Explainable artificial intelligence model to predict acute critical illness from electronic health records

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Clinical Decision Support

Background:

• Acute critical illness prediction often relies on traditional Early Warning Scores (EWS) with limited sensitivity and specificity.
• Existing Artificial Intelligence (AI) systems show high predictive performance for early critical illness detection using electronic health records (EHR).
• Lack of transparency in AI decision-making hinders clinical translation of these advanced systems.

Purpose of the Study:

• To develop and present an explainable AI early warning score (xAI-EWS) system.
• To improve the early, real-time prediction of acute critical illness.
• To facilitate clinical translation by providing interpretable insights into AI predictions.

Main Methods:

• Development of an explainable AI (xAI) model trained on electronic health records (EHR) data.
• Integration of explainability features to accompany AI-driven predictions.
• Evaluation of the xAI-EWS system for early detection of acute critical illness.

Main Results:

• The xAI-EWS system demonstrates potential for early detection of acute critical illness.
• The system provides explanations linked to EHR data, enhancing transparency.
• Explainability is shown to potentiate the clinical translation of AI-based early warning systems.

Conclusions:

• Explainable AI offers a pathway to overcome the clinical translation barrier for AI-based critical illness prediction.
• The xAI-EWS system represents a significant advancement in real-time patient monitoring.
• Integrating interpretability into AI systems is crucial for building clinical trust and utility.