A machine learning approach to classifying New York Heart Association (NYHA) heart failure

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Heart failure is a growing global health crisis, with patient classification crucial for treatment.
• The New York Heart Association (NYHA) Functional Classification is widely used but relies on subjective physician assessment, introducing potential bias.
• Developing objective tools is essential to improve the accuracy and reliability of heart failure patient stratification.

Purpose of the Study:

• To develop and evaluate machine learning models for unbiased assessment of heart failure severity using the NYHA classification.
• To compare the performance of Decision Tree, Random Forest, and Voting Classifier models in stratifying heart failure patients.
• To assess the potential of machine learning as a supplementary tool to reduce bias in clinical practice.

Main Methods:

• A dataset of 434 heart failure patients was used for model training and evaluation.
• Supervised learning was employed to train a Decision Tree model.
• Ensemble learning techniques, including Voting Classifier and Random Forest, were utilized to enhance predictive accuracy.
• Model performance was rigorously assessed using 10-fold cross-validation with stratification.

Main Results:

• The Voting Classifier achieved the highest accuracy at 99.54%, followed by Random Forest at 96.77%, and Decision Tree at 76.28%.
• Both Random Forest and Voting Classifier demonstrated perfect accuracy (100%) in classifying NYHA Class II patients.
• The Voting Classifier showed high accuracy for NYHA Classes I (98.7%) and III (100%), while Random Forest achieved perfect accuracy for NYHA Class IV.

Conclusions:

• Machine learning models, particularly the Voting Classifier and Random Forest, show significant promise in accurately and objectively classifying heart failure patients based on NYHA functional status.
• These models can serve as valuable, unbiased tools to support clinicians, potentially reducing diagnostic bias and improving patient management.
• Further research should explore additional variables and datasets to refine these models and deepen the understanding of factors influencing heart failure progression.