The objective of ensuring safe food for the World ’ s constantly growing population has been a major preoccupation of governments, international organizations (e.g. WHO/FAO CODEX, ILSI, ISO, ICMSF, etc.) and professional and trade bodies over many years. Yet, in deprived areas of the world, there remains a basic need to ensure a reliable food supply. In all countries, especially in developed consumer-oriented countries, the need is to ensure that foods do not present an unacceptable risk to the health and well being of the consumer. At times, therefore, there is a clash of priorities: foods exported from third world countries help to support their national economy, but the foods are required to comply with the rules imposed by international trade particularly the import regulations of developed countries. Meanwhile, the indigenous population often consume foods that do not meet those criteria. Attempts to improve food quality and safety are important for all consumers – but if you are starving the importance of quality and safety appears less important that having enough food for your family. People in developed countries who constantly demand ever-increasing quality and safety standards often overlook this paradigm.
Following the publication by Accum (1820) of his ‘ Treatise on Adulteration of Food and Culinary Poisons ’ and subsequent work in the middle of 19th century by the Lancet Analytical Sanitary Commission and other bodies (see Amos, 1960 ) the need to improve food safety in the United Kingdom led to the introduction of food legislation concerned with such diverse areas as food composition, food additives, chemical contaminants and, eventually, on microbiological contamination. In recent times legislation has been concerned primarily with controlling those aspects of food production that are necessary to ‘ ensure ’ the safety of foods at all stages from ‘ farm to fork ’ . In the area of food microbiology, such legislative control has been aimed at improving the safety and quality of foods processed by the food industry, or supplied by catering outlets, since it is rightly perceived that industrial scale production impacts on many more consumers than does traditional domestic production. Yet it is often the small producer or caterer who presents the greatest risk to consumer well being.
RISK ASSESSMENT, FOOD SAFETY
OBJECTIVES AND MICROBIOLOGICAL
CRITERIA FOR FOODS
Throughout the world, the law imposes a duty of care and responsibility for the safety and quality of foods on those business organizations involved in the procurement, process-ing, distribution and retail sale of the products. For instance, in Europe, the basic premise of food law is enshrined in a General Regulation ( Anon., 2002a ) on food safety, with sub-sidiary legislation on key issues, including microbiological aspects of safety. One facet of modern food legislation is the requirement for risk assessment by governments in order to provide the legislative framework within which food producers, processors, caterers and all others concerned with food must operate.
FOOD SAFETY OBJECTIVES AND RISK ASSESSMENT
Modern approaches to food safety include the identifi cation of actual, or potential, hazards from microbial contamination, assessing the risk that such contamination may cause dis-ease in the consumer, and then seeking to employ processes that will control and minimize such risks. ‘ Hazard ’ can be defi ned as something that has the potential to cause harm, for instance the contamination of food by pathogenic bacteria. ‘ Risk ’ is defi ned as the like-lihood of harm in a defi ned situation; for instance, consumption of food contaminated with specifi c pathogenic microorganisms and/or their toxins. Risk assessment of foods is therefore concerned with assessing the potential risk that consumption of a food may cause harm to consumers. As is amply demonstrated by ICMSF (2002) , risk assessment requires an understanding of microbial contamination per se and also that both food process opera-tions and domestic food handling practices may reduce or increase the risk from a defi ned hazard for a defi ned group of consumers (infants, children, the aged, the immuno-compro-mised, etc.).
Rather than seeking to control food safety on an ad hoc basis, ‘ perceived wisdom ’ requires that specifi c objectives should be set to ensure, so far as is practicable, that food does not threaten the health and well being of consumers. The Phytosanitary Measures (SPS) of the World Trade Organisation (WTO) require member states to ensure that their sanitary and phytosanitary requirements are based on scientifi c principles whilst not unnec-essarily restricting international trade. This means that member countries must establish appropriate measures on the basis of the actual risks likely to be involved; originally, this requirement was primarily for those risks arising from chemical contaminants. The concept of Quantitative Microbiological Risk Assessment (QMRA) was introduced in the 1990s fol-lowing development of predictive models for growth and survival of pathogenic, and other, microbial populations in foods based on fundamental studies of microbial growth and sur-vival (see e.g. Haas, 1983, 2002 ; Roberts and Jarvis, 1983 ; Whiting, 1995 ). This approach is based on the application and extension of the concept of microbiological compositional analysis ( Tuynenburg-Muys, 1975 ) and uses the wide availability and computational power of desktop computers, often using dedicated microbial modelling software.
The WTO/SPS proposed the concept of an ‘ appropriate level of protection ’ (ALOP) defi ned as, ‘ the level of protection deemed appropriate by the member (country) to establish a sanitary or phytosanitary measure to protect human, animal and plant life or health within
its territory ’ . Subsequently, the Codex Committee on Food Hygiene (CCFH) developed consensus protocols for risk analysis of pathogens in food (see for instance Anon., 2000a, b, 2002a ) and produced a report on the ‘ Principles and Guidelines for the Conduct of Micro-biological Risk Management ’ (MRM) ( Anon., 2002b, c ). A key output was the redefi nition of the WTO/SPS defi nition, as ‘ ALOP refers to a level of protection of human health estab-lished for a food-borne pathogen ’ . However, there was little guidance on the nature of an ALOP or how it might be established. Several alternative approaches for setting an ALOP were debated including the concept of ‘ as-low-as reasonably achievable ’ (ALARA) based on the performance of the available risk management options.
All aspects of control require the defi nition of criteria for the ‘ disease burden ’ that public health can accept for a population, for instance ‘ the number of cases per year per 100,000 population for a specifi c hazard in a specifi c food commodity ’ . But even this leaves much room for debate. In 1979, Mossel and Drion had proposed the concept of a ‘ lifetime tol-erable risk ’ for botulinum and other toxins, but other objectives were more constrained both in relation to the time span and the defi ned population. So what is the tolerable risk to which any consumer should be exposed and over what timeframe? Is the ‘ population at risk ’ the total population, the most susceptible group(s) or only that proportion of the population who actually consume a particular food? Does the ALOP include specifi c demo-graphic groups of the population? Are related health concerns linked together (e.g. all cases of salmonella food poisoning) or considered only in relation to specifi c foods? What is the impact of alternative transmission routes, that is transmission by food handlers; cross-con-tamination between foods, due to poor storage and handling practices, and transmission of pathogens from food to consumers; or even, person-to-person transmission? Is the risk timeframe a defi nable period (e.g. a year) or a lifetime? Havelaar et al. (2004) propose a defi nition for an ALOP as ‘ no more than x cases of acute gastroenteritis per 100,000 popu-lation per year associated with hazard Y and food Z ’ . This ALOP concept provides a useful target measure for public health policy but is of limited use in implementing food chain safety measures.
The International Commission on Microbiological Specifi cations for Food (ICMSF) introduced the concept of ‘ Food Safety Objectives ’ (FSOs) that was adopted subsequently by CODEX (CCFH) as part of its MRM document. An FSO provides a means to convert public health goals into parameters that can be controlled by food producers and monitored by government agencies. It is defi ned ( ICMSF, 2002 ) as, ‘ the maximum frequency and/or concentration of a microbial hazard in a food considered tolerable for consumer protec-tion ’ . ICMSF (2002) notes that FSOs are ‘ typically expressions of concentraprotec-tions of micro-organisms or toxins at the moment of consumption ’ . Concentrations at earlier stages of the food chain are considered to be performance criteria. Hence, an FSO seeks to take account of hazards arising both during commercial processing and from unpredictable effects asso-ciated with retail and domestic food storage and handling. By contrast, performance criteria relate to the requirement to control hazards at earlier stages of the food chain.
