Gradient boosting machine predicted 12-month changes in six-minute walk distance in PAH patients with the highest accuracy (r = 0.72, R2 = 0.51, RMSE = 0.38) among the tested machine learning models.
Observational (n=181)
No
Can machine learning algorithms predict individual 12-month changes in six-minute walk distance in patients with pulmonary arterial hypertension?
Machine learning algorithms, particularly gradient boosting machines, can accurately predict individual 12-month changes in six-minute walk distance in PAH patients using baseline clinical data.
Effect estimate: r = 0.72
Objective: To evaluate whether machine learning (ML) algorithms can predict individual 12-month changes in six-minute walk distance (6MWD) in pulmonary arterial hypertension (PAH) using baseline clinical data. Design and method: This retrospective single-center pilot study included 181 patients with confirmed PAH (117 female, 64 male) evaluated between 2010 and 2022. Baseline assessment comprised 92 demographic, hemodynamic, laboratory, and functional variables obtained during standardized diagnostic workup including right heart catheterization. The primary endpoint was the 12-month change rate in 6MWD compared with baseline. Six supervised ML models (Lasso regression, Ridge regression, k-nearest neighbors, decision tree, random forest, gradient boosting machine) were trained using an 80/20 train–test split. Hyperparameters were optimized by grid search with 5-fold cross-validation; missing data were imputed using median values. Performance was assessed on the test set using RMSE, MAE, median AE, R2, and Pearson correlation (r). Results: Gradient boosting machine showed the highest predictive accuracy (r = 0.72, R2 = 0.51, RMSE = 0.38). Random forest (r = 0.69, R2 = 0.47, RMSE = 0.40) and Ridge regression (r = 0.56, R2 = 0.32, RMSE = 0.41) also performed well. Lasso regression (r = 0.49, R2 = 0.24, RMSE = 0.44), k-nearest neighbors (r = 0.38, R2 = 0.14, RMSE = 0.45), and decision tree (RMSE = 0.48) showed lower performance. Overall, ML models trained on baseline variables predicted clinically relevant 12-month changes in 6MWD with moderate to strong correlations and low prediction errors. Conclusions: ML algorithms can predict individual changes in 6MWD at 12-month follow-up in PAH using baseline clinical data alone. Gradient boosting machine achieved the best performance. Predictive modeling may complement current risk assessment tools and support personalized, data-driven management strategies. External validation in multicenter cohorts is required.
Mira 等（周五）在肺动脉高压中进行了一项观察性研究（n=181）。评估了机器学习算法（梯度提升机）与其他机器学习模型在6MWD与基线相比的12个月变化率（r = 0.72）。梯度提升机在测试的机器学习模型中，以最高的准确性预测了肺动脉高压患者六分钟行走距离的12个月变化（r = 0.72，R2 = 0.51，RMSE = 0.38）。