Machine Learning‐Based Prediction of Mortality in Patients With Mitral Regurgitation: Results From Two Nationwide, Prospective Studies

Figures

• 

  FIGURE 1. Flowchart of cohort selection.

• 

  FIGURE 2. Comparative performance metrics across predictive models in the internal and external validation sets. (A) Internal validation set performance metrics and (B) external validation set performance metrics. NPV, negative predictive value; PPV, positive predictive value.

• 

  FIGURE 3. Comparison of different ML methods and currently available risk scores. (A) Training set ROC curves and forest plot of each model AUC score; (B) internal validation set ROC curves and forest plot of each model AUC score; and (C) external validation ROC curves and forest plot of each model AUC score. AUC, area under the curve; ROC, receiver operating characteristic.

• 

  FIGURE 4. Kaplan–Meier survival curves stratified by predicted risk group. (A) Internal validation population and (B) external validation population. CI, confidence interval; HR, hazard ratio.

• 

  FIGURE 5. Clinical utility and calibration of the CatBoost model in the internal and external validation datasets. Clinical impact curves in the (A) internal and (B) external validation sets; calibration curves in the (C) training, (D) internal validation, and (E) external validation sets.

• 

  FIGURE 6. SHAP analysis for the CatBoost model. (A) Feature importance ranked by mean absolute SHAP value. (B) SHAP summary plot illustrating the relationship between feature values and their impact on mortality prediction. Red indicates high feature values; blue indicates low feature values. ACEI/ARB, angiotensin‐converting enzyme inhibitor/angiotensin receptor blocker; AS, significant aortic stenosis; BMI, body mass index; CKD, chronic kidney disease; SHAP, SHapley Additive exPlanations; LVEF, left ventricular ejection fraction; NYHA, New York Heart Association functional class (III/IV vs. I/II); SBP, systolic blood pressure; Tbil, total bilirubin.