A stacking ensemble model for predicting the occurrence of carotid atherosclerosis

Area of Science:

• Cardiovascular Medicine
• Machine Learning
• Endocrinology

Background:

• Carotid atherosclerosis (CAS) is a major risk factor for cardio-cerebrovascular events.
• Predicting CAS occurrence is crucial for preventative strategies.

Purpose of the Study:

• To enhance the prediction of CAS occurrence using stacking ensemble machine learning.
• To incorporate a wide range of predictors, including endocrine-related markers.

Main Methods:

• Developed five individual prediction models (LR, RF, SVM, XGBoost, GBDT) for CAS.
• Integrated base models using a stacking ensemble algorithm to improve prediction and address overfitting.
• Utilized SHAP value method for in-depth analysis of variable importance, focusing on endocrine factors.

Main Results:

• The ensemble model achieved superior performance over individual models, with AUCs of 0.893 (testing) and 0.861 (validation).
• Optimal accuracy, precision, recall, and F1 score were observed in the validation set.
• Carotid stenosis, age, and endocrine-related factors were identified as significant predictors.

Conclusions:

• The ensemble model demonstrates enhanced accuracy and generalizability in predicting CAS risk.
• This approach underscores the critical role of endocrine dysfunctions in CAS development.
• The model serves as a promising tool for identifying high-risk individuals to prevent CAS and related diseases.