Analysis of simulated outbreak data and spatial analysis of highly pathogenic avian influenza for preparedness planning and policy

The first objective of this research was to develop and evaluate an approach to analyze and communicate the results of a large number of simulated outbreaks of highly pathogenic avian influenza (HPAI) to decision-makers and policy-makers, using the North American Animal Disease Spread Model (NAADSM), and to make recommendations on the most effective HPAI control policy for Ontario, Canada, specifically, on the effect of stamping-out and ring-culling strategies on the magnitude of an HPAI outbreak. Negative binomial regression analysis was used to identify significant predictors of the number of farms infected for each scenario. Interaction plots were developed from the output of the negative binomial regression analysis, to facilitate communication of simulation results to policy-makers and to analyze the relationship between movement restrictions and destruction strategy. Negative binomial regression analysis was appropriate for handling the right-skewed count data of the simulated HPAI outbreaks in Ontario, while interaction plots were an appropriate visualization tool for communication to policy-makers. For policy development, the modeling results suggested that stamping-out of the infected/detected flocks, without ring-culling, in combination with movement restrictions on direct and indirect contacts, would be the most appropriate policy for Ontario.

The second objective was to compare the results of simulated outbreaks of HPAI using randomly generated point locations and flock sizes and compare the results to those obtained when real data, for Ontario, Canada, were used. The NAADSM requires farm point locations and flock sizes; however, real location data are often unavailable. Therefore random location and flock size datasets are typically used. Three datasets were developed consisting of: 1) a real-industry dataset - real flock size and location data; 2) a "random-industry" dataset - using industry data for random point locations and flock sizes; and 3) a "random-census" dataset - using Statistics Canada agricultural census data for randomly generated point locations and industry data for flock sizes. Four production types were used (commercial chicken meat, commercial eggs, commercial turkey, and hobby poultry) for the analyses. Four outbreak scenarios were investigated for comparison of the real data versus both sets of randomly generated data, considering both a weighted median and weighted maximum number of contacts per day between farms, including various control strategy options (e.g. movement restrictions and destruction strategies). Negative binomial regression analysis and a Kolmogorov-smirnov (K-S) equality-of-distributions test were carried out to determine if any significant difference existed between the three datasets. For both tests, in the majority of the scenarios, there were significant differences between the datasets. The main difference was the real data had a larger maximum number of farms infected compared to the two random datasets, when a median contact structure was used but not when a maximum contact structure was used. Overall, under the conditions set in this particular study, randomly generated flock size and location data were found to be a suitable replacement for real-industry data.