A novel recommender framework with chatbot to stratify heart attack risk.

Area of Science:

• Cardiovascular disease research
• Artificial intelligence in healthcare
• Machine learning for risk prediction

Background:

• Cardiovascular diseases are a leading cause of death, necessitating early detection and intervention.
• Machine learning models show promise for identifying heart attack risk, but often lack transparency.
• Understanding specific risk factors and their associations is crucial for clinical diagnosis.

Purpose of the Study:

• To develop an explainable artificial intelligence (XAI) system for heart attack prediction and risk stratification.
• To enhance transparency in machine learning-based cardiovascular risk assessment.
• To integrate a conversational AI for patient consultation and support.

Main Methods:

• Applied the CatBoost classifier for initial heart attack risk prediction.
• Utilized SHAP (SHapley Additive exPlanations) for transparently explaining model predictions.
• Integrated the BioMistral Large Language Model (LLM) for a digital doctor chatbot functionality.
• Developed a Django-based web application with Google Maps API integration.

Main Results:

• Achieved high accuracy in predicting patient risk levels, with an average AUC of 0.88.
• The system provides both group-based and patient-specific explanations for risk classifications.
• The integrated LLM chatbot offers consultation, answers user questions, and provides hospital location services.
• The system demonstrates improved patient management and potential for lowering heart attack risk.

Conclusions:

• The developed XAI recommender system offers accurate and transparent heart attack risk prediction.
• The system enhances patient understanding and engagement in managing cardiovascular health.
• The combination of predictive modeling, explainability, and conversational AI represents a significant advancement in digital health tools.
• Timely intervention facilitated by such systems can aid in avoiding subsequent disabilities from cardiovascular events.