Predictors of Paravalvular Leakage After Transcatheter Aortic Valve Replacement in Patients With BAV: A Machine...

Area of Science:

• Cardiovascular Medicine
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Bicuspid aortic valve (BAV) anatomy is linked to increased paravalvular leakage (PVL) risk post-transcatheter aortic valve replacement (TAVR).
• Accurate prediction of PVL is crucial for optimizing TAVR outcomes in BAV patients.

Purpose of the Study:

• To develop and validate a predictive model for post-TAVR PVL specifically in patients with BAV.
• To identify key anatomical and procedural predictors of PVL in this patient cohort.

Main Methods:

• A random forest (RF) model was developed using data from 1,080 patients undergoing TAVR (373 BAV type 0, 707 BAV type 1).
• Logistic regression was used for comparative analysis. Model performance was validated on an independent cohort of 109 patients.

Main Results:

• The RF model identified 7 significant predictors of PVL, including calcification volumes and anatomical measurements of the left ventricular outflow tract.
• The RF model demonstrated high predictive accuracy (AUC 0.982 in derivation, 0.975 in validation), outperforming logistic regression.
• Clinical utility was confirmed through decision curve and clinical impact analyses.

Conclusions:

• A 7-predictor machine learning model effectively identifies patients with BAV at high risk for post-TAVR PVL.
• This model can assist in procedural planning and strategies for PVL prevention in BAV patients.