ICMSF (2002) provides the following simplistic equation to describe the performance criteria concept: H 0 R I FSO. The terms in the equation (i.e. the performance criteria) are: the initial level of the specifi c hazard ( H 0 ) associated with raw materials and
ingredients, the cumulative decrease in hazard level due to all processing factors ( R ), and the cumulative increase in hazard as a consequence of post process microbial growth ( I ). The symbol implies that the cumulative effect should be less than, or at least not greater than, the FSO expressed in terms of log 10 units for a specifi c organism. Suppose for instance that the FSO for a specifi c pathogen in a defi ned food is considered to be not more than 1 organism/10 g (i.e. 1 log 10 organisms/g) at the time of consumption. Suppose further that the maximum initial contamination level is likely to be 100 organisms/g (2 log 10 organisms/g) and the maximum reduction is likely to be obtained by a combination of thermal and other processes is 3 log 10 units. Then, to ensure that the FSO is not exceeded, the risk of re-growth between processing and consumption ( I ) must be zero, that is substituting H 0 2, R 3 and FSO 1, in the rewritten equation I FSO R H 0 1 3 2 0. Thus conditions of storage and handling post-processing must be such as to prevent any growth of surviving organisms. However, if the initial level of contamination of the product ( R ) were only 10 organism/g (1 log 10 organisms/g) then the re-growth allowance would be not greater than 1 log 10 unit : I 1 3 1 1.
Such calculations are, of course, based on ‘ point ’ values and make no allowance for ations in microbial distribution within or between batches of food ingredients or for vari-ations in process effi ciency. Nonetheless they do provide a simple way to demonstrate how an FSO can be used to assess risk for a specifi c products and processes.
So, the FSO is defi ned as ‘ the maximum likely level of hazard that is acceptable ’ following the integration of several stages in food processing, based on knowledge of microbial asso-ciations of foods, processing hurdles ( Leistner and Gould, 2001 ; ICMSF, 2005 ) which may result in death or inhibition of microorganisms and of the likelihood of re-contamination and/or re-growth of organisms during subsequent storage and handling. Thus the FSO con-cept relates also to the use of the Hazard Analysis and Critical Control Point (HACCP) concept for controlling the effectiveness of food processing operations.
For any manufacturing process, HACCP requires analysis of potential hazards and the identifi cation and monitoring of control points that are critical to elimination or reduction of each hazard (CCPs). Furthermore, the concept requires that each CCP will be moni-tored using simple, indirect methods to ensure the process is operated correctly and that the effectiveness of monitoring is verifi ed also by appropriate microbiological examination. HACCP is now required by law in many countries and forms part of a wider Total Quality Management (TQM) procedure that is used within a business to ensure that all foods pro-duced conform to criteria that defi ne acceptable quality and safety standards.
The underlying themes of ALOP, ALARA, FSOs, HACCP, TQM and GMP imply that all persons responsible for production, distribution, sale and preparation of foods work together to ensure that potential hazards are identifi ed and controlled effectively in order to minimize ‘ so far as is achievable ’ the risks to consumer health. This requires knowledge and experience of potential microbiological hazards and the likely effects of both acceptable and unacceptable practices at all stages from ‘ farm to fork ’ .
An FSO differs from a microbiological criterion; an FSO is not applicable to individual ‘ lots ’ and does not specify sampling plans. Rather an FSO is a statement of the level of control expected for a food processing operation that can be met by the proper application
of GMP, HACCP systems, performance criteria, process/product criteria and/or acceptance criteria ( ICMSF, 2002 ). Ideally, an FSO should be quantitative and verifi able, though not necessarily by microbiological examination of food.
An FSO provides a means by which control authorities can communicate clearly to industry what is expected for foods produced in properly managed operations, for instance by specifying the frequency or concentration of a microbial hazard that should not be exceeded at the time of consumption. An FSO therefore provides a basis for the establish-ment of product criteria that can be used to assess whether an operation complies with a requirement to produce safe food.
FSOs can be established using risk evaluation by an expert panel using quantitative risk assessment. In all cases, the fi rst step is the identifi cation of hazards associated with spe-cifi c foods by epidemiological or other means. Next an exposure assessment is required to estimate the probable prevalence and levels of microbial contamination at the time of con-sumption. Such assessment requires information about the amount of product consumed by different categories of consumers and use of mathematical models that take account of the prevalence of the organism(s), the nature of the food processing operations, the probability for growth of the organisms in the food both before and after processing and the impact that food-handling practices will have on the levels of organisms likely to be consumed.
Characterization of the hazard ( Anon., 2000c ) requires assessment of the severity and duration of adverse effects resulting from exposure of individuals to a specifi c pathogen. A dose–response assessment provides a measure of the potential risk. The likelihood of expo-sure is dependent not only on the characteristics of specifi c strains of microbe, but also on the susceptibility of the host and the characteristics of the food that acts as the carrier for the organisms. Finally, risk characterization combines the information to produce a risk assessment that indicates the possible level of disease (usually as the number of cases per 100,000 people per year) likely to result from the given exposure. Risk characterization needs to be validated by comparison with epidemiological and other data and should refl ect the distribution of risk associated with the many facets affecting contamination, survival and growth of a specifi c organism in food processed in a specifi c way.
The overall process of establishing an FSO for any one specifi c food/pathogen combina-tion is very challenging. It is clear from the FAO/WHO Expert consultacombina-tion ( Anon., 2000b ) that hazard and risk characterization even for a limited scenario (e.g. salmonellae in eggs and broiler chicken) requires data of the highest quality and use of effective mathematical models to interpret those data (cf. views of Roberts and Jarvis, 1983 ). This is not to suggest that the approach is invalid or unattainable – rather it shows how little we really under-stand about those microorganisms in foods that are responsible for many apparently com-monplace causes of human food-borne disease. However, Szabo et al. (2003) have described the development of a system for achieving an FSO for control of Listeria monocytogenes in fresh lettuce.
CODEX has now published a Guide for National Food Safety Authorities ( Anon., 2006a ) that explains the whole concept of Food Safety Risk Analysis. This report covers all aspects of risk assessment for foods including providing guidance on the four stages of a risk management procedure ( Table 14.1 ). Microbiological risk assessment is based on the
use of ‘ quantitative microbiological metrics ’ as a risk management option. ‘ Quantitative metrics ’ is defi ned as the ‘ quantitative expressions that indicate a level of control at a spe-cifi c step in a food safety risk management system … the term “ metrics ” is used as a col-lective for the new risk management terms of food safety objective (FSO), performance objective (PO) and performance criteria (PC), but it also refers to existing microbiologi-cal criteria ’ . The report provides a case study for Listeria monocytogenes in ready-to-eat foods and recognizes the desirability of using FSOs, POs and PCs in the development of risk-based microbiological criteria; however, methods for achieving these objectives are still under development.
Statistical methods play a major role in the use and interpretation of mathematical mod-els of microbial contamination, growth and survival in relation to the epidemiology of food-borne disease, but it is not appropriate to consider such matters here. The reader is recommended to consult publications that consider this matter in more detail ( Jouve, 1999 ; ICMSF, 2002 ; Szabo et al., 2003 ; Havelaar et al., 2004 ; Anon., 2002c, 2005a, 2006a ; Rieu et al., 2007 ). However, we do need to consider the implications that arise in the application of quantitative and qualitative data in food control situations.
