Section V

Food Composition Data and Population Studies (continued)

The Effects of Australian, US and UK Food Composition Tables on Estimates of Food and Nutrient Availability in Australia

Karen M. Cashel

School of Human and Biomedical Sciences, University of Canberra, P.O. Box 1, Belconnen, ACT 2616, Australia

Heather Greenfield

Department of Food Science and Technology, The University of New South Wales, Sydney, NSW 2052, Australia

Until the late 1980s, Australia used national food composition tables that were compiled in the late 1960s, predominantly from overseas sources, or foreign tables, particularly those of the UK or the USA. New tables, The Composition of Foods, Australia (COFA), based on an ongoing national analytical program, have been progressively released from 1989. The quantity and adequacy of the foods and nutrients available for consumption in Australia, 1990–91, calculated on the basis of the new Australian tables are compared here with those obtained using the US or UK tables. There are marked differences in the edible weights of foods and the amounts of nutrients available for consumption when the different databases are used. The most marked effect is on the quantity, type and sources of fat in the food supply, assessed as at least 60 per cent higher from meats, and 15–22 per cent higher in total using the data from the US or UK. Iron and zinc are all higher and retinol activity, vitamin C and magnesium lower using the foreign data. Calcium is 35 per cent higher when UK data are used and thiamin 59 per cent higher when US data are used.

Food composition databases are essential components of nutritional monitoring and surveillance, and of much health-related research, yet many countries have traditionally relied on the United States of America (US) or United Kingdom (UK) tables rather than develop their own national tables. Many individual users also rely on non-local data as their source of information. Inappropriate food composition data have the potential to undermine or misdirect the research or nutrition effort, but few studies have been done to provide quantitative evidence of this.

Until the late 1980s, Australia, like many other countries, relied on a national food composition database which incorporated data from a variety of sources, including from overseas tables, scientific publications and food industry information (1). By the late 1970s, the inadequacies of the information provided in the range of food items and nutrients, had many major users, particularly researchers, turning to other sources of data. In Australia, in the main, users were either developing their own databases by supplementing the Australian tables with data from overseas tables, the food industry and journal publications, or using overseas computer-based packages as their principal source. This approach was exacerbated by the growing availability of overseas databases, including in software packages, well in advance of their Australian print only counterpart. The most widely introduced and used overseas food composition data in Australia were those from UK (2), or the US (3), available in print or on computer tape and/or incorporated into software packages. The use of US or UK data in Australia was usually justified by arguments that the health problems and food patterns were similar, and the Australian tables were too limited in their coverage of foods and nutrients.

In 1989, revised food composition data for Australia (4), began to be released. These data were based entirely on an ongoing national food analysis program initiated at the beginning of the 1980s (5). The previous national tables (TCAF) (6, 7) included fewer than 650 food items, and just 16 nutrients, while in 1993, the new tables (COFA) (4, 8–12) include some 1400 food items and a greatly expanded range of nutrients, including data on fatty acids, sugars, amino acids and organic acids. This database continues to grow on an annual basis.

The new analytical data on Australian foods provide a unique opportunity to compare the gross and nutrient composition data of local foods with data from overseas sources for apparently similar foods, and to assess the effect of using local data on the determination of foods and nutrients consumed rather than overseas data. In this paper, the foods available for consumption per capita (13) are used to demonstrate and compare the US, UK and Australian tables.

Data on the per capita food supply have provided the only consistent measure of trends in foods and nutrients consumed in Australia. The food supply data are used to monitor the nutritional adequacy of the food supply, and, in the absence of more specific consumption data at household or individual level, have provided the basis for developing a range of public health nutrition policy and programs, including the nutrition component of the National Health Goals and Targets (14, 15).

The food supply data represent foods as available, rather than as prepared and consumed (i.e. in cooked and/or mixed form). This level of definition of “food consumed” allows ready identification of the scope and source of any differences found specific using alternate sources food composition data. Some of these differences may be difficult to identify, or may be overlooked in foods as consumed due to the effects of different methods of food preparation and combination.

In this paper, the effects of using US or UK rather than Australian national food composition databases are assessed. Specifically, factors influencing the quantity and adequacy of the foods and nutrients available for consumption in Australia will be determined and compared.

