Evaluation of post-mortem diagnostic tests for bovine tuberculosis using Bayesian latent class analysis

This study aimed to evaluate the performance of real-time PCR (qPCR), ELISA IDEXX™, and bacterial isolation as post-mortem diagnostic tests in animals with lesions compatible with bovine tuberculosis detected by Brazilian Federal Inspection Service as part of the bovine tuberculosis active surveillance. Bayesian latent class models were used to estimate diagnostic tests' sensitivity, specificity, correlations, predictive values and frequency of infected animals. Samples of tuberculosis-suggestive lesions collected by FIS sanitary inspection routine in slaughterhouses from 11 Brazilian states were analyzed. Isolation was the most sensitive technique, 94.54% (95% Credible Interval (CrI) 90.09%–97.65%), qPCR was 64.69% (95% CrI 54.41%–74.15%) sensitive and ELISA IDEXX™ 26.74% (95% CrI 22.82%–30.97%). Tests' specificities were 98.19% (95% CrI 95.75%–99.45%), 93.49% (95% CrI 79.28%–99.66%), 95.53% (95% CrI 91.71%–98.02%) respectively. Despite its low sensitivity, ELISA IDEXX™ was able to identify positive samples that were not detected by the other techniques. These samples had high probability to be true positives given ELISA's positive predictive value. The correlations between qPCR and isolation were neither biologically nor statistically significant. The low sensitivity of the qPCR is a limiting factor to its use as a post-mortem diagnosis in bovine tuberculosis suggestive lesions. Its use could be recommended in situations of high prevalence, or in parallel association with other tests, such as ELISA IDEXX™. ELISA IDDEX™ should not be used as a unique test, or in substitution of the other tests, for the post-mortem diagnosis of bovine tuberculosis due to its sensitivity.