First, it must be realized that no amount of quantitative or qualitative testing can ever con-trol the safety and quality of manufactured foods (or other materials). Rather, such testing
TABLE 14.1
The Stages of Risk Assessment, Control and Management (based on Anon., 2006a)
Step Process
1 Preliminary Risk Management Activities 1.1 Identify and defi ne the food safety issue
1.2 Develop a risk profi le
1.3 Establish broad management goals
1.4 Decide whether a risk assessment is necessary 1.5 Establish a risk assessment policy
1.6 Commission the risk assessment 1.7 Review the results of the risk assessment 1.8 Rank the food safety issues and set priorities a 2 Select Risk Management Options
2.1 Identify available risk management options 2.2 Evaluate identifi ed risk management options 2.3 Select risk management option(s)
2.4 Identify a desired level of consumer health protection 2.5 Decide on preferred risk management option(s) 2.6 Deal with ‘ uncertainty ’ in quantitative risk and policy 3 Implement Risk Management Decisions
4 Monitor and Review
indicates whether or not a production process, including all sources of contamination, is adequately controlled in terms of process conditions, process hygiene, pre- and post-process storage and distribution, etc. End-point testing in a factory environment provides data for feedback control of a process. More satisfactory, is the use of Quality Assurance pro-grammes such as those which seek to identify and control potentially hazardous stages of a process, including HACCP described, for instance by Bauman (1974) , Ito (1974) , Peterson and Gunnerson (1974) , ICMSF (1988) , Wallace and Mortimer (1998) , Jouve (2000) . In the HACCP system, end-point testing is used to validate process controls for raw material qual-ity; time–temperature relationships for heating, cooling, freezing, etc.; process plant clean-ing and disinfection; operator hygiene and the many other factors critical to the production of foods under GMP.
A different aspect of food control relates to the assessment of actual or potential health risks associated with particular food commodities or products, whether imported or home produced; and the assessment of ‘ quality ’ of foods in retail trade in so far as this may be required by food legislation. For such purposes, it is not possible to ‘ control ’ the process or the post-process distribution and storage, although inspection of process plants, includ-ing those in exportinclud-ing countries, is now the ‘ norm ’ . Testinclud-ing by enforcement bodies is tar-geted to assess whether foods on sale have the necessary qualities expected of them and/or whether they constitute a (potential) risk to the health of the consumer.
Consequently, various qualitative and quantitative microbiological criteria have been derived to provide guidance both for production personnel within industry and for enforce-ment authorities. Such criteria may not have legislative status but properly devised criteria can be of enormous value in ensuring compliance with GMP.
MICROBIOLOGICAL CRITERIA
Microbiological criteria can be defi ned as, ‘ limits for specifi c or general groups of micro-organisms that can be applied in order to ensure that foods do not present a potential health hazard to the consumer and/or that foods are of a satisfactory quality for use in commerce ’ . This defi nition is deliberately vague since it encompasses a wide range of types of criteria: 1. A Microbiological Guideline is used to provide manufacturers, and others, with
an indication of the numbers of organisms that should not be exceeded if food is manufactured using Good Manufacturing Practices (GMP) and stored during its ‘ normal ’ shelf life under appropriate conditions.
2. A Microbiological Specifi cation defi nes the limits that would be considered appropriate for a particular food in a particular situation and may be used in contractual
commercial agreements or may be recommended by national or international agencies as a means of improving the quality and safety of foods.
3. A Microbiological Standard is that part of national, or supra national, legislation that aims to control the safety and, in some cases, the quality of foods manufactured in, or imported into, that country.
Microbiological standards therefore have mandatory effect whilst specifi cations and guidelines do not. It is not intended to consider here the arguments for and against the use of legislative microbiological standards for foods since these have been well argued else-where ( Thatcher, 1955, 1958 ; Wilson, 1955 ; Ingram, 1961, 1971 ; Charles, 1979 ; Baird-Parker, 1980 ; Mossel, 1982 ). It is worthwhile, however, to summarize the principles for establishment of ‘ Microbiological Reference Values ’ (i.e. criteria) described in more detail by Mossel (1982) :
1. The number of criteria should be strictly limited so that the maximum number of samples can be examined for a given laboratory capacity.
2. The choice of criteria must be based on ecological considerations and be related to organisms of importance in public health and/or quality.
3. Criteria should be carefully formulated in justifi able quantitative terms.
4. Species, genera or groups of organisms to which the criteria are applied should be described in appropriate taxonomic terms.
5. Test methods should be described in suffi cient detail to permit their use in any reputable laboratory, and should be applied only after full validation including inter-laboratory trials.
6. Numerical values should be derived only as a result of adequate surveys to establish technological feasibility.
In 1979, Codex Alimentarius adopted the principles summarized above and agreed mod-ifi ed defi nitions of the various types of criteria to include reference to the use of sampling plans and standardized methodology as a prerequisite for setting criteria ( Anon., 1997 ). Consideration of most of these points is outside the scope of this chapter and the reader is referred to the publications of Mossel (1982) , ICMSF (1986, 2002) , and the UK Health Protection Agency ( Gilbert et al., 2000 ).
However, principles (3) and (6) are clearly the raison d ’ etre for this chapter which is concerned with interpretation of laboratory data in relation to quantitative and qualitative microbiological criteria.
How Should Microbiological Criteria Be Set for Quantitative Data? Collection of Data
Numerical criteria should be derived from surveys undertaken to determine what is techni-cally feasible. Having decided upon the tests to be done and the methodology and culture media to be used, a survey should be made of products from several manufacturers who operate their process plants under GMP. As a prerequisite, an inspection should be made of the manufacturing processes together with microbiological examination of products so that any defi ciencies in the process can be identifi ed and corrected before the survey is done.
Samples should be obtained at various times during the working day. They should be stored and transported to the laboratory under appropriate (e.g. ambient, refrigerated or
frozen) conditions for examination using suitable microbiological reference methods. Ideally the laboratory examination should be undertaken within a few hours of sample pro-curement. Sampling is then repeated over a period of several days in the same and, ideally, also in other factories producing the same generic product, until suffi cient data (not less than 100 data sets) have been collected. A distribution curve is prepared of the log 10 colony
counts against the frequency of occurrence of counts, for example, the data for a cooked meat product shown in Fig. 14.1 ; the 95th percentile ( ) value (i.e. the log count which is exceeded only by colony counts on 5% of the samples) is also shown on the graph (see also Fig. 2.1 for colony counts on sausages manufactured in two different process plants).
Such curves can also be drawn in an idealistic manner ( Fig. 14.2 ) to indicate the mean, mode and the 95th percentile value ( ). In the case of the cooked meat data ( Fig. 14.1 ) the 95th percentile is 3.5 log 10 cfu/g indicating a well-controlled process, whereas for the data
in Fig. 2.1, the 95th percentile is slightly below 7.0 log cfu/g, which is unacceptably high even for a raw meat product.
