A hierarchical Bayesian approach to estimate the most probable number (MPN) concentration of Salmonella in raw...

Estimating bacterial concentrations in food products is a key element in quantitative microbial risk assessment (QMRA). A Bayesian probabilistic approach with the most probable number (MPN) technique was developed to estimate the concentration of Salmonella in chicken meat based on the qualitative data. Salmonella contamination within a batch of food samples tends to lack homogeneity, and a zero-inflated distribution, i.e., zero-inflated Poisson (ZIP) or zero-inflated negative binomial (ZINB) distribution, was subsequently assumed as the underlying distribution of the target microorganism in food. Meanwhile, observations at low contamination levels are often censored due to the limit of determination of the microbiological detection methods. A hierarchical structure was included in the Bayesian estimation method in order to take into account the majority of the censored data. The performance of these methods was compared in two observed Salmonella datasets. The qualitative detection rate of Salmonella in dataset A was 31.25% and that in dataset B was 30.00%. On one hand, the results showed that the estimation method would generate biased estimated parameters once the censored data were ignored. On the other hand, in case of heavy censoring, the ZIP distribution was more adaptable to estimating Salmonella concentration from qualitative data than the ZINB distribution. The mean concentrations of Salmonella obtained by the proposed ZIP-based hierarchical Bayesian method was 0.43 MPN/g for dataset A and 0.44 MPN/g for dataset B. Our alternative method may be preferred for using qualitative data in QMRA, especially with a low bacterial contamination level and a large number of censored data.