J.M. Holden
Joanne M. Holden is a nutritionist at the Food Composition Laboratory, Beltsville Human Nutrition Research Center, United States Department of Agriculture, Beltsville, Maryland, USA.
Accurate values for energy, nutrients and other food attributes are required to calculate levels of intake, to monitor the adequacy of the food supply relative to human nutrient requirements, to formulate and label new products and to facilitate trade. For accuracy, a specific estimate must be statistically representative of the universe of all values for the component in the food product of interest. Serious bias in the estimate can lead to erroneous conclusions and costly mistakes affecting dietary assessment and trade.
The Food Composition Laboratory of the United States Department of Agriculture (USDA) has conducted research to develop strategies for sampling the United States food supply to estimate values for components in many foods, This work requires description of foods in terms of the food source, ingredients, preparation methods, preservation state, cultivar and other factors that may influence component levels. Marketing and demographic data can be used to identify parameters that are potential sources of variance. Demographic data may indicate the use of foods by different population subgroups which may be distinguished by ethnicity, gender, age or dietary habits (for example, vegetarians). In addition, protocols for sample handling and chemical analysis should be standardized for reproducibility and to minimize the impact of variability in the measurement process.
In many countries the average daily diet may contain ten to 25 different food items. Since a nation's food supply is a complex mixture of processed and non-processed products, each food item has diverse forms representing many brands, formulations and geographical sources. Since it is not possible or desirable to analyse every available package of a food, it is necessary to develop a strategy for selecting units of foods to obtain a representative sample of the population of all available units. Limits on time, personnel and fiscal resources and the abundance of new and different food products in the marketplace in any country require that strategies for selecting representative units of foods for analysis be based on well-defined statistical principles as well as scientific objectives incorporating the intended use of the food composition data.
OBJECTIVES FOR SAMPLING
There are a number of possible specific objectives for sampling. These include:
· the development of a national food composition database;
· the determination of aflatoxin levels in a load of grain;
· the determination of pesticide levels in a food product;
· quality control of food manufacturing;
· the determination of significant differences in the vitamin content of different animal muscles;
· comparisons among regions or brands,
The definition of the objective helps to determine the most appropriate sampling strategy (Holden et al., 1987). If the objective is to develop a national food composition database, then two major questions need to be answered. What foods should be selected for analysis? What nutrient(s) or component(s) should be measured? Food analysis projects can estimate levels of a single component (for example, selenium, (3-carotene or total fat) in foods consumed by a population of individuals, or they may focus on a single food (for example, beef, milk or carrot) and its major components.
WHAT COMPONENTS SHOULD BE ANALYSED?
The components of interest may be nutrients, additives, natural toxicants or contaminants, Each component or class of components represents a unique sampling challenge, In general, three factors determine the selection of components. First, the component should be important in terms of actual or suspected public health effects. Second, the available analytical methods for the component(s) of interest should be robust, valid, capable of producing accurate data and economically feasible, Third, in view of fiscal and personnel limitations, analytical priorities should include those components for which available data are unacceptable (Greenfield and Southgate, 1992).
WHAT FOODS SHOULD BE SAMPLED?
In the selection of foods, priority should be given to those that are the major contributors of the component to the diet. In general, a limited number of foods (five to 100) contribute 50 to 90 percent of a single component to the diet of a population (Beecher and Matthews, 1990; Schubert, Holden and Wolf, 1987), Foods for which data are unacceptable or unavailable should be selected.
Since consumer preferences for foods or for specific forms of those foods change in response to food availability, market pressures, socio-economic factors and scientific information, foods should be analysed as eaten, Food consumption data can be used to identify the forms and methods of preparation of foods as well as their frequency of consumption, For example, in the United States some sectors of the population now often choose grilled or roasted chicken parts instead of deep-fat fried parts. Therefore, accurate and current data are needed for grilled or roasted chicken to reflect the impact of new dietary trends, In some cultures chicken or fish may be prepared whole by stewing or baking until the bones soften, and all parts, including the bones, are eaten. In this case, the food should be prepared as consumed before analysis.
