Data on the composition of foods are essential for a variety of purposes in many different fields of work, for example the assessment of energy and nutrient intake in individuals and groups. They are also necessary to assess the effect of diet on health and disease outcomes in epidemiological research and they have a wide variety of other applications. However, food composition data has its limitations, and an understanding of these limitations is fundamental to using the data correctly.

Variation in nutrient composition of foods

All foods are biological materials and therefore show natural variability in their composition. This variation may be due to differences in the animal or plant species assessed, variations in agricultural practices or environmental factors (e.g. soil and climate) as well as the storage, processing and preparation of foods. Even processed foods which are produced under very controlled conditions show some variability due to differences in the composition of ingredients and variations in processing, packaging and storage.

Widdowson and McCance (1943) wrote: “There are two schools of thought about food tables. One tends to regard the figures in them as having the accuracy of atomic weight measurements; the other dismisses them as valueless on the grounds that a foodstuff may be so modified by the soil, the season, or its rate of growth that no figure can be a reliable guide to its composition. The truth, of course, lies somewhere between these two points of view.”

The degree of variation in nutrient composition also varies for different nutrients. Micronutrient (vitamin and mineral) values vary more widely than macronutrient (carbohydrate and protein) values, with the exception of fat, as the fat content of different cuts of meat is very variable. Vitamin C and folate show wide natural variability and are particularly unstable, being affected by heat and light. This is such that nutrient databases can only give an approximate or typical indication of the levels of these nutrients. The water content of foods is another major source of variation in nutrient composition, particularly in the case of staple foods such as rice. The water content of cooked rice can vary between 65-80% which can make a considerable difference to the energy and nutrient composition per 100g.

Ideally, food composition databases should provide average (or median) values for each nutrient, together with a statistical measure of variability which would give users some idea of the level of accuracy that can be expected from the database. The predictive accuracy of nutritional analysis calculations generally improves with the number of foods consumed and the length of the study, however accurate food consumption data collection is also important.

Data quality

Data quality is a measure of whether the data is fit for the purpose for which it will be used. An understanding of the ways in which the database will be used is therefore a necessary prerequisite to any discussion about data quality in food composition databases. Databases are used in many different ways and therefore providing high quality data for every type of user is a difficult and challenging task. However, there are a number of requirements in common, which are discussed in further detail below. These include:

• That the compositional data are representative of the foods consumed in the country or region for which the database was developed.

• That the methods chosen for analysis are appropriate and have been carried out correctly according to defined protocols.

• That the data accurately represent the composition of the foods.

The requirements for accuracy vary somewhat between users. For example, those who use food composition databases for educational purposes are using the data in a semi-quantitative way and therefore require a lower degree of accuracy in food composition data compared to nutritional epidemiologists, whereas those who use food composition databases for analysing dietary intake in metabolic studies require a degree of accuracy that is very hard to achieve.

Sampling

Most users of food composition databases assume that the compositional data will be representative of the foods consumed by the population being studied, however this is difficult to achieve. The individual foods consumed by respondents in a study are rarely representative of the same food for the country as a whole. Furthermore, databases are often used over a long period of time and the sampling may have taken place several years earlier, over which time the composition of foods may have changed (manufacturer’s formulations often change over time e.g. to change the fat or sodium content). All sampling is associated with a degree of sampling error (uncertainty) but database compilers must at least aim to ensure that the samples are representative of foods consumed by a particular population at the time of sampling, in order to determine relationships between nutrient intake and disease/health status with an appropriate degree of confidence. The variability in nutrient composition of individual foods determines the number of duplicate samples that need to be collected. The sampling protocol of foods also needs to take into account any seasonal, regional or other variations (as described above). Knowledge of the foods being studied, for example the methods used in its production, storage and marketing, will help to determine whether a pattern of sampling is needed (e.g. stratification by season, which is done for milk).

