2017 ESA Annual Meeting (August 6 -- 11)

COS 57-5 - The role of spatial coupling in the outbreak of avian influenza

Tuesday, August 8, 2017: 2:50 PM
D137, Oregon Convention Center
Gabriel Gellner, Colorado State University, Colleen T. Webb, Department of Biology, Colorado State University, Fort Collins, CO and Kim M. Pepin, USDA
Background/Question/Methods

Understanding the factors that govern the dynamics of an ongoing outbreak of avian influenza is fundamental to the management and control of a potentially devastating disease, economically as well as for human health and security. Our research develops and applies a novel data-driven model of the H7N2 lowpath outbreak that occurred in the poultry industry in Virginia in 2002, in which we examine the key factors that were operating. Goals of this work are: identify factors that facilitate transmission, quantify their relative role, and estimate unobserved epidemiological quantities (e.g. delay between infection and notification/detection).

We present the model and its application to this case study. Potential transmission factors included in the model are: the role of spatial versus aspatial rates of disease transmission between farms, the movement data of feed trucks during the outbreak, the potential influence of the surveillance team movement between farms, and the company and production types of each farm operation. Beyond the factors that influence the transmission rate of the disease, the model also gives insight into the time between infection and notification, as the detection of clinical signs in poultry livestock is often after the onset of the infection. This unknown delay between infection and detection can have important impacts on the spread within and between farms.

Results/Conclusions

Utilizing a Bayesian approach, we found that, in the case of the Virginia outbreak, the local transmission is significant, having the biggest impact at less than 3 km. Including the full model with aspatial and spatial transmission as well as the company type organized by production type gave the best overall fit to the data. The full model gives more realistic delay between notification and infection, with uncertainty in the delay being lowest for models with company, production type, or company and production type.