Submitted to: Journal of Food Processing and Preservation Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/19/2021
Publication Date: 7/28/2021
Citation: Oscar, T.P. 2021. Development and validation of a neural network model for growth of salmonella newport from chicken on cucumber for use in risk assessment. Journal of Food Processing and Preservation Research. e15819. https://doi.org/10.1111/jfpp.15819.
Interpretive Summary: Salad is a popular side dish served with chicken. However, cross-contamination of salad with Salmonella from utensils (cutting board, knife, hands) used to process raw chicken for cooking followed by growth of the pathogen on salad before serving could lead to foodborne illness. Growth of a low number (7 cells) of a chicken isolate of Salmonella Newport on cucumber and Romaine lettuce as a function of times (0 to 8 h) and temperatures (16 to 40C) encountered during meal preparation and serving was investigated and modeled. The resulting computer model can be used to predict the amount of growth of Salmonella on salad under different scenarios of meal preparation and serving and thus, can be used to assess the safety of salad before consumption. In addition, the model fills an important data gap in risk assessments conducted by regulatory agencies to develop policies that protect the food supply and public health.
Technical Abstract: Cross-contamination of ready-to-eat salad fruits and vegetables with Salmonella from utensils used to prepare raw chicken for cooking followed by pathogen growth during holding of salad at room temperature before serving is an important risk factor for salmonellosis. The current study was undertaken to develop and validate a model for growth of Salmonella on salad ingredients for use in quantitative microbial risk assessment (QMRA) and to provide a perspective review of the approach used. Portions (0.2 g) of cucumber (mesocarp) and Romaine lettuce with native microflora were inoculated with a low initial number (0.85 log MPN/portion) of Salmonella Newport (chicken isolate). Kinetic data for model development and validation were obtained using an automated, whole sample enrichment, miniature most probable number (WSE-mMPN) method. A replicated (n = 10), full 5 x 7 factorial design of time (0, 2, 4, 6, 8 h) and temperature (16, 20, 24, 28, 32, 36, 40C) was used for model development, whereas a replicated (n = 6), full 4 x 6 factorial design of time (1, 3, 5, 7 h) and temperature (18, 22, 26, 30, 34, 38C) was used for model validation (interpolation). A neural network (multiple layer feedforward with two hidden layers of two nodes each; MLF22) model was developed in an Excel spreadsheet using NeuralTools, a spreadsheet add-in program. The neural network (MLF22) model was validated using the test data, model performance, and model validation criteria of the Acceptable Prediction Zones (APZ) method in the Validation Software Tool (ValT) for predictive microbiology. When the proportion of residuals (observed log MPN/portion – predicted log MPN/portion) in the APZ (pAPZ) was = 0.70, the model was considered to provide predictions with acceptable accuracy and bias for the test data evaluated. The model was developed and validated using the combined data for cucumber and Romaine lettuce because only minor differences in growth of Salmonella Newport on the two food matrices was observed. The data for model development (n = 350) and validation (n = 144) met all criteria for test data in the APZ method. The overall pAPZ was 0.94 for model development and 0.92 for model validation (interpolation) and there were no local prediction problems. Therefore, the neural network (MLF22) model was considered to provide predictions with acceptable accuracy and bias (validated) for growth of a low initial number (0.85 log MPN/portion) of Salmonella Newport on salad fruit (cucumber) and vegetable (Romaine lettuce) portions (0.2 g) with native microflora for the times (0 to 8 h) and temperatures (16 to 40C) used in model development and validation and can be used with confidence in QMRA.