Optimising the use of new data streams for making epidemiological inferences in veterinary epidemiology : a thesis presented in partial fulfilment of the requirements for the degree of PhD in Veterinary Epidemiology at Massey University, Manawatu, New Zealand

Abstract

Many ‘big data’ streams have recently become available in animal health disciplines. While these data may be able to provide valuable epidemiological information, researchers are at risk of making erroneous inferences if limitations in these data are overlooked. This thesis focused on understanding the better use of two data streams—livestock movement records and genetic sequence data. The first study analysed national dairy cattle movement data in New Zealand to explore whether regionalisation of the country based on bovine tuberculosis risk influenced trade decisions. The results suggested that the observed livestock movement patterns could be explained by the majority of, but not all, farmers avoiding purchasing cattle from high disease risk areas. The second study took an alternative approach—qualitative interviews—to understanding farmers’ livestock purchasing practices. This study suggested that farmers are not necessarily concerned with disease status of source farms and that it may be the reliance on stock agents to facilitate trade that creates the observed livestock movement patterns in New Zealand. The findings from this study also implied that various demographic and production characteristics of animals may influence farmers’ livestock selling practices, which were quantitatively verified in the third study analysing livestock movement data and animal production data. These studies not only showed that analyses based solely on ‘big data’ can be misleading but also provided useful information necessary to predict future livestock movement patterns. The final study evaluated the performance of various genetic sequence sampling strategies in making phylodynamic inferences. We showed that using all available genetic samples can be not only computationally expensive, but also may lead to erroneous inferences. The results also suggested that strategies for sampling genetic sequences for phylodynamic analyses may need to be tailored based on epidemiological characteristics of each epidemic.