Calculating contamination risk

There has been greater understanding of how microorganisms grow and thrive in foods. But despite advances, microbes continue to elude technologies, human illness spills over borders, and product spoilage still incurs economic loss.

The proper detection of pathogenic microbes has so concerned the WTO, that in 1995, it erected two agreements—the Sanitary and Phyto-Sanitary (SPS) Agreement and the Technical Barriers to Trade (TBT) Agreement—to contain food-borne illnesses with scientific knowledge and processes that determine risks. Both agreements forever changed the way that food safety is determined and the way that food is traded. They did this by relying on techniques that estimate changes in numbers of food-borne pathogens from their point of production or harvest—through processing, distribution, storage, sale and final preparation and, ultimately, consumption.

While this “farm-to-fork” approach is already well-established in HACCP, the use of predictive microbiology is relatively new. Translating these principles into practical approaches—under different processing conditions, across different nations—has proven to be a challenge. Now, a range of free online software tools can help manufacturers quantify their worst risks.

In this context, ‘risk’ has a specific number, giving meaning to the likelihood of an adverse event (eg a consumer gets food-borne illness) and the severity of that event. This differs from the idea of a hazard, which is simply the possibility that something could go wrong (eg salmonella could be in a food and it could make someone ill). Risk quantifies and ranks the importance of hazards, giving focus and resources to areas of greatest need, shifting attention from less probable events.
Risk plays a role in biosecurity, but its significance extends beyond preventing the introduction of exotic plant and animal pathogens. In essence, risk creates a measure of equivalence across different nations, making it easier to understand a risk’s potential effect on consumer health. Under the SPS and the TBT, the only valid reason to restrict import of foods would be a demonstrated, higher public health risk from the imported product, in comparison to its domestic counterpart; or the potential introduction of an exotic pest species.

Conceptually, the farm-to-fork assessment of food safety risk is relatively simple. It is summarized in “The ICMSF Equation”:
Ho - ∑R + ∑I ≤ PO (or FSO) (Equation 1)

In plain language, the equation says that for the food to be safe for human consumption, the sum of the initial microbial contamination level (H0) less the sum of reductions in microbial load (“∑R”, eg due to dilution, inactivation, etc) plus the sum of increases (“∑I”, eg due to recontamination, concentration, growth, etc) should remain below levels that are defined (explicitly or implicitly) in a Performance Objective (PO) or Food Safety Objective (FSO). A Food Safety Objective defines the acceptable (microbiological) status of the food at the point of consumption, whereas a Performance Objective defines the maximum frequency and/or concentration of a hazard in a food at a specified step in the food chain at some point before consumption. The FSO specified should be below the level that would lead to human illness. Conformance with the PO should mean that the FSO is achieved.

While these concepts are easy to define, setting a FSO and translating it into practical POs for industry depends on extensive knowledge of the ecology of microbial pathogens in foods. Microbes can grow, survive, or die depending upon the conditions they experience. Those same conditions can affect different organisms in different ways. Each species has it own limits to growth and conditions for death, for different environmental factors. Growth, death or survival depends on the specific organism and its strain type; food composition and additives (eg pH, water activity, presence of organic acids, etc); presence of other microbes in the food; processing steps; and storage and packaging conditions (eg temperature, gas mixtures, vacuum packing, etc). Collectively, responses to these factors constitute the ecology of the microbe in the food, and their interactions and effects can be complex. (In lightly preserved seafood or ready-to-eat processed meat, six to eight factors, including lactic-acid bacteria, can control the growth of Listeria monocytogenes.) Consider also, that death and growth are not instantaneous processes. They are also governed by the time or duration that each set of environmental conditions prevails.

The aim of this new model is still the same: keep consumers safe. However, how this is achieved is not written in stone. The new model does not spell out exact methods in the form of prescriptive regulations; rather it allows any alternative methods of processing and preservation, provided they can give the same level of product safety and integrity as current methods. This allows food manufacturers to be inventive, without jeopardizing public health. This may actually benefit consumers in the form of better-quality products, at more economic prices.

