AI-assisted longitudinal cardiac phenotyping facilitated identification of domain-specific remodeling trajectories in fabry cardiomyopathy
K T Sung, W C Yu, M E Liu, P L Lin, Y H Lai, H W Yang, S H Chang, C L Hung, D M NiuAbstract
Background
Fabry cardiomyopathy involves structural hypertrophy, mechanical dysfunction, and electrical conduction abnormalities that often evolve asynchronously. Whether enzyme replacement therapy (ERT) modifies multidomain longitudinal remodeling patterns rather than isolated parameters remains unclear.
Methods
We performed a multicenter retrospective longitudinal study of 38 patients with genetically confirmed Fabry disease, with paired baseline and follow-up evaluations over a median of 5.3 years (IQR 2.8–7.1). Structural remodeling was indexed by left ventricular mass index (LVMi), mechanical remodeling by left ventricular global longitudinal strain (LVGLS), and electrical remodeling by QRS duration on 12-lead ECG. Echocardiographic measurements were quantified using an AI-assisted deep learning platform to ensure standardized longitudinal analysis. Within-patient directional changes were classified as adverse or non-adverse according to domain-specific clinical meaning, and electrical–mechanical remodeling phenotypes were constructed.
Results
Electrical remodeling showed significant longitudinal progression, with QRS duration increasing from 131.11 ± 34.07 ms to 138.22 ± 36.30 ms (Δ +7.11 ms, 95% CI 0.68–13.53; p=0.031). Mechanical remodeling demonstrated a trend toward improved deformation, with LVGLS changing from −14.74 ± 5.83% to −16.69 ± 4.44% (Δ −1.95%, 95% CI −4.05 to 0.14; p=0.067). Structural remodeling remained stable, with LVMi changing from 177.35 ± 74.48 to 181.67 ± 73.92 g/m² (Δ +4.32 g/m²; p=0.413). Based on combined directional changes in QRS and LVGLS, four electrical–mechanical phenotypes were identified: concordant adverse (n=18, 47.4%), electrical-leading adverse (n=12, 31.6%), mechanical-leading adverse (n=5, 13.2%), and concordant favorable (n=3, 7.9%). Untreated patients were predominantly characterized by concordant adverse remodeling (81%), whereas ERT-treated patients exhibited redistribution toward electrical-leading remodeling (55%) and exclusive occurrence of concordant favorable remodeling (p<0.001 for distribution difference).
Conclusions
Fabry cardiomyopathy is characterized by frequent multidomain, non-parallel remodeling over long-term follow-up. ERT does not produce uniform structural and functional recovery but is associated with a redistribution of longitudinal remodeling trajectories across domains. These findings challenge the assumption of parallel disease progression and support a trajectory-based framework for longitudinal assessment in Fabry cardiomyopathy.