Sampling, sample preparation, logistics

The sampling of foodstuffs in order to estimate their nutritional composition is an area in which the use of statistical models and regulatory or normative texts is a complex matter. However, general recommendations arising from the scientific literature can be used. Knowledge of fishing and farming practices is essential to supplement these general recommendations. A set of specifications has been drawn up to accurately describe methods of sample taking, preparation and transportation. The average contents presented in the nutritional forms on this site are averages obtained in the five analyses performed with different batches. The part analysed was the edible part (for example, the skinless raw fillets for the fish).

Once the choice of species and nutrients to be analysed is made (see the corresponding methodology forms), the aim is to define the sampling plan from the practical point of view: how many samples to select and how? Where and when to carry out the sampling? The samples must then be prepared and sent for analysis under circumstances to be specified.

1.a. Is it possible to use a statistical model?

In order to construct a rigorous sampling plan, and particularly to determine the number of analyses and individuals to be sampled, we can reference a statistical and/or regulatory model.

As regards the number of analyses, if we take the hypothesis that the content of a nutrient among individuals (in this case, the fish from a particular species, for example) is distributed around the average according to a normal distribution, the mathematical formula which gives the number of samples to take is: 

n = (z(a)/r)2*(CV)2

where z (a) and CV are defined as follows:

•  z (a) = value associated with the probability of error a (located in a table)
• r = desired precision (for example: 10%, 5%...)
• CV = variation coefficient, in other words standard deviation quotient divided by the average level of the nutrient being considered in the population in question.

The issue of knowing the standard deviation and the average content of the population in question is therefore raised... Not knowing these data, we may replace them with the calculated estimations of these parameters in more limited populations. By way of an exercise, we took as an estimation the values determined by the United States Department of Agriculture (USDA, http://www.nal.usda.gov/fnic/foodcomp/search/) for fish eaten in the United States. This approach has three limitations:

1. We are in no way certain that that the levels of nutrients in aquatic products consumed in the United States are identical to the levels of nutrients in aquatic products consumed in France, and that these levels follow similar distributions.
2. This calculation is not possible for all nutrients and all species because the USDA does not systematically provide the standard deviation (or standard error). Sometimes, the number of nutrient levels determined by the USDA is not sufficient to calculate a significant standard deviation.
3. When standard deviation and average content are not available, for some nutrients the calculation of the number of analyses to be performed according to the above formula gives a result in the order of hundreds or even thousands of analyses (even for a low level of precision), which is totally unrealistic, given the analytical cost.

For these three reasons, this statistical approach was not used.

1.b. Do normative or regulatory texts enable us to determine a sampling plan?

It is clear that there are no standards (AFNOR, CEN, Codex Alimentarius, GAFTA ISO) defining sampling plans. However, the aim of these standards is to determine the commercial quality of products in relation to a target value and not to determine the average levels of a nutrient. Furthermore, European directives specify sampling methods, but for contaminants (heavy metals, dioxins, pesticides, etc.) or GMOs, for which, again, the issues and the starting hypotheses are completely different to those in our project.

1.c. Finally, the sampling of foodstuffs in order to analyse their nutritional characteristics is based on a pragmatic approach.

Not being able to establish a sampling plan on statistical bases or according to normative or regulatory texts, we drew on the recommendations in the landmark work in the production of data on nutritional composition: Food composition data, production, management and use by H. Greenfield and D. A. T. Southgate, second edition, FAO Rome 2003. A set of sampling, preparation and logistical specifications was drawn up.

The samples must be as representative as possible of consumption; selective sampling was performed, taking into account specific product characteristics:

• For wild fish, their sites and periods of supply were selected during peak catch periods and in the ports were the greatest tonnages are landed.
• For fish which vary greatly in their fat composition over the year (“fatty” fish, such as the Atlantic herring and the Atlantic horse mackerel), sampling was performed over two periods: when the fish is at its biggest and when it is at its smallest.
• For fish produced by French aquaculture, such as the sea bass, bream, trout, turbot and carp, the samples were taken from different farms. 
• For other fish produced by aquaculture, such as salmon, Nile tilapia, Basa and Nile perch, the different samples were taken from different fish wholesale traders or dealers in fresh or frozen form. 
• For shellfish and crustaceans

For oysters and mussels, samples were taken in the main production sites. Mussels were cooked by the technical centres before analysis.

