Comparative diagnostic performance of machine learning models and traditional scores for HFpEF in older adults - PubMed
4 hours ago
- #HFpEF Diagnosis
- #Machine Learning
- #Cardiology
- Machine learning (ML) models, especially Random Forest (RF) and XGBoost, outperform traditional HFpEF diagnostic scores in older adults.
- The study involved training ML models on three derivation cohorts (N=1474) and validating them in two independent cohorts (N=542).
- RF and XGBoost showed the highest diagnostic accuracy (AUC: RF 0.98, XGBoost 0.96) compared to HFA-PEFF (0.86) and H2FPEF (0.79).
- Natriuretic peptides were the most influential feature in the ML models, accounting for 36% of model explainability.
- The findings suggest integrating ML-based tools into clinical workflows could improve early HFpEF diagnosis and treatment initiation.