Richard Earnshaw Department of Microbiology, Campden & Chorleywood Food Research Association, Chipping Campden, Glos, G155 6LD, UK
Introduction
For many years, expert food microbiologists have relied on a range of broad rules and personal experience to assist in the development of new products and the management of microbiological risk. Biological systems such as those controlling microbial survival and growth are complex. Reliance on simple rules and experience can have limited value in many of the day-to-day situations encountered by industrial microbiologists, particularly those, which involved the interplay of several factors in process and product formulation. Traditionally, food microbiologists would use slow and expensive challenge tests or microbiological experiments to assist in making critical decisions concerned with multiple factor microbial control systems. There was clearly a need to provide a flexible and reliable model to assist in critical decision making. Computer simulation of microbial growth and survival characteristics was an obvious way forward and over the last 10 years many of the world's top food microbiologists have embarked on a productive exercise to establish useful computer based growth/survival models or predictive models as they are more frequently called. Mathematical models of microbial growth and inactivation are not new. Indeed, the pioneers of the science of microbiology were successful in defining microbial growth and death kinetics as mathematical expressions under simple conditions. The revolution we have seen over the last 10 years is principally based on two factors:
- the commitment of the food microbiology community to systematically provide reliable models which can be influenced by food formulation and process variables and
- the provision of these tools as easy to use computer applications.
The alternatives to predictive modelling
In order to understand the benefits of the predictive modelling approach we could consider the alternatives:
- Expensive and slow experiments
- "Expert" estimation
- Comparison with the published literature
- Trial and error
- Keep to the "tried and trusted" approach - do not develop new products
- Get it wrong and pay the consequences
When viewed from this perspective, predictive modelling looks like an attractive and commercially desirable approach but it needs to be used with care. Predictive modelling is not the panacea for all the food microbiologist's problems. Models are only as good as the understanding upon which they are based and if used incorrectly it is possible to abuse the tools they provide. Predictive models need to be used with full appreciation of how they have been constructed and the potential pitfalls. In essence, predictive microbiological models will not answer all your questions but in some situations they can be used as effective decision making tools. In many cases limited experiments will be needed to back up critical decisions.
What is a predictive food microbiology model?
Mathematical models can take many forms and it is not the purpose of this paper to analyse the different mathematical approaches. In the majority of cases the models used in predictive food microbiology are kinetic descriptions of growth or death based on the principles of fitting curves to real experimental data. Mathematical formulae are then derived to predict how different factors interact and give rise to specific outcomes. Obviously this approach is time consuming and involves the collection of many data from experiments in controlled systems. In the majority of cases these experiments are performed in microbiological media and not real foods. At first sight this might seem to be an over simplistic and risky approach. Fortunately, in the majority of cases, validation studies which compare the descriptions of growth in controlled laboratory media studies with the behaviour of microbes in real foods show that such models are highly reliable and can perform useful functions. This element of model validation is very important if any degree of acceptance is to be placed on the predicted outcome of any model. The majority of commercially available models have been thoroughly validated and a knowledge of the validation and reliability of models when used for particular purposes is an important aspect of their use and application.
In addition to kinetic descriptions of growth and death, other models have been produced which present the user with a probability value for a particular circumstance eg. growth or death of a microbial population. These models are much cheaper to produce and can have useful merits under certain circumstances eg. predicting the probability of growth or survival of certain high risk microbes such as Clostridium botulinum or the likelihood of spoilage occurring.
What can predictive models tell you?
Depending on how models have been constructed there are several possible outputs from a predictive food microbiology model:
- Growth/death curve
- Lag time
- Time taken for a set level of increase/decrease in population
- Time take for a particular population level increase or decrease
- Probability of growth/survival/toxin production
In general, graphical representations of microbial growth or destruction are the most commonly used outputs from a predictive model because they provide a good overview of microbial behaviour and can be used to make effective decisions in a commercial context. A typical growth prediction is shown in Figure 1.
What can you change in terms of control parameters for any prediction?
Predictive models are essentially simulation techniques, which allow the determination of outcome for hypothetical circumstances. When applied to analyse problems it is obviously of benefit to be able to modify those circumstances to evaluate a range of options. The controlling parameters vary from model to model but in general, the commonly used predictive models allow modification of two or three factors which will include, temperature, pH, water activity and in some
circumstances levels of preservatives or protective gas atmosphere used in food packaging. Depending on needs, the model criteria can be changed to deliver the required information.
How do you obtain predictive models so that you can use them?
