A similarity measure to assess the stability of classification trees
Microbial food safety process risk models are simplifications of the real world that help risk managers in their efforts to mitigate food safety risks. An important tool in these risk assessment endeavors is sensitivity analysis, a systematic method used to quantify the effect of changes in input variables on model outputs. In this study, a novel sensitivity analysis method called classification and regression trees was applied to safety risk assessment with the use of portions of the Slaughter Module and Preparation Module of the E. coli O157:H7 microbial food safety process risk as an example. Specifically, the classification and regression trees sensitivity analysis method was evaluated on the basis of its ability to address typical characteristics of microbial food safety process risk models such as nonlinearities, interaction, thresholds, and categorical inputs. Moreover, this method was evaluated with respect to identification of high exposure scenarios and corresponding key inputs and critical limits. The results from the classification and regression trees analysis applied to the Slaughter Module confirmed that the process of chilling carcasses is a critical control point. The method identified a cutoff value of a 2.2-log increase in the number of organisms during chilling as a critical value above which high levels of contamination would be expected. When classification and regression trees analysis was applied to the cooking effects part of the Preparation Module, cooking temperature was found to be the most sensitive input, with precooking treatment (i.e., raw product storage conditions) ranked second in importance. This case study demonstrates the capabilities of classification and regression trees analysis as an alternative to other statistically based sensitivity analysis methods, and one that can readily address specific characteristics that are common in microbial food safety process risk models.