Development and External Validation of an Echocardiography-Based AI Model for Predicting Aortic Stenosis Progression

Area of Science:

• Cardiology
• Medical Informatics
• Predictive Analytics

Background:

• Aortic stenosis (AS) is a common valvular heart disease in older adults.
• Current surveillance methods for AS lack personalization, leading to inefficient resource use and delayed interventions.
• Variability in AS progression necessitates improved prediction models.

Purpose of the Study:

• To develop and externally validate predictive models for progression to severe AS.
• To establish fixed-horizon prediction models for mild or moderate AS.
• To improve the timeliness and efficiency of AS management.

Main Methods:

• Retrospective cohort study utilizing structured echocardiographic reports from two tertiary centers.
• Development of three models using 21 routine echocardiographic and demographic variables.
• Prediction of progression to severe AS at 1-, 3-, and 5-year horizons, evaluated by AUROC and F1 score.

Main Results:

• Models showed consistent predictive performance in internal testing and external validation.
• Internal test set AUROCs ranged from 0.90 to 0.91 across prediction horizons.
• External validation AUROCs were high, ranging from 0.88 to 0.91, confirming model robustness.

Conclusions:

• Routinely available structured echocardiographic data can effectively predict AS progression.
• Developed models demonstrated consistent performance in an independent external cohort.
• These findings support the use of risk prediction for personalized AS surveillance strategies.