• Methods

Food Composition Data

The data used are the food composition tables, or series of tables developed for national use in Australia, the US and the UK. These are, in Australia, Composition of Foods, Australia (4,8–12); for the US, the USDA series Composition of Foods (3); and from the UK the 1978 HMSO edition of McCance and Widdowson's The Composition of Foods and the subsequent supplements released in the 1980s (2,16,17).

The official printed data sources rather than commercial packages were used. As many of the commercial computer-based packages have modified or extended databases, this approach was to ensure that only the official data were used. Further, the printed versions include detailed information and explanatory notes and appendices to assist the user to interpret and apply the data.

• Food Consumption Data

The quantity and type of foods available for consumption per capita (AC) in 1990– 91 (13) in Australia are used (AC). The edible weight of foods and associated nutrients available are calculated using the most appropriate data selected from the three data sources. For example, the edible portion factors (EPF) for carcase meats should allow for losses at the level of both the butcher (carcase to retail meats) and the consumer (retail to raw edible meat) (18).

To assess the adequacy of the food supply to meet the nutrient requirements of the population, the calculated nutrients available per capita are compared with the weighted population recommended dietary intakes (WPRDI). Prior to this comparison, thiamin and vitamin C are adjusted to make some allowance for losses during food processing and cooking and niacin equivalents are calculated (13). The WPRDI is derived by calculating the sum of nutrients needed to provide the RDIs (19) for the proportion of the population in each age and sex group, and the WPRDI is then expressed per capita.

Analyses undertaken

Quantity of Food. The effect of differences in edible portion factors (EPF) was assessed using fruit, vegetable and meat data from the three sources. These factors reflect the proportion of the food that is edible and usually eaten by the population. For example, for a fruit such as the raw orange, the COFA EPF of 0.74 indicates that 74 per cent of the food item is considered edible flesh, the other 26 per cent (in this instance, skin, seeds, pith) is usually discarded.

Nutrients Available — Effect of Differences in EPF. Meats and vegetables were used as the basis for this assessment. The EPF of each of the three data sources were used to calculate the nutrients available from the determined edible weights of meats and vegetables using COFA nutrient composition data.

Nutrients Available — Effect of Differences in Nutrient Composition. For this example, the COFA EPF were used as the basis for determining the edible weight of meats and vegetables. Nutrients available in the food supply from these foods were then calculated using each of the three nutrient composition data sources.

Table I. Effect of different edible portion factors from different food composition tables on the weight of fruits, vegetables and meats available for consumption (kg per capita per year)

a = fresh equivalent weight
b = edible portion factor
c = edible weight
d = includes FEW of grapes to be dried
e = apricots, figs, plums, berries, figs, cherries, custard apples, mangoes, pawpaws, strawberries, olives
f = rounded from more detailed individual data items
g = total edible weight/total FEW
h = beetroot, beans, cucumber, eggplant, marrows, mushrooms, sweet potato etc
j = data in brackets derived using EP factors for beef
k = data in brackets dervied using EP factors for composite boneless meat

Nutrients Available — Effect of Differences in EPF and Nutrient Composition. The effects of differences in edible portion and nutrient composition data in each of the three data sources were assessed for all food groups, including the meat and vegetable groups. The relative contributions of the macronutrients to total energy were calculated.

Nutrient Adequacy. The total nutrients available in the food supply were then assessed for adequacy against the WPRDI. The proportion of energy contributed by the macronutrients was also determined. The range of nutrients included was selected on the basis of consistency across all data sources, and on the basis of those for which there are RDIs for use in Australia (19).

• Results

In this paper, the COFA data are used as the basis for all comparisons made.

Quantity of Food Consumed

Each of the three food composition tables provides EPF for foods such as fruits, vegetables and meats. Table I shows the EPF for a range of raw fruits, vegetables and carcase meats from each of the three data sources. On a weight basis, the fruit and vegetable items comprise 93 per cent of all fruits, and 92 per cent of all vegetables available for consumption in Australia. The remaining items from these food groups are included in the “other” category. For both fruits and vegetables there are marked differences in the EPF for individual foods reported in each source.

Despite differences in the EPF of up to plus 27 per cent or minus 32 per cent of the edible weight (EW) of the individual fruits and vegetables, the impact on the total edible weight of these commodities available for consumption is much smaller (minus 10 per cent to plus 2 per cent). The use of the UK and US data gives a total EW of vegetables 101.9 per cent and 93.9 per cent of that obtained using COFA. The effect on total EW of fruits is greater, with results 90.9 per cent and 94.0 per cent using UK and US data compared to COFA.

