Functional capacity remains an important prognostic marker in pulmonary arterial hypertension (PAH), although predicting future changes in six-minute walk distance (6MWD) at an individual patient level remains challenging. A retrospective pilot study presented at the ESH 2026 evaluated whether machine learning (ML) algorithms could predict 12-month changes in 6MWD using baseline clinical data.
The analysis included 181 patients with confirmed PAH who underwent comprehensive clinical assessment between 2010 and 2022. Six supervised ML models were trained and tested to predict the rate of change in 6MWD over 12 months.
Findings
- The gradient boosting machine model demonstrated the strongest predictive performance for 12-month changes in 6MWD (r=0.72; R²=0.51; RMSE=0.38).
- Random forest and Ridge regression models also showed moderate predictive accuracy, while Lasso regression, k-nearest neighbors, and decision tree models performed less well.
- Overall, ML models trained using baseline clinical variables demonstrated moderate-to-strong correlations with observed 12-month changes in 6MWD.
The findings suggest that ML-based models may help estimate future changes in 6MWD in patients with PAH using routinely collected baseline clinical variables, although further external validation is needed.
