Delignette-Muller M L, Rosso L, Flandrois J P
Laboratoire de Bactériologie, CNRS URA 2055, Faculté de médecine, Oullins, France.
Int J Food Microbiol. 1995 Oct;27(2-3):139-46. doi: 10.1016/0168-1605(94)00158-3.
The results of growth predictions using square root and polynomial models published in 14 papers were studied. Errors on quantities of practical interest such as lag time, generation time or the time required to reach a given increase in number of cells, are analyzed. The distribution of these errors was examined with the perspective of the practical use of predictive models in food industry. Highly unsafe predictions and significant average errors were observed in some cases. A good knowledge of predictive models accuracy seems essential for their efficient and safe use, for example to predict the shelf life of a product. Yet, authors generally gave no pragmatic information on such things as the average relative error or the range of errors on predicted variables. Problems of robustness of models when tested in different conditions were noticed, which corroborates the necessity of a systematic validation of models on new data.
