Objective:

To develop and validate an interpretable machine learning-based prediction model for atrial fibrillation risk in middle-aged and older patients with coronary heart disease using electronic medical records data from hospitalized patients.

Approach:

• Study Design: Retrospective cohort study analyzing 47,617 hospitalized CHD patients from January 2020 to December 2025.
• Data Processing: Random forest imputation and LASSO screening were used to identify predictors.
• Model Training: Eight machine learning algorithms were trained and validated with a 7:3 split of the dataset.
• Performance Evaluation: Model performance was evaluated using AUC, calibration curves, and decision curve analysis, with SHAP and LIME for interpretability.

Key Findings:

• LASSO regression identified 16 predictors, with pulse rate, total cholesterol, systolic blood pressure, creatinine, and triglycerides as the top contributors.
• XGBoost achieved AUCs of 0.867 (95% CI: 0.862–0.872) in the training set and 0.813 (95% CI: 0.802–0.823) in the validation set.
• Nonlinear dose–response relationships were observed for multiple continuous variables.

Interpretation:

The XGBoost model provides individualized AF risk assessment in middle-aged and older CHD patients, facilitating early identification and targeted intervention.

Limitations:

• The study is based on a single-center dataset, which may limit generalizability.
• The retrospective design may introduce biases inherent to historical data, potentially affecting the results.

Conclusion:

The interpretable XGBoost model offers a practical tool for assessing AF risk in CHD patients.