Submitted to: Institute of Food Technologists
Publication Type: Abstract Only
Publication Acceptance Date: 2/1/2004
Publication Date: 8/1/2004
Citation: Oscar, T.P. 2004. Performance of growth models for salmonella and other pathogens. Institute of Food Technologists. 114D-2.
Technical Abstract: Performance evaluation of models for pathogen growth involves verification and validation for interpolation and extrapolation. Bias (Bf) and accuracy (Af) factors are the most common measures of model performance for food pathogens. Limitations of Bf and Af are that they do not detect some forms of prediction bias, no growth data are excluded from calculation of Bf resulting in overestimation of model performance and Bf and Af are mean values that are subject to bias by outliers. The objective of this study was to evaluate performance of growth models for Salmonella and other pathogens using a new performance factor, pBf, which overcomes limitations of Bf and Af. Observed and predicted lag times (LT) and maximum specific growth rates (SGR) were compared using Bf, Af and pBf, which was the proportion of Bf in a safe prediction zone from 0.7 (fail-safe) to 1.2 (fail-dangerous). Comparisons were made with data used in model development (verification) and data not used in model development (validation) but within (interpolation) or outside (extrapolation) the matrix of model variables. Performance of models for LT was less (P < 0.05) than performance of models for SGR as indicated by higher Af and lower pBf. Most models for growth of Salmonella had acceptable Bf (0.7-1.15), Af (1-1.45) and pBf (> 0.7) for verification, interpolation and extrapolation data. However, some models for growth of other pathogens had unacceptable performance even for verification data. Of the three performance indices, pBf was the most reliable indicator of model performance because it was better able to detect performance problems than Bf and Af. Models with pBf greater than 0.7 for verification, interpolation or extrapolation provided accurate and unbiased predictions of pathogen growth and thus, could be used with confidence in the food industry to predict food safety under the conditions tested.