An explainable model for predicting Worsening Heart Failure based on genetic programming

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Heart Failure (HF) presents a significant healthcare burden, with early detection of Worsening Heart Failure (WHF) being crucial for improved patient outcomes.
• Current AI models for cardiovascular diseases often lack transparency, leading to physician reluctance due to their 'black box' nature.

Purpose of the Study:

• To develop a novel, interpretable diagnostic model for predicting WHF.
• To provide a tool that assists clinicians in identifying high-risk patients for timely intervention.

Main Methods:

• A threshold-based binary classifier was developed using a mathematical model derived from Genetic Programming (GP).
• The model identifies key predictors of WHF, including creatinine, sPAP, and CAD, and offers a 3D graphical representation for enhanced understanding.
• Retrospective data from 519 HF patients were analyzed.

Main Results:

• The GP-based classifier achieved an average score of 96% across all evaluation metrics.
• The model demonstrated superior performance compared to commonly used machine learning algorithms.
• The interpretable nature of the model facilitated clinical staff's understanding and decision-making.

Conclusions:

• The developed GP model effectively predicts WHF with high accuracy and interpretability.
• This approach has the potential to significantly impact HF management by enabling earlier diagnosis, targeted treatment, and reduced hospitalizations.