Abstract
BACKGROUND: Aortic stenosis (AS) is a common disease with delays in diagnosis and treatment. Wearable sensors may provide a scalable, noninvasive method to diagnose, assess severity, and monitor AS. OBJECTIVES: The primary objective was to quantify the correlation between Doppler-derived aortic valve (AV) acceleration time (AT) and sensor-derived AT; the secondary objective was to train machine learning models using sensor data to predict maximal transvalvular velocity (Vmax). METHODS: This prospective, single-center case-control study enrolled 40 subjects with severe (n = 10), moderate (n = 10), and mild (n = 10) AS according to echocardiography guidelines, along with controls without AS (n = 10) who were age- and sex-matched to the severe AS cohort. At the time of echocardiography, patients wore a sensor simultaneously recording electrocardiogram, seismocardiogram, and phonocardiogram signals. RESULTS: The mean age across groups was 77.8 years, and 55% were female (n = 22). Sensor-derived AT showed a strong correlation with Doppler AT (r = 0.85; 95% CI: 0.73-0.92; P < 0.001). Groupwise comparisons demonstrated progressively longer AT from controls through increasing AS severities (P < 0.01). Predictions on unseen patient data from a machine learning model trained on a limited subset of AS patient recordings (n = 32) correlated with AV Vmax (r = 0.76; 95% CI: 0.68-0.83; P < 0.001). CONCLUSIONS: A chest-worn multimodal wearable sensor accurately measures AV AT and predicts AV Vmax, supporting the feasibility of accessible, lower-cost tools for early detection and longitudinal monitoring of AS.