Experimental Design and Mathematical Modelling Methods for the Study of Anthelmintic Resistance in UK Livestock Populations

Love, Johnathan and Kelly, Louise and Robertson, Chris (2019) Experimental Design and Mathematical Modelling Methods for the Study of Anthelmintic Resistance in UK Livestock Populations. PhD thesis, UNSPECIFIED.

[img]
Preview
Text (Love-Strathclyde2019-Experimental-Design-and-Mathematical-Modelling-Methods-for)
Love_Strathclyde2019_Experimental_Design_and_Mathematical_Modelling_Methods_for.pdf
Final Published Version

Download (9MB)| Preview

    Abstract

    The Faecal Egg Count Reduction Test (FECRT) is the most widely used field-based method for estimating anthelmintic efficacy and as an indicator of anthelmintic resistant nematodes in cattle. In this thesis, statistical aspects such as the analysis of cattle faecal egg count (FEC) data and the identification of statistically robust experimental study designs for the FECRT, are examined. Using field study cattle FECRT data, the validity of current guidelines on parameter estimates for evaluating percentage estimates and confidence intervals (CIs), were assessed. For FECs obtained using sensitive counting techniques, percentage estimates are recommended to be evaluated using arithmetic group means. For FECs obtained by less sensitive counting techniques, the maximum likelihood estimator of zero inflated distributions is recommended when evaluating percentage estimates. It would not be recommended however, to use CIs that assume FECs to be normal, and it is therefore recommended that relevant intervals for percentage estimates be obtained using a Bootstrap or Bayesian framework. A simulation study was conducted using Bootstrap methodology to assess the coverage probability of 95% percentile intervals, associated with different percentage estimates. The coverage was considered for scenarios involving various diagnostic sensitivities, treatment group sizes and classifications of pre-treatment group means. Very few scenarios consisted of 95% Bootstrapped percentile intervals with adequate coverage probabilities. A further simulation study was then carried out with Bayesian methodologies being employed. The accuracy of percentage estimates was examined under the scenarios described above. In the majority of scenarios: in order to obtain the most accurate percentage estimates, one would only need to adopt a paired study design involving a positive treatment group.