K.I. Baghurst and P.A. Baghurst
Dr Katrine Baghurst and Dr Peter Baghurst are Principal Research Scientists at CSIRO Division of Human Nutrition, PO Box 10041, Gouger Street, Adelaide SA 5000.
From the late 1940s until the early 1970s, a series of reviews and studies were published assessing values obtained by use of food tables compared to chemical analyses (Bransby & others 1948, Groover & others 1967, Buzina & others 1966, Pekkarinen 1967, Whiting & Leverton 1960, Eagles & others 1966, Marr 1971). The views expressed in these papers about the validity of food tables for assessment of nutrient intake varied widely but where tables were available for locally produced food products, correlations between calculated and analysed data were generally considered to be good. For example, the early studies of Bransby and co-workers in the 1940s, which compared the results obtained for 33 adults using chemical analyses of three-day weighed records with food table analyses, produced correlation coefficients of 0.93 for energy, 0.91 for protein, 0.81 for fat, 0.89 carbohydrate, 0.80 for calcium but only 0.49 for iron. There was some bias in the data, in that the calculated values for energy, carbohydrate and fat tended to be higher and for protein, calcium and iron, lower, than the analysed figures. However, the differences were small, again with the exception of iron (2053 vs 2088 calories; 76 g vs 68 g protein; 81 g vs 84 g for fat; 255 g vs 265 g for carbohydrate; 1.1 g vs 1.0 g for calcium and 19 g vs 11 g for iron). An analysis of the relative ranking of subjects using tertiles also showed that 70–80% of subjects remained in the same extreme tertile of intake comparing analysed and calculated values. Again, however, the matching for iron was poor with only 50% remaining in the same extreme tertiles and 20% being grossly misclassified. Marr, in her review in 1971, concluded that Bransby's data and the other data available to that time, showed that, whilst absolute agreement for every individual is not achieved with the use of food tables, there is a high degree of correlation for most nutrients, such that it is possible to compare groups of individuals in terms of their broad distributional properties and to rank them, two common needs in epidemiologic or population-based research.
In recent years, nutritional methodology issues have continued to receive wide coverage both in the epidemiological and nutritional literature. There have been extensive discussions on techniques and the validity of various methods to assess usual or current intake and how best to express the resulting data. However, despite the large number of recent publications and the many national and international workshops and symposia on these topics in recent years (Baghurst & Baghurst 1981, Block 1982, Stuff & others 1983, Byers & others 1985), there has been relatively little discussion of food composition data bases, their use and abuse.
The nature of epidemiological work
The lack of discussion about the validity and use of food tables in the epidemiological field, probably relates, in part, to the fact that no realistic alternative approach to the analysis of population-based dietary intake data exists. Epidemiological work, by its nature, involves studies of large numbers of people. These studies can be cross-sectional or prospective in nature, in which case a measure of current usual dietary intake is often required, or they can be retrospective, in which case past usual intake is the item of interest. The studies may involve a case-control or cohort design but in both cases, the main purpose is usually to compare relatively large groups of individuals. Therefore the need for absolute accuracy is not as great as when trying to assess, for example, the carbohydrate intake of a single diabetic patient in a clinical setting. A systematic error in the food data base, such as may occur when an analytical technique for a given nutrient gives consistently high or low values across all foods, will not markedly affect the final outcome of a case-control or cohort study where comparison between groups is the issue, but may cause problems to the user wishing to assess absolute intakes in an experimental or clinical setting. On the other hand, random errors or biases in the food composition data base can cause problems even in comparative studies. For example, if one particular food or category of foods is eaten more often in one group than in the other, errors in analyses for those foods will bias the comparison, making differences between the groups bigger or smaller than they are truly.
Computerisation of data bases
A major area of concern in relation to the use of food tables in epidemiological work is that they are often used by people not trained in any way in their use and, indeed, often with no nutritional training at all. The increasing availability of predesigned computer packages for analysing diets will only compound this problem by further distancing the user from the producers of the tables.
Food compositional data bases, in their printed form, often have extensive notes detailing the source of data, selection of items for analysis, the range of values found, the analysis techniques used and any assumptions made when calculating values for items or nutrients not directly measured. They also carefully annotate missing data for a given food. This information is rarely available in computer packages. Indeed in many computer packages missing values are entered as zero rather than using best estimate figures. In a major recent review of the use of dietary intake data, Anderson (1986) estimated that use of zero figures for missing data was the major potential error source in food tables. Unless the users of tables have an original copy of the annotated data, and unless it is also possible for them to amend errors in any computerised data base they may use, biases will inevitably be introduced into their dietary intake analyses. Unfortunately, many users of computerised food tables will not avail themselves of the printed or original computerised tables because of the additional costs.
