Structure and Organization of Centres

Abstract

Increasing evidence suggests that persistence of Listeria monocytogenesin food processing plants has been the underlying cause of a number ofhuman listeriosis outbreaks. This study extracts criteria used by foodsafety experts in determining bacterial persistence in the environment,using retail delicatessen operations as a model. Using the Delphimethod, we conducted an expert elicitation with 10 food safety expertsfrom academia, industry, and government to classify L. monocytogenespersistence based on environmental sampling results collected over sixmonths for 30 retail delicatessen stores. The results were modeled usingvariations of random forest, support vector machine, logisticregression, and linear regression; variable importance values of randomforest and support vector machine models were consolidated to rankimportant variables in the experts' classifications. The duration ofsubtype isolation ranked most important across all expert categories.Sampling site category also ranked high in importance and validationerrors doubled when this covariate was removed. Support vector machineand random forest models successfully classified the data with averagevalidation errors of 3.1% and 2.2% (n = 144), respectively. Ourfindings indicate that (i) the frequency of isolations over time andsampling site information are critical factors for experts determiningsubtype persistence, (ii) food safety experts from different sectors maynot use the same criteria in determining persistence, and (iii) machinelearning models have potential for future use in environmentalsurveillance and risk management programs. Future work is necessary tovalidate the accuracy of expert and machine classification againstbiological measurement of L. monocytogenes persistence.