Setting Criteria Limits
For quantitative data, it is necessary to decide on the maximum number of organisms (the critical level: M ) that could be permitted in any circumstances. This will normally be related to the minimum spoilage level (MSL) or, in the case of pathogens, to the minimum infective dose (MID). This critical level is the maximum count which is acceptable for foods manu-factured under GMP and will usually be set at least one log cycle above the 95th percentile, provided that such value does not approach too closely the MSL or MID. The data from the survey of sausage manufacture (Fig. 2.1) indicate that many of the colony count values
30 25 20 15 10 5 0 0.25 0.75 1.25 1.75
Colony count (log cfu/g) 2.25 2.75 3.25
w
3.75 4.25 4.75
F
requency (%)
FIGURE 14.1 Frequency distribution of colony counts on cooked meats, overlaid with a ‘ normal ’ distribution curve and showing the 95% percentile ( ) at 3.49 log 10 cfu/g (based on data of Kilsby et al., 1979 ).
are unacceptable; no attempt should be made to use such data to set critical values – rather the manufacturing processes must be improved, before the survey is repeated.
The lower critical limit ( m ) is set at some point above the 95th percentile ( ) and below the maximum ( M ) such that the value of m represents counts that can be achieved most of the time in product manufactured under GMP. The tolerance ( ) between and m should be set to take account of the imprecision in microbiological measurement, the variability likely to occur in good quality raw materials, the manufacturing process, etc. For industrial control purposes, m is frequently established at, or very close to, the 95th percentile value. Figure 14.2 illustrates the establishment of a three-class sampling plan based on the results of a survey.
In establishing criteria for ‘ point of sale ’ or import sampling (i.e. by enforcement author-ities), due attention is required not only to what can be achieved in GMP but also to the effects of subsequent storage and distribution. This requires understanding of the micro-biological consequences of storage time and temperature and of the effects of the intrinsic properties of the food (i.e. pH, aw , presence of antimicrobial agents, etc.). In addition, from a public health viewpoint, the potential for consumer mishandling during storage, cooking and serving needs to be considered; and the vulnerability of particular consumer groups. Hence the use of rapid indirect methods to indicate that a product has been adequately processed may be of more value than quantitative microbiological methods per se ( Charles, 1979 ).
For qualitative testing (e.g. for presence/absence of pathogens), the survey of product manufactured under conditions of GMP must be done using realistic sizes (e.g. at least 25 g) and numbers of samples in order to determine the prevalence of the target organism in the product. As explained in Chapter 5, the value of quantal tests is dependent not only on the relative sensitivity and specifi city of the test procedure, but also on the number of sam-ples examined. To obtain meaningful results, at least 100 samsam-ples should be examined from
Accept-ance plan
Verdict based on results of survey
Accept Conditional release n 10 c 2 Reject Survey F requency of counts λ φ Mode Median m M MSL
FIGURE 14.2 Distribution plot of the results of microbiological surveys on a given type of food (modifi ed from Mossel, 1982 ).
each of a number of manufacturers. If the target organism is likely to be present in the test material, then the survey should be done using several quantities of each replicate sample, possibly using an MPN test in order to ascertain both the prevalence and level of potential contamination.
Establishing Sampling Plans
Having established critical values it is necessary to consider the sampling plan requirements. In the fi rst instance, a decision concerning the choice of Attributes or Variables systems is required. As described previously (Chapter 5) an Attributes scheme merely requires a ‘ go-no go ’ decision based upon the analytical fi ndings and takes go-no go-note of microbial distribu-tion or methodological imprecision except in so far as tolerance may have been built in to the proposed values for m and M . By contrast, the Variables scheme assumes that the transformed data conform to a ‘ normal ’ distribution and incorporates into the specifi cation a tolerance related to the number of samples to be examined and the standard deviation of data determined on the replicate samples examined.
Traditionally organizations such as ICMSF and Codex Alimentarius have recommended the adoption of Attributes systems based on three-class sampling plans for quantitative data and two-class plans for quantal data (see Chapter 5). ICMSF (1986, 2002) acknowledge the benefi ts of Variables sampling plans in those circumstances where the microbial distri-bution in foods is known, as for instance, in food manufacture where the process is stable. However, they argued against the use of Variables plans for criteria that might be applied, for instance, in food control situations where knowledge of the microbial distribution is unknown (e.g. examination of samples taken at a port-of-entry). Whilst there is logic in this argument, increasingly legislative criteria are introduced for use by the food production industry where such knowledge does exist. In addition, the use of microbiological meas-urement uncertainty now provides a means of assessing the variability between replicate random samples of a food, although such variability may not refl ect the true microbial dis-tribution in the food.
European legislation on microbiological criteria for foods ( Anon., 2005b ) follows the Attribute sampling concept, albeit with some variations in the process hygiene criteria. Two types of criteria are published: for ‘ food safety ’ and ‘ process hygiene ’ . The food safety cri-teria are mainly two-class sampling plans for particular pathogens, for which the cricri-teria require ‘ absence ’ (i.e. non-detection by the specifi ed method) of the target organism in a defi ned quantity of sample. The process hygiene criteria are mainly, but not exclusively, based on three-class sampling plans. An important variation introduced into the interpreta-tion of the process hygiene sampling plans for Enterobacteriaceae and aerobic colony counts on carcasses of meat animals is that a product is defi ned as ‘ satisfactory ’ if the daily mean log 10 colony count m, ‘ acceptable ’ if the daily mean log 10 count is between m and M , and ‘ unacceptable ’ if the daily mean M . All of the other three-class process hygiene sampling plans conform to the normal scheme where ‘ satisfactory ’ requires all counts m , acceptable if not more than c/n counts m but M, and unacceptable if any count M .
So far as I am aware, no national or international agency has adopted a Variables scheme for microbiological examinations although such schemes have been adopted in legislation for quantitative determination of chemical contaminants (see, for instance, the European criteria for mycotoxins such as afl atoxin, ochratoxin and T2 toxin; Anon., 2006b ). However, there is a difference in the type of sampling plan used for such micro-chemical analyses. A number of representative samples are drawn, in accordance with the ‘ lot ’ size; these samples are then blended and comminuted before drawing replicate analytical sam-ples. The blending of the primary samples is intended to minimize ‘ between sample ’ vari-ance and complivari-ance with the criteria is dependent upon the analytical variation between the analytical samples, which should be representative of the entire ‘ lot ’ . In these analyses allowance is made for the measurement and sampling uncertainty (see also below). Because of the possible risks of environmental contamination, blending of samples prior to drawing a portion for microbiological examination is not generally attempted.
Sampling Plans for Attributes or Variables?
In setting any sampling plan, it is essential to recognize that the level of sampling which would be required to provide a high degree of protection to both the manufacturer and the consumer cannot be achieved in practice. From the data presented earlier (Chapter 5), for a ‘ lot ’ consisting of, say, 2000 units, an Attributes Single Sampling Plan would require 125 sample units to be examined (Table 5.4). For an Acceptable Quality Level (AQL) of 1%, not more than two defective samples can be permitted (i.e. n 125, c 2), yet there would still be a producer risk (at 5% probability) of accepting up to 1.1% defective items and a 10% consumer risk of accepting 5.3% defectives. For microbiological examination, such a level of sampling is not viable from either a practical or an economical standpoint. However, if the level of sampling plan is not fi xed in relation to the ‘ lot ’ size and the AQL, but by the economic and practical constraints associated with testing, both the producer and consumer risks increase considerably.
