Bayesian latent class analysis of diagnostic sensitivity and specificity of tests for surveillance for bacterial kidney disease in Atlantic salmon Salmo salar

Abstract

Bacterial kidney disease (BKD) diagnosis is often carried out by IFAT, ELISA, and PCR because these tests yield rapid results compared with bacterial culture for the causative agent, Renibacterium salmoninarum. However, the diagnostic sensitivity (DSe) and specificity (DSp) of these tests is generally unknown, particularly in subclinically infected Atlantic salmon. Bayesian models were used to analyse test results from 454 field samples sourced from salmon populations with different infection status and submitted to independent laboratories. Results showed that DSe was highly dependent on the presentation of infection. When applied to fish with BKD-associated gross lesions, the highest to lowest posterior medians for DSe were: PCR2 (0.98), PCR1 (0.97), bacterial culture (0.86), ELISA (0.82) and IFAT (0.75). When applied to fish without lesions, the posterior medians for DSe were: PCR1 (0.82), PCR2 (0.82), bacterial culture (0.29), IFAT (0.29) and ELISA (0.21). The posterior medians for DSp were similar in populations with and without lesions. ELISA had the highest DSp (0.99), followed by culture (0.98–0.99), IFAT (0.96), PCR1 (0.94–0.96) and PCR2 (0.83–0.84), where different numbers reflect the estimates for populations with or without lesions respectively. Among tests that were evaluated for intra-laboratory repeatability (ELISA, PCR1 and IFAT), ELISA had the highest repeatability (kappa=0.82). We concluded that PCR tests were the fittest of the evaluated tests for the purpose of screening subclinically-infected adult Atlantic salmon.