Bayesian Estimation for Diagnostic Testing of Biosecurity Risk Material in the Absence of a Gold Standard when Test Data are Incomplete

Abstract

Diagnostic testing is used by biosecurity officers for the detection and identification of plant and animal pathogens, often informing high-consequence decisions such as restricting the entry of trade goods. It is rare that such tests can be considered gold standards; however, uncertainty can be reduced by using the results of other tests, measuring performance on samples of known status and incorporating prior knowledge from expert judgement. This article presents an extension to the methods of Joseph et al. (Am J Epidemiol 141:263–272, 1995), and Dendukuri and Joseph (Biometrics 57:158–167, 2001) for Bayesian estimation in the absence of a gold standard test, which allows for the use of incomplete test data. This extension is demonstrated with a novel application: the case study of myrtle rust from Holliday et al. (Plant Dis 97:828–834, 2013), which involves samples from potential biosecurity risk material on importation pathways to Australia. The samples were tested at two laboratories, and prior estimates for pathway prevalence were obtained by expert elicitation. The Bayesian estimation was based on a model with and without covariances for the test results to assess the assumption of conditional independence. The results show that pathogen prevalence, diagnostic sensitivity and diagnostic specificity can be estimated using all available data even where some samples have been subject to only one of two available tests. The results also indicate the importance of consideration of the assumption of conditional independence. The findings enable diagnostic testing laboratories and decision makers to make use of all test results and to explicitly incorporate prior knowledge to estimate pathogen prevalence and test accuracy.