Can a machine learning-based diagnostic model incorporating novel plasma protein biomarkers accurately diagnose heart failure with preserved ejection fraction?
A novel machine learning diagnostic model incorporating specific plasma proteins and clinical features showed high accuracy (AUC 0.895) for identifying HFpEF, offering a potential new diagnostic tool.
The pathophysiology of heart failure with preserved ejection fraction (HFpEF) remains incompletely understood. This study aimed to identify potential protein biomarkers for the accurate diagnosis and phenotyping of HFpEF and to construct a machine learning-based diagnostic model incorporating these biomarkers and key clinical features. In a cross-sectional study of 249 cardiac patients, HFpEF-associated plasma proteins were identified using Olink PEA and validated by ELISA. A machine learning nomogram was developed and its diagnostic performance was evaluated. Analysis identified 92 plasma proteins,among which Serine protease 27(PRSS27), P-selectin glycoprotein ligand 1 (PSGL-1), Biregional Cell Adhesion Molecule-related (BOC), NF-κB essential modulator (NEMO), Glyoxalase 1(GLO1))) were specifically expressed in HFpEF group. Enrichment analysis indicated these differential proteins were primarily involved in inflammatory response, immune response, and the Phosphatidylinositol 3-kinase-AKT serine/threonine kinase (PI3K-AKT) signaling pathway. A diagnostic model integrating three proteins with clinical features (LDL-C, ALB) demonstrated excellent performance (AUC: 0.895), showing strong discriminatory power, good calibration, and potential clinical applicability. This study identifies potential protein biomarkers for HFpEF diagnosis, provides new insights into its pathophysiology, and offers a practical diagnostic tool for clinical use.
Yimei et al. (Tue,) studied this question.