Construction of a machine learning-based prediction model for mitral annular calcification

Abstract

Objective To develop a risk prediction model for mitral annular calcification (MAC) using various machine learning algorithms to enable early identification and risk assessment of MAC. Methods A total of 500 patients who were hospitalized and underwent echocardiography at Zhongda Hospital, Southeast University, from July 2022 to March 2024, were selected as subjects, including 250 patients with MAC and 250 without. Clinical data, such as general characteristics and laboratory test indicators, were collected. The subjects were randomly divided into a training set (350 cases) and a test set (150 cases) at a 7∶3 ratio. Nine machine learning algorithms, including logistic regression, relaxed support vector machines (RSVM) , decision tree, elastic net, multilayer perceptron, K-nearest neighbors, random forest, extreme gradient boosting (XGBoost) , and light gradient boosting machine (LightGBM) , were used to build prediction models for MAC. The performance of the models was evaluated using the area under thereceiver operating characteristic curve (AUC) , and the best-performing model was selected. The Shapley additive explanations (SHAP) method was used to assess feature importance, and feature selection was performed to construct the final model. Results In the test set, the random forest model had the largest AUC (AUC＝0.913) , with a sensitivity and specificity of 89.2% and 75.0%, respectively. After feature selection, a simplified random forest model containing three important features, triglyceride-glucose (TyG) index, estimated glomerular filtration rate (eGFR) and age, was built, and the final model had an AUC of 0.896 in the test set, with high prediction accuracy. Conclusion The random forest model performed best among the machine learning-based MAC risk prediction models, and the simplifiedmodel was able to efficiently predict the occurrence of MAC. This method provides a convenient clinical tool for earlyrisk assessment of MAC.

Keywords

• mitral annular calcification
• machine learning
• insulin resistance
• triglyceride-glucose index
• shapley additive explanations