Validation software tool (VAlT)for predictive microbiology based on the acceptable prediction zone method

Authors: Oscar, Thomas

Submitted to: International Journal of Food Microbiology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/15/2020
Publication Date: N/A
Citation: N/A

Interpretive Summary: The food industry does an admirable job of producing food with low prevalence and numbers of human bacterial pathogens. However, improper storage of food during distribution and home storage can provide conditions for growth of these pathogens to high and dangerous levels. Models that predict growth of pathogens in food are valuable tools for assessing risk of illness from food as function of storage conditions. Validation of these models is an important step in their development because it provides users with confidence that predictions are reliable. A software tool called “vault” (ValT) developed in the present study will make it easier for model developers to properly validate their models using a comprehensive set of criteria developed by the U. S. Department of Agriculture. The impact will be safer food for the consumer.

Technical Abstract: The current study was undertaken to develop a validation software tool (ValT) for predictive microbiology that is based on the test data and model performance criteria of the acceptable prediction zone method developed by the U. S. Department of Agriculture. The new software application was developed in Excel. A published tertiary model for growth of Salmonella Typhimurium definitive phage type 104 on chicken skin was used as a case study. A model prediction was considered acceptable when its residual (observed – predicted) was in an acceptable prediction zone from -1 log (fail-safe) to 0.5 log (fail-dangerous). A model was considered to provide predictions with acceptable accuracy and bias when the proportion of residuals in the acceptable prediction zone (pAPZ) was = 0.700 and there were no local prediction problems. A local prediction problem occurred when pAPZ was < 0.700 for a single level of an independent variable (time or temperature). The overall pAPZ were 0.823 for dependent data (n = 384), 0.826 for interpolation data (n = 178), and 0.786 for extrapolation data (n = 196) to another food matrix (kosher chicken skin). Although overall pAPZ were acceptable, the model failed validation for dependent data, interpolation, and extrapolation because the dependent data and interpolation data did not satisfy all criteria for test data and because of local prediction problems for all sets of test data.