Heart failure & cardiomyopathiesDiagnostic or prognostic study
Comparative diagnostic performance of machine learning models and traditional scores for heart failure with preserved ejection fraction (HFpEF) in older adults
- Diagnostic accuracy study
- derivation and independent validation cohorts
Authors' conclusion
Machine learning algorithms, particularly RF and XGBoost, demonstrated superior diagnostic accuracy compared to established HFpEF scoring systems.
Study context
- Population
- Individuals aged 60 to 80 years; HFpEF cohorts and community-based cohorts without HFderivation N = 1474
- Participants
- 2,017 individuals
- Comparison
- Machine learning (ML) models vs HFpEF scores
Performance
As reported in the abstract
Diagnostic performance for HFpEF (AUC)
RF
0.98
AUC
XGBoost
0.96
AUC
HFA-PEFF
0.86
AUC
H2FPEF
0.79
AUC
ΔC-index
- RF vs H2FPEF
- +0.20
- XGBoost vs H2FPEF
- +0.18
- RF vs HFA-PEFF
- +0.12
- XGBoost vs HFA-PEFF
- +0.10
- RF vs HFpEF-ABA
- +0.17
- XGBoost vs HFpEF-ABA
- +0.15
Further findings
Most influential feature (RF and XGBoost)
- Elevated natriuretic peptides, share of model explainability
- 36%
Original publication
Comparative diagnostic performance of machine learning models and traditional scores for HFpEF in older adults. Eur J Heart Fail. Published 20 March 2026. doi:10.1093/ejhf/xuag039
Summarised from the published abstract; every number is checked against it