gms | German Medical Science

Introduction: Today healthcare systems are confronted with Hospital-acquired infections (HAIs), especially emerging multi-drug resistant pathogens. These pathogens could be spread by patients visiting multiple hospitals. In epidemiological concept, hospitals are linked by patient transfers. According to Donker et al. [1], [2], the hospital connectedness critically impacts the number of HAIs. Based on hospitalization data from AOK Lower Saxony (2008-2015), AOK Bavaria (2007-2016), AOKPLUS Thuringia and Saxony (2010-2016) and coming TK data of entire Germany, we construct a theoretical hospital network for Germany. Within this network, we attempt to identify the best effective control and detect measurements for HAI. On the other hand, health companies have different coverages of the population across the federal states. With respect to our data, health insurance coverage from the AOK ranges between 30% (AOK Lower Saxony) and 40% (AOK Bavaria). Since incomplete insurance data could potentially lead to underestimation of epidemic spread and misspecification of hospital network, we intend to evaluate the influences of the incompleteness and find the best way to estimate the missing data and its effect on hospitalization network.

Methods: We use the direct referral sub-datasets [3] from original datasets. To compare the structures of networks, we employ the centrality measures: betweenness, closeness, and modularity detection algorithms. From the network structure, the surrogate patient information could be compensated based on the known information in diverse resampling ways. By simulating the spread of hospital-associated infections between hospitals based on SIS, SIR models, we explore how disparate initial assumption settings and the incompleteness in the network impact the pathogen spread model. We compose and analyze patient networks according to the following:

• Comparing for each dataset the heterogeneous patient characteristics and trajectories.
• Calculate network measures like betweenness, closeness and modularity, on datasets with a changing fraction of patients removed.
• Comparing different resampling methods based on patient removed datasets and network measures with the originals.
• Extrapolating coverage to 100% with the best resampling method.
• Simulating the spread of HAI with epidemic models on the final datasets for different infectious diseases.

Results: In our preliminary results, four communities are detected in Lower Saxony network. Whereas the AOKPLUS network is formed by two main communities almost exactly corresponding to the two federal states Thuringia and Saxony. The values of various network measures are varying through the changing sizes of randomly dropped-out fractions (10% to 70%) and the different resampling algorithms. Conducting the simulation of epidemic models reveals that distinct resampling methods indicate different time-changing infected rate curves.

Discussion: According to the publication [4], the community structures detected in French network indicate high geographical resolution. To alleviate the influences of missing data, conducting the resampling experiments and determining a relatively effective resampling method in aspect of network measures and infected rates are both crucial.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.