SPECIAL SITUATIONS
Some foods may be important sources of nutrients or other components for certain population subgroups (for example, young children and pregnant women) but may not be important in the diets of the general public, Such foods (for instance, infant formula) may be the primary source of the component in the diets of the subgroup. Therefore, accurate data for representative samples are required, Some foods may contain high levels of specific components but may not be consumed frequently or in large amounts, For example, beef, liver and oysters are high in copper, Sampling plans could be developed to provide analytical data to assess the levels of copper intake by specific groups or individuals, Similar data may be needed to assess suspected contamination of specific foods or to identify certain rich sources of nutrients for therapeutic use.
DATA FROM OTHER SOURCES
Available composition data may pertain to forms of foods not actually or currently consumed by the population groups of interest. For example, agricultural research concerning crop yields, cultivar development or fertilization experiments may yield composition data for specific components such as nitrogen in the plant as harvested. Similarly, data may be available for components of forms of a food material after partial processing, for example, minerals in raw cane sugar or levels of fat in animal carcasses. These data should be considered with caution, since further processing and preparation of the food before consumption will cause changes in component levels because of changes in levels of volátiles (primarily moisture), the discarding of inedible parts and gain or loss of components that are destroyed or lost along with lipid or water-based juices (sometimes called drip-loss), Conversely, the hydration of some cereal grains and their products (for example, macaroni or noodles) affects the level of component per defined gram weight. Dissection and preferential selection of specific parts of a food (for example, removal of separable visible fat on meat slices) by the consumer group will also affect component levels.
SELECTING THE ANALYTICAL SAMPLE
In statistical terms, the group of items or units selected from the population of interest and used to represent that population is called the sample (Cochran, 1977). This terminology should not be confused with the use of the term "sample" at the laboratory level, where it refers to the individual unit or container to be analysed, "Analytical sample" or "analytical unit" is the preferred term, The word "population" describes the large and dynamic collection of items relevant to the objective from which the sample subset is chosen. Generally, the population of interest is very large, while the sample represents a smaller subset of all items or units.
Figure 1 illustrates the concept of sample and population, using carrots as an example. The universe for carrots would include all units, brands and forms of carrots from all geographical areas. Experimental cultivars not yet in the market would also be included. Within that infinite universe the population might include all forms of carrots actually in the food supply of the human population of interest. Therefore, the experimental cultivars and the units grown or traded in geographical areas that do not intersect with the areas where the consumers live would be excluded, The actual sample would be the subgroup of all units and brands of interest. The units to be selected would be drawn randomly in keeping with the probabilities of occurrence for particular types and brands in the population to assure the representativeness of the sample (Cochran, 1977) (Figure 2), The probabilities of occurrence can be determined by reviewing existing data for production, marketing and consumption, The lack of a statistically based sampling protocol can contribute to bias in the estimate of central tendency for a component in a food or food product (Figure 3), Use of biased estimates can result in faulty conclusions about the population of interest. It may also lead to the formulation of erroneous public policy.
The investigator can define the characteristics of the food product, which may influence the composition and the variability of the components) of interest. Relevant characteristics include the primary food source and species (for example, the source may be wheat or corn, coconut or sunflower seed, or beef or pork), the part of the plant or animal used, preservation state, food processing treatments or added ingredients, For some components, geographical source and ripening practices are important, For others, packaging type, pH and storage are sources of variability. For foods that vary by ingredients from one shop to another, such as pizza, one would require the formulation or recipe. For this reason detailed food descriptions are necessary for the class of foods under consideration.
Following the definition of relevant descriptors for a food, it is necessary to identify the specific major sources of that food as it is consumed by the people of interest, In addition, the distribution and marketing schemes need to be identified. For brand products, sales volume data and product information are important to the selection of representative units. For commodity products, such as meats, eggs and milk, it is possible to identify the major breeds, agricultural production regions or cultivars, as well as major commercial purveyors of the products and an approximation of their sales ranking. In regions where food production is localized, the major outlets for products (for example, butchers and bakeries) or ingredients (for instance, flour mills and refineries) can be identified. Some products may be manufactured in one location and distributed nationwide, while others may be formulated in many regions from different sources of raw ingredients, By defining the form of the product and its sources, the investigator can begin to determine which specific products will need to be included in the sample as well as the time and location for sampling.
