Spatiotemporal prediction of vancomycin-resistant Enterococcus colonisation

Area of Science:

• Infectious disease epidemiology
• Hospital infection control
• Computational epidemiology

Background:

• Vancomycin-resistant enterococci (VRE) poses significant patient health and economic burdens.
• Predicting VRE colonization is challenged by the often-overlooked influence of antibiotic use in interconnected patient populations.
• This study examines the role of patient inter-ward movements and antibiotic exposure in VRE colonization.

Purpose of the Study:

• To investigate the relationship between patient flow, antibiotic usage, and VRE colonization within hospital wards.
• To develop predictive models for daily VRE colonization probability at the ward level.
• To assess the efficacy of machine learning models in predicting VRE spread.

Main Methods:

• Utilized intrahospital patient movement, antibiotic use, and PCR screening data.
• Applied the PageRank algorithm to derive ward-level centrality measures for patient and antibiotic flow.
• Employed decision tree and random forest models to predict daily VRE colonization probability.

Main Results:

• Patient movement and antibiotic use centrality measures effectively predict VRE colonization at the ward level.
• A decision tree model provided interpretable rules for VRE risk assessment.
• The random forest model achieved a high AUC of 0.883, outperforming the decision tree model (AUC 0.755).

Conclusions:

• Patient mobility and antibiotic exposure patterns within hospitals are significant predictors of VRE colonization.
• Developed centrality measures offer a novel approach to quantify ward-level infection transmission dynamics.
• Findings support the development of an early warning system for VRE to enhance infection prevention and outbreak management.