The application of research synthesis and Bayesian methods to evaluate the accuracy of diagnostic tests for Salmonella in swine
This thesis presents the results of three complementary studies which were carried out to evaluate the accuracy of diagnostic tests for Salmonella in pigs. First, a research synthesis method approach, which included a systematic review, meta-analysis and meta-regression, was used to map out existing primary research investigating the accuracy of bacterial culture, antibody or antigen -capture ELISA, and PCR for Salmonella in pigs under field conditions.. Large statistical variability, limited methodological soundness and reporting precluded a quantitative synthesis of findings from multiple studies. The meta-regression identified significant factors, such as variations in test protocols, which explained much of the variability of reported estimates of test accuracy. The need for consistent use of a standard reference test is essential to ensure comparability of results generated in future studies. In the second study, the accuracy of a bacterial culture, real-time (RT) PCR, and a mix-ELISA for Salmonella in were evaluated in western Canadian nursery and grow-finish pigs using traditional and Bayesian statistical methods. Ten farrow-to-finish pig farms from Alberta and Saskatchewan were purposively selected based on their presumptive Salmonella status. Bacteriological culture, RT-PCR and a mix-ELISA were performed on feces and blood samples collected from grow-finish (n=294) pigs and pens. Bayesian estimates of test sensitivity (Se) and specificity (Sp) at the individual pig level were similar to traditional statistical estimates. Sensitivity of culture and RT-PCR ranged from 65-75%, PCR Sp was 98-99% and ELISA Se and Sp at a cutoff of OD¡Ý20% ranged from 59-63% and 84-87%, respectively. In the third study, Salmonella serovar distribution and risk factors for Salmonella shedding were investigated in breeding, nursery, and grow-finish pigs using the same 10 herds. Among 418 Salmonella isolates, most common serovars were Derby (28.5%), Typhimurium, var. Copenhagen (19.1%), and Putten (11.8%). More Salmonella were detected in pooled pen than individual pig samples, confirming that the use of pooled samples is more effective for detecting the full range of serovars that may be present on Canadian pig farms. Sows shed significantly more Salmonella than nursery or grow to finish pigs, suggesting that the breeding herd is an important source of Salmonella persistence. Pelleted feed and nose-to-nose pig contact through pens were also associated with increased Salmonella prevalence, indicating that these factors are relevant as control targets. The main advantages of research synthesis methods are increased power and precision in effect estimates and identification knowledge gaps and areas requiring further research. Bayesian methods for evaluating test accuracy are useful when there is no known "gold standard", which is often the case for zoonotic and food-borne pathogens. Both research synthesis and Bayesian methods are valuable tools for evaluating diagnostic test accuracy and should be more frequently used when developing monitoring and control programs in food safety.
DegreeDoctor of Philosophy (Ph.D.)
DepartmentLarge Animal Clinical Sciences
ProgramLarge Animal Clinical Sciences
Copyright DateSeptember 2009