Longitudinal ECG trajectories identify high-risk aortic stenosis phenotypes years prior to TAVR

Abstract Background Early detection of aortic stenosis (AS) remains challenging with many patients presenting late in their disease course. While echocardiography is the diagnostic standard, its use as a screening tool is limited by cost and accessibility. Electrocardiogram (ECG) analysis, particularly using automated interpretation tools, could provide a scalable, cost-effective screening approach. Purpose To evaluate the utility of longitudinal ECG analysis as a screening tool for AS by assessing whether a previously validated AI ECG-based AS model (AK-AVS) can identify high-risk patients years before transcatheter aortic valve replacement (TAVR). Methods We analyzed 7,860 ECGs from 2,040 TAVR recipients collected up to 10 years pre-procedure. The temporal distribution of AK-AVS scores was evaluated using various thresholds (0.5-0.9) in the overall cohort. Unsupervised clustering was applied to identify distinct longitudinal AK-AVS trajectories and their relationship with clinical outcomes was assessed using Cox models adjusted for clinical factors. Predictive utility of the STS and EuroSCOREII scores were evaluated via Harrel’s C-index for models with and without clusters. Results In the overall cohort, 90.8% of patients exceeded an AK-AVS threshold of 0.5 at 6 months pre-TAVR, 82.4% exceeding 0.6, and 50% exceeding 0.8. Among the 450 patients with 5+ ECGs prior to TAVR, AK-AVS scores 0.6 were detected 4.51 years before TAVR, 4.07 years for 0.7, and 3.67 years for 0.8. Unsupervised clustering revealed three distinct trajectories: Persistently High (57.1%, n=1,165), Accelerated Progression (23.6%, n=482), and Stable Low (19.3%, n=393) (Fig. 1A). Significant differences between groups were observed in age (Persistently High: 77.8, Accelerated Progression: 78.4, Stable Low: 72.6 years, p0.001), aortic valve area (0.77, 0.76, 0.80 cm², p=0.011), and permanent pacemaker rates (17.3%, 16.4%, 10.7%, p=0.008) (Fig. 2). No differences were observed across sex, mean gradient, coronary disease, diabetes, COPD, atrial fibrillation, peripheral vascular disease, ejection fraction, Agatston score, and quality of life scores. Both Persistently High and Accelerated Progression patterns independently predicted higher 1-year mortality (adjusted HR 1.48 and 1.40, respectively) compared to Stable Low (Fig. 1B). Adding cluster information improved 1-year mortality risk prediction: EuroSCOREII + clusters (C-index=0.613 vs 0.607) and STS + clusters (C-index=0.607 vs 0.580). Conclusion Longitudinal ECG analysis detects AS-related changes up to 4.5 years before TAVR, with distinct trajectories predicting mortality. These findings support ECG's role as a systematic screening tool for early AS detection and risk stratification.