The increasing use of food data bases by non-specialists also reinforces the need to annotate tables and programs carefully to prevent common coding errors such as the use of codes for dried items (coffee grounds, cocoa powder, gravy powder or custard powder) when the appropriate code is for the rehydrated form. There is also a need to avoid ambiguous labelling such as ‘beef, fat’, a code descriptor used in one widely used table to mean the fat from beef. In our experience, the code for ‘beef, fat’ has often been used mistakenly when attempting to code fatty beef. This problem may be compounded in computerised programs if abbreviated food descriptors or abbreviated food tables are used. It should be possible to overcome these problems by careful naming of items or by the use of warning messages where known potential problems exist. In short, computer programs which include the use of food tables need to be more user-proof.
Specific needs of epidemiologists
The needs that nutritional epidemiologists have in relation to food compositional data bases are, in most respects, similar to those of other users. The data base would contain accurate and comprehensive nutrient and non-nutrient analyses (nitrates, nitrites, oxalates, phytic acid, additives, contaminants etc) for all major food items eaten by Australians, including a wide range of foods commonly consumed by ethnic communities and by the more traditional aboriginal groups.
The range of nutrients and non-nutritive components of the food supply that are of interest to epidemiologists, and especially to cancer and cardiovascular epidemiologists, is far in excess of those currently available or indeed immediately planned for the Australian data base. For example, despite the fact that Australia has recently developed recommended levels of intake for nutrients such as vitamin E and selenium, we have little or no data on the levels of these nutrients in Australian foods. Indeed, there are few overseas data on these nutrients or other non-nutritive components such as nitrites, nitrates and nitrosamines, which are of great interest in cancer epidemiology. The relevance of the overseas data that do exist is difficult to assess as items such as selenium and nitrates and nitrites are highly dependent on the soils in which foods are grown and on related farming practices. Others such as vitamin E are also dependent on the commercial usage of various fats and oils. Other nutrients which would be of particular interest to cancer epidemiologists for similar reasons, include manganese and various carotenoids (including not only those which are precursors of vitamin A such as β- and α-carotene and β-cryptoxanthin, but also non-provitamin A carotenoids such as lutein and lycopene). In the cardiovascular field, the pressing need is for further data relating to fatty acid profiles of fats and oils from various sources, including local varieties of fish, and to the various dietary fibre components of foods (when measurement issues relating to the latter have been resolved).
Many nutritionists involved in population-based research would prefer to have accurate data for a wide range of nutrients in key or basic foodstuffs than to have limited nutrient information on a wide range of foods, some of which are very rarely eaten and only contribute marginally to overall nutrient intake in the population group of interest. In this regard, it is particularly important to have these accurate analytical data, together with details about their reliability and the range of values for specific nutrients in given foods, where those foods are major contributors to population intakes of the nutrient. For example, it is more important for the assessment of population figures of β-carotene intake to have exhaustive data for β-carotene in carrots which are the major and dominant population source of this nutrient. Small inaccuracies in tables due to analyses of samples limited in terms of their source, cultivar, stage of maturity etc will have a marked effect on the accuracy of population intake figures for carotenes whereas similar or even larger inaccuracies related to other minor sources such as nectarines would have a negligible effect. The same concern relates to the representativeness of analytical data for dietary fats and oils in relation to fatty acid profile and vitamin E for which they are dominant sources; to liver and kidney for items such as retinol and vitamin B12; and to eggs for cholesterol. It is also essential to have accurate and detailed data for basic foodstuffs such as breads which can come in a wide variety of types and which contribute substantially to a wide range of nutrients.
Food tables are often used by epidemiologists and other research workers to aid in the design of questionnaires relating to food intake. In many instances, the researcher is seeking to determine which food to include in a questionnaire, or which to take into consideration in computing dietary intakes, and would be aided in this by the inclusion, in both printed and computerised data bases, of nutrient content per usual serve size as well as per 100 g. In addition, many inexperienced users need an indication of the relative importance of consumption of various foods when assessing nutrient intakes at the population level. This would prevent the common occurrence of a key food or foods being omitted from questionnaires on the basis of relatively low levels of nutrient per 100 g or serve, when in fact the frequency of consumption of that food is high and the usual serve size is large. In short, it must be remembered that the nutrient contribution that a food makes to a given diet, is a function not just of its chemical composition, but also serve size and frequency of consumption.