For quantal data, Attributes plans provide the only option but this is not the case for quantitative data such as colony counts. Hence, one must question the wisdom of using Attributes schemes for microbiological criteria based on quantitative data. Of course, mak-ing a decision as to whether a specifi c colony count exceeds, or not, a predetermined limit is simple to understand and requires no statistical appraisal of the data. Calculation of the measurement uncertainty of counts on replicate samples is not required for Attributes plans. Yet, increasingly, those responsible for food examination are required by contract, or legis-lation, to derive and report values for measurement uncertainty to provide a measure of the precision of results (see Chapters 10 and 11). When only a limited number of representa-tive samples has been examined, the risks of making wrong decisions are very high, so it is sensible to use of all the available analytical data to assess whether or not a set of results is compliant with a defi ned limit. I would argue therefore that this justifi es the adoption of Variables sampling schemes for quantitative data, provided that knowledge exists of the microbial distribution in the foods. However, whichever scheme is adopted, it is essential to defi ne the degree of ‘ risk ’ that is acceptable to the manufacturer and the consumer.
Example 5.4 (based on Kilsby et al., 1979 ) illustrates a special version of a Variables sampling plan for which the food safety criterion was defi ned as ‘ reject with a 95% proba-bility any “ lot ” where the proportion exceeding C is ⱸ 10% ’ . The criterion is set, for n 5, such that the producer can expect to reject with 95% probability any ‘ lot ’ where 10% or more of the samples might have colony counts that exceed an upper limit of C. But, on average, the scheme recognizes that there is still a 5% risk that a ‘ lot ’ would be accepted if more than 10% of samples have counts that exceed C . The scheme also sets a GMP crite-rion that accepts with 90% probability those ‘ lots ’ where the proportion exceeding a lower limit ( C m ) is less than 20%; conversely, there is a greater than 10% risk of accepting a ‘ lot ’ where the proportion exceeding C m is greater than 20%.
EXAMPLE 14.1 COMPARISON OF ATTRIBUTES AND VARIABLES SAMPLING PLANS
The level of samples for examination necessary to provide a high degree of protection to both the manufacturer and the consumer cannot be achieved in practice. In Chapter 5, it was shown that for a ‘ lot ’ consisting of, say, 2000 product units, an Attributes Single Sampling Plan would require 125 sample units to be examined (Table 5.4). If the level of sampling is fi xed not in relation to the ‘ lot ’ size but by economic and practical constraints associated with testing, both the producer and consumer risks increase considerably.
Suppose that the data on colony counts used in Example 5.4 were applied to an Attributes sampling plan, defi ned as n 5, c 2, m 5 log 10 cfu/g and M 6 log 10 cfu/g.
The colony count data are:
Lot Log 10 cfu/g Mean ( )x SD ( s )
A 4.52; 4.28; 4.79; 4.91; 4.50 4.60 0.25
B 4.98; 5.02; 5.28; 4.61; 5.11 5.00 0.25
For an Attributes plan, each individual result is required to comply with the criteria, thus for lot A, since each of the counts is below the marginally acceptable limit ( m ), the ‘ lot ’ would be deemed acceptable. For lot B, where the counts have the same distribu-tion as lot A, two counts exceed m , two counts equal m (after ‘ rounding ’ 4.98 and 5.02 both equal 5.0) and one count is lower than m . This ‘ lot ’ would also be acceptable by the Attributes plan. But as we saw in Example 5.4, in the Variables scheme ‘ lot B ’ would have been acceptable on the basis of the safety/quality criterion but not in respect of the GMP requirement.
The acceptance criteria for the Variables sampling plan were:
● Reject the ‘ lot ’ with 95% probability if 10% or more of the colony counts exceed M ( C M
in the scheme) (i.e. a producer ’ s risk of 5%).
● Accept the ‘ lot ’ with a 90% probability if less than 20% of the colony counts exceed m ( C m in the scheme) (i.e. a consumer risk of 10%).
Consider an Attributes sampling plan based on similar criterion limits (i.e. m 5 log 10
cfu/g and M 6 log 10 cfu/g) and assume that no ‘ unacceptable ’ results (i.e. greater than 6
log 10 cfu/g) are obtained by examining fi ve replicate samples. For such a plan there would
be a 59% chance of accepting the ‘ lot ’ with a true incidence of 10% defectives, a 33% chance of accepting a lot with 20% true defectives and even a 5% chance of accepting a ‘ lot ’ with 45% defectives. Hence, the Attributes plan based on the same results but ignoring the variance of the test results, offers a lower standard of manufacturer and consumer pro-tection than does the Variables plan.
THE RELEVANCE OF MICROBIAL MEASUREMENT UNCERTAINTY TO MICROBIOLOGICAL CRITERIA
The CODEX Committee on analysis and sampling discussed an outline proposal ( Anon., 2004 ) that included a recommendation that ‘ Codex Commodity Committees concerned with commodity specifi cations and the relevant analytical methods should state, inter alia, what allowance is to be made for measurement uncertainty when deciding whether or not an analytical result falls within specifi cation ’ . The proposal also noted ‘ this requirement may not apply in situations where a direct health hazard is concerned, such as for food pathogens ’ . Whilst indicating a need to have a standardized approach to the use of uncer-tainty measures in interpreting microbiological data, no specifi c approach was proposed.
Previously, the UK Accreditation Service ( Anon., 2000c ) had provided guidance on how laboratories might cite and interpret analytical data and the European DG SANCO
Now for a three-class Attributes sampling plan for which the criteria limits are m 5.0 and M 6.0, the AQL will be dependent upon the number of replicate samples examined ( n ) and the number of results ( c ) that are marginally acceptable (i.e. 5.0 but 6.0). The proportion of defective or marginally defective samples that are likely to be accepted with a probability of 95% can be derived for various sampling plans using the method of Bray et al. (1973) , as described in Chapter 5 and illustrated in Table 5.10. If we assume zero defectives (i.e. no colony counts M ), then for a sampling plan with n 5, c 2 0 and either c 1 1
or 2, the AQL for acceptance of marginal defectives with a 90% probability would approxi-mate to 10% or 25%, respectively. However, there is also a 90% probability that either plan could accept at least 1% actual defectives with 5% or 20% marginal defectives. It is not pos-sible to determine with any certainty the proportion of marginal defectives that would be rejected on 90% of occasions but it is much larger ( 40% marginal defectives).
Hence, an Attributes plan for an identical number of samples ( n 5, c 1 2), would be
less discriminatory and would be more likely to accept a higher number of defective or marginally samples than would the Variables plan.
For this reason alone, one must question the justifi cation for basing microbiological criteria on Attributes schemes for legislative purposes. When only a limited number of samples can be analysed, best use of the analytical data can be achieved only by adoption of a Variables sampling scheme. However, whichever scheme is adopted, it is essential to defi ne at an early stage the degree of ‘ risk ’ that would be acceptable.
( Anon., 2003 ) assessed the problem associated with differing national interpretations of uncertainty in relation to compliance with defi ned limits. The DG SANCO report ( Anon., 2003 ) recommended that ‘ … . measurement uncertainty (should) be taken into account when assessing compliance with a specifi cation ’ . The report continued, ‘ In practice, if we are considering a maximum value in legislation, the analyst will determine the analytical level and estimate the measurement uncertainty of that level, subtract the uncertainty from the reported concentration and use that value to assess compliance. Only if that value is greater than the legislation limit will the control analyst be sure “ beyond reasonable doubt ” that the sample concentration of the analyte is greater than that prescribed by legislation ’ . Note that this recommendation is dependent upon expressing uncertainty on an additive scale, where the expanded uncertainty ( U ) can be subtracted from the estimated value to give the lower 97.5 % confi dence boundary.
