Submitted to: Journal of Food Protection
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/6/2016
Publication Date: 1/1/2017
Citation: Oscar, T.P. 2017. Neural network model for thermal inactivation of Salmonella Typhimurium to elimination in ground chicken: Acquisition of data by whole sample enrichment, miniature most-probable-number method. Journal of Food Protection. 80(1):104-112. doi: 10.431510362-028x.jfp-16-199.
Interpretive Summary: Poultry meat and egg products are often linked to cases of salmonellosis. If ground chicken is not thoroughly cooked, the deadly pathogen, Salmonella, can survive and cause salmonellosis. Predictive models are valuable tools for assessing food safety. However, existing cooking models for death of Salmonella in ground chicken do not provide predictions at the recommended final cooked temperature of 165F. In addition, they do not predict when all Salmonella are dead. The predictive model developed in this research successfully addresses these limitations of existing models. Thus, the new model will be a valuable tool for the chicken industry and consumers to predict and manage this important risk to public health.
Technical Abstract: Predictive models are valuable tools for assessing food safety. Existing thermal inactivation models for Salmonella and ground chicken do not provide predictions above 71 degrees C, which is below the recommended final cooked temperature of 73.9 degrees C. They also do not predict when all Salmonella are eliminated without extrapolating beyond the data used to develop them. Thus, a study was undertaken to develop a model for thermal inactivation of Salmonella to elimination in ground chicken at temperatures above those of existing models. Ground chicken thigh portions (0.76 cm3) in microcentrifuge tubes were inoculated with 4.45 +/- 0.25 log MPN of a single strain of Salmonella Typhimurium (chicken isolate). They were cooked at 50 to 100 degrees C in 2 or 2.5 degrees C increments in a heating block that simulated pan frying. A whole sample enrichment, miniature most probable number (WSE-mMPN) method was used for enumeration. The lower limit of detection was one Salmonella per portion. MPN data were used to develop a multiple layer feedforward neural network model. Model performance was evaluated using the acceptable prediction zone (APZ) method. The proportion of residuals in an APZ (pAPZ) from -1 log (fail-safe) to 0.5 log (fail-dangerous) was 0.913 (379/416) for dependent data and 0.910 (162/178) for independent data for interpolation. A pAPZ greater than or equal to 0.7 indicated that model predictions had acceptable bias and accuracy. There were no local prediction problems as pAPZ for individual thermal inactivation curves ranged from 0.813 to 1.000. Independent data for interpolation satisfied the test data criteria of the APZ method. Thus, the model was successfully validated. Predicted times for a one log reduction ranged from 9.6 min at 56 degrees C to 0.71 min at 100 degrees C. Predicted times for elimination ranged from 8.6 min at 60 degrees C to 1.4 min at 100 degrees C. The model will be a valuable new tool for predicting and managing this important risk to public health.