A machine learning predictive model based on conventional two-dimensional echocardiography and serum biomarkers for early detection of ascending aorta dilation in BAV patients

TL;DR

A machine learning model combining echocardiography and blood markers helps detect early aortic dilation in bicuspid aortic valve patients more accurately than traditional methods.

Contribution

This is the first model integrating hemodynamic and metabolic markers for early detection of aortic dilation in BAV patients.

Findings

The model achieved an AUC of 0.825 and 74.5% accuracy in predicting aortic dilation.

Key predictors included age, HDL-C, ApoB, left ventricular mass index, and AAoV.

The model outperformed traditional anatomical indicators in predicting AAD.

Abstract

In order to address the challenge of early detection of ascending aortic dilation (AAD) in patients with bicuspid aortic valve (BAV), a machine learning prediction model integrating ultrasound hemodynamics and serum markers was developed to break through the limitations of traditional anatomical indicators.

A total of 51 patients with BAV were prospectively enrolled and divided into ascending aortic dilation group (BAV-D, n = 25) and non-dilated group (BAV-ND, n = 26). Two-dimensional echocardiographic parameters [ascending aorta maximum flow rate (AAoV), mean pressure difference (AAoMPG)] and blood lipid markers [High-Density Lipoprotein Cholesterol (HDL-C), ApoB, etc.] were collected, and the key predictors were screened by the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm, and the logistic regression model was constructed and the nomogram was visualized. Leave one cross-validation (LOOCV) was used to evaluate the robustness of the model.

AAoV, AAoMPG and HDL-C in the BAV-D group were significantly higher than those in the BAV-ND group (all P < 0.05). LASSO screened out five core predictors: age, HDL-C, ApoB, left ventricular mass index (LVMI), and AAoV. The AUC of the model was 0.825 (95% CI: 0.694–0.933), the accuracy was 74.5% (sensitivity 72.0%, specificity 76.9%), and the nomogram verification AUC was 0.809.

The machine learning model constructed by integrating hemodynamics (AAoV) and metabolic markers (HDL-C and ApoB) for the first time can accurately quantify the risk of AAD in BAV patients, and its performance is significantly better than that of a single anatomical parameter, providing a visual decision-making tool for early intervention.

Figures