Machine learning models, specifically k-nearest neighbors, predicted WHO functional class at 12 months (r = 0.401) to forecast healthcare costs in patients with pulmonary arterial hypertension.
Observational (n=181)
No
Can machine learning models predict WHO-FC at 12 months and forecast healthcare costs in patients with PAH using baseline clinical data?
Machine learning models using baseline clinical data can predict WHO functional class at 12 months in PAH patients, enabling early estimation of healthcare costs.
Effect estimate: r = 0.401
Abstract Background Pulmonary arterial hypertension (PAH) is a rare, progressive disease associated with high morbidity, mortality, and healthcare costs. Rising cost pressure in European healthcare systems increases the need for tools enabling early economic planning and efficient resource allocation. The World Health Organization functional class (WHO-FC), a validated indicator of disease severity in PAH, serves as a proxy for healthcare utilization and expenditure. Purpose To develop and validate machine learning (ML) models that forecast healthcare costs in PAH by predicting WHO-FC 12 months after therapy initiation using baseline clinical data, supporting preventive resource management and cost-efficient care. Methods This pilot study analyzed data from 181 patients with invasively confirmed PAH. Fifty-six baseline variables, including demographic, echocardiographic, hemodynamic, functional, and laboratory data, were used to train six ML models, including regularized and tree-based approaches. The primary outcome was WHO-FC at 12-month follow-up. Predicted WHO-FC values were linked to published real-world per-patient-per-month (PPPM) cost data (pharmacy, medical, total), stratified by WHO-FC (I–IV) and inflation-adjusted to 2024. Models were trained with an 80/20 train–test split and validated using five-fold cross-validation. Results WHO-FC improved from 2. 9 ± 0. 4 to 2. 3 ± 0. 5, while mean total PPPM costs remained stable (12, 329 vs. 12, 332). The k-nearest neighbors (k-NN) model achieved the best predictive performance (r = 0. 401; R² = 0. 161; RMSE = 0. 125; MAE = 0. 106). Predicted total costs ranged from 12, 481 to 12, 614 across models. Conclusions ML-based prediction of WHO-FC enables early cost estimation and patient stratification in PAH using routinely available clinical data. This approach may assist preventive health economic planning and improve efficiency in healthcare resource use. The concept warrants multicenter validation to confirm generalizability and practical utility for preventive resource allocation and cost-effective management in PAH.
Kramer et al. (Mon,) conducted a observational in Pulmonary arterial hypertension (PAH) (n=181). Machine learning models (k-nearest neighbors) was evaluated on WHO-FC at 12-month follow-up (r = 0.401). Machine learning models, specifically k-nearest neighbors, predicted WHO functional class at 12 months (r = 0.401) to forecast healthcare costs in patients with pulmonary arterial hypertension.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: