Use of a third class in latent class modelling for the diagnostic evaluation of five infectious salmon anaemia virus...

In the absence of a reference standard, a latent class model (LCM) was used in this study to assess diagnostic sensitivity (DSe) and specificity (DSp) of a recently developed reverse-transcription polymerase chain reaction (RT-PCR) for infectious salmon anaemia virus (ISAV). The study included 4 populations of Atlantic salmon, and to ensure the identifiability of the LCM, four additional detection methods were used in parallel including real-time RT-PCR (qRT-PCR), virus isolation (VI), indirect fluorescent antibody test (IFAT), and a lateral flow immunoassay (LFI). While a conventional LCM assumes DSe and DSp to be constant across the populations, Nérette et al. (2008) previously reported concerns about non-constant DSp of RT-PCR, which detects viral RNA from both active and inactive viral particles. It was suspected that some ISAV recovered fish may carry residual RNA and may be more likely to test positive compared to naïve fish. The various mixture distributions of the two sub-classes of non-infected fish would lead to a non-constant combined DSp estimate across populations. Within a Bayesian framework, the conventional two-class LCM was extended to three classes of infection stages (naïve non-infected, recovered non-infected carrying RNA, and infected). The resulting analysis confirmed the existence of three classes of fish with substantially different test performances for ISAV. For infected fish, DSe of RT-PCRs and VI approximated 90%, and antibody based assays were the least sensitive (DSe around 65%). Regardless of the test, the DSp estimates on naïve fish were all above 98% with LFI being in average the most specific. Only RT-PCR and qRT-PCR tested positive with the additional class of recovered fish (DSp around 30%). The true infectious status of this sub-class (i.e. viral RNA carriers) is debatable and requires further knowledge about ISAV infection dynamics at the fish level. Promising applications of multiple class estimates require adjustments of traditional test interpretation and further epidemiological knowledge of the infection dynamics at the population level.