Context, targets and scientific issues
Nowadays a qualitative hazard-based approach of food safety is implemented by food business operators. This can result in control measures that are either more or less efficient than needed. The first case can be costly in terms of finance and/or deterioration of the nutritional or organoleptic properties of the food. The second case can be costly too, mostly in terms of consumers’ safety.
Experience rather than a scientific approach is still too often the base for decisions made by food business operators. At a time where the new European Food Law is enforced, more rigorous methods are required to assess and validate the outcome of control measures. Indeed this is true not only for food process operators but also for the competent authority
- To better set the context and its current evolution a quote from a paper from the European Network of Excellence Med-Vet-Net might be useful: “Strategies to control hazards in food have evolved from generic food hygiene requirements to more targeted systems based on the HACCP system. In parallel, food safety management by public authorities evolved from rigid ‘command and control’ systems to a more flexible approach. The role of governments is changing from prescribing how food business should operate towards setting targets for what they should achieve. Key targets from a governmental perspective are related to public health goals. Setting of public health targets and risk-based food safety management systems should be a science based process, according to international agreements.” (Havelaar, 2007).
Targets
- Targets established by food business operators for their microbiological hazard control measures, in the context of the application of the HACCP System (Codex Alimentarius General Principles of Food Hygiene), are called “critical limits”. Thus, at the level of each CCP, a measurement has to be performed and compared to a critical limit, defined as a criterion which separates acceptability from unacceptability. “Critical limits must be specified and validated” by “obtain[ing] evidence that the elements of the HACCP plan are effective”. Yet the term “acceptability” is loosely defined. The standard EN ISO 22000:2005 indicates “The [acceptable] level shall take into account established statutory and regulatory requirements, customer food safety requirements, the intended use by the customer and other relevant data”. Nowdays no information easily accessible to operators is given as to the acceptable levels underlying the food safety criteria in the EU Regulation 2073/2005 on microbiological criteria. Furthermore not all pathogenic micro-organisms are included in the latter. For example microbiological criteria are not applicable to highly virulent hazards the prevalence and concentration of which should remain low, such as shiga-toxinproducing Escherichia coli (STEC), since impracticable sample size would be needed. Yet “food shall not be placed on the market if it is unsafe” (EU Regulation 178/2002). Therefore the food business operators are faced with a difficult situation.
- Targets established by governments in relation to public health goals could be “The maximum frequency and/or concentration of a hazard in a food at the time of consumption” (food safety objective, FSO) or “at a specified step in the food chain before the time of consumption” (performance objective, PO) (Codex Alimentarius Procedural Manual, 15th ed. 2004). FSOs and Pos should be established “to provide or contribute to the appropriate level of protection” (ALOP) to which the regulatory body does refer. Targets might be also expressed so as to account for the distribution of the hazard concentration, as suggested in the report of a meeting of the WHO Collaborating Centre on Risk Assessment at RIVM (Bilthoven, The Netherlands, 2006-12-01).
-
At the moment there is some reluctance among risk managers to enter into the process of articulating the ALOP. This reluctance extends to the establishment of FSOs. Nevertheless the PO concept seems to gain more recognition. Two other concepts strongly linked to the PO one are widely accepted as they were in fact used for a long time: “process criteria” and “product criteria” describing the effect on microbial concentration achieved respectively by a process or by food composition factors so as to achieve the PO. According to the Codex Alimentarius or to FAO/WHO Consultations, Pos could be established by the competent authority or the operators at the retail step, or by the operators at any earlier step in the process up to the primary production.
-
Despite the actual risk-based tendency, the link between hazard-based critical limits and risk based objectives does not seem to be established so far.
