Title:

Structure and geography of a hospital patient transfer network

Authors:

Alex Stivala (Swinburne University of Technology and Universita della Svizzera italiana)
Maksym Byshkin (Universita della Svizzera italiana)
Francesca Pallotti (University of Greenwich)
Alessandro Lomi (Universita della Svizzera italiana and University of Melbourne)

Abstract:

We describe a secondary analysis of an interhospital patient transfer
network, in which the nodes represent hospitals and the directed arcs
between them represent transfers of critically ill patients. There are
over 3000 nodes (hospitals) in this network. Each of the hospitals has
a geographic location, and hence this is a spatially embedded network.
We describe the power law distribution and "small world" properties of
this network, and discuss the relationship between its network
community structure and geography.

We describe the difficulties and incremental progress in estimating an
exponential random graph model (ERGM) for this network. Using
currently available estimation methods we were only able to estimate
models for subnetworks of the network (based on census
divisions). With a recently published improved sampling algorithm, we
were able to estimate a simple model of the entire network, when
converted to an undirected network. However using the Equilibrium
Expectation algorithm for estimating ERGMs for large directed
networks, we were able to estimate models for the entire network
relatively quickly. We present several models of the network,
incorporating geographical information. Explicitly controlling for
geographic distance between hospitals has a significant effect on
certain other model parameters, in some cases reversing their sign.
We discuss the implications of the models for the effect of hospital
attributes and geography on interhospital patient transfer network tie
formation.