After marketing and distribution variables have been defined, consumption patterns should be assessed to determine where to select the samples, If the objective is a national database, then it is necessary to sample food products on the basis of the population distribution and product use, Several questions should be answered, Is the food consumed frequently and in significant amounts by the population of interest? In which regions or populations is the food consumed? Is the food consumed more often in rural areas than in towns? If the food is widely consumed by many subgroups, what is the distribution of the population in the country or region of interest? Major population centres within a country can be identified and used as locations for sample selection.
This method is illustrated by a study of selenium in approximately 200 foods conducted by the USDA. Samples of the foods were purchased in two major supermarkets in each of nine cities (two to three cities in each of four regions of the country) (J, M, Holden, in preparation). For each of the major contributors of dietary selenium (that is, beef, white bread, pork, chicken, eggs) approximately 100 analytical samples were prepared. For less important foods, five to 25 samples were chosen. The investigators assumed that in choosing samples of the products with the highest volume of sales (that is, brands within the largest supermarkets in major metropolitan areas), the most frequently consumed and representative products were selected for a specific food.
HOW MANY ANALYTICAL SAMPLES ARE NEEDED?
The number of samples analysed will determine, in part, the statistical power of the estimate, Statistical equations for calculating the number of samples required (Cochran, 1977) are based on the assumption of normality (or approximate normality) for the distribution of values in the population, with equal variances for the subpopulations. In food composition research normality can usually only be assumed because the distribution is rarely known. Although statistical models for calculating the required number of samples can be complex and multi-tiered, the appropriate number of samples, n, is based on four parameters: n³ (ts)2/ (ry)2.
The first parameter, t, is the abscissa of the normal curve that cuts off an area a at the tails of the distribution, indicating the desired confidence level. The standard error of the estimate is denoted by s, while the sample mean is denoted by y. The mean and standard error can be obtained from previously published data or pilot studies.
3 The effect of bias on the estimate of the mean/L'effet du biais sur l'estimation de la moyenne/Efecto del sesgo sobre la estimación de la media
Some existing handbooks of food composition data publish standard deviations or standard errors of the mean and can be used as rough estimates of sample size. The coefficient of variation, if known, can substitute for s/y. The limit of the desired relative error in the estimate, that is, the proximity of the estimate to the "true" mean, for example, within 10 percent, is represented by r.
The calculation of the appropriate number of samples is an iterative process which begins with the investigator's approximation or "guess" of the suitable number of samples based on preliminary cost estimates or the capabilities of the analytical laboratory. After the initial calculation the estimate of the number of samples is refined by recalculation until successive trial values of n are approximately the same. The cost of sampling can be included in the equation as well (Cochran, 1977).
In the past, the mean or average value has been used as the estimate of the level of a component in the food, However, the use of the mean presumes that the statistical distribution of all values for that component in the food follows the Gaussian or normal distribution (Schubert, Holden and Wolf, 1987).
In the United States, the Food Composition Laboratory, in collaboration with the National Cancer Institute, recently compiled and published a food composition table of the levels of five carotenoids in important fruit and vegetable contributors (Mangels et at., 1993), The values were collected from published and unpublished analytical sources, Because of the apparent skewed distribution for several foods and the limited amount of available data (one to 14 acceptable sources per food), median values were used in the table and database, However, the use of the median precludes the simple calculation of a variance indicator.
More research is needed to evaluate the characteristics of statistical distributions that result from broad-based original sampling as well as those that result from compilations of data from different sources and to compare the use of means and medians in food composition databases, In particular, caution is required when estimating food composition values from small data sets.
HOW GOOD DO THE DATA HAVE TO BE?
