Abstract
Early detection of piscirickettsiosis is an important purpose of government- and industry-based surveillance for the disease in Atlantic salmon farms in Chile. Real-time qPCRs are currently used for surveillance because bacterial isolation is inadequately sensitive or rapid enough for routine use. Since no perfect tests exist, we used Bayesian latent class models to estimate diagnostic sensitivity (DSe) and specificity (DSp) of qPCR and culture using separate two-test, single-population models for three farms (n = 148, 151, 44). Informative priors were used for DSp (culture (beta(999,1); qPCR (beta(98,2)), and flat priors (beta 1,1) for DSe and prevalence. Models were run for liver and kidney tissues combined and separately, based on the presence of selected gross-pathological signs. Across all models, qPCR DSe was 5- to 30-fold greater than for culture. Combined-tissue qPCR median DSe was highest in Farm 3 (sampled during P. salmonis outbreak (DSe = 97.6%)) versus Farm 1 (DSe = 85.6%) or Farm 2 (DSe = 83.5%), both sampled before clinical disease. Median DSe of qPCR was similar for liver and kidney, but higher when gross-pathological signs were evident at necropsy. High DSe and DSp and rapid turnaround-time indicate that the qPCR is fit for surveillance programmes and diagnosis during an outbreak. Targeted testing of salmon with gross-pathological signs can enhance DSe.