Samples drawn from a ‘ lot ’ should, but may not, be truly representative of the ‘ lot ’ ( Anon., 2002d ) and the results of any examination will provide only an estimate of the true microbial population. As discussed previously, even in a well-controlled laboratory, quan-titative microbiological methods are subject to many sources of error that often lead to substantial estimates of measurement uncertainty. The expanded repeatability measurement uncertainty (i.e. the 95% confi dence boundaries) of an analysis can frequently extend to 5% of the mean colony count (i.e. up to 0.25 log cfu) from replicate tests on a single sample, and inter-sample and inter-laboratory variation results in even wider limits (see e.g. Corry et al., 2007b ; Jarvis et al., 2007a,b ). Furthermore, the intrinsic errors associated with MPN and other tube dilution methods are even greater, and the 95% confi dence lim-its (CL) are so wide, that these methods should generally be restricted to use in industrial quality assessment except where one is looking for specifi c changes in an otherwise stable situation (e.g. in potable water analysis).
The level of uncertainty attributable to sampling is but poorly documented. For chemi-cal analyses on foods the uncertainty due to sampling is greater than the measurement uncertainty ( Lyn et al., 2002, 2003 ) and limited data from Jarvis et al. (2007b) indicate that a similar situation may apply to microbiological analyses. Hence, the use of crite-ria limits that take no account of such vacrite-riation cannot be considered to be scientifi cally sound. For instance, if a mean log count is (say) 5.8 and the 95% measurement uncertainty limit is 0.5 log cycles then, on 19 occasions out of 20, the mean count indicates that the ‘ true ’ result lies within the log range 5.3–6.3, and a true result outside this range could be expected on one occasion in 20. An Attributes scheme makes no allowance for such varia-tion except in so far as the posivaria-tioning of the control limits.
There are different ways in which measurement uncertainty estimates could be applied in relation to criteria limits. For upper control limits, the suggestions of various bodies include:
1. Subtracting the expanded 95% uncertainty value from a mean result before comparing the ‘ corrected ’ result with the limit, to ensure ‘ compliance without reasonable doubt ’ (Anon., 2003).
2. Applying a ‘ guard band ’ to the limit by adding the uncertainty value to the limit before comparing the actual result (essentially the same effect as in (1) but avoiding the necessity to ‘ correct ’ each result) ( Anon., 2007 ).
3. Applying the general statistical procedures for CL of a mean value to permit comparison of the possible range of the true value with the limit.
4. Ignoring uncertainty estimates completely because wide estimates demonstrate poor laboratory technique – not necessarily true but arguable.
Some of these options are illustrated in Example 14.2.
EXAMPLE 14.2 CAN MEASUREMENT UNCERTAINTY BE USED IN ASSESSING COMPLIANCE WITH A CRITERION? Compliance of mean colony counts
EC legislation ( Anon., 2005b ) on microbiological criteria for swabs of meat carcases (other than pig for which different limits are set) uses a two-class plan and sets criteria limits for Enterobacteriaceae (2.5 log 10 cfu/cm 2 ) and aerobic colony count (5.0 log 10 cfu/cm 2 ) that
require compliance of the daily mean colony count with the absolute limit ( M ).
Suppose that replicate swabs of carcases are examined daily for aerobic colony count and mean values and estimates of expanded measurement uncertainty for the data are deter-mined. Suppose, also, that the expanded (95% probability) intra-laboratory reproducibility uncertainty is 0.5 log 10 cfu/cm 2 and the mean colony counts (as log 10 cfu/cm 2 ) on four
sepa-rate series of replicate examinations are A 4.0, B 4.6, C 5.4 and D 6.0 log 10 cfu/cm 2 .
Figure 14.3 illustrates the upper and lower bounds of the distribution limits of these four sets of mean colony counts together in relation to the criterion limit value ( M 5.0 log 10 cfu/cm 2 ). A 7.0 6.0 5.0 4.0 3.0 B Scenario Colon
y count (log cfu/cm
2)
C D
FIGURE 14.3 Implication of measurement uncertainty on mean values in relation to a crite-rion limit. Four scenarios showing mean counts (log 10 cfu/cm 2 ) of A 4.0, B 4.6, C 5.4 and
D 6.0 and a limit value of 5.0 log 10 cfu/cm 2 with upper and lower confi dence Intervals based on
Scenario A shows a mean colony count of 4.0 log 10 cfu/cm 2 with the upper boundary of
the 95% CL at 4.5 log 10 cfu/cm 2 ; here there can be no dispute – the test result is compliant
with the limit of 5.0 log 10 cfu/cm 2 . Scenario B shows a mean colony count of 4.6 log 10 cfu/
cm 2 but the upper boundary of the 95% CL (5.1 log
10 cfu/cm 2 ) exceeds the criterion limit.
Scenario C shows a mean colony count (5.4 log 10 cfu/cm 2 ) that is above the criterion limit
but the lower boundary of the 95% CL (4.9 log10 cfu/cm 2 ) is below the criterion limit –
are these data compliant with the limit or not compliant? Scenario D shows both the mean colony count of 6.0 log 10 cfu/cm 2 and the lower boundary of 5.5 log 10 cfu/cm 2 is also
above the criterion limit – here there can be no dispute; the results do not comply with the criterion.
Whilst the decisions regarding Scenarios A & D are clear-cut, the situation is less clear for data in scenarios B & C. Some might consider Scenario B to be acceptable, since the mean value is below the criterion limit and, if one follows the proposal of DG SANCO ( Anon., 2003 ) to deduct the expanded uncertainty value from the mean value, the ‘ cor-rected ’ mean value (i.e. the lower boundary of the 95% CL) is signifi cantly below the limit value. Hence it can be argued that the colony count is compliant with the limit. However, others might deem this to be unacceptable since the upper boundary of the 95% CL exceeds the criterion limit. For Scenario C , the mean value is above the criterion limit and might be interpreted as non-compliant; however, by deduction of the expanded uncertainty value the ‘ corrected ’ mean count is below the criterion limit, so the count is again com-pliant with the limit. In both cases, it would be ‘ beyond reasonable doubt ’ that the test data comply with the criterion. However, this approach is not generally accepted at this time.
Compliance of individual colony counts
Where a criterion requires that each individual result does not exceed a criterion limit, as in a normal two-class or three-class Attributes sampling plan, then the situation would be different. Suppose that, for the data from Scenario C above (mean count 5.4 log 10 cfu/cm 2 ),
the individual colony counts on fi ve replicate samples were 5.6, 5.0, 5.1, 5.5 and 5.8 log 10
cfu/cm 2 . If the two-class sampling plan was, for instance, n 5, c 1, M 5.0, then four
of the fi ve counts exceed the criterion limit value (5 log 10 cfu/cm 2 ) so the data set would
be deemed to be non-compliant. But if the expanded measurement uncertainty were to be subtracted from each individual value, the ‘ corrected ’ values would be 5.1, 4.5, 4.6, 5.0 and 5.3 log 10 cfu/cm 2 , respectively. Hence, although the corrected mean value is below the
limit value (and therefore compliant), the individual colony counts would be deemed to be non-compliant since two of the individual corrected values would exceed the criterion limit.
Guidelines have been published for expression and use of measurement uncertainty in chemical analyses (see for instance Anon., 2007 ), but none exist for data on micro-biological examination. The microbiology community needs to consider this matter on an international basis and to come forward with proposed Guidelines on compliance with criteria.
