Animal Health & Welfare

Testing the tests

Effective diagnostic tests must have the capability to accurately identify both positive and negative individuals. Where animals are to be vaccinated, then a test must Differentiate between Infected and Vaccinated Animals (i.e. be a DIVA test), since the regulatory impact and effect on population health of being unable to do so will be unacceptably severe. BioSS was tasked, in collaboration with a Contract Research Organisation (CRO) and the Universities of Cambridge and Aberystwyth, to design a field trial to establish the properties of a DIVA test for Bovine TB, recognising that both specificity and sensitivity are likely to differ between vaccinated and unvaccinated sub-populations.

For Bovine TB there is no perfect "gold standard" test. The standard field test lacks sensitivity, and the best test is expensive, requiring the animal to be killed and examined post mortem. Understanding what trial data could be available and how to analyse these effectively were critical in designing the study.

BioSS proposed that, for a small group of animals, both DIVA tests and post mortem results should be used to provide direct information on sensitivity and specificity. Subsequently, data from the two independent DIVA tests would be collected in parallel from larger populations of interest. Analysing the additional data by Bayesian latent class analysis, while using the initial results to define informative priors, would allow estimation of the test sensitivities and specificities in vaccinated and unvaccinated animals, along with the infection prevalence in both populations.

This hybrid method of data analysis supports identification of study designs that have sufficient numbers to quantify test characteristics to the prescribed precision, but which avoid the greater costs associated with collecting large amounts of post mortem data.

box plot graph of specficity estimates against sample sizeThe effect of increasing second phase sample size on precision and its variability for estimates of test specificity using two independent DIVA tests, given an initial 30 positive animals and 100 negative animals tested by both DIVA tests and post mortem examination. The upper (white) boxplot denotes the potential range of values for the upper limit of the 95% confidence interval (CI) of estimated specificity, given a specified sample size. The lower (blue) boxplot is the equivalent for the lower limit of the 95% CI. As the sample size increases, the limits of the 95% CI are more likely to be close to the true value: the horizontal red line.

Further details from: Giles Innocent and Iain McKendrick

Article date 2015

Consultancy Advice & Collaboration

Plant Science

Animal Health & Welfare

Ecology & Environmental Science

Human Health & Nutrition