Controlling nosocomial infection based on structure of hospital social networks

Nosocomial infection raises a serious public health problem, as implied by the existence of pathogens characteristic to healthcare and hospital-mediated outbreaks of influenza and SARS. We simulate stochastic SIR dynamics on social networks, which are based on observations in a hospital in Tokyo, to explore effective containment strategies against nosocomial infection. The observed networks have hierarchical and modular structure. We show that healthcare workers, particularly medical doctors, are main vectors of diseases on these networks. Intervention methods that restrict interaction between medical doctors and their visits to different wards shrink the final epidemic size more than intervention methods that directly protect patients, such as isolating patients in single rooms. By the same token, vaccinating doctors with priority rather than patients or nurses is more effective. Finally, vaccinating individuals with large betweenness centrality is superior to vaccinating ones with large connectedness to others or randomly chosen individuals, as suggested by previous model studies. [The abstract of the manuscript has more information.]

💡 Research Summary

**
The authors constructed two empirical social‑contact networks from medical records of a 482‑bed tertiary teaching hospital in Tokyo, one collected on a weekday and the other on a weekend. The networks comprise 605 individuals (388 patients, 94 nurses, 123 physicians) and 3,046 undirected, unweighted edges that represent patient‑patient contacts within the same room, nurse‑nurse contacts within the same ward, patient‑nurse, patient‑physician, and nurse‑physician contacts derived from shared medical records, and physician‑physician contacts within the same clinical team. The resulting graphs exhibit a small‑world topology (average shortest‑path length L ≈ 4.84, clustering coefficient C ≈ 0.53) and a pronounced hierarchical‑modular organization: dense sub‑graphs correspond to rooms, wards, and departments, while inter‑ward links are primarily mediated by physicians.

Using a stochastic SIR model with a frequency‑dependent transmission rate λ and a unit recovery rate, the authors simulated epidemic spread on the static networks. They identified a critical transmission rate around λ ≈ 0.13 at which the basic reproduction number R₀ crosses unity, leading to a phase transition from minor to major outbreaks. Above this threshold the distribution of final epidemic sizes becomes bimodal, reflecting the modular structure: a small peak corresponds to outbreaks that die out early, while a large peak represents widespread epidemics that must traverse narrow physician‑mediated bridges (e.g., a psychiatry ward connected to the rest of the hospital by only two junior doctors).

The study then evaluated several intervention strategies. First, “physician‑movement restriction” (limiting the number of wards a doctor visits) was compared with “patient isolation” (assigning each patient to a single room). Simulations showed that restricting physicians’ cross‑ward activity reduces the final epidemic size far more effectively than isolating patients, because physicians act as the primary conduits linking otherwise loosely connected modules.

Second, vaccination strategies were examined. Prioritizing physicians for vaccination outperformed prioritizing patients or nurses, again reflecting the central role of doctors in network connectivity. Moreover, targeting individuals with the highest betweenness centrality (i.e., those that lie on many shortest paths) yielded a larger reduction in epidemic size than targeting those with the highest degree (most contacts) or random vaccination. This result aligns with the notion that breaking “bridges” in a modular network is more impactful than merely immunizing “hubs.”

Overall, the paper demonstrates that the hierarchical‑modular architecture of hospital social networks fundamentally shapes nosocomial disease dynamics. Effective control measures should focus on limiting the cross‑ward movements of physicians and on vaccinating those who occupy bridge positions (high betweenness), rather than relying solely on patient isolation or degree‑based immunization. These insights provide a data‑driven basis for designing more efficient infection‑control policies in healthcare settings.