The only way to fully guarantee that food is free from pathogens or their toxins is to test it just before consumption. Clearly this is not a practical approach. Instead, just as it is done with HACCP, a better approach is to understand the ecology and physiology of particular microbes, in order to estimate their potential growth or death in food. This is done by measuring product and process conditions (temperature, mixing ratios, separation of fat and aqueous phases, pH, added salt, nitrite, organic acids, distribution conditions, storage conditions, etc) and durations of each step. For each set of conditions experienced, the rates of growth or death are then multiplied by the duration of the respective processes or steps in food-processing operations, along the total ‘farm-to-fork’ chain. Other simple mathematical calculations are undertaken for steps in which microbes are concentrated (eg through evaporation, separation of aqueous and lipid components) or diluted (eg mixing, hydration etc). In this way, the total microbial increase or decrease can be estimated (ie the calculation summarized in Equation 1).

But how can we know the rate of growth or death for every single possible combination of food composition and every organism of interest? A large part of the answer lies in a series of databases and software now available free-of-charge on the internet. These data and tools are the result of ‘predictive microbiology’, the area of food microbiology science that became very active in the early 1980s. The earliest predictive models are for thermal inactivation of spores of Clostridium (ie the “botulinum cook”) and that are still used as the basis of design of thermal processes for foods.

Predictive microbiology is based on the notion that while the microbial ecology of foods is complex, microorganisms respond in a predictable way. Once the responses of a particular microbe have been described, quantified and summarized under one environment, it possible to determine how the same type of microbe will behave under similar conditions. Because predictive microbiology assumes that the environment dictates microbial behavior, foods are usually described in terms of important environmental factors (pH, water activity etc) and storage conditions (temperature, gaseous atmosphere, etc). This means that actual type of food is less critical.

The data on microbial responses can be generated by purpose-designed experiments or by gathering existing data from the literature, or industry or government records. From that data, the patterns of microbial responses to temperature, water activity, pH, organic acids and other factors in food are discerned and quantified. To make that information more useful to other people it is summarised as mathematical models, and to make the models even more useful to others, they are usually incorporated into ‘user-friendly’ software. Such mathematical models can be considered as condensed knowledge of the microbial ecology of foods.

The user starts off by selecting the organism of concern and “enters” the important properties of the food and the storage conditions. The software then generates a prediction of how fast the organism will grow or die, or forecasts the amount of growth that will occur in a specified amount of time. Usually, models are based on a lot of data, and the predictions of the model are validated (or checked for consistency against other data not used to generate the model) before the model is released.

These types of tools can provide the answers needed to complete the ICMSF equation (Equarion 1) and are used in formal risk assessment, usually in combination with stochastic simulation modelling software (which is briefly discussed below).

• ComBase/ ComBase Predictor
www.combase.cc

ComBase is a database which includes growth and death rates for pathogenic and spoilage microbes (Figure 1). The user can search for data on specific pathogens in particular foods or types of foods or food formulations. ComBase currently contains about 35,000 records on growth and survival of pathogens, and about 4,000 on spoilage organisms. Of those datasets, approximately 22,000 are full log-count curves; and about 15,000 report growth or death rates, but without the ‘raw’ data. The database encompasses both data from the UK government-sponsored Food MicroModel project and the USDA-funded Pathogen Modeling Program, as well as data abstracted from the published scientific literature and from collaborating laboratories. As new collaborators join the ComBase consortium, the amount and diversity of data continues to increase, and any user can add their own data (subject to quality checks). ComBase Predictor is a suite of predictive models that is integrated with ComBase to translate data into predictive models.

• Pathogen Modeling Program
http://pmp.arserrc.gov/PMPOnline.aspx

Pathogen Modeling Program (Figure 2) is a suite of predictive models based on studies conducted by USDA researchers, as well as some taken from published literature. The program (accessible online or by download) contains many models for the growth of food-borne pathogenic bacteria; models for inactivation of pathogens due to various treatments; and useful information about predictive microbiology in general, through the “predictive microbiology portal” (http://portal.arserrc.gov/).

