Modelling the risk of AMR bacterial colonization in France’s hospitalized-patient network
Abstract
Background: Antimicrobial-resistant (AMR) bacteria can spread silently in hospitals because colonized patients are often asymptomatic and missed by routine screening. We assessed how patient movement and ward sharing shape colonization risk across France’s acute-care system.
Methods: Using 2023 French hospital discharge data (PMSI-MCO), we reconstructed a temporal patient-based network linking patients who shared at least one overnight stay in the same ward. The network included 4,086,926 patients, 1,278 hospitals and 95,952,349 co-location links. We ran 2,000 stochastic susceptible-infected simulations over 365 daily layers, each seeded with five random colonized patients on a randomly chosen start date. For each patient, colonization probability was estimated as the proportion of simulations in which they became colonized, then aggregated by residence, hospital and ward type. Spatial clustering was measured using Moran’s I, and generalized additive models identified associated factors.
Results: Simulated colonization risk was highly heterogeneous. Residential-area risk showed strong spatial clustering (Global Moran’s I=0.80, 95% CI: 0.78-0.81, p < 0.001), with hotspots in urban and rural settings. Hospital-level risk varied widely but showed little spatial clustering (Moran’s I=0.012), indicating that patient pathways and hospital organization were more important than geographic proximity. Patient risk increased with longer ward stays, repeated admissions and movement across wards. Hospital risk was higher with longer stays, larger admission volumes, bed capacity and readmitted patients.
Conclusions: This pathogen-agnostic framework requires no microbiological data, can be applied to any health system with routine discharge records, and translates patient trajectories into national colonization risk maps. It can help identify high-exposure patients, wards, hospitals and communities, and prioritize surveillance, screening and prevention.
Key messages: • Patient movement and ward co-location reveal where simulated AMR colonization risk concentrates within hospitals and may be redistributed to communities through discharged patients.
• Nationwide patient mobility data can support risk-based surveillance, screening and early prevention when pathogen-specific data are limited.
Related
- Evidence for patterns of colonization risk in the nationwide hospitalized patient network in France
- Risk Factors for Colonization With Antibiotic-Resistant Bacteria Among Hospitalized Patients in Bangladesh: An Antibiotic Resistance in Communities and Hospitals (ARCH) Study
- Maternal Vaginal Pathogenic Bacterial Colonization and Early-Onset Neonatal Sepsis in Mwanza City, Tanzania: A Prospective Cohort Study.
- From respiratory cultures to empirical treatment: Integrating local resistance data, colonization risk, and patient-level factors in hospital-acquired pneumonia
- Serum and cerebrospinal fluid biomarkers for risk stratification of hospitalized neonates and young infants with suspected serious bacterial infections: a secondary analysis of a prospective cohort study
BibTeX
@article{bhuria2026modelling,
title = {{Modelling the risk of AMR bacterial colonization in France’s hospitalized-patient network}},
author = {S Bhuria and P Crépey},
journal = {The European Journal of Public Health},
year = {2026},
month = oct,
doi = {10.1093/eurpub/ckag169.979},
url = {https://doi.org/10.1093/eurpub/ckag169.979},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Modelling the risk of AMR bacterial colonization in France’s hospitalized-patient network
AU - S Bhuria
AU - P Crépey
PY - 2026
DA - 2026/10/01
JO - The European Journal of Public Health
DO - 10.1093/eurpub/ckag169.979
UR - https://doi.org/10.1093/eurpub/ckag169.979
AB - Background: Antimicrobial-resistant (AMR) bacteria can spread silently in hospitals because colonized patients are often asymptomatic and missed by routine screening. We assessed how patient movement and ward sharing shape colonization risk across France’s acute-care system. Methods: Using 2023 French hospital discharge data (PMSI-MCO), we reconstructed a temporal patient-based network linking patients who shared at least one overnight stay in the same ward. The network included 4,086,926 patients, 1,278 hospitals and 95,952,349 co-location links. We ran 2,000 stochastic susceptible-infected simulations over 365 daily layers, each seeded with five random colonized patients on a randomly chosen start date. For each patient, colonization probability was estimated as the proportion of simulations in which they became colonized, then aggregated by residence, hospital and ward type. Spatial clustering was measured using Moran’s I, and generalized additive models identified associated factors. Results: Simulated colonization risk was highly heterogeneous. Residential-area risk showed strong spatial clustering (Global Moran’s I=0.80, 95% CI: 0.78-0.81, p < 0.001), with hotspots in urban and rural settings. Hospital-level risk varied widely but showed little spatial clustering (Moran’s I=0.012), indicating that patient pathways and hospital organization were more important than geographic proximity. Patient risk increased with longer ward stays, repeated admissions and movement across wards. Hospital risk was higher with longer stays, larger admission volumes, bed capacity and readmitted patients. Conclusions: This pathogen-agnostic framework requires no microbiological data, can be applied to any health system with routine discharge records, and translates patient trajectories into national colonization risk maps. It can help identify high-exposure patients, wards, hospitals and communities, and prioritize surveillance, screening and prevention. Key messages: • Patient movement and ward co-location reveal where simulated AMR colonization risk concentrates within hospitals and may be redistributed to communities through discharged patients. • Nationwide patient mobility data can support risk-based surveillance, screening and early prevention when pathogen-specific data are limited.
SN - 1101-1262
SN - 1464-360X
ER -