Analysis

The quality of analytical data is of primary importance for all purposes for which foods are analysed. The methods chosen for analysis must be appropriate as different methods yield different results. For example, there are different methods for analysing the fibre content of foods. Greenfield and Southgate (1994) have developed principles for the generation of data for food composition databases. It is essential to establish a Quality Assurance Scheme (QAS) that applies to all aspects of data generation, for example procedures for handling samples when received, the preparation of analytical samples and the storage of samples need to be set out in detail. Documentation is also a vital element in the implementation of a QAS, so that samples can be tracked through the laboratory analysis, from beginning to end. As well as the work underway within EuroFIR, the International Network of Food Data Systems (INFOODS) project is also working on the standardisation of analytic methods.

Compilation

Database compilers have responsibility for making sure that the quality of the food composition databases meets the needs of the end-users. More often than not, compilers have to make use of data from other published or unpublished sources that they have not had any control over producing. These data still need to be scrutinised, by considering the sampling protocol, analytical methods used and quality assurance procedures that are in place. One of the difficulties with database compilation is that relatively few published sources of data describe the sampling or quality assurance procedures used in sufficient detail to assess the quality of the data and this is something that needs to be addressed in the future. Standardisation is needed at all levels of the compilation process.

Conclusion

Data quality is an integration of sampling, analysis and compilation procedures. Assessing the quality of analytical data has been addressed for some nutrients, such as iron, selenium and carotenes with the assignment of ‘quality codes’. This is also being extended to include other nutrients and to sampling (see Holden et al. 2002). As described earlier, all foods whether of animal or vegetable origin show natural variability in their composition and therefore even the best sampling protocol can only provide a sample with a composition within the confidence limits assigned. Analytical differences vary with individual nutrients and confidence limits tend to be wider for micronutrients than macronutrients. All calculations carried out from food composition databases are therefore associated with a certain degree of inaccuracy which users should be aware of.

Future directions

Food composition data are currently being used in multi-centre studies and ways which were not previously anticipated (e.g. specialised fields in health, agriculture, environmental sciences and economics) and therefore some of the conventional criteria for judging the quality of food composition data may need to be re-examined, expanded or adjusted. Furthermore, recent environmental events such as diminishing crop biodiversity, the development of genetically modified organisms (GMOs) and climate change will also require the attention of database compilers. Key aspects that need to be re-examined include the representativeness of foods, completeness of food composition databases (i.e. the need to generate data for missing foods) and an integrated approach to food composition databases development internationally (see Burlingame 2004). Comparability and reliability of national food composition databases are also needed to minimise artificial differences in food composition data due to systematic and random errors associated with analytical and compilation processes.

EuroFIR

One of the objectives of the EuroFIR project has been to develop a pan-European quality framework for food composition databases. This helps to establish procedures and a quality infrastructure to enable a common understanding among database producers, compilers and users of the requirements for quality assurance and to guarantee better data management and data interchange at a national and international level. The quality of existing data and all new data generated as part of the EuroFIR programme is also being critically assessed, using the established framework, prior to its acceptance. This work (WP 1.3) is led by INSA (Instituto Nacional de Saúde Dr. Ricardo Jorge) in Lisbon, Portugal. Further details can be found elsewhere on the EuroFIR website.

References

Burlingame B (2004) Fostering quality data in food composition databases: visions for the future.

Journal of Food Composition and Analysis 17: 251-258.

Greenfield H and Southgate DAT (1994) Food Composition Data; Production, Management and

Use. Revised edition. Chapman & Hall, London.

Holden JH, Bhagwat SA and Patterson KY (2002) Development of a multi-nutrient data quality

evaluation system. Journal of Food Composition and Analysis. 15: 339-348.

Southgate DAT (2002) Commentary: Data Quality in Sampling, Analysis and Compilation. Journal

of Food Composition and Analysis 15: 507-513.

Widdowson EM and McCance RA (1943) Food tables, their scope and limitations. Lancet i: 230-

232