Statistical tools for spatio-temporal epidemiology, with application to veterinary diseases : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Statistics at Massey University, Palmerston North
Open Access Location
In epidemiology we are concerned with disease occurrence and its associated explanatory factors. Through analysis of the patterns in disease spread, in space and/or time, we are able to obtain information about possible risk factors and transmission mechanisms. The main focus in spatial epidemiology has been human health. However, economic costs and the concern about zoonoses has fuelled a growing field of veterinary epidemiology. Veterinary epidemiology has the added complication of the `human effect'. For a disease to be recorded we require humans to detect and report the disease, and once reported human intervention is generally applied. This can lead to the true level of disease being under-represented with the loss of information impeding modelling and model predictions. The reliability of statistical analyses depends on the quality of the underlying data. Anomalies could introduce significant bias and lead to inappropriate decision making. Residual analysis is often used to detect anomalous data, but with hierarchical models (common within epidemiology) the highly exible representation of variation can mask outliers. We propose the use of exceedance probabilities as a tool for identifying and assessing anomalous data in spatio-temporal models for routinely collected areal disease count data. We illustrate this methodology through a case study on outbreaks of foot-and-mouth disease (FMD) in Viet Nam for the time period 2006 to 2008. The exceedance probabilities identify several provinces where the number of infected communes was unexpectedly low. These findings are particularly interesting as these provinces are located along major cattle movement pathways within Viet Nam. With epidemic data, the primary interest is the understanding of the transmission of the disease and the effectiveness of intervention strategies. While epidemic curves provide an excellent representation of the temporal patterns, we propose the additional use of a new graphical tool, the `cluster curve' to summarise the changes in spatial clustering through time. The cluster curve is based on the inhomogeneous K-function, and provides a means for summarizing the progression of clustering in infectious disease outbreak data taking into consideration spatial variation in the underlying population. We look at the application of the cluster curve to two outbreaks of FMD in England (2001) and Japan (2010) and to the 2007 epidemic of Equine Influenza (EI) in Australia. By comparing our knowledge of the actual course of the outbreak with the insight provided by the cluster curve we are able to showcase the effectiveness of our tool. Throughout the progression of the outbreak several time windows obtained small sample sizes. Therefore, we also look at the inclusion of significance indicators to definitively differentiate between true clustering and noise due to these small sample sizes. The epidemic outbreaks studied all had intervention methods applied. The impact of intervention strategies was investigated through the simulation (via InterspreadPlus) of five intervention methods on outbreaks of FMD in two geographical regions. Using the cluster curve, we found that intervention methods that created buffer zones were found to have particular characteristics of spatial spread. We found that non-buffer methods were less effective in controlling local spread. This is most likely due to infection transmission prior to clinical signs. This kind of analysis demonstrates the practical importance of having effective tools for describing changes in the spatial patterns of disease during an epidemic outbreak.
Veterinary epidemiology, Epidemiology, Statistical methods, Foot and mouth disease, England, Viet Nam, Equine influenza, Australia