A quantitative assessment of the validity of animal-health surveys using stochastic modelling.

This paper presents a stochastic simulation model to evaluate the efficacy of regional or national surveys aimed at identifying infection in populations of animals. The process of evaluation involves specification or calculation of cluster-level test sensitivity and specificity, which are derived from two probability distributions of the number of individual-level positive tests expected from non-infected and infected clusters, respectively. Probability distributions for the number of positive clusters expected in a situation of freedom from infection and under various levels of cluster prevalence are specified and used to determine survey properties (the survey being considered a diagnostic system), and ROC curves are drawn. Likelihood ratios allow investigators to state the extent to which a survey result is more likely to be observed if the region or country is infected at a given prevalence than if it is free from infection. The result of a survey carried out to investigate the presence of porcine reproductive and respiratory syndrome (PRRS) in Switzerland is used to illustrate this approach. The model can be adapted to a wide range of survey designs.