Phenotyping heart failure using model-based analysis and physiology-informed machine learning.

Area of Science:

• Cardiovascular Physiology
• Computational Modeling
• Machine Learning in Medicine

Background:

• Heart failure with reduced ejection fraction (HFrEF) and preserved ejection fraction (HFpEF) have distinct mechanisms.
• HFpEF is a heterogeneous condition, making diagnosis and treatment challenging.
• Current clinical data may not fully capture the mechanistic differences in HFpEF.

Purpose of the Study:

• To phenotype mechanistic differences between HFrEF and HFpEF using a computational cardiovascular model.
• To identify distinct subgroups within HFpEF based on hemodynamic parameters.
• To explore potential biomarkers for differentiating HFpEF subgroups.

Main Methods:

• A closed-loop cardiovascular model was coupled with patient-specific transthoracic echocardiography (TTE) and right heart catheterization (RHC) data.
• Model simulations were optimized to match patient hemodynamic measurements.
• Machine learning techniques (PCA, k-means, hierarchical clustering) were applied to optimized model parameters.

Main Results:

• Reduced left ventricular (LV) contractility is the primary driver of HFrEF.
• HFpEF exhibits a heterogeneous phenotype, with three distinct subgroups identified: HFpEF1 (HFrEF-like), HFpEF2 (classic HFpEF), and a non-clustering group.
• Elevated LV systolic and diastolic volumes were identified as potential biomarkers for HFrEF-like HFpEF (HFpEF1).

Conclusions:

• Cardiovascular modeling can delineate HFpEF subgroups not distinguishable by clinical data alone.
• HFpEF dysfunction involves the entire cardiovascular system, unlike the cardiac-focused dysfunction in HFrEF.
• Identifying distinct HFpEF phenotypes may lead to patient-specific treatment strategies.