Impact of microbial distributions on food safety I. Factors influencing microbial distributions and modelling aspects

Abstract

Relatively little is known about exactly how microorganisms are physically distributed in foods, yet these distributions determine the likelihood that a foodstuff will cause illness and the consequential public health burden. When a batch of food is sampled to determine the microbiological status of the batch, the effectiveness of the sampling programme is also related to the spatial distribution of the microorganisms that are being sampled for. In the absence of exact knowledge, generalising assumptions are often made as to the nature of the distributions. Better insight into the actual microbiological distributions may help to improve designing sampling plans and food safety management decision-making. This study discusses mechanisms influencing the spatial distributions of microorganisms in foods, three types of spatial distributions, i.e. regular, random, and clustered, the relationship between spatial distribution and frequency distributions, and the suitability of statistical distributions employed to model microbial distributions in foods. Commonly used statistical distributions, namely the Normal distribution, various types of the Poisson distribution, the Lognormal distribution, the Gamma distribution, the Negative Binomial distribution, and the Poisson-Lognormal distribution are examined and their strengths and weaknesses evaluated. Five specific criteria are proposed to assess the suitability of statistical distributions to model microbial distributions. These criteria require model outcomes to be non-negative, to allow zeros, to be discrete, to approximate Poisson and to approximate Lognormal. Especially the ability to model spatial clustering is investigated. It is concluded that the Poisson-Lognormal and the Negative Binomial are the most suitable statistical distributions given the suitability criteria proposed. However, the ultimate choice of the most suitable one should also depend on how well they fit actual observations.