Scientific issues
The microbiological quality of a final product is the result of numerous operations along the process where different ingredients are progressively aggregated and transformed in different equipments in which the process parameters (temperature, flow rates, etc.) are defined according to specific constraints (safety, quality, acceptance of the final product, etc.). If everything were perfect and completely defined one could imagine deduce the quality of the final product by a deterministic relationship from the characteristics of the inputs. This is not the case! Many factors lead to reject such a simplified vision and oblige to consider also some randomness of the inputs and outputs. Roughly speaking, there is a need to introduce the variance besides the main tendencies (expectation) of a modelled random variate. The sources of variation are numerous and have to be described for each specific case. Nevertheless two main categories are usually distinguished: (i) variability, i.e. the true heterogeneity of the system, including fixed effects (e.g. the manufacturer can choose between different ways of production), and random effects (e.g. the microbiological status of a given ingredient varies from one batch to another, and may be described by a Poisson distribution) and (ii) uncertainty, including parameter uncertainty (e.g. the parameter of the Poisson distribution is unknown by the modeller), scenario uncertainty (e.g. the modeller is not sure all possible ways of production have been described), and model uncertainty (this can be the consequence of not knowing the precise mechanisms acting in the process, or most often the impossibility of observing some important covariates, for instance the evolution of the temperature, the exact proportion of mixed ingredients, etc.).
In the classical quantitative risk assessment (QRA) approach, it is recommended to separate variability on the one hand from uncertainty on the other hand. Quite schematically, the idea is (i) to fix uncertain parameters of the model to their central values (for instance mean or median) getting a most probable distribution of the output; also (ii) to let them take different values according to their uncertain distribution, getting for each value a possible distribution (this distribution of distributions allows to assess the confidence we can have in the most probable distribution). However, as stated in a European report (European commission, 2003), "Separation of variability and uncertainty in QMRA models (so-called second order models) has up to now rarely been made, a reflection of the fact that this can be a daunting task." This observation, which was definitely true in 2002, still partly applies, even if new attempts have since been published, including some from the partners of this proposal (AFSSA, 2002, 2006; Albert, 2005; Crépet, 2007; Pouillot 2003, 2004, 2007; Delignette-Muller, 2006).
Nevertheless, notice that things are not so clear-cut, as the limits between variability and uncertainty depend on the point of view, and the scope of the model. When modelling at the level of a country, the fixed effect of the chosen process will be probably converted as a random effect since the modeller may decide to describe that not all the manufacturers will have chosen the same way (instead of modelling the effect of each way). If the quality of an input can be controlled, then it could become a fixed effect. This must be decided when defining the scope and the aim of the model.
To date, this modelling approach is usually applied from the point of view of risk managers in an objective of risk assessment, i.e. applied to the consumption of a whole country, as those prepared by national food safety agencies / authorities, or even applied to a theoretical world-vide consumption, as those prepared at international level and published by the Joint FAO/WHO Consultations on Microbiological Risk Assessment (JEMRA). We list hereafter only those published since 2002 by the full partners to this project: vegetable purees (Afchain, 2007), cold-smoked salmon (AFSSA, 2006; Pouillot, 2007), fresh vegetables (Crépet, 2007); smeared cheese (Aziza, 2006), a turkey based product (Bemrah, 2003), potted meat (Cornu, 2006), chicken (AFSSA, 2004; Denis, 2006), water (Afssa, 2002; Pouillot, 2004), soft cheese made from raw milk (Sanaa, 2004). They focus on hazard characterization, exposure assessment and risk characterization. The detailed description of processes is partial or even not done and control measures along the food chain, whether at the primary production or food production lines, are poorly or not described. Some of them provide interesting insight on the processes, but still limited indication on control measures that could be applied by the manufacturers in food manufacture premises and on production lines. Therefore they do not provide assistance to the food business operators as they were not designed to do so. More importantly, they do not provide a mean for the food business operators to adapt their targets to Pos, whether set by themselves or the competent authority.
When this project is written, only a few attempts to do the latter have been published. Of course many articles present the concepts in quite general terms (see for example Food Control 16 (9), 2005). But as regards application to practical cases, we can only quote recent documents from the Codex Committee on Food Hygiene. These documents are not intended at the food business operators. They offer examples of models either deterministic (heat inactivation of Salmonella in eggs) or pseudo-stochastic (Listeria monocytogenes in cold-smoked salmon). Yet, we consider that these attempts present only simplistic examples of the underlying principles. Practical recommendations, simple to implement but accounting for variability, and applicable by the food business operators are clearly needed.