Eurachem ( Anon., 2007 ) recommends the approach in which ‘ guard bands ’ are set around a criterion limit to show the level of tolerance that can be given for a particular form of analysis. The width of the guard band is determined by knowledge of acceptable limits to the measurement uncertainty of the test method and forms part of the decision process in setting criteria. The benefi t of this approach is that there is a clear statement of intent with regard to the measurement uncertainty that is deemed acceptable for a specifi c form of analysis. Such an approach provides a demonstration of greater transparency in setting and interpreting criteria.
It can be argued that a three-class sampling plan for microbiological criteria already incorporates the guard band concept since the lower limit ( m ) is the actual acceptance limit whilst the upper criterion limit ( M ) is an absolute that must never be exceeded. However, there is a signifi cant difference in approach, since microbiological three-class plans are pred-icated on the interpretation of each of n results against the criteria, rather than the average of the n colony counts.
This illustrates one of the many issues concerning measurement uncertainty that still needs to be addressed internationally for microbiological examination of foods. It again raises the issue of whether it is justifi able to convert variables data into ‘ simple ’ ‘ go-no go ’ values for use in an Attributes sampling scheme; or whether the time is now right to change to use of a Variables sampling scheme that takes account of all the data relevant to a set of colony counts on replicate samples drawn from a ‘ lot ’ . Failure to make a clear decision on this approach will hinder future development on microbiological criteria for foods.
Criteria based on presence or absence of a particular organism or group of organisms are affected not only by the distribution of the target organism(s) within the food but also by the adequacy, or otherwise, of the test procedures. Hence, there is necessity to use accred-ited methods with the level of sensitivity and specifi city suitable for the intended purpose. Even then, tests giving negative results on a number of individual replicate samples do not guarantee freedom from the organism in question; they merely indicate that the probability for occurrence of the organism in the ‘ lot ’ is within certain tolerances. The tolerance will be dependent upon the number of samples tested and, of course, on the effi ciency of the method used. However, in the same way that a scheme showing apparent absence of specifi c organisms cannot guarantee total absence of the organisms from the product, the detec-tion of one or more positive samples could also arise by chance. Hence, product of equiva-lent quality could be rejected on one occasion yet be accepted on another occasion if only a small number of samples are tested.
For a lot having, say, 1% defectives (i.e. 1% of the 25 g sample units derived from a lot contain at least one detectable salmonella), then if 10 samples are tested on average one would expect to detect salmonellae and therefore reject the lot (if c 0) on 10 occasions out of 100, yet not detect them and accept the lot on 90 occasions (Table 5.1). However, if the true preva-lence of defectives is only 0.1%, on average, then tests on 10 samples would expect to detect salmonellae only once in every 100 tests. The analyst is unable to differentiate between false negative and true negative results when carrying out quantal tests on real-life samples, whilst detection of confi rmed positive results indicates that the presumption for a high prevalence of contamination is probably well founded.
CONCLUSION
In establishing microbiological criteria, including sampling plans, due cognizance is required of the effects of sample numbers on the effi ciency of the laboratory examination and on the overall work of the laboratory. The potential cost of intensifying testing schemes needs to be balanced against the costs of unnecessarily rejecting valuable food materials and/or of increasing the risk of spoilage or food-borne disease by accepting defective products. Those responsible for setting criteria must defi ne the AQL, Producer ’ s Risk and Consumer ’ s Risk and must also recognise the imprecision of microbiological methods. It is important, there-fore, that a ‘ transparent ’ approach to the setting of criteria is adopted including a more generally accepted set of decision rules for criteria. It is also essential that internationally acceptable Guidelines should be published for use of measurement uncertainty in compli-ance assessment. Until such decisions have been taken it is dubious whether meaningful criteria can be set for microbiological examination of foods.
Effective microbiological control comes from use of GMP and HACCP and other control strategies at all stages of food production, distribution and storage, based on knowledge of the microbial ecology of particular foods under different process and storage conditions. Such control strategies also require effective inspection assessments of manufacturing proc-esses. End-point testing on manufactured foods is effective only as a means of retrospective assessment of the process and storage conditions. However, SPC provides trend analysis of microbiological data as an additional means of monitoring change in a manufactur-ing process. The distribution of organisms in foods and the statistical variation associated with methods of enumeration lead to the conclusion that at present microbiological criteria should not be used other than as guidelines and specifi cations. Except in the case of high levels of contamination by pathogens, there is little or no evidence to show any benefi ts to food safety of the imposition of legislative microbiological criteria for foods, other than in the retrospective investigation of food poisoning incidents.
References
Accum , F ( 1820 ) A Treatise on Adulteration of Food and Culinary Poisons . Longman, Hurst, Rees, Orme & Brown , London .
Amos , AJ ( 1960 ) Pure Food and Pure Food Legislation . Butterworths , London .
Anon. (1997) Principles for the establishment and application of microbiological criteria for foods. 3rd edition. Codex Alimentarius Guidelines, Report No CXG 21e. WHO/FAO, Rome.
Anon. (2000a) The interactions between assessors and managers of microbiological hazards in food. Report of a WHO Expert consultation, Kiel, Germany, 21–23 March 2000 . WHO, Geneva. Anon. (2000b) Joint FAO/WHO Expert Consultation on Risk Assessment of Microbiological Hazards
in Foods FAO Food and Nutrition Paper 71 . WHO/FAO, Rome.
Anon. ( 2000 c ) The Expression of Uncertainty in Testing. UKAS Publication LAB 12 . Laboratory of the Government Chemist , London .
Anon. ( 2002 a ) Regulation (EC) No. 178/2002 of the European Parliament and of the Council of 28 January 2002 laying down the general principles and requirements of food law, establishing the
European Food Safety Authority and laying down procedures in matters of food safety . Offi c. J.
Eur. Comm. , L31 , 1 – 24 . 1 February 2002.
Anon. (2002b) Principles and guidelines for incorporating microbiological risk assessment in the development of food safety standards, guidelines and related texts. Report of a joint FAO/WHO
Consultation , Kiel, Germany, 18–22 March 2002 . FAO/WHO/Institute for Hygiene and Food Safety of the Federal Dairy Research Centre, Kiel .
Anon. ( 2002 c ) Proposed draft principles and guidelines for the conduct of microbiological risk
man-agement. Joint FAO/WHO Food Standards Programme. Codex Committee on Food Hygiene 35th session. CX/FH 03/7 . FAO and WHO , Rome and Geneva .
Anon ( 2002 d ) Guide on Sampling for Analysis of Foods. NMKL Procedure No. 12. NMKL . National Veterinary Institute , Oslo, Norway .
Anon. (2003a) The relationship between analytical results, the measurement uncertainty, recovery fac-tors and the provisions in EU food and feed legislation. Report to the EU Standing Committee on
the Food Chain and Animal Health Working Group Draft, 5 June 2003.
Anon. (2004) The use of analytical results: sampling plans, relationship between the analytical results, the measurement uncertainty, recovery factors and provisions in Codex standards. CODEX Report ALINORM 04/27/23 Appendix VII . FAO/WHO, Geneva.
Anon. (2005a) Food safety management systems – requirements for any organisation in the food chain. ISO 22000, 2005 .
Anon. ( 2005 b ) Commission Regulation No. 2073/2005 of 15th November 2005 on Microbiological Criteria for Foodstuffs . Offi c. J. Eur. Union , L338 , 1 – 26 of 22 December 2005.