An additional aid to the user would be the inclusion of condensed lists of nutrients as well as the more common page-by-page, screen-by-screen or line-by-line food lists. For example, a few pages or screens listing fat levels in all foods, either per 100 g or per serve, would be a time-saving device for those wishing to concentrate on fat intakes.
Continuous updating of tables
One problem which epidemiologists, and other users in Australia, will have to face in the coming years will be how to cope with a data base in continuous evolution. This is a particular problem in epidemiological work where changes in intake over time are being monitored or where one wishes to compare the results from various studies. It will clearly be necessary for all research workers to describe in detail in their publications, which of the various versions of the Australian or other tables they have used in their analyses, while those undertaking longitudinal studies should use a consistent data base with the possibility of having to recompute earlier data sets.
The rolling nature of data release in Australia may also cause problems with the use of computer packages for dietary analysis. Due to their high cost, purchasers will be reluctant, or unable, to repurchase such packages each time NUTTAB, the computerised form of the Australian food tables (Commonwealth Department of Community Services & Health 1989), is updated especially if the need for reprogramming has increased the costs. Clearly a system that can be updated readily and cost-effectively will be needed in the coming years to avoid non-standardised usage.
Conclusion
In conclusion, a reliable and comprehensive food composition data base for Australian foods is essential for nutritional epidemiology research and for other population-based surveys. Although it is possible to assess nutrient status, in some instances, using biomedical indices or chemical analysis of duplicate diets, the size of the sample populations and the wide range of nutrients usually of interest, prohibits this approach in all but a small number of applications. The usefulness of the new Australian food tables (Cashel & others 1989) could be enhanced considerably if funds were available to speed up the release of data for other major food groups and to include analyses, at least for key foods or food groups, for components such as vitamin E, manganese, selenium and nitroso-compounds for which there are no local data and few relevant overseas data available.
References
Anderson, SA (ed). 1986. Guidelines for use of dietary intake data. Bethesda MD: Federation of American Societies for Experimental Biology.
Baghurst, KI & Baghurst PA. 1981. The measurement of usual dietary intake in individuals and groups. Trans. Menzies Found. 3: 139–60.
Block, G. 1982. A review of validations of dietary assessment methods. J. Am. Diet. Assoc. 87: 43–7.
Bransby, ER, Daubney, CG & King, J. 1948. Comparison of nutrient values of individual diets found by calculation from food tables and by chemical analysis. Br. J. Nutr. 2: 232–6.
Buzina, R, Keys, A, Bordarec, A & Fidanza, F. 1966. Dietary surveys in rural Yugoslavia. III. Comparison of three methods. Voeding 27: 99–105.
Byers, T, Marshall, J. Fiedler, R, Zielezny, M & Graham, S. 1985. Assessing nutrient intake with an abbreviated dietary questionnaire. Am. J. Epid. 122: 51–4.
Cashel, K, English, R & Lewis, J. 1989. Composition of foods, Australia. Canberra: AGPS.
Commonwealth Department of Community Services & Health. 1989. NUTTAB89. Nutrient data table for use in Australia. Disk format. Canberra: AGPS.
Eagles, JA, Whiting, MG & Olsen, RE. 1966. Dietary appraisal — problems in assessing dietary data. Am. J. Chin. Nutr. 19: 1–9.
Groover, ME, Boone, L, Houk, PC & Wolf, S. 1967. Problems in the quantitation of dietary surveys. J. Am. Med. Assoc. 201: 8–10.
Marr, J. 1971. Individual dietary surveys: purposes and methods. World Rev. Nutr. Diet. 13: 105–64.
Pekkarinen, M. 1967. Weighing methods in dietary surveys. Voeding 25: 26–31.
Stuff, JE, Ganza, C, Smith, ED, Nichols, BJ & Montandon, CM. 1983. A comparison of dietary methods in nutritional studies. Am. J. Clin. Nutr. 37: 300–5.
Whiting, MG & Leverton, RM. 1960. Reliability of dietary appraisal: comparisons between laboratory analysis and calculation from food tables. Am. J. Publ. Health 50: 815–23.
