| Food Composition Data: A User's Perspective (1987) |
|Report and recommendations of the conference|
Food composition data
There are three important aspects of food composition data themselves: (a) what data exist, (b) how good these data are, and (c) how easy these data are to obtain.
What Food Composition Data Exist?
The number of foods that have been analysed for their content varies tremendously around the world. Tables which include these data are available in a variety of forms
(note that these tables are not entirely independent, since many of the basic data are shared among them):
- international tables (e.g. Platt's Tables of Representative Values of Foods Commonly Used in Tropical Countries);
- regional tables (e.g. FAO and USHEW's Food Composition Tables for Use in Africa);
- national tables (e.g. USDA's Handbook No. 8);
- food industry data bases (many major food companies have their own data bases);
- commercial data bases (there are a large number of diet analysis programs, which include data bases, available for purchase by individual consumers);
- local, special-purpose tables (many hospitals maintain data bases for menu planning and nutrition guidance);
- journal articles (a number of journals, such as the Journal of the American Dietetic Association and Ecology of Food and Nutrition, frequently publish papers containing food composition data).
While there is no complete index to all the food composition data that exist, several partial directories are available. The Food and Agriculture Organization (FAO) of the United Nations published a listing in 1975 which covered international, regional, and national tables . This is now out of date and FAO has no plans for its updating. In 1986 INFOODS issued a similar directory of tables currently used . EUROFOODS (paper 5) and NORFOODS (paper 16) have prepared listings of data available within their regions. Within the United States, Loretta Hoover of the University of Missouri-Columbia annually issues a Nutrient Data Bank Directory, which includes characteristics and contents of currently available data bases . Additionally, Darlene Hildebrandt, of the University of Washington in Seattle, issues a listing of Computer Programs and Databases in the Field of Nutrition .
With respect to the data that are available, in North America and Western Europe it is usually possible to find basic nutrient composition data for most common foods. However, there are many regions throughout the world where data on the composition of even the most frequently consumed foods do not seem to be available, or when available are seriously out of date (papers 14 and 15).
Beyond the problem of determining what data exist is the problem of determining what the available data represent. While often an introduction to printed tables will provide some indication of the analytic methods used, rarely is sufficient information given on how the food samples were gathered and analysed, and on how the data themselves were scrutinized and manipulated. Moreover, only infrequently is there any indication of the variability inherent either in the food or in the analytic method.
In terms of the data that are not available to potential users, no food composition data system contains values for all the components or foods desired by all users, and it is unlikely that any table or data base ever will, because of the rapid expansion of the number of foods and nutrients of interest.
The components of foods that are most frequently missing include:
- nutrients (especially trace minerals, some B vitamins, and lipid-soluble vitamins);
- subcomponents of nutrients and classes of nutrients, such as retinoids, carotinoids, fatty acids, starches, specific sugars, etc.;
- non-nutrients such as dietary fibres, xanthines, allergens, toxins, and selected con taminants;
- ingredients including additives.
While the situation is often that there are not good, reliable methods for assaying certain components [10,12], many of the data do not exist simply because of the magnitude of the task of collecting them. Users with specific needs have two options: (a) they can generate the data themselves, gathering representative samples of the foods of interest and assaying them for the desired components, or (b) they can estimate (impute) the missing values from known data on similar foods and components. The first option requires resources that users rarely have available, while the second requires clear and well-defined rules for estimation, rules which do not currently exist. An associated problem is that rarely do data produced by individual users enter into the public store of food composition data, with or without appropriate documentation.
While data on new, manufactured foods and foods infrequently consumed are often missing from food tables, there is a major and significant gap concerning the composition of "foods as consumed." Many of the data in tables represent foods that are "raw," but many (if not most) foods are eaten after being processed, stored, and/or prepared in various ways that may each affect at least some of the nutrient levels . Moreover, often the data on prepared dishes in the tables are not the results of analyses but have been estimated by the compilers of the tables.
Two important types of "foods as consumed" are mixed dishes, such as stews and curries, and foods that are purchased already prepared, such as those obtained in a restaurant. For these foods, composition data, based on either analysis or estimation, must start from a recipe. However, it is often difficult to define, much less obtain, "standard" or representative recipes for most of these foods. If a recipe can be selected, one must then address the issues of labile or soluble nutrients, cooking losses, nutrient interactions, and fluid or fat loss (or gain) which can significantly alter nutrient concentration per unit weight. Additional problems arise from shifts in the availability and costs of ingredients which frequently lead to modifications of the recipes.
Efforts are proceeding in two general directions with respect to adding data to food tables. First, more analytic methods are being developed and analyses being conducted, and, second, discussions and research are being carried out to develop guidelines for making estimation more accurate. However, both these efforts must be greatly expanded, and co-ordinated, before users can devote their efforts to the using of food composition data rather than to the finding and completion of food composition data bases.
