Adding the spatial dimension to the social network analysis of an epidemic: investigation of the 2007 outbreak of equine influenza in Australia

Abstract

Equine influenza is a highly contagious and widespread viral respiratory disease of horses and other equid species, characterised by fever and a harsh dry cough. In 2007, in the first reported outbreak in Australia, the virus spread through the horse populations of two states within 4 months. Most of the geographic spread occurred within the first 10 days and was associated with the movement of infected horses prior to the implementation of movement controls. This study applies social network analysis to describe spread of equine influenza between horse premises infected in the early outbreak period, identifying spread occurring through a contact network and secondary local spatial spread. Social networks were constructed by combining contact-tracing data on horse movements with a distance matrix between all premises holding horses infected within the first 10 days of the outbreak. These networks were analysed to provide a description of the epidemic, identify premises that were central to disease spread and to estimate the relative proportion of premises infected through infected horse movements and through local spatial spread. We then explored the effect of distance on disease spread by estimating the range of local spread (through direct contact, transmission on fomites and windborne transmission) based on the level of fragmentation in the network and also by directly estimating the shape of the outbreak's spatial transmission kernel. During the first 10 days of this epidemic, 197 horse premises were infected; 70 of these were included in the contact-traced network. Most local spread occurred within 5 km. Local spread was estimated to have occurred up to a distance of 15.3 km - based on the contact-and-proximity network - and at a very low incidence beyond this distance based on the transmission kernel estimate. Of the 70 premises in the contact network, spread to 14 premises (95% CI: 9, 20 premises) was likely to have occurred through local spatial spread from nearby infected premises, suggesting that 28.3% of spread in the early epidemic period was 'network-associated' (95% CI: 25.6, 31.0%). By constructing a 'maximal network' of contact and proximity (based on a distance cut-off of 15.3 km), 44 spatial clusters were described, and the horse movements that initiated infection in these locations were identified. Characteristics of the combined network, incorporating both spatial and underlying contact relationships between infected premises, explained the high rate of spread, the sequence of cluster formation and the widespread dispersal experienced in the early phase of this epidemic. These results can inform outbreak control planning by guiding the imposition of appropriate control zone diameters around infected premises and the targeting of surveillance and interventions.

Fig. 1

Infected horse movements during the first 10 days of the equine influenza outbreak of 2007 in Australia. (a) 17–19 August 2007, (b) 20–24 August 2007, (c) 25–26 August 2007. Nodes are coloured yellow and labelled if their out-degree ≥ 1. Horse events where transmission is known to have occurred are denoted by triangles. Numerical labels are unique node identifiers. The red node (ECQS) denotes Eastern Creek Quarantine Station (Sydney) where clinical signs were first observed in a horse in quarantine on 17 August 2007. (For interpretation of the references to colour in this text, the reader is referred to the web version of the article.)

Fig. 2

Contact network representing infected horse movements between premises holding horses infected in the first 10 days of the 2007 equine influenza outbreak in Australia. (a) Graphed in arbitrary space, excluding 135 isolates. (b) Time-ordered dendrogram, excluding isolates. Nodes are coloured yellow if their out-degree ≥ 1. Horse events where transmission is known to have occurred denoted by triangles. Node labels: Centennial Parklands Equestrian Centre Sydney (CP), Maitland event (M), Moonbi event (Mo), Narrabri event (N), Parkes show (P), Warwick event (W), other labels are unique premises identifiers. (For interpretation of the references to colour in this text, the reader is referred to the web version of the article.)

Fig. 3

Proximity networks representing spatial relationships between premises holding horses infected in the first 10 days of the 2007 equine influenza outbreak in Australia. (a) Distance cut-off dichotomised at 5 km, (b) ‘Maximal network’ with distance cut-off of 15.3 km. Graphed in arbitrary space, including isolates. Nodes are coloured yellow if their out-degree ≥ 1. Horse events where transmission is known to have occurred denoted by triangles. Node labels: Centennial Parklands Equestrian Centre Sydney (CP), Maitland event (M), Moonbi event (Mo), Narrabri event (N), Parkes show (P), Warwick event (W), other labels are unique premises identifiers. (For interpretation of the references to colour in this text, the reader is referred to the web version of the article.)

Fig. 4

Contact-and-proximity networks representing spatial relationships and infected horse movements between all premises holding horses infected in the first 10 days of the 2007 equine influenza outbreak in Australia. (a) Distance cut-off dichotomised at 5 km. (b) ‘Maximal network’ with distance cut-off of 15.3 km, all nodes are incorporated into the same weak component. Graphed in arbitrary space, including isolates. Red links represent the movement of infected horses, grey links represent spatial proximity with distance cut-off set at 5 km, blue links signify the movement of infected horses over a distance less than the cut-off. Nodes are coloured yellow if their out-degree ≥ 1, triangles denote horse events. Centennial Parklands Equestrian Centre Sydney (CP), Maitland event (M), Moonbi event (Mo), Narrabri event (N), Parkes show (P), Warwick event (W), other labels are unique premises identifiers. (For interpretation of the references to colour in this text, the reader is referred to the web version of the article.)

Fig. 5

Estimates of the effective range of local spread in the first 10 days of the 2007 equine influenza outbreak in Australia. The inverted fragmentation index represents the proportion of nodes included in the giant weak component of the combined contact-and-proximity network when the distance cut-off is varied. The directly estimated transmission kernel for the same time period is overlaid for comparison. Most local spread occurred within 5 km of an infected premises, with very little local spread beyond 15 km.

Fig. 6

Clusters of infected premises described by the ‘maximal’ contact-and-proximity network (distance cut-off dichotomised at 15.3 km) in the first 10 days of the 2007 equine influenza outbreak in Australia. (a) Cluster boundaries (red) coalesced from 15.3 km buffers around premises grouped according to the components of the maximal proximity network, and infected premises (closed black circles). (b) Block-modelled network of infected horse movements between clusters with node sizes relative to number of premises per cluster, labelled and coloured yellow if out-degree ≥ 1. (For interpretation of the references to colour in this text, the reader is referred to the web version of the article.)