Campylobacter Risk Analysis: Do models speak louder than data?

In several countries quantitative microbiological risk assessments (QMRAs) have been performed for Campylobacter in chicken meat. The models constructed for this purpose provide a good example of the development of QMRA in general and illustrate the diversity of available methods. Despite the differences between the models, the most prominent conclusions of the QMRAs are similar. These conclusions for example relate to the large risk of highly contaminated meat products and the insignificance of contamination from Campylobacter positive flocks to negative flocks during slaughter and processing. Nonetheless, there seems to be a discrepancy between model predictions and the accumulating microbiological data. For example, a recent study in the Netherlands showed that model predictions on the efficacy of “testing and scheduling” of broiler flocks as a control strategy, could not be confirmed by quantitative data. The easy way out of this problem is to conclude that the models are inaccurate. However, this may be too simple. The models are based on the best knowledge and insights of a diverse set of scientists and poultry processing experts, combined with some logical argumentation. This implies that if model results cannot be validated, this knowledge and insights are somehow insufficient. Also, microbiological data may suffer from a variety of errors, related to isolation of the pathogens, heterogeneity of the samples, the sampling procedure, test sensitivity, recovery, disturbing flora, etc. In this presentation the widely spread trust in data among microbiologists will be challenged, and compared with the sound logic of QMRA. It will be discussed how bridging the gap between data and models should be considered a joint effort of microbiologists and modellers. The strength of mathematical models in generating and testing hypotheses in food microbiology will be advocated.