Cardiac energetics and ventriculo-arterial interaction-based phenotyping in heart failure: a machine learning...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Medicine

Background:

• Cardiac power output (CPO), stroke-work index (LVSWI), and ventriculo-arterial coupling (VAC) are key indicators of cardiac energetics and ventricular-vascular interaction.
• The prognostic value of these Doppler-derived indices at the bedside in acute decompensated heart failure (ADHF) is not well-established.
• Conventional echocardiography may not fully capture the complex hemodynamic profiles in ADHF patients.

Purpose of the Study:

• To determine if machine learning (ML) phenotyping using LVSWI, CPO, and VAC can refine risk stratification in ADHF.
• To compare the prognostic performance of ML-based phenotyping against traditional echocardiographic assessments.
• To identify distinct hemodynamic phenotypes within the ADHF population.

Main Methods:

• A prospective study of 500 ADHF patients with LVEF < 40% was conducted.
• LVSWI, CPO, and VAC were calculated from admission echocardiography and standardized.
• K-means clustering identified phenotypes, and Kaplan-Meier curves/Cox regression analyzed outcomes. A random forest classifier predicted mortality.

Main Results:

• Two phenotypes emerged: low-output/uncoupled (n=262) and preserved-output/coupled (n=238).
• The low-output group exhibited higher mortality (21.4% vs. 11.8%, p=0.03) and worse hemodynamic parameters despite similar LVEF.
• An augmented logistic model incorporating the three indices outperformed the traditional model (AUC 0.76 vs. 0.67). The random forest model achieved AUC 0.81.

Conclusions:

• ML phenotyping using Doppler-derived CPO, LVSWI, and VAC identifies physiologically distinct ADHF phenotypes.
• These indices offer complementary, physiology-based risk stratification beyond conventional echocardiography.
• Integration into routine practice may guide personalized, hemodynamic-targeted therapies for ADHF patients.