Among 68,714 UK Biobank participants, three AI-derived digital biomarkers — a cine-CMR structural index (DASSi), an AI-electrocardiogram signal, and phase-contrast CMR peak aortic velocity — each independently predicted prevalent aortic stenosis (AS) and prospectively forecast aortic valve replacement. Crucially, genetic and transcriptomic profiling showed partially distinct heritable architectures: peak aortic velocity mapped closely onto known clinical AS genetic risk, while DASSi and AI-ECG together defined a myocardial-remodeling axis largely orthogonal to conventional AS susceptibility loci.
Aortic stenosis has long been treated diagnostically as a single-dimensional lesion, reducible to valve-area or gradient measurements. This preprint — not yet peer-reviewed — challenges that framing by positioning AS as a multidimensional remodeling syndrome with separable structural, electrical, and hemodynamic components, each carrying distinct genomic signal. The practical implication is significant: patients flagged by myocardial-remodeling axes (DASSi + AI-ECG) but not by hemodynamic measures may represent an under-recognized, high-risk subphenotype currently missed by standard echocardiographic surveillance. For longevity-focused medicine, identifying cardiac remodeling before symptomatic valve disease develops could meaningfully shift the intervention window. Limitations include the observational UK Biobank design, which skews toward healthier, predominantly European participants, limiting generalizability. Causal inference remains unestablished. If validated in independent clinical cohorts post-peer review, this digital phenotyping framework could reshape AS screening protocols and stratify patients for earlier, targeted interventions.