The different types of prawns were bought cooked from different producers.

• for transformed (preserved) products, the brands to be sampled were selected on the basis of data from the 2004 SECODIP household purchase panel.

For each product, five samples were prepared from five different batches (one sample per batch). Each batch is made up of at least five individuals. 

A batch may be composed of:

• for wild products, a trawl line, a boat…
• for aquaculture products, a producer, a pond…
• for transformed products, a brand, a place of origin…

The figure of five batches was chosen because it would give a snapshot of the variability in levels of each nutrient within the species under consideration and satisfy budgetary constraints. This figure may at first glance seem low. But if we consider the data currently available in the CIQUAL or foreign composition tables, we will see that it is quite rare to use five values with this level of representativeness to determine an average value. Using five analyses per species and per nutrient therefore represents very significant progress in relation to data quality levels previously available in French composition tables.

Each sample was itself constituted from a minimum of ten individuals taken at random from the batch. Here again, we arrived at the figure of ten empirically. Using a starting sample of at least ten individuals to build a homogenous sample is an attempt to limit intra-batch variability. Within a batch of wild fish, there may be individuals of different size and sex, which may impact on nutritional composition. 

When sampling, other precautions were taken: operators at the technical centres (experienced professionals) checked the scientific name, the fishing area, the fishing date, the freshness criteria (these had to be "extra" or "A", or sometimes "B" but for a maximum of two parameters. Finally, digital photos of different species were taken, with a ruler used to indicate their size. 

The sampled aquatic products were generally bought raw (except prawns and, of course, the preserved products). Preparation did not involve cooking, following the results of the study "Etude de l’influence des traitements de conservation et de transformation alimentaires sur les acides gras du poisson" (Study of the infuence of food preservation and transformation treatments on fatty acids in fish) ACTIA/CEVPM/AFSSA/IFREMER/IPL/ITERG/ Pôle Filière Produits Aquatiques (National Cluster for Aquatic Products) carried out in 2002 and 2005. (With the exception of the part in which AFSSA (the French Food Safety Agency) was involved). This study indicated that cooking did modify the nutrient levels in foodstuffs (which comes as no surprise, especially given water transfers), but that it was very difficult to carry out repeatable, reproducible cooking procedures. Taking into account cooking in our study would have added a very significant variability factor which would have been very difficult to control. However, should some data users want to know the nutrient contents of a cooked product, we recommend that they perform calculations based on data obtained as part of this project and a coefficient of nutrient retention during cooking produced in studies carried out by the United States Department of Agriculture and available at the following address http://www.nal.usda.gov/fnic/foodcomp/Data/retn6/retn06.pdf

Only certain products were cooked by the technical centres (mussels, whelks, pink shrimps)

Fish bought whole will be superficially peeled. Indeed, fillets are most often eaten without skin, and leaving it would not produce standard homogenates. The brown muscle is in general left on the fillets, except for large fish like tuna and halibut. For these species the type of peeling used should be specified because there are significant fat deposits under the skin. Ideally, so that the sample analysed is as representative as possible of what is available on the market, machine peeling would be preferable, but as this cannot be done by the technical centres, we used manual peeling.

Trimming will also be performed: in other words, the removal of the stomach wall. The flanks will be removed.

As many bones as possible will be removed from large and small fish. For small fish like sardines, it is however possible that in many cases bones remain, which may increase the calcium content of the samples. However, this is the same way as they are eaten.

Shellfish are of course analysed without their shells, and prawns are also shelled, since the analysis must focus on the edible part.

In order to produce a homogenous batch, the products were crushed, vacuum-packed in plastic bags and kept in the dark, before being sent to different laboratories.

Fresh aquatic products have quite short lifespans, in order to best preserve their quality and facilitate logistics. The samples were frozen before being sent to the different laboratories.

The products were frozen to -20°C and transported in specialist containers at -20°C.

When they arrived at the different laboratories, the temperature of the products was checked to ensure that they were not defrosting.

For defrosting before analysis, the defrosting protocol used was the one specified in the COFRAC 80 programme (0°C maximum air temperature).