Three-year risk prediction of aortic stenosis using routine medical records: derivation and validation in 919 954...

Area of Science:

• Cardiology
• Biomarkers
• Machine Learning

Background:

• Untreated moderate to severe aortic stenosis (AS) has high mortality (33-42%).
• Early identification is crucial for effective treatment like transcatheter aortic valve replacement.
• Machine learning shows promise for predicting cardiovascular outcomes.

Purpose of the Study:

• To develop and validate a machine learning model (ASrisk) for predicting 3-year risk of aortic stenosis.
• To evaluate ASrisk's ability to identify AS sequelae and echocardiographic outcomes in undiagnosed individuals.
• To assess ASrisk's capacity for enrichment and predicting aortic valve area reduction.

Main Methods:

• Developed and validated a machine learning model using serum biomarkers and vital signs.
• Utilized data from 919,954 participants across two cohorts (Mount Sinai Data Warehouse and UK Biobank).
• Analyzed AS sequelae, echocardiographic findings, and aortic valve area changes in relation to ASrisk scores.

Main Results:

• ASrisk demonstrated a quantitative increase in AS sequelae odds across increasing deciles.
• Higher ASrisk scores correlated with increased odds of echocardiographic findings in undiagnosed individuals.
• Individuals with ASrisk > 0.95 showed an 11-fold enrichment for AS diagnosis and significant aortic valve area reduction within 3 years.

Conclusions:

• The ASrisk model accurately predicts the risk of aortic stenosis three years prior to diagnosis in the general population.
• ASrisk can identify individuals at high risk for AS and associated cardiovascular complications.
• This tool supports early intervention strategies for aortic stenosis.