Enhanced machine learning prognostic model for functional capacity improvement in pulmonary arterial hypertension patients undergoing cardiac rehabilitation

Background Cardiac rehabilitation (CR) enhances functional capacity in pulmonary arterial hypertension (PAH) as measured by the 6-minute walk test (6MWT). A gain of ≥33 meters in 6MWT correlates with reduced PAH-related hospitalizations, while smaller gains indicate worse clinical outcomes. Purpose This study employs machine learning with feature selection and synthetic data augmentation to identify predictors of a ≥33-meter improvement in 6MWT with CR. Methods Twenty-nine PAH patients (mean age: 50.8±13.5/years; 19 Idiopathic PAH, 10 Chronic Thromboembolic PAH; mean pulmonary arterial pressure (PAP): 51±19 mmHg) completed an 8-week exercise-based Phase II CR program. A ≥33-meter improvement in the 6MWT defined success. Baseline clinical data, lung functions, 6MWT and accelerometer data were preprocessed, with Gaussian noise augmentation applied for model robustness. Feature selection used Chi-Squared (χ²) and Mutual Information (MI), while Stratified 5-Fold Cross-Validation assessed Random Forest, MLP, SVM, and Gradient Boosting models. Performance was evaluated via accuracy, precision, recall, and F1-score, with feature importance analyzed (Figure 1A). Results Figure 1B presents classification model performance using features selected by Chi-Squared (χ²) and Mutual Information (MI). MI-selected features yielded the highest accuracy, with Random Forest, MLP, and Gradient Boosting reaching 76.7%. Disease duration was the most important predictor across models, followed by step count, chronotropic response index, WHO functional class, BMI, and PAH Clinical Stage, with varying importance based on selection method and model. Conclusion Our study analyzes patient data to identify key factors associated with functional capacity improvement through CR in PAH. Our model achieved promising performance in predicting a ≥33-meter improvement in the 6MWT. The most influential predictors include disease duration, step count, chronotropic response index, WHO functional class, BMI, and PAH clinical stage.Feature importance scores and models