Explainable Machine Learning Based Prediction of Severity of Heart Failure Using Primary Electronic Health Records.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Heart Failure (HF) is a critical global health issue affecting over 64 million individuals.
• Early and accurate diagnosis of HF is vital for effective patient management and improved outcomes.

Purpose of the Study:

• To investigate the efficacy of machine learning (ML) models in predicting heart failure severity.
• To utilize primary Electronic Health Records (EHRs) for developing predictive models.

Main Methods:

• Employed machine learning algorithms including Gaussian Naive Bayes, Random Forest, and CatBoost.
• Utilized a public dataset comprising EHRs from 2008 heart failure patients.
• Evaluated model performance for predicting HF severity.

Main Results:

• The CatBoost model exhibited superior performance in predicting heart failure severity compared to other methods.
• Feature importance analysis for tree-based models aligned with clinically significant parameters, indicating model trustworthiness.

Conclusions:

• Machine learning models, particularly CatBoost, show significant promise for the timely and efficient diagnosis of heart failure.
• Leveraging primary EHR data with ML offers a reliable strategy for enhancing HF patient assessment.