.8% (κ = 0.516; 95% CI, 0.395-0.631); 6- to 8-day intraclass correlation coefficients were 0.51-0.54. The Disrupted phenotype was associated with higher LAVI (pooled mean difference, 5.54 mL/m²; 95% CI, 1.11-9.98; P = .014) and greater odds of LA enlargement (LAVI ≥40 mL/m²; pooled odds ratio, 3.68; 95% CI, 1.45-9.29; P = .006). Per 1-SD higher ΔAge, LAVI was lower (β, -2.89 mL/m²; P = .007) and LA enlargement less likely (odds ratio, 0.53; P = .004).

Conclusion: Wearable-derived AI-ECG age exhibits a reproducible day-night pattern. A Disrupted phenotype is associated with adverse LA remodeling. Because reproducibility remains insufficient for individual-level classification, prospective outcome-based validation is warranted.