A Bayesian modelling framework to estimate Campylobacter prevalence and culture methods sensitivity: Application to a chicken meat survey in Belgium

Aims: To estimate the true prevalence of Campylobacter and the diagnostic sensitivity of routine detection methods by applying a Bayesian modelling approach. Methods and Results: Results from a Belgium-wide survey of Campylobacter contamination in chicken meat preparations (n = 656 samples) showed that Campylobacter was detected in 24.2% of the samples by enrichment, compared with 41% detected by direct plating. Combining positive results from both methods increased the apparent prevalence to 48.02%. Bayesian model was set up in WinBUGS software, the model estimates Campylobacter prevalence as 60% (95% Credibility interval (CI): 47-82%), and the sensitivity of enrichment culture and direct plating as 41% (95% CI: 31-52%) and 69% (95% CI: 50-85%), respectively. Conclusions: The parallel use of direct plating and enrichment culture adds value for Campylobacter detection from chicken meat preparations, but the false-negative results from each culture method must be taken into account. Significance and Impact of the Study: Monitoring data could be strongly biased by the microbiological techniques used to generate it. To circumvent this bias, we describe an applied Bayesian framework for better interpretation of Campylobacter survey data in view of the imperfect test characteristics of routine culture methods.