Accuracy of testing strategies using antibody-ELISA tests on repeated bulk tank milk samples and/or sera of individual animals for predicting herd status for Salmonella dublin in dairy cattle.

There is currently no perfect test for determining herd-level status for Salmonella Dublin in dairy cattle herds. Our objectives were to evaluate the accuracy, predictive ability, and misclassification cost term of different testing scenarios using repeated measurements for establishing the S. Dublin herd status. Diagnostic strategies investigated used repeated bulk tank milk antibody-ELISA tests, repeated rounds of blood antibody-ELISA tests on non-lactating animals or a combination of both approaches. Two populations hypothesized to have different S. Dublin prevalences were included: (i) a convenience sample of 302 herds with unknown history of infection; and (ii) a cohort of 58 herds that previously tested positive to S. Dublin. Bulk milk samples were collected monthly for 6-7 months and serum were obtained from 10 young animals on two occasions, at the beginning and end of bulk milk sampling period. A series of Bayesian latent class models for two populations and comparing two tests were used to compare bulk milk-based to serum-based strategies. Moreover, Monte Carlo simulations were used to compared diagnostic strategies combining both types of samples. For each diagnostic strategy, we estimated the predictive values using two theoretical prevalences (0.05 and 0.25). Misclassification cost term was also estimated for each strategy using these two prevalences and a few relevant false-negative to false-positive cost ratios. When used for screening a population with an expected low prevalence of disease, for instance for screening herds with no clinical signs and no previous S. Dublin history, a diagnostic strategy consisting of two visits at 6 months interval, and with herd considered positive if bulk milk PP% ≥ 35 and/or ≥ 1/10 animals are positive on one or both visits could be used to confidently rule-out S. Dublin infection (median negative predictive value of 0.99; 95% Bayesian credible intervals, 95BCI: 0.98, 1.0). With this approach, however, positive results should later be confirmed with more specific tests to confirm whether S. Dublin is truly present (median positive predictive value of 0.36; 95BCI: 0.22, 0.57). The same diagnostic strategy could also be used confidently to reassess the S. Dublin status in herds with a previous S. Dublin history. When use for such a purpose, the predictive value of a positive result could be greatly improved, from 0.78 (95BCI: 0.65, 0.90) to 0.99 (95BCI: 0.94, 1.0) by requiring ≥ 1 positive result on both visits, rather than at any of the two visits.