Food composition data must be good enough to permit the careful assessment of food consumption patterns and their impact on the health of population groups and subgroups. Similarly, the data must be of sufficient quality that other scientific and economic objectives defined by investigators can be accomplished. The required level of quality will be determined by the specific objective for compiling data, For analytical data, quality is determined, in part, by the representativeness of selected units. The validity of analytical
methods, the use of quality control programmes, the appropriateness of the sampling plan, the nature of the sample-handling protocol and the number of samples analysed are other factors that affect quality (Mangels et at., 1993).
EXPERT SYSTEMS
An expert systems approach has been developed to evaluate the quality of analytical data according to these factors, Each factor is rated from 0 for the poorest to 3 for the best (Table 1). For example, a score of 3 for analytical method indicates the use of a valid, well-documented analytical technique, Conversely, a rating of 0 is assigned when there is no evidence of validation of the method, little documentation about methodological procedures or no evidence of the understanding of the execution of the method, or when there is evidence of critical errors. Ratings are assigned for all five factors for each value for each food presented in the source document. Then the ratings are averaged across all five factors to yield a quality index per source. A worksheet illustrating the assignment of ratings is shown in Table 2.
In the expert system, ratings are assigned by the computer as information passes along a "decision pathway" (Figure 4). Ratings for each factor are the product of successive decisions made relative to critical information gleaned from the source.
When all sources for a given combination of food components have been evaluated and stored in an ASCII file, the quality indices for acceptable sources are summed to yield a total, referred to as the "quality sum", From the quality sum the data set for a food and component, for example, b -carotene in raw carrots, is assigned a "confidence code", an indicator of the relative confidence the user can have when using that datum (Table 3). The individual mean values from single acceptable sources can be combined in any of several ways to yield an aggregate component value for the specific food. Alternatively, the individual acceptable sources can be used without aggregation and distinguished by geographical source or some other attribute, The USDA expert system has been adapted to compile data, without aggregation, for individual carotenoids from international sources (West and Poortvliet, 1993).
It is recommended that the confidence codes for component values for specific foods be given in databases to guide the users and compilers in the appropriate use of the data. The determination and availability of data quality indicators will assist investigators in setting priorities for new analyses, Updating of databases will be facilitated by the ease of evaluating new sources and executing the algorithms to determine the acceptability of new data, In addition, the individual sources of acceptable data used to compute the mean or median value should be available to users so that they can monitor the process of data selection for critical uses of the data. To date, evaluations of values have been included in databases for selenium, copper and five carotenoids (Schubert, Holden and Wolf, 1987; Lurie et at., 1989; Mangels et al, 1993).
TABLE 1 - Summary of factors for evaluating quality of data/Resume des critères de qualité des données/Resumen de los criterios relativos a la calidad de los datos
|
Factor |
Rating |
|||
|
3 |
2 |
1 |
0 |
|
|
Analytical method |
Published documentation with validation for foods analysed; use of appropriate reference material with results within acceptable range or 95-100 percent recoveries on similar food and use of same sample under other method or laboratory with agreement within 10 percent; exemplary processing and saponification of sample and identification and quantification of carotenoid. |
Some documentation; incomplete validation studies; 90-110 percent recoveries on similar foods or use of same sample under other method or laboratory with agreement within 10 percent; adequate processing and saponification of sample and identification and quantification of carotenoid. |
Some documentation; minimal validation; <8 percent CV for repeatability or 8-20 percent CV for repeatability along with 80-120 percent recoveries on food similar to sample or use of related food under other method or laboratory with agreement within 10 percent; minimally acceptable processing and saponification of sample and identification and quantification of carotenoid. |
No documentation of method, no reference or inaccessible reference given; non-chromatographic method used; no validation studies or failure to achieve acceptable results with reference material, repeatability (³ 20 percent CV), recovery (<80 or >120 percent), or comparison method or laboratory; inadequate processing or saponification of sample or identification or quantification of carotenoid. |
|
Analytical quality control |
Optimum accuracy and precision of method monitored and indicated explicitly by data. |
Documentation of assessment of both accuracy and precision of method; acceptable accuracy and precision. |
Some description of minimally acceptable accuracy and/or precision. |
No documentation of accuracy or precision; unacceptable accuracy and/or precision. |
|
Number of samples |
>10; SD, SE or raw data reported |
3-10 |
1 -2 or not specified |
- |
|
Sample handling |
Complete documentation of procedures including analysis of edible portion only, validation of homogenization method, details of food preparation, and storage and moisture changes monitored. |
Pertinent procedures documented including analysis of edible portion only; procedures seem reasonable but some details not reported. |
Limited description of procedures including evidence of analysis of edible portion only. |
Totally inappropriate procedures or no documentation of criteria pertinent to food analysed. |
|
Sampling plan |
Sampling of multiple geographical areas with description of and statistical basis for sampling; sample representative of brands/ varieties consumed or commercially used. |
At least two geographic regions sampled; sample is representative. |
One geographic area sampled; sample is representative of what some eat. |
Not described or sample is not representative. |
The clear definition of guidelines for assessing data quality can inform analysts in the public and private sectors as well as journal editors about the requirements for data quality.