Anon. (2006a) Food Safety Risk Analysis: A Guide for National Food Safety Authorities. FAO Food
and Nutrition Paper 87. WHO/FAO, Rome .
Anon. ( 2006 b ) Commission Regulation (EC) No. 1881/2006 of 19 December 2006 setting maximum levels for certain contaminants in foodstuffs . Offi c. J. Eur. Union , L364 , 5 – 24 of 20 December 2006. Anon. (2007) Eurachem/CITAC Guide: Use of Uncertainty Information in Compliance Assessment . Ellison, S L R and Williams, A (eds). Eurachem, LGC, Teddington ( http://www.eurachem.org/ guides/Interpretation_with_expanded%20uncertainty_2007_v1.pdf ).
Baird-Parker , AC ( 1980 ) Food microbiology in the 1980s . Fd. Technol., Australia , 32 , 70 – 77 . Bauman , HE ( 1974 ) The HACCP concept and microbiological hazard categories . Food Technol. ,
28 ( 9 ) , 30 – 34 and 74.
Bray , DF , Lyon , DA , and Burr , IW ( 1973 ) Three class attributes plans in acceptance sampling . Technometrics , 15 ( 3 ) , 575 – 585 .
Charles , RHG ( 1979 ) Microbiological standards for foodstuffs . Health Trend , 11 , 1 – 4 .
Corry , JEL , Hedges , AJ , and Jarvis , B ( 2007 b ) Measurement uncertainty of the EU methods for micro-biological examination of red meat . Food Microbiol. , 24 , 652 – 657 .
Gilbert , RJ , de Louvois , J , Donovan , T , Little , C , Nye , K , Ribeiro , CD , Richards , J , Roberts , D , and Bolton , FJ ( 2000 ) Guidelines for the microbiological quality of some ready-to-eat foods sampled at the point of sale . Commun. Dis. Public. Health , 3 , 163 – 167 .
Haas , CN ( 1983 ) Estimation of risk due to low doses of microorganisms, a comparison of alternative methodologies . Am. J. Epidemiol. , 118 , 573 – 582 .
Haas , CN ( 2002 ) Conditional dose-response relationships for microorganisms: Development and application . Risk Anal. , 22 , 455 – 463 .
Havelaar , AH , Nauta , MJ , and Jansen , JT ( 2004 ) Fine-tuning Food Safety Objectives and risk assess-ment . Int. J. Food Microbiol. , 93 , 11 – 29 .
ICMSF ( 1986 ) Microorganisms in Foods. 2 Sampling for Microbiological Analysis: Principles and
ICMSF ( 1988 ) Microorganisms in Foods. 4. Application of the Hazard Analysis Critical Control
Point (HACCP) System to Ensure Microbiological Safety and Quality . Blackwell , Oxford .
ICMSF ( 2002 ) Microorganisms in Foods. 7. Microbiological Testing in Food Safety Management . Kluwer Academic/Plenum Publishers , New York, USA .
ICMSF ( 2005 ) Microorganisms in Foods. 6. Microbial ecology of Food Commodities . Kluwer Academic/ Plenum Publishers , New York, USA .
Ingram , M ( 1961 ) Microbiological standards for foods . Food Technol. , 15 ( 2 ) , 4 – 16 .
Ingram , M ( 1971 ) Microbiological Standards for foods . Food Industry South Af. , 24 ( 2 ) , 8 – 13 . Ito , K ( 1974 ) Microbiological critical control points in canned foods . Food Technol. , 28 , 46 – 48 . Jarvis , B , Corry , JEL , and Hedges , AJ ( 2007 ) Estimates of measurement uncertainty from profi ciency
testing schemes, internal laboratory quality monitoring and during routine enforcement examination of foods . J. Appl. Microbiol. , 103 , 462 – 467 .
Jarvis , B , Hedges , AJ , and Corry , JEL ( 2007 ) Assessment of measurement uncertainty for quantita-tive methods of analysis: Comparaquantita-tive assessment of the precision (uncertainty) of bacterial colony counts . Int. J. Food Microbiol. , 16 , 44 – 51 .
Jouve , J-L ( 1999 ) Establishment of food safety objectives . Food Contr. , 10 , 303 – 305 .
Jouve , J-L ( 2000 ) Good manufacturing practice, HACCP and quality systems , In Lund , BM , Baird-Parker , AC , and Gould , GW (eds.) The Microbiological Safety and Quality of Food . Vol. II ch 58. As per, Publisher , Gaithersburg, Maryland . pp. 1627–1655
Kilsby , DC , Aspinall , LJ , and Baird-Parker , AC ( 1979 ) A system for setting numerical microbiological specifi cations for foods . J. Appl. Bacteriol. , 46 , 591 – 599 .
Leistner , L and Gould , GW ( 2001 ) Hurdle Technology: Combination Treatment
for Food Stability, Safety and Quality . Kluwer Academic/Plenum Publishers , New York .
Lyn , JA , Ramsey , MH , and Wood , R ( 2002 ) Optimised uncertainty in food analysis: application and comparison between four contrasting ‘ analyte–commodity ’ combinations . Analyst , 127 , 1252 – 1260 .
Lyn , JA , Ramsey , MH , and Wood , R ( 2003 ) Multi-analyte optimisation of uncertainty in infant food analysis . Analyst , 128 , 379 – 388 .
Mossel , DAA ( 1982 ) Microbiology of Foods: The Ecological Essentials of Assurance and Assessment
of Safety and Quality , 3rd edition . University of Utrecht , Utrecht .
Mossel , DA and Drion , EF ( 1979 ) Risk analysis. Its application to the protection of the con-sumer against food-transmitted diseases of microbial aetiology . Antonie van Leeuwenhoek , 45 , 321 – 323 .
Peterson , AC and Gunnerson , RE ( 1974 ) Microbiological critical control points in frozen foods . Food
Technol. , 28 , 37 – 44 .
Rieu , E , Duhem , K , Vindel , E , and Sanaa , M ( 2007 ) Food Safety Objectives should integrate the vari-ability of the concentration of pathogen . Risk Anal. , 27 , 373 – 386 .
Roberts , TA and Jarvis , B ( 1983 ) Predictive modelling of food safety with particular reference to Clostridium botulinum in model cured meat systems . In Roberts , TA and Skinner , FA (eds.) Food
Microbiology: Advances and Perspectives . Academic Press , London .
Szabo , EA , Simons , L , Coventry , MJ , and Cole , MB ( 2003 ) Assessment of control measures to achieve a Food Safety Objective of less than 100 cfu of Listeria monocytogenes per gram at the point of consumption for fresh pre-cut iceberg lettuce . J. Food Protect. , 66 , 256 – 264 .
Thatcher , FS ( 1955 ) Microbiological standards for foods: Their function and limitations . J. Appl.
Bacteriol. , 18 , 449 .
Thatcher , FS ( 1958 ) Microbiological standards for foods. II. What may microbiological standards mean? . Food Technol. , 12 , 117 – 122 .
Tuynenburg-Muys , G ( 1975 ) Microbial safety and stability of food products . Antonie van
Leeuwenhoek , 41 , 369 – 371 .
Wallace , CA and Mortimer , S ( 1998 ) HACCP – A practical approach , 2nd edition . Aspen Pub , Gaithersburg, Maryland .
Whiting , RC ( 1995 ) Risk Analysis. A Quantitative Guide . Wiley , Chichester, UK .
Wilson , GS ( 1955 ) Symposium on food microbiology and public health: general conclusion . J. Appl.