How Good Are the Data that Do Exist?
There is considerable variability in the quality of food composition data, and rarely is information about data quality available to the user (paper 18). It appears that the individual data that make up food tables and data bases have often undergone only limited scrutiny. While major tables choose their sources carefully and document these sources, this is expensive and time consuming, and many data-base compilers do not give sufficient attention to this problem, leaving the responsibility of data quality to those from whom they acquire data. Similarly, estimation of data to fill in gaps in tables is frequently not performed with sufficient care, nor are these procedures documented, partly because of the lack of accepted guidelines. Clearly one conclusion that must be drawn from these considerations is that users must use food composition data cautiously.
Before considering how to improve the quality of food composition data, it must be pointed out that most variability in that data is not due to analytic error (papers 2 and 19). Of the number of factors that influence the observed levels of components in foods, it is true that several can be considered error, and their contributions to the overall variability of food composition data evaluated, categorized and, in some cases, minimized. For example, the analytical procedures introduce variability which can be minimized by following good laboratory techniques .
However, many other sources of nutrient variation are inherent in the foods themselves. These include geographical region of production, cultivar/species, changes in fortification levels, and agricultural practices in general . Studies are needed to identify, characterize, and evaluate these several sources of variation to permit data compilers to provide users with food composition data that are less variable, perhaps through subdivisions of existing food categories regional tables, dated values, and so forth.
The two general areas where major efforts are needed to improve the situation with regard to the quality of food composition data are: (a) improvement of the quality of the data per se, and (b) improvement of communications so that the user will be able to determine the quality of specific data of interest.
Improving Quality of Data
The long-range improvement of the quality of food composition data can best be achieved through amelioration of the measurement system (improvement of sampling techniques and analytic methods, development of standards for generating food composition data, development of training programmes in food analysis, and use of biological reference standards) and standardization of the procedures for manipulating data, including those for estimating data that are not directly available as analytic determinations. These are all essential efforts that will contribute to the reduction of the errors in food composition data.
Improving the Documentation of Quality of Data
Another area in which the field of food composition data demands a major effort is that of documenting more carefully the "context" of the data - those factors which can, and do, contribute to the variability of the data. This is essential so that users can be made aware of the potential problems of the data, and be given enough information to judge for themselves whether the data are of sufficient quality for their needs.
Additionally, there is a need for the development of an overall scheme to indicate the reliability of data (see paper 18 for a detailed discussion of this topic). For example, Exler  has described, and used, a procedure for evaluating existing data against fixed standards to produce a score, or "confidence code," for each data set. These confidence codes not only give the users of the data an indication of their reliability but also inform data generators where new data are needed, as well as providing data compilers with tools to rationally combine new food composition data with existing data.
Thus, food composition data are of uneven and often unknown quality, and users approach them with due recognition of this problem. Moreover, they must be aware of both the inherent variation in food composition data and the variation that can be introduced by the gathering and manipulating of data. Every effort must be made to make available to the user information concerning the food and its analysis that will provide insight into the reliability of the data and its suitability for a particular purpose.
How Accessible Are the Data?
The accessibility of food composition data is obviously essential to its usage, and has three key aspects: (a) finding the data (if they exist), (b) obtaining those data, and (c) determining the precise meaning of the data obtained.
Finding the Data
Determining whether the desired data exist and finding where they are located is discussed above and, as pointed out, represents a significant problem. Currently there is no complete, up-to-date, global catalogue of food composition data. It should be emphasized that the effort involved in compiling and keeping current such a catalogue, which includes enough information to be widely useful to the various user groups, is a major undertaking.
Obtaining the Data
The question of moving the data around - data interchange - is also a major problem because of the time and effort that must be devoted to the actual acquisition of food composition data tables must be entered manually into the user's system, or programs custom-written to read specifically formatted tapes or disks. As the situation becomes more complex, with more generators, compilers, and users of data, the problems of data interchange will increase; and therefore dealing with them will consume more of the users' resources. Thus a major, essential task is the development of standardized guidelines for food composition data interchange.
Identifying the Data
Precisely identifying the data - determining exactly what food and what nutrient the numbers represent (the question of standardized terminology) - is key to the criticaluse of food composition data (paper 6). A standardized food-naming and classification system is critical to data entry, interchange, and retrieval, and currently no acceptable scheme exists. Although there are common elements that appear in the naming systems in most food composition tables, true compatibility does not exist even among the most commonly used data sets. The development of a standardized global terminology for food composition data which addresses the associated problem of classification is an important task that needs to be initiated and accomplished as soon as possible.