M ost validation studies for WSSV PCR have been limited to analytical experiments and those that used diagnostic validation procedures, followi ng the pathway prescribed by the World Organisation for Animal Health (OIE) in chapter 1.1.2. of the Manual of Diagnostic Tests for Aquatic Animals, assumed that the reference test was perfect. One example of the latter is the two PCR tests for WSSV that are certified as "fit for purpose" in the OIE test registry. Bayesian latent class models (LCM), which don't require an assumption of a perfect reference test, provide a practical solution for statistical analysis of the accuracy of 2 tests (++, +-, -+, and --) in at least 2 distinct populations and allow for inclusion of relevant prior knowledge about diagnostic sensitivity (DSe) and specificity (DSp) of the tests under evaluation. The analysis can be readily implemented in open-source software such as OpenBUGS (http://www.openbugs.net/w/Downloads) but users should ensure that model assumptions of conditional independence or dependence, constant diagnostic accuracy across populations, and distinct prevalences are critically assessed in each analysis and appropriate code is used to run the model.
T he purpose of the present study was to diagnostically valid ate two PCR (OIE test and CSIRO Taqman qPCR) for purpose of surveillance in apparently-healthy shrimp and shrimp products assuming neither test was perfect. Previously, both tests had been analytically validated to the end of stage 1 of the OIE test validation pathway. The DSe and DSp of the two qPCR assays were estimated for 2 conditionally-dependent tests in 2 populations assuming that neither test was perfect (Branscum et al. Prev. Vet Med 2005: 68:145-163 ). A conditional dependence model was preferred to an independence model because both PCRs target similar sequences albeit in different locations in the WSSV genome. The conditional dependence model requires informative prior information on at least 2 parameters to help ensure model identifiability (i.e. a unique set of parameter values generated the joint test results , namely the 4 cell counts in the cross-classified data tables ). Informative beta (a,b) distribution s were specified for the DSe (beta 130.7, 15.4 ) and DSp (beta 15.7, 1.3) of the OIE qPCR based on the expert opinion of a co-author (NM). The priors for the OIE qPCR were readily justifiable as that assay had been used in the AAHL Fish Diseases Laboratory in Geelong since 2010 and relevant test accuracy data had been obtain ed in a preliminary study. Priors for the other model parameters were flat (beta 1,1), indicating that any value between 0% and 100% was equally likely.
The median DSe and DSp values and 95% probability intervals were estimated for 3 testing scenarios that would be expected if the assays were used in other laboratories including duplicate wells that were both positive for a positive interpretation. The Bayesian model estimates of DSe for the 2 assays were almost identical: 88.9% (CSIRO) and 89.9% (OIE) and both tests had DSp of 97.6%. In conclusion, both assays had comparable DSe and DSp and were deemed fit for the purpose of surveillance for WSSV in apparently-healthy infected prawns and prawn products.