Use of social network analysis to characterize the pattern of animal movements in the initial phases of the 2001 foot and mouth disease (FMD) epidemic in the UK

Abstract

Aggregated movement data do not take into account the relative position of the units within a higher-level structure. Social network analysis (SNA) and graph theory provide a tool to organise and analyse relational data overcoming the limitations of standard methods where the position of individuals/observations does not affect the result of the analysis. Some recorded movements of cattle and sheep during the initial phase of the 2001 foot and mouth disease (FMD) outbreak in the UK, before the ban on animal movements was imposed, are analysed descriptively using SNA. With the data available, a directed dichotomized network with 653 nodes and 797 arches was analysed. Most of the 10 nodes with the highest betweenness (3 farms, 4 markets and 3 dealers) were identified as key players in the initial spread of the infection. Three groups of nodes with distinctive proportion of k < or = 2 neighbours would result in three different theoretical outbreak dimensions assuming that the infection is only disseminated by the movements included in the network: no spread, spread up to 7% and around 25%. There are three hierarchical clusters with 308, 215 and 130 nodes, respectively. Farms in cluster 1 appear to be more similar in their movement patterns to non-farm holdings than to farms in clusters 2 and 3. Relative betweenness, k-neighbours and structural equivalence using hierarchical clustering were able to identify key actors in the evolution of the initial phases of the FMD outbreak such as markets, dealers and farms with atypical movement patterns. Holdings with high betweenness, large number of k < or = 2 neighbours and with movement pattern as in cluster 1 should be targeted in disease control activities once primary actors like markets, dealers and slaughter houses have been contained.