A random forest machine learning model using 7 predictors effectively identified post-TAVR paravalvular leakage risk in patients with bicuspid aortic valve, achieving an AUC of 0.975 in validation.
Observational (n=1,189)
Can a machine learning model accurately predict post-TAVR paravalvular leakage in patients with bicuspid aortic valve stenosis?
A machine learning model incorporating anatomical and procedural predictors can highly accurately predict the risk of paravalvular leakage after TAVR in patients with bicuspid aortic valve stenosis.
BACKGROUND: The specific anatomy of bicuspid aortic valve (BAV) increases the incidence of ≥ mild paravalvular leakage (PVL) after transcatheter aortic valve replacement (TAVR). OBJECTIVES: The purpose of this study was to develop and validate a model for predicting post-TAVR PVL in patients with BAV. METHODS: A total of 1,080 patients (373 cases with type 0 and 707 cases with type 1) undergoing TAVR were analyzed. The random forest (RF) model was established to predict PVL. Logistic regression analysis was used to investigate predictive differentiation and accuracy values based on the receiver-operating characteristic curve. Additionally, 109 cases further verified the reliability and stability of the model. RESULTS: The RF model of the derivation data set produced 7 predictors, including the corrected calcification volume (Corr-CV) of leaflets, the Corr-CV of annular surroundings, the Corr-CV of the left ventricular outflow tract, the perimeter of the left ventricular outflow tract, the plane/perimeter of the supra-annular plane, horizontal aorta, annular ellipticity, and postdilation. The area under the curve (AUC), accuracy, sensitivity, and specificity of the RF and the logistic model were 0.982, 0.994, 0.997, 0.993, and 0.978, 0.951, 0.917, 0.957, respectively. The AUC of the RF model in the validation data set was 0.975, whereas the AUC of the logistic analysis model was 0.988. Furthermore, the decision curve analysis and clinical impact curve analysis verified the clinical applicability and net benefit. CONCLUSIONS: The machine learning model with 7 predictors effectively identifies post-TAVR PVL risk in patients with BAV stenosis, aiding procedural planning and PVL prevention. (Study on Standard Evaluation System and Optimal Treatment Path of Senile Valvular Heart Disease; NCT05044338).
Mao et al. (Tue,) conducted a observational in Bicuspid aortic valve stenosis (n=1,189). Random forest machine learning model vs. Logistic regression model was evaluated on Post-TAVR paravalvular leakage. A random forest machine learning model using 7 predictors effectively identified post-TAVR paravalvular leakage risk in patients with bicuspid aortic valve, achieving an AUC of 0.975 in validation.