VARIABILITY
The generation of accurate food composition data requires that variability inherent in the food be accurately quantified while variability inherent in the measurement process be minimized. In general, the major sources of statistical variability in dietary estimates are the food consumption data (captured by the dietary assessment tool) and the food composition data. Variability for food composition data is attributable to sampling plan (for example, product, brand, geography), sample handling (for example, temperature, homogenization, storage, changes in water content), analytical method and analytical quality control. Inherent differences in water content for different samples would also be an effect of sampling plan.
Overall variability can be partitioned by an analysis of variance into the sources of variability and can be quantified (Holden et al., 1991), The assessment of the sources and magnitude of variability for food composition data can indicate areas where improvements in the measurement process need to be made. While sampling is only one source of variability, a lack of representative sampling can increase the degree of bias in the estimates of central tendency and cause errors in the estimates of variance, As previously mentioned, a small number of foods may contribute the major proportion of a component in the diet of the population. Therefore, it is recommended that sampling resources be dedicated to obtaining statistically sound estimates for those foods that are major contributors.
TABLE 2
Worksheet evaluating quality of data for b -carotene content in winter squash/Données relatives au bêta-carotène dans le potiron/Hoja de trabajo para determinar el valor del betacaroteno en la calabaza grande
|
Product description |
Referencea |
Number of samples |
Rating |
Quality indexb |
b -carotene value |
||||
|
Number of samples |
Analytical method |
Sample handling |
Sampling plan |
Quality control |
|||||
|
Acorn, microwave 8 min. |
a |
3 |
2 |
2 |
1 |
1 |
1 |
1.4 |
490 |
|
Frozen, commercial |
b |
4 |
2 |
1 |
1 |
1 |
1 |
1.2 |
2670±6 |
|
Butternut, cooked 20 or 40 min. |
c |
4 |
2 |
1 |
1 |
1 |
1 |
1.2 |
4570±12 |
|
Canned |
d |
2 |
1 |
1 |
1 |
1 |
1 |
1.0 |
923 |
|
Pressure-cooked |
e |
2 |
1 |
1 |
1 |
1 |
1 |
1.0 |
2800 |
|
Canned |
f |
3 |
2 |
1 |
1 |
0 |
0 |
0.8 |
1250±180 |
|
Frozen, cooked |
g |
3 |
2 |
1 |
1 |
2 |
0 |
1.2 |
2400±570 |
|
Frozen, cooked |
h |
3 |
2 |
1 |
1 |
2 |
0 |
1.2 |
1400±600 |
|
Acorn, frozen |
i |
1 |
1 |
1 |
1 |
1 |
0 |
0.8 |
300 |
|
Butternut, frozen, cooked |
j |
2 |
1 |
1 |
1 |
2 |
0 |
1.0 |
850±350 |
|
Butternut, frozen, cooked |
k |
3 |
2 |
1 |
1 |
2 |
0 |
1.2 |
3600±1 600 |
|
Frozen, cooked |
l |
2 |
1 |
0 |
1 |
1 |
1 |
0d |
800 |
|
Quality sume |
|
|
|
|
|
|
|
10.4 |
|
|
Confidence codef |
|
|
|
|
|
|
|
a |
|
|
Mediang |
|
|
|
|
|
|
|
|
2400 |
|
Minimumg |
|
|
|
|
|
|
|
|
490 |
|
Maximumg |
|
|
|
|
|
|
|
|
4570 |
a While data, including ratings, are authentic, the references are coded.b A quality index ³ 1 is required for a datum to be considered acceptable.