There are several approaches to accessing predictive models. Perhaps the simplest is to find a suitable model described in the scientific literature but this is not an easy option for the non-expert. From a practical point of view, the best way to use a model is to obtain the appropriate software package and run it on your own computer. There are some disadvantages here too, you will need some expertise to challenge the model with appropriate "questions" and you will need to consider the validation status of any prediction you might want to use for decision making. There are several commercially available software packages such as MicroModel which provides a collection of models for the important food pathogens and a few spoilage, microbes for which an annual fee is paid to use the software package on your own computer. The United States Department of Agriculture (USDA) has produced a collection of pathogen growth predictive models called the Pathogen Modelling Programme (PMP) and this is available free of charge from their Internet site (http://www.arserrc.gov).
The generic model packages that are available try to cover the, majority of possible applications but they cannot cover all eventualities. Before buying any model it is important to check that it will cover the organisms and control factors you wish to use.
Spoilage and shelf life
The majority of public domain predictive models focus on the growth and survival of pathogenic microbes for obvious reasons. It is also important to be able to predict the growth of food spoilage organisms when considering the likely stability and shelf life of food products. Campden & Chorleywood Food Research Association (CCFRA) has addressed this need by developing a collection of models which can be used to assess spoilage rates or likely stability. This collection of models is called Forecast and is available to potential users via an enquiry service (+44 (0)1386 842071) which runs the models on behalf of clients after a detailed consultation with respect to their needs. The consultancy aspect of this approach also allows subsequent expert interpretation and consideration of model validation status. Table 1 shows the range of models available within Forecast.
Table 1
|
Table 1
Current options for CCFRA Forecast Models |
| Model |
pH |
Salt (% w/v) |
Temperature °C |
| Bacillus spp |
4.0-7.0 |
0.5-10.0 |
5-25 |
| Pseudomonas spp |
5.5-7.0 |
0.0-4.0 |
0-15 |
| Enterobacteriaceae |
4.0-7.0 |
03-10 |
0-30 |
| (Additional models for yeast and lactic
acid bacteria to be added in late 1997 |
CCFRA is also developing predictive models focused on the needs of pickle and sauce manufacturers so that they can quickly assess the safety and stability of potential new formulations of differing pH, water activity, preservative concentration and specified thermal process. It is intended that these models will go some way to replacing the unreliable and inflexible CIMSCEE formula which is commonly used in this sector of the industry.
How reliable are the models?
As stated in the introduction, predictive food microbiology models will never be perfect mimics of real situations but they can be good enough to be of great practical benefit.
An important aspect to consider is the measure of reliability that any model provider can establish for similar circumstances or food categories. Assessing the reliability of a predictive food microbiology model can be a relatively simple comparison of predicted values versus real experimental data from food. It will not be possible to achieve total correlation but a good relationship should be expected. A typical validation plot is shown in Figure 2.
Model measurements can vary from food type to food type. Good models should be validated in a range of foods, which reflect their likely application and use.
Practical applications of models
Figure 1 shows how a model has been put to practical use by comparing predicted values of numbers against predetermined standards for termination of shelf life. Many other potential applications exist:
- What level of microorganisms will be present under different temperatures of storage?
- How much salt is needed to restrict microbial numbers to a pre-set level after 1 week storage at 8°C?
- What effect will a 10°C increase in cooking temperature have in terms of microbial reduction?
- How can the lag time for a particular pathogen be increased?
- How low does the water activity of a particular food need to be to stop yeast spoilage?
The list of potential queries is endless but it is always sobering to remember that an inappropriate "question" will present a user with an inappropriate "answer". The need to have an expert microbiologist involved in the operation and application of predictive models is self-evident.
Predictive models should be used as decision making tools to allow productive focusing of effort in process and product development and risk and hazard assessment. They can be of great value in complex HACCP studies if used correctly. They should be followed up with targeted practical trials and challenge tests. Used in this way, predictive models can be powerful tools for industrial food microbiologists.
Predictive models also have a role to play in education in that they allow simple demonstration of microbial behaviour and risk without the need for expensive laboratory exercises. Their use in the training of non-experts in the food industry remains to be explored but some university courses are beginning to cover the topic and provide students with opportunity to use and learn from the models.
The Golden Rules of predictive microbiology
- Use predictive models to give initial estimations of microbiological behaviour - apply them to focus attention on the useful domains of control variables
- Always consider the reliability and validation status of any model -remember that no model is perfect
- Use limited experimentation/challenge tests to reinforce critical decisions made on the strength of model predictions
- Involve an expert microbiologist when planning the use of models, interpreting model predictions and making important microbiological decisions
- Consider the use of predictive models if you want to speed up decision processes and reduce costs
The future
It is certain that a greater diversity of useful models with greater reliability with become available in future years. The current interest in quantitative risk assessment will undoubtedly stimulate this development. As we learn more about model construction and use it is likely that the cost of production will come down opening up many new opportunities. Sector specific models are starting to emerge and the popularity of the internet and on-line services will make predictive models more widely available in the coming years.
Conclusions
Predictive models are powerful tools which can be used in product and process development, shelf life management, risk assessment, HACCP and education of microbiologists and technologists.
|
