Added value of experts' knowledge to improve a quantitative microbial exposure assessment model--Application to...

In a previous study, a quantitative microbial exposure assessment (QMEA) model applied to an aseptic-UHT food process was developed [Pujol, L., Albert, I., Magras, C., Johnson, N. B., Membré, J. M. Probabilistic exposure assessment model to estimate aseptic UHT product failure rate. 2015 International Journal of Food Microbiology. 192, 124-141]. It quantified Sterility Failure Rate (SFR) associated with Bacillus cereus and Geobacillus stearothermophilus per process module (nine modules in total from raw material reception to end-product storage). Previously, the probabilistic model inputs were set by experts (using knowledge and in-house data). However, only the variability dimension was taken into account. The model was then improved using expert elicitation knowledge in two ways. First, the model was refined by adding the uncertainty dimension to the probabilistic inputs, enabling to set a second order Monte Carlo analysis. The eight following inputs, and their impact on SFR, are presented in detail in this present study: D-value for each bacteria of interest (B. cereus and G. stearothermophilus) associated with the inactivation model for the UHT treatment step, i.e., two inputs; log reduction (decimal reduction) number associated with the inactivation model for the packaging sterilization step for each bacterium and each part of the packaging (product container and sealing component), i.e., four inputs; and bacterial spore air load of the aseptic tank and the filler cabinet rooms, i.e., two inputs. Second, the model was improved by leveraging expert knowledge to develop further the existing model. The proportion of bacteria in the product which settled on surface of pipes (between the UHT treatment and the aseptic tank on one hand, and between the aseptic tank and the filler cabinet on the other hand) leading to a possible biofilm formation for each bacterium, was better characterized. It was modeled as a function of the hygienic design level of the aseptic-UHT line: the experts provided the model structure and most of the model parameters values. Mean of SFR was estimated to 10×10(-8) (95% Confidence Interval=[0×10(-8); 350×10(-8)]) and 570×10(-8) (95% CI=[380×10(-8); 820×10(-8)]) for B. cereus and G. stearothermophilus, respectively. These estimations were more accurate (since the confidence interval was provided) than those given by the model with only variability (for which the estimates were 15×10(-8) and 580×10(-8) for B. cereus and G. stearothermophilus, respectively). The updated model outputs were also compared with those obtained when inputs were described by a generic distribution, without specific information related to the case-study. Results showed that using a generic distribution can lead to unrealistic estimations (e.g., 3,181,000 product units contaminated by G. stearothermophilus among 10(8) product units produced) and emphasized the added value of eliciting information from experts from the relevant specialist field knowledge.