Coronary heart disease risk prediction based on GAIN imputation and interpretable machine learning

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Coronary heart disease (CHD) is a major global health burden.
• Existing diagnostic methods face limitations in cost, accessibility, and interpretability.
• There is a need for robust and explainable predictive models for CHD.

Purpose of the Study:

• To develop a machine learning model for predicting CHD risk.
• To ensure the model is interpretable and clinically applicable.
• To identify key predictors of CHD.

Main Methods:

• Retrospective analysis of hospitalized patients with and without CHD.
• Utilized Generative Adversarial Imputation Network (GAIN) for missing data.
• Developed and compared multiple machine learning models, including XGBoost.

Main Results:

• XGBoost model achieved an Area Under the Curve (AUC) of 0.9053.
• SHAP values provided global and local model interpretability.
• Key predictors identified: respiratory rate, age, hs-cTnI, and hypertension.

Conclusions:

• The developed XGBoost model offers robust and interpretable CHD risk prediction.
• The model can be deployed in EHRs for automated screening.
• Potential for use in mobile health applications for patient self-monitoring.