Use of sensitivity analysis to aid interpretation of a probabilistic Bacillus cereus spore lag time model applied to heat-treated chilled foods (REPFEDs)

The microbiological safety and quality of REfrigerated Processed Foods of Extended Durability (REPFEDs) relies on a combination of mild heat treatment and refrigeration, sometimes in combination with other inhibitory agents that are not effective when used alone. In this context, the output of a probabilistic model predicting the lag time of heat-treated Bacillus cereus spores under realistic heat-treatment profile and chilled supply-chain conditions, has been investigated using a sensitivity analysis technique. Indeed, knowing that there was uncertainty in the model (e.g. due to lack of data to build the model input probability density function), the objective of the analysis was to evaluate if the variability associated with some inputs (e.g. the consumers' refrigerator temperature values reported in Europe and US markets were different) had a significant impact on the model output, i.e. on the expected lag time of heat-treated B. cereus spores in REPFEDs. To do so, the uncertainty and variability associated with the various model inputs have been identified and then separated using a second order Monte Carlo decomposition. Concerning the variability, there was a significant difference between the chilled supply-chains (Europe, US) and between the raw material groups (low, medium or high contamination levels). For example, in the European market, after a heat treatment of 90 °C for 10 min, with a high raw material contamination level, the predicted 5th percentile of the lag time was 17 days, while it was 35 days with a low raw material contamination level. This was confirmed with an ANOVA. The impact of the uncertainty on the lag time has been illustrated graphically by building confidence intervals around its 5th percentile. A sensitivity analysis based upon uncertainty and variability decomposition is clearly a complex and time consuming exercise; however, it provides a greater confidence (greater transparency and better understanding) in the model output when making food safety decisions (e.g. determining the safe shelf-life of REPFEDs).