A bulletin for the Australian Food Industry December 1996
Contents: Enterohaemorrhagic E. coli outbreak in Japan | Cross-contamination and EHEC | Repercussions from food poisoning incidents | Statistical methods and food safety | National food hygiene standard | Botulism linked to cheese
Statistical methods and food safety
Three areas of food safety which require the use of statistical methods are predictive microbiology, inter-laboratory trials and sampling.
Predictive microbiology
Predictive microbiology is concerned with the construction of mathematical models which allow a food manufacturer to predict the microbiological safety and shelf life of a food, given certain food properties and environmental factors. These would include, for instance, the pH and water activity of the food and the range of temperatures the food is subject to during storage.
Statistical techniques such as non-linear regression and response surface models are used to consolidate and extrapolate experimental data. For example sigmoid curves may be fitted to logarithms of microbial counts versus storage time. Two such sigmoid curves are generated by the Gompertz and Baranyi models and figure 1 compares the fit of these models under stated conditions to the observed growth of salmonellae.
FIGURE 1
Example of sigmoid curves - growth
rates for salmonellae
Quality assurance in the laboratory
As a quality assurance measure, machines and personnel involved in routine testing for food hazards (microbial or chemical) are subject to interlaboratory trials. Different laboratories each receive sub-samples of the one sample and are asked to determine the presence or level of a known food hazard e.g. salmonellae, lead, aflatoxin, the testing for which they routinely conduct. The laboratory measurements should of course agree with one another within acceptable experimental error, for the common method used.
Laboratories producing atypical results can be identified clearly using statistical techniques. For example, if each laboratory has been asked to determine the number of coliforms present in two samples of milk powder, then a Youden diagram (figure 2) can identify any outlying results from any laboratory.
FIGURE 2
Milk powder coliform count
While HACCP systems are now seen as the basis for food safety assurance, there are many circumstances when end product testing is still necessary. One such case is when a large batch of canned foods is imported from a manufacturer who may not have a HACCP system in place or when a purchase of a consignment is made through a third party.
If 'x' cans, where 'x' is greater than 'c' (a critical value) out of an initial sample of 'n' cans are unsatisfactory, then the import authority or purchaser may decide to reject the total batch or order a more stringent sampling procedure. The values of 'n' and 'c' are chosen on the basis of statistical theory which allows the probability of acceptance of a defective product being accepted to be determined, given the actual percentage of defective cans in the batch.
The stringency of sampling will of course be determined by the risk posed by the particular defect detected.
Food Science Australia
PO Box 52, North Ryde 1670. Tel +61 2 9490 8397 Fax +61 2 9490 8499
Email enquiries@csiro.au