Explainable artificial intelligence models in predicting major cardiovascular events: insights from the PolyIran and...

Area of Science:

• Cardiovascular disease research
• Artificial intelligence in healthcare
• Biomedical informatics

Background:

• Cardiovascular diseases (CVDs) are a leading global health concern, necessitating advanced predictive modeling.
• Existing CVD risk prediction models often lack explainability, limiting clinical trust and application.
• Accurate and interpretable models are crucial for effective CVD prevention strategies.

Purpose of the Study:

• To develop and evaluate explainable artificial intelligence (AI) models for predicting major cardiovascular events (MCVE) within 60 months.
• To compare the performance of various machine learning algorithms in CVD risk prediction.
• To identify key predictors of MCVE using explainable AI techniques.

Main Methods:

• Secondary analysis of two large cohort studies (Golestan and Pars Cohorts) with 9,769 participants.
• Application of machine learning models including XGBoost, logistic regression, SVM, Random Forest, KNN, and MLP.
• Utilized SMOTE for class imbalance, SHapley Additive exPlanations (SHAP) for model interpretability, and standard performance metrics (AUC, accuracy, specificity).

Main Results:

• XGBoost demonstrated the best and most stable performance across pooled and individual cohorts, achieving high AUC (0.84 pooled).
• Key predictors identified by SHAP analysis included age, creatinine, and systolic blood pressure.
• Higher creatinine levels, age, SBP, FBS, BMI, LDL/HDL ratio, male sex, and smoking increased MCVE risk; PolyPill use showed protective effects.

Conclusions:

• Explainable AI, specifically XGBoost, offers high predictive accuracy for MCVE and enhances understanding of risk factors.
• Creatinine emerged as a significant, previously underutilized predictor for CVD risk stratification.
• Integrating kidney function assessment into CVD risk prediction models can improve preventative healthcare strategies.