c Mean (± standard deviation).
d Because of zero rating for analytical method, quality index is zero.
e The sum of the quality indices for acceptable references; it serves as the basis of the confidence code.
f The confidence code is derived from the quality sum.
g Based on the acceptable means. In this case, values from references f, i and I were unacceptable.
CONCLUSION
To assure accuracy in food composition data, it is crucial that statistically based sampling protocols be developed. Each food category of interest comprises specific products, The probabilities of occurrence for each of these products must be incorporated in the sampling protocol, By sampling randomly from the marketplace, manufacturer and farms according to the carefully planned protocol, the investigator can obtain representative samples for analysis. The analysis of representative samples provides database values that can be used within a specific level of statistical confidence for the purpose of the investigation.
TABLE 3 - Assignment and meaning of confidence codes/Attribution et signification des codes de fiabilité/Asignación y significado de los códigos de confianza
|
Sum of quality indices |
Confidence code |
Meaning of confidence code |
|
>6.0 |
a |
The user can have considerable confidence in this value |
|
3.4 to 6.0 |
b |
The user can have confidence in this value; however, some problems exist regarding the data on which the value is based |
|
1.0 to <3.4 |
c |
The user can have less confidence in this value because of limited quantity and/or quality of data |
References
Beecher, G.R. & Matthews, R.H. 1990. Nutrient composition of foods. In Present knowledge in nutrition, p. 430-439. Washington, DC, International Life Sciences Institute. 6th ed.
Cochran, W.G. 1977. Sampling techniques. New York, Wiley. 3rd ed.
Greenfield, H. & Southgate, D.A.T. 1992. Food composition data: production, management, and use. London. Elsevier Applied Science,
Holden, J.M., Gebhardt, S., Davis, C.S. & Lurie, D.G. 1991. A nationwide study of the selenium contents and variability in white bread, J. Food Compos. Anal,. 4:183-195,
Holden, J.M., Schubert, A., Wolf, W.R. & Beecher, G.R. 1987. A system for evaluating the quality of published nutrient data: selenium, a test case. Food Nutr. Bull., 9: 177-193.
Lurie, D.G., Holden, J.M., Schubert, A., Wolf, W.R. & Miller-Ihli, N. 1989, The copper content of foods based on a critical evaluation of published analytical data, J. Food Compos. Anal., 2:298-316.
Mangels, A.R., Holden, J.M., Beecher, G.R., Forman, M.L. & Lanza, E. 1993, Carotenoid content of fruits and vegetables: an evaluation of analytic data. J. Am. Diet. Assoc., 93: 284-296.
Schubert, A., Holden, J.M. & Wolf, W.R. 1987. Selenium content of a core group of foods based on a critical evaluation of published analytical data. J. Am. Diet. Assoc., 87: 285-299.
West, C.E. & Poortvliet, E. 1993. The carotenoid content of foods with special reference to developing countries. Arlington, Virginia, USA, International Science and Technology Institute, Vitamin A Field Support Project (VITAL).
Les approvisionnements alimentaires comprennent de nombreux produits transformés et non transformés, et il est impossible d'analyser chaque article pour obtenir des données sur sa composition. Il est donc nécessaire d'adopter une stratégie de sélection des denrées permettant d'obtenir un échantillon représentatif de la consommation. Pour obtenir des données exactes sur la composition des aliments, il faut que la variabilité inhérente aux aliments soit correctement quantifiée, tandis que la variabilité résultant du processus de mesure doit être réduite au minimum. Il est nécessaire de disposer de données exactes, non biaises, pour éviter les conclusions erronées ainsi que des erreurs coûteuses dans l'évaluation des régimes alimentaires et le commerce.
