Exploring Guidelines for Classification of Major Heart Failure Subtypes by Using Machine Learning.

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Heart failure (HF) has subtypes, often classified by ejection fraction (EF) and end-diastolic volume.
• Current EF cut-offs (e.g., 50%) lack clear justification and create a 'gray zone' (40% < EF < 50%) for diagnosis.
• Existing classification methods struggle with ambiguity and defining clear transitions between HF phenotypes.

Purpose of the Study:

• To develop and validate machine learning (ML) models for classifying HF subtypes.
• To explore classification using ventricular volumes beyond EF.
• To address diagnostic ambiguities in the HF 'gray zone' and improve classification guidelines.

Main Methods:

• Applied unsupervised and supervised ML models to classify HF subtypes.
• Utilized a training set of 48 HF patients and 403 Monte Carlo-generated surrogate patients.
• Analyzed classification performance with varying EF cut-offs and explored HF candidates outside current rules.

Main Results:

• The Support Vector Machine model achieved the best performance (4.06% test error).
• End-systolic volume (ESV) proved to be a more effective discriminator than EF.
• The ML approach successfully classified HF patients within the 'gray zone' and other relevant candidates.

Conclusions:

• ML models driven by ventricular volume data show promise for HF subtype classification, including the 'gray zone'.
• ESV is a key metric for developing improved HF classification guidelines.
• The curvilinear relationship between EF and ESV challenges the utility of linear EF dividers in HF diagnosis.