Comparison of bacterial culture, polymerase chain reaction, and a mix-enzyme-linked immunosorbent assay for the detection of Salmonella status in grow-to-finish pigs in western Canada with a Bayesian approach.

Abstract

Among grow-to-finish pigs from 10 herds in Alberta and Saskatchewan, 23 (16%) of 144 fecal samples were culture-positive and 40 (28%) of 144 pigs were seropositive for Salmonella. With a Bayesian model specifying dependence between the 2 tests, the sensitivity (Se) of culture and real-time polymerase chain reaction (RT-PCR) was 79% to 86%, depending on the cut-off value for the enzyme-linked immunosorbent assay (ELISA). Culture specificity (Sp) was assumed to be 100%; RT-PCR Sp was found to be 94%. The ELISA Se was 76% and 51% at optical density cut-off values ≥ 20% and ≥ 40%, respectively; the Sp was 94% at each cut-off value. The model showed some sensitivity to ELISA prior information, the ELISA Se being approximately 8% lower when informative prior information was specified in the model. When there was no adjustment for dependence between culture and RT-PCR, the posterior estimates for both culture and RT-PCR Se were 11% higher than with the conditional-dependence model and had considerably narrower probability intervals, which suggests that correlation between culture and PCR is important and should be adjusted for in future studies.