Development and validation of an interpretable machine learning model for predicting atrial fibrillation risk in middle-aged and older patients with coronary heart disease

Figures

• 

  Figure 1. LASSO regression for predictor selection. (A) Cross-validation deviance as a function of the log-transformed penalty parameter (λ). Vertical dashed lines denote the λ value at minimum error (left) and the λ at one standard error from the minimum (right, lambda.1se = 0.006045492). (B) Coefficient shrinkage paths of the candidate variables. Each curve represents the trajectory of a coefficient as λ increases, illustrating the order of variable entry and the effect of regularization strength on model sparsity.

• 

  Figure 2. Spearman correlation heatmap of LASSO-selected continuous predictors.

• 

  Figure 3. Dose-response relationships between LASSO-selected continuous variables and AF risk based on restricted cubic spline (RCS) regression. (A) Age; (B) pulse rate; (C) SBP; (D) DBP; (E) TC; (F) TG; (G) CREA; (H) UA; (I) PLR; (J) LMR; (K) FAR; (L) CRP/ALB.

• 

  Figure 4. Receiver operating characteristic (ROC) curves and area under the curve (AUC) of eight ML models. (A) Training set; (B) validation set.

• 

  Figure 5. Calibration curves for predicted versus observed AF risk across eight ML models. (A) Training set; (B) validation set.

• 

  Figure 6. Decision curve analysis (DCA) comparing the net benefit of eight ML models across a range of threshold probabilities. (A) Training set; (B) validation set.

• 

  Figure 7. SHAP (shapley additive explanations) analysis for interpreting the XGBoost model. (A) SHAP-based feature importance bar chart. (B) SHAP summary dot plot: red and blue colors indicate high and low feature values, respectively; SHAP values to the left of the vertical axis decrease predicted AF risk, while those to the right increase it. (C) Force plot for an individual patient, demonstrating the contribution of each feature to the patient-specific risk prediction.

• 

  Figure 8. LIME (local interpretable model-agnostic explanations) analysis for individual patient predictions from the XGBoost model. (A) Feature contributions for a positive case. (B) Feature contributions for a negative case. Feature values are standardized; green/red bars indicate increased/decreased AF risk relative to the local baseline.