L'abondance de produits alimentaires nouveaux et différents et les disponibilités limitées de ressources impliquent que les stratégies d'échantillonnage doivent se fonder sur des principes statistiques bien définis et viser des objectifs scientifiques tenant compte de l'utilisation prévue de données sur la composition des aliments. Le présent article décrit les stratégies mises au point par le Département de l'agriculture des Etats-Unis pour échantillonner les approvisionnements alimentaires afin de réaliser des estimations des éléments entrant dans la composition de nombreux aliments.
On peut avoir besoin de données sur un certain élément qui se retrouve dans de nombreux aliments, ou au contraire sur les principaux composants d'un aliment donné. Le plus souvent, les composants sont choisis en fonction de leurs effets réels ou supposés sur la santé publique, de la disponibilité de méthodes d'analyse sûres et praticables, et du manque de données acceptables. De façon générale, la priorité devrait être donnée aux aliments qui apportent la plus grande quantité de l'élément choisi au régime alimentaire, à l'exception toutefois des aliments qui sont importants pour les sous-groupes de population vulnérables, des aliments qui contiennent des teneurs élevées de certains composants spécifiques, des aliments dont on pense qu'ils contiennent des contaminants, ou qui peuvent avoir une valeur thérapeutique.
L'article examine les façons de caractériser des produits qui peuvent avoir une influence sur la composition et la variabilité des composants des aliments, et expose une méthode de sélection des échantillons à analyser. Il décrit aussi l'approche systèmes experts pour évaluer la qualité des données analytiques. L'analyse d'échantillons représentatifs fournit des données chiffrées qui peuvent être utilisées avec un niveau de fiabilité statistique déterminé à des fins spécifiques.
Los suministros alimentarios están integrados por numerosos productos elaborados y no elaborados, y es imposible analizar cada una de las partidas para obtener datos sobre la composición de los alimentos. Por ello es necesaria una estrategia para seleccionar unidades de alimentos con el fin de obtener una muestra representativa de la población. Para que los datos sobre composición de alimentos sean exactos, es necesario cuantificar con precisión la variabilidad inherente a los alimentos, evitando al mismo tiempo en lo posible la variabilidad resultante del proceso de medición. Son necesarios datos exactos y no sesgados para evitar conclusiones equivocadas y errores costosos en las evaluaciones de la dieta y en el comercio.
Tanto la abundancia de productos alimenticios nuevos y diferentes como la disponibilidad de recursos limitados exigen que las estrategias de muestreo se basen en principios estadísticos bien definidos así como en objetivos científicos que incorporen el uso previsto de los datos sobre la composición de los alimentos. El Departamento de Agricultura de los Estados Unidos ha elaborado una metodología para tomar muestras de los suministros alimentarios con el fin de efectuar estimaciones relativas a los componentes de muchos alimentos.
En ocasiones pueden ser necesarios datos sobre un único componente que se incluye en muchos alimentos, mientras que otras veces se explorará un único alimento y sus principales componentes. Habitualmente los componentes se seleccionan por razón de sus efectos reales o presuntos sobre la salud pública, porque se dispone de métodos de análisis bien fundados y viables y porque faltan datos aceptables. En general se dará prioridad a los alimentos que más contribuyen a la dieta. Sin embargo son excepción a esta regla los alimentos que son importantes para grupos vulnerables de la población, los que contienen un volumen elevado de componentes específicos, los que podrían contener contaminantes o los que tienen una utilidad terapéutica.
Existen diversos modos de determinar las características de los productos que pueden influir en la composición y variabilidad del componente o los componentes de los alimentos, y se dispone de una metodología para seleccionar las muestras que se han de analizar, También se esta aplicando el concepto de sistemas expertos para evaluar la calidad de los datos analíticos, El análisis de muestras representativas permite obtener valores para la base de datos que pueden emplearse con un fin determinado dentro de un margen específico de confianza estadística.