• Seafood Spoilage Predictor
www.dfu.min.dk/micro/sssp/Home/Home.aspx

Seafood Spoilage Predictor is series of tools for the prediction of growth of pathogens and spoilage organisms relevant to fresh and lightly-preserved fish products. Developed by workers at the Danish Fisheries Research Institute, it calculates the effects of fluctuating temperature on microbial population growth, with a technique called ”time-temperature-function integration”. It also includes models for the interaction between L. monocytogenes and lactic-acid bacteria in lightly-preserved, long shelf-life products, like cold smoked salmon.

• The Refrigeration Index
www.mla.com.au

(before download please contact Mr Ian Jenson: ijenson@mla.com.au)
The Refrigeration Index is an Australian product, and is used both by industry and regulatory agencies, having been endorsed for regulatory use under the Australian Export Meat Orders. The Refrigeration Index is a mathematical model that predicts the growth of E coli as a function of pH, water activity, and lactic-acid concentration and temperature. The model is used to assess potential growth of E coli on red meat by time-temperature-function integration. It does this by the continuous monitoring of a carcass’ temperature, and then feeding that information into the software to generate the total predicted E coli-growth during processing, chilling and distribution (Figure 3). The predicted extent of growth determines whether the product is acceptable for sale or not.

• Homemade software
In addition to the software tools described above, there are many more models in the published literature. It is a relatively straightforward task to translate those models into user-friendly software tools using spreadsheet software.

Applications
There is a wide range of applications for predictive-microbiology models and predictive-microbiology tools: product and process design, the development of cost-effective HACCP plans, decision making about the microbial safety of foods after a loss of control, and demonstration of equivalence of alternative food-safety processes.

Predictive microbiology models can greatly reduce the need for challenge testing or shelf-life testing. In some ways, predictive models are more useful than these approaches. The responses of a wider range of strains can be considered, and often there is more data available to answer a shelf-life and/or safety question than could realistically be generated by challenge trials or shelf-life trials. In general, however, predictive modellers recommend that a small number of challenge tests are undertaken to confirm the model predictions. Predictive models should be used to fine tune the range of variables, to minimize the number of challenge trials undertaken (eg by eliminating formulation options that clearly will not be effective).

Predictive models can identify Critical Controls Points (CCP) for HACCP plans, and specify their critical limits. These models quantify the importance of different processing steps in terms of their potential for microbial growth or inactivation. Therefore predictive models can help to discriminate processing steps where control may be beneficial but not crucial, from steps and conditions that are critical to product safety.
Similarly, predictive models can assess the significance of loss of control at a CCP and the options for corrective actions. For example, if a loss of temperature control occurred in a perishable product or component, the model could be used to estimate how much growth of a variety of pathogens might have occurred. From this, it would be possible to determine the significance of the loss of control (eg no potential for growth, minor growth, or extensive growth) and, thus, whether re-processing could correct the problem, if needed, or whether the product should be discarded.
By using Equation 1 in reverse, it is also possible to specify a PO from a given FSO. A PO is something that a food processor needs to achieve, based on a public-health goal expressed in the Food Safety Objective. Given the product formulation, and usual storage times and conditions after manufacture, it is possible using predictive methods to estimate how much growth could occur if a pathogen were initially present in the food. The PO is then set so that microbial levels at that point are such that, even after the microbial growth that could be expected to occur during product distribution and storage, the FSO (at the point of consumption) would not be exceeded.
Predictive models can also be used to aid in the design of new-product formulations that enable specified “best-by” or “use-by” times to be achieved. Because models often include the combined effect of many hurdles, combinations of mild levels of multiple hurdles can be determined, rather than relying on the high level of a single hurdle.

Still, these mathematical models are not perfect predictive tools and should be used within their limits, which are usually specified within the software packages. A good overview of the appropriate use of predictive microbiology models can be found at the Predictive Microbiology Portal http://portal.arserrc.gov/)

Food-safety risk assessment as the means to evaluate equivalence of food-safety management procedures in different nations and techniques for microbial food-safety risk assessment are being developed internationally by various organizations. The FAO and the WHO have been charged jointly with this responsibility by the Codex Alimentarius Commission. They have conducted risk assessments for the Codex Committee on Food Hygiene and member countries, and are developing guidelines for conducting risk assessments (http://www.fao.org/ag/agn/agns/jemra_index_en.asp).

