A digital biomarker for aortic stenosis development and progression using deep learning for two-dimensional echocardiography

BACKGROUND: The timely identification of aortic stenosis (AS) and disease stage that merits intervention requires frequent echocardiography. However, there is no strategy to personalize the frequency of monitoring needed. OBJECTIVES: To explore the role of AI-enhanced two-dimensional-echocardiography in stratifying the risk of AS development and progression. METHODS: This was a multicenter study of 12,609 patients without severe AS undergoing transthoracic echocardiography in New England (n=8,798, 71 [IQR 60–80] years, n=4250 [48.3%] women) & Cedars-Sinai, California (n=3,811, 67 [IQR 54–78] years, 1688 [44.3%] women). We examined the association of an AI-derived Digital AS Severity index (DASSi; range 0–1) with i) longitudinal changes in peak aortic valve velocity (AV V(max); m/sec/year), and ii) all-cause mortality or aortic valve replacement (AVR) incidence, using multivariable generalized linear and Cox regression models, respectively, adjusted for age, sex, race/ethnicity, and baseline echocardiographic measurements. RESULTS: The median follow-up was 4.1 [IQR 2.3–5.4] (New England) and 3.8 [IQR 3.1–4.4] years (Cedars-Sinai). Within each cohort, higher baseline DASSi was independently associated with faster progression rates in AV V(max) (for each 0.1 increment: +0.033 m/s/year [95%CI: 0.028–0.038, p<0.001], n=5,483 & +0.082 m/s/year [95%CI 0.053–0.111], p<0.001, n=1,292, respectively). Furthermore, there was a dose-response association between higher baseline DASSi and the incidence of death/AVR (adj. HR 1.10 [95%CI: 1.08–1.13], p<0.001 & 1.14 [95%CI 1.09–1.20], p<0.001, respectively). Results were consistent across severity strata, including those without hemodynamically significant AS at baseline. CONCLUSIONS: An AI model built for two-dimensional-echocardiography can stratify the risk of AS progression, with implications for longitudinal monitoring in the community.