An interpretable LightGBM model for predicting coronary heart disease: Enhancing clinical decision-making with...

Area of Science:

• Cardiovascular Disease Research
• Machine Learning in Healthcare
• Biomedical Data Science

Background:

• Coronary Heart Disease (CHD) poses a significant global health burden.
• Traditional CHD diagnostic methods like angiography and ECG have limitations including cost, subjectivity, and misdiagnosis.
• There is a need for accurate and interpretable CHD risk prediction tools.

Purpose of the Study:

• To develop and validate a novel prediction framework for Coronary Heart Disease (CHD) risk.
• To enhance the accuracy and interpretability of CHD risk assessment using machine learning.
• To create a user-friendly scoring system for clinical risk stratification.

Main Methods:

• Utilized three public datasets (BRFSS_2015, Framingham, Z-Alizadeh Sani) for model training and validation.
• Employed the LightGBM algorithm for efficient and high-performance CHD prediction.
• Integrated SHAP (SHapley Additive exPlanations) values for model interpretability and developed a CHD scoring system.

Main Results:

• The LightGBM model achieved high performance, with accuracy up to 90.61% and AUROC up to 81.11% on the BRFSS_2015 dataset.
• Validation on Framingham and Z-Alizadeh Sani datasets showed improved accuracy (up to 85.26% and 80.33% respectively) and AUROC (up to 67.37%).
• SHAP analysis identified age, smoking status, diabetes, hypertension, and high cholesterol as critical predictors of CHD risk.

Conclusions:

• The proposed LightGBM framework offers an accurate and interpretable approach to CHD risk prediction.
• SHAP analysis provides valuable insights into the key factors driving CHD risk.
• The developed CHD scoring system serves as a practical tool for clinicians in assessing and managing patient risk.