The automated collection of non-specific data from livestock combined with current techniques of data mining and time series analyses facilitate the development of veterinary syndromic surveillance. This type of approach may enhance traditional surveillance of animal diseases. An example involves the continuous analysis of fallen cattle data, which are registered at farm level. However, further research is needed to incorporate such monitoring processes within an early warning system. This study presents a process aimed at 1) fitting automatically the parameters of the classical AutoRegressive Integrated Moving Average models (ARIMA) including patterns of trend and seasonality aggregated at different spatial levels, 2) predicting the mortality at n-ahead period; and 3) detecting mortality peaks. The application of this work is illustrated in the context of fallen dairy cattle data sets from two regions of Spain. The mortality levels registered by week are modelled at county, province and region levels between 2006 and 2013. Using these models the mortality is predicted between January 2014 and June 2015. Values of mortality that are out of the predicted confidence limits are identified as mortality peaks. The causes of such mortality peaks in some affected farms are assessed using data from expert's reports held by associated insurance companies This work compares patterns of fallen dairy cattle in populations with disparate management and environmental conditions with the aim of illustrating a novel approach to obtain information from mortality data at different administrative levels.
|Number of pages||13|
|Journal||Epidemiologie et Sante Animale|
|Publication status||Published - 24 Mar 2017|
- dairy cattle
- syndromic surveillance
- hierarchical time series
- ARIMA models