AI modeling for outbreak prediction: A graph-neural-network approach for identifying vancomycin-resistant...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Infectious Disease Epidemiology

Background:

• Nosocomial transmission of vancomycin-resistant enterococci (VRE) is a significant patient safety concern.
• Current infection control measures lack early detection methods for VRE carriers, necessitating improved screening strategies.
• Standardized VRE screening criteria are absent, posing challenges for infection prevention and control.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI)-based approach for early identification and prediction of patients at risk for VRE carriage.
• To assist infection prevention and control staff by providing a human-in-the-loop predictive tool.
• To enhance patient safety and reduce nosocomial VRE transmission through proactive, targeted interventions.

Main Methods:

• Utilized data from 8,372 patients, including over 125,000 hospital movements and patient-related information.
• Created time-dependent graph sequences representing hospital dynamics.
• Applied graph neural networks (GNNs) to classify patients as VRE carriers or noncarriers, integrating clinical diagnosis (ICD) and operations/procedures (OPS) codes as node features.

Main Results:

• The AI model achieved a macro F1 score of 0.880, with a sensitivity of 0.808 and specificity of 0.942 in predicting VRE carriage.
• Clinical diagnosis (ICD) and operations/procedures (OPS) codes were identified as the most impactful parameters for prediction.
• Modeling hospital dynamics with GNNs proved effective for early VRE carrier detection.

Conclusions:

• AI-based tools combining heterogeneous data types can accurately predict VRE carriage, offering high sensitivity.
• The developed GNN approach provides a promising foundation for automated infection prevention and control systems.
• Such systems can improve patient safety, manage antimicrobial resistance, and enable cost-efficient interventions.