Bayesian estimation of the diagnostic accuracy of a multiplex real-time PCR assay and bacteriological culture for 4 common bovine intramammary pathogens

Bacteriological culture (BC) is the traditional method for intramammary infection diagnosis but lacks sensitivity and is time consuming. Multiplex real-time PCR (mr-PCR) enables testing the presence of several bacteria and reduces diagnosis time. Our objective was to estimate bacterial species-specific sensitivity (Se) and specificity of both BC and mr-PCR tests for detecting bacteria in milk samples from clinical mastitis cases and from apparently normal quarters, using a Bayesian latent class model. Milk samples from 1,014 clinical mastitis cases and 1,495 samples from apparently normal quarters were analyzed by BC and mr-PCR. Two positive culture definitions were used: >= 1 cfu/0.01 mL and >= 10 cfu/0.01 mL of the specified bacteria. The mr-PCR was designed to simultaneously detect Staphylococcus aureus, Streptococcus uberis, Escherichia coli, and Streptococcus agalactiae. The priors used in our Bayesian model were weakly informative, with BC priors using the best available error data. Results were compared with those obtained using uniform priors for mr-PCR to test robustness. Weak and uniform priors gave about the same posterior distributions except for Strep. uberis from normal quarters and Strep. agalactiae. Multiplex real-time PCR Se on milk from clinical mastitis were lower than mr-PCR Se on milk from normal quarters. Multiplex real-time PCR Se was higher than BC on milk from normal quarters. Multiplex real-time PCR Se was generally lower than BC on milk from clinical mastitis and it varied by clinical severity. The estimate specificities of detection for all pathogens were >= 99%, regardless of sample type. The effect of milk sample preservation before testing was evaluated and may have been a factor that affected our observed results. A significant association was observed between sample age and mr-PCR results leading to reduced detection of E. coli and Strep. agalactiae in nonclinical samples. Differences in sample age between conduct of BC and of mr-PCR, did not concur with any apparent differences between Se estimates of the 2 tests. Further work should be done to extend these results to other PCR-based tests for detecting bacterial species in milk samples, for which presented results could be used as prior parameter distributions. Limits of sample handling and storage and the potential existence of substances in clinical case samples that may interfere with PCR reactions also are worth further investigation.