New paper in the J. Roy. Soc. Interface on optimizing surveillance for livestock diseases June 11, 2012
We present a first quantitative attempt to include the full dynamical nature of bovine displacements in the study of animal infectious disease spreading aimed at characterizing the vulnerability of the livestock trade system to design novel and optimized surveillance systems:
Optimizing surveillance for livestock disease spreading through animal movements
Paolo Bajardi, Alain Barrat, Lara Savini, Vittoria Colizza
J. Roy. Soc. Interface doi: 10.1098/rsif.2012.0289 (2012).
The data on cattle trade movements used in the present study is obtained from the Italian National Bovine database and provides a daily description of the movements of each bovine in Italy, specifying the premises of origin and destination and the date of the movement for each animal (identified through a unique ID). The dataset can be described through a dynamical network where the nodes correspond to premises and a directed link represents a displacement of bovines between two premises.
The small white dots (i.e. the nodes of the network) represent the Italian livestock premises and an arc connecting two of them represents the exchange of batches of bovines. If an infectious disease outbreak occurs, the epidemic may propagate spatially, from one animal holding to another, through the movements of infected animals. Taking advantage of extensive computer simulations, it is possible to compare different seeds of the outbreak in terms of the spreading patterns they produce, and group into clusters the nodes that infect similar sets of premises along their invasion paths. Here, two clusters are shown as examples, and for each of them three snapshots are reported that reproduce the invasion paths of nodes belonging to the same cluster.
Through simulations on the fully dynamic network, we have studied the role of the initial conditions in shaping the propagation process. Clusters of seeds emerge that lead to similar spreading patterns in terms of infected premises (as in the examples reported in the figure), and are also characterized by similar epidemic profiles and peak times. These clusters cannot be identified from purely structural or geographical considerations. The reduction of the degrees of freedom in the initial conditions through clustering also allows us to define a novel method to identify premises characterized by a large vulnerability, an important knowledge for risk assessment analysis. Indeed, although the displacement network is characterized by a large temporal variability, intrinsically altering the centrality role of nodes from a given observation time to another, it is possible to identify sentinel nodes representing premises that are often reached by the disease and, when detected as infected, are able to provide valuable information on the seeding farms of the outbreak and thus on the likely spreading path. The proposed method can be used in order to optimize surveillance systems and define rapid and efficient containment strategies.
Check out the manuscript for additional details and the complete description of the novel methodology proposed for seeds clustering and sentinel identification.