While waterfowl are considered the primary reservoir for avian influenza, the virus could potentially be spread by an array of peri-domestic wildlife species.  As such, the virus poses a re-emerging threat to wildlife, agriculture, and human health.

We developed a contact network model for backyard poultry farms in Colorado.  Our primary objectives were 1) to assess the potential movement of avian influenza virus between backyard poultry farms as a function of wildlife movements, and 2) to identify highly connected farms likely to be involved in virus spread. 

The contact network was based on a database of backyard poultry farm locations developed for avian influenza surveillance.  Movement probabilities were modeled using a distance decay function and, in general, the probability of arriving on farm j from farm i was a function of distance between farms, the probability of moving a particular distance and direction, and the probability of crossing roads and streams.  We assessed individual farm connectivity by iteratively removing farms from lowest to highest degree where degree was calculated as the sum of movement probabilities for all edges emanating from a particular node.  Farm connectivity results were then used to simulate the influence of connectivity on outbreak detection and control.

Results/Conclusions

We identified the network minimum spanning tree (MST) which is the least cost path that includes every node in the network.  The MST identifies the farms that form the backbone of the network and reveals likely spreading patterns of avian influenza within the network.  In our assessment of individual farm connectivity patterns, we found large differences in connectivity between farms, with some farms exhibiting very high probabilities of wildlife movement between neighboring farms and other farms showing very low connectance.  This variability suggests a large differential in the likelihood of an individual farm being involved in the spread of avian influenza and has important implications for designing surveillance and control strategies.  The model also identified farms that have the potential to act as bridges between farm clusters.

Simulations of avian influenza outbreaks indicated that surveillance strategies based on connectivity scores can reduce the time to outbreak detection compared to random testing.  Similarly, weighting control measures by connectivity can decrease disease spread compared to commonly applied circular buffer control schemes.  Network models can be an important tool in the optimization of disease surveillance strategies and the efficient identification of farms targeted for biosecurity and outbreak control measures.