Validation of Lag Time and Growth Rate Models for Salmonella Typhimurium: Acceptable Prediction Zone Method

摘要

The prediction bias (B_f) and accuracy (A_f) factors are the most widely used measures of performance of predictive models for food pathogens. However, B_f and A_f have limitations that can produce inaccurate assessments of model performance. Consequently, an objective of the current study was to develop a method for quantifying model performance that overcomes limitations of B_f and A_f Performance of published lag time and growth rate models for Salmonella Typhimurium were evaluated for data used in model development and for data not used in model development but that were inside (interpolation) or outside (extrapolation) the response surface of the models. In addition, performance of published models for growth of Escherichia coli O157:H7 was evaluated for data used in model development. Observed and predicted values were compared using B_f, A_f, and pRE, a new performance factor that quantified the proportion of relative errors (RE) in an acceptable prediction zone from an RE of -0.3 (fail-safe) to 0.15 (fail-dangerous). A decision diagram based on criteria for test data and model performance was used to validate the models. When B_f and A_f were used to quantify model performance, all models were validated. In contrast, when pRE was used to evaluate model performance, 2 models for S. Typhimurium and both models for E. coli O157:H7 failed validation. Overall, pRE was a more sensitive and reliable indicator of model performance than B_f and A_f because unacceptable pRE, which indicated a performance problem, were obtained for 8 of 20 evaluations, all of which had acceptable B_f and A_f. A limitation of pRE was the inability to distinguish between global and regional prediction problems. However, when used in combination with an RE plot, pRE provided a complete evaluation of model performance that overcame limitations of B_f and A_f.