How do patients move within the Norwegian hospital system? A comprehensive ward- and hospital-level network analysis

ABSTRACT Patient movements within and between hospitals create networks that can facilitate the spread of nosocomial infections. Many relevant pathogens have long-lasting carriage, allowing colonised patients to move through multiple wards across successive admissions, thereby indirectly linking wards. Yet system-wide ward-level analyses based on individual patient trajectories remain rare, even though they can reveal important features for understanding and simulating the system. We analysed ward- and hospital-level networks using individual patient trajectories from the Norwegian Patient Registry (3.6 million registrations), covering hospital care for ~55% of the population over one year (2012). We characterised the global network structure and, at ward level, calculated multiple centralisation measures and assessed percolation-based connectivity. From these, we identified central wards using directed K-core decomposition combined with Gaussian mixture modelling clustering. Analyses were performed separately for inpatients and all patients, and extended by linking episodes across increasing time gaps to explore how assumptions about episode continuity influence inferred connectivity. Various patient-based statistics were also analysed. Patient movements generated sparse, regionally structured networks with clear core-periphery organisation. Inflow K-cores were larger than outflow cores, with central wards dominated by major hospital referral specialities and medical wards acting as key entry and exit points. Community structure differed by patient type: all-patient networks were largely locally contained, whereas inpatient networks spanned hospitals and regions. Temporal linking increased hub dominance only in all-patient networks, while inpatient structure remained comparatively stable. Dynamic diffusion simulations and a real outbreak analysis independently supported the identified structural backbone, demonstrating that central network hubs also represent the dominant potential pathways of patient-mediated spread under simplified transmission assumptions. The results reveal a robust hierarchical organisational structure of patient movements that can help prioritise surveillance, while underscoring the need for pathogen-specific epidemiological and contextual data for predictive modelling.