Analysis of Faecal Egg Count (FEC) and Faecal Egg Count Reduction Test (FECRT) datasets is frequently complicated by a high degree of variability between observations and relatively small sample sizes. In this thesis, statistical issues pertaining to the analysis of FEC and FECRT data are examined, and improved methods of analysis using Bayesian Markov chain Monte Carlo (MCMC) are developed. Simulated data were used to compare the accuracy of MCMC methods to existing maximum likelihood methods. The potential consequences of model selection based on empirical fit were also examined by comparing inference made from simulated data using different distributional assumptions. The novel methods were then applied to FEC data obtained from sheep and horses. Several syntactic variations of FECRT models were also developed, incorporating various different distributional assumptions including meta-population models. The inference made from simulated data and FECRT data taken from horses was compared to that made using the currently most widely used methods. Multi-level hierarchical models were then used to partition the source of the observed variability in FEC using data intensively sampled from a small group of horses. The MCMC methods out-performed other methods for analysis of simulated FEC and FECRT datasets, particularly in terms of the usefulness of 95% confidence intervals produced. There was no consistent difference in model fit to the gamma-Poisson or lognormal-Poison distributions from the available data. However there was evidence for the existence of bi-modality in the datasets. Although the majority of the observed variation in equine FEC is likely a consequence of variability between animals, a considerable proportion of the variability is due to the variability in true FEC between faecal piles and the aggregation of eggs on a local scale within faeces. The methods currently used for analysis of FEC and FECRT data perform poorly compared to MCMC methods, and produce 95% confidence intervals which are unreliable for datasets likely to be encountered in clinical parasitology. MCMC analysis is therefore to be preferred for these types of data, and also allows multiple samples taken from each animal to be incorporated into the analysis. Analysing the statistical processes underlying FEC data also revealed simple methods of reducing the observed variability, such as increasing the size of individual samples of faeces. Modelling the variability structure of FEC data, and use of the inferred parameter values in precision analysis and power analysis calculations, allows the usefulness of a study to be quantified before the data are collected. Given the difficulties with analysing FEC and FECRT data demonstrated, it is essential that such consideration of the statistical issues pertaining to the collection and analysis of such data is made for future parasitological studies.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:513286 |
Date | January 2010 |
Creators | Denwood, Matthew James |
Publisher | University of Glasgow |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://theses.gla.ac.uk/1837/ |
Page generated in 0.0064 seconds