Under the Codex model, food-safety risk assessment involves identification of the food-borne hazard; the explanation of the context of the hazard; the determination of human exposure to the hazard in food, or water (“exposure assessment”); characterization of the hazard, including details of the type and severity of illness caused and the relationship between the dose ingested and disease severity; and the synthesis of all the abovementioned information (called “risk characterization”) to produce an estimate of the risk (disease burden, or the frequency and severity of illness due to the hazard in the food) among the population exposed to the food. A wide range of resources for food-safety risk assessment can be found at www.foodrisk.org, hosted by USA’s Joint Institute of Food Safety and Applied Nutrition.
The tools of predictive microbiology are useful for exposure assessment, because data describing contamination levels and frequencies at the time of consumption are not usually available. In this case, changes in microbial levels along the farm-to-fork pathway are estimated using predictive microbiology techniques as described above.
As noted earlier the microbial ecology of foods can be complex, particularly when considering all the potential changes along the food chain. There is variability in many of the factors that can affect the probability of pathogens being present in a food, and the levels likely to be present at consumption. Codex recommends that risk assessment be as quantitative as possible and, to fulfil that aim, most risk assessors recommend the use of stochastic-modelling techniques to fully characterize the variability in human exposure to the hazard. In this approach—now widely used in many areas of science, government and industry—each factor that can affect risk is described not as a single representative value, but as a distribution of the possible values of that variable.

The factors are combined in a mathematical equation (or series of equations) that describes how all the factors interact and which, when evaluated, provides a numerical estimate of the risk. To get the best estimate of risk, the risk is calculated repeatedly—but each time using a different value from the range of possible values for each factor. This is essentially a complete set of “what-if” calculations for all possible combinations of values of each of the factors. The outcome is a distribution of possible levels of risk, which can be characterized by a range of values and a most probable value. Clearly this process will entail a large number of calculations. Fortunately, there are commercial software packages that automate this process, improve on it, as well as provide many tools for more detailed analysis of the results. The process is called “Monte Carlo simulation” and is best achieved using stochastic-simulation modelling software, including those such as @Risk (www.palisde.com), Crystal Ball (www.crystalball.com), Analytica (www.lumina.com).

Fully quantitative risk assessments can be very complex and time-consuming tasks. Often they require teams of experts and many person-years of commitment. Nonetheless, risk-based approaches offer many benefits to all sectors of the food industry and there have been efforts to generate simpler food safety risk-assessment tools or tools that would allow risks from different sources to be ranked and prioritized. These include decision trees and various risk-matrix schemes.
A simple, spreadsheet-based tool for ranking of microbial food safety risks has also been published and is available online (Figure 4). It guides users through the risk-assessment process by requiring the user to select answers to a series of questions about the product, pathogen and process to generate a risk ranking. It uses similar logic to more complex models and, while it does not enable stochastic modelling, it is a good tool for helping to teach the principles of microbial food-safety risk assessment, including the contributions or interplay of factors. It can be also be used to undertake “what-if” scenarios by automating the risk-calculation process. The spreadsheet can be downloaded from www.foodsafetycentre.com.au/riskranger.htm. However, all accompanying documentation MUST be read, before attempting to use the application.

As food production and processing becomes more centralized and as the scale of processing plants increases, even very low levels and frequencies of contamination may cause illness and food-borne outbreaks. Modern microbial food safety systems have to rely on knowledge of microbial ecology, not testing, and to incorporate that knowledge in the design of processes and systems that minimise the potential for contamination and growth of microbial pathogens in foods. This approach is already well established in HACCP but can be improved by use of predictive microbiology and quantitative, risk-based, approaches that focus on the hazards that are most severe and/or most likely to occur.

A range of data and software tools are now available, without cost, on the internet that can assist food scientists and foods safety managers to attain the benefits of these new approaches to microbial food safety management.
Figure 1

Figure 2

Figure 3

Figure 4