The other major food group on which EPFs have a marked effect is the meats. Table I shows that for beef, lamb and pigmeat there is a consistently higher EPF for carcase meats in the UK and US databases than in COFA. The effect of using these EPFs to calculate the raw EW of meat available for consumption is an EW of meats and poultry of 119.1 per cent and 118.9 per cent when UK or US factors are used, respectively, compared to COFA. The basis for the revised EPF for Australian carcass meats is reported elsewhere (18).

Nutrients Available for Consumption

Effect of Differences in EPF. Using COFA nutrient data as a constant in all calculations, Table II indicates the effect of the different EPF from each of the three sources on the nutrients calculated as available for consumption from vegetables and meat. The effects are generally consistent with the differences in total EW shown in Table I. The exception to this is retinol activity contributed by the meats. The EPF for meats (beef, veal, lamb, pigmeat) are low in COFA compared to all other sources, but the EPF for offal are similar. While small amounts of retinol are contributed by muscle meats, as shown in brackets in Table II, offal is the major determinant of the quantity of retinol contributed by meat. This is responsible for the similar retinol contributed by all three sources.

Effect of Differences in Nutrient Composition per 100 g Edible Portion. Table III shows the impact of the differing nutrient data from the three data sources on the contribution of vegetables and meats to the nutrients available in the food supply. The quantities of EW of food are calculated using the COFA EPF. This table shows that the fat contribution from meats was 147 per cent and 133 per cent of that of COFA using the UK and US data, respectively. Combined with the associated variations in protein levels, this results in the energy contributed from meats also being 124 per cent and 121 per cent, respectively, compared to COFA. Using fat-trimmed composite data for boneless meats when available in the US databases reduces this to 115 per cent of the COFA fat contribution. The energy contributed by vegetables using UK and US nutrient data are similarly higher compared to COFA due mainly to the considerably higher carbohydrate levels. Calcium, riboflavin and thiamin contributed by vegetables are higher when US and UK data are used. Magnesium and retinol activity levels from vegetables are also higher than COFA when the US or UK data are used, as is niacin from meats. These results reflect the generally higher levels of these nutrients reported in these data sources. Retinol activity, however, is lower from meats when data sources other than COFA are used. Differences in the composition of offal are primarily responsible for the variation obtained. While offal is the main source of retinol activity contributed by the meats group, the use of US data suggest a considerably greater contribution from other meats, particularly poultry, and a considerably lower contribution when the UK data are used.

Table II. Effect of differences in source of EPF on nutrient contribution from vegetables and meats, quantity per capita per day. COFA used as nutrient composition source

^a = values in parentheses are for retinol activity from non-offal meats

Effect of Differences in EPF and Nutrient Composition Data. Table IV provides a similar comparison for all food groups, except that the different source data EPF factors have also been applied. For vegetables, for example, Table I showed that while the EPF from the different data sources varied considerably for any particular vegetable, the differences were small for the total weight of vegetables. The combination of differences in EPFs and nutrient data at the individual vegetable level, however, result in very different estimates of the nutrients available for consumption from vegetables. The carbohydrate contribution from vegetables compared to COFA is higher using the UK (154 per cent) and US (138 per cent) data. The minerals all show variation with data source. Calcium and magnesium levels are higher when the UK and US food composition tables are used, being up to 156 per cent that of COFA for calcium from vegetables when UK data are used. Zinc contributed by vegetables using US data, is nearly twice that obtained using COFA. The use of UK or US data also suggest a considerably higher total retinol activity from vegetables compared to COFA: 129 per cent and 151 per cent, respectively. This is also seen with thiamin and riboflavin (120 per cent and 125 per cent, respectively, that of COFA when UK data are used). In contrast, vitamin C is around 90 per cent that of COFA when UK or US data are used.

Table III. Effect of differences of nutrient composition on nutrients available from vegetables and meats, quantity per capita per day. COFA used as an EPF source

^a = values in parentheses are for retinol activity from non-offal meats

For meats, the differences in both EPF and nutrient composition in the three data sources further exacerbate the trends observed in Tables II and III. Fat, energy, and the minerals calcium, iron and zinc are all higher when data other than COFA are used. The fat contribution from meats is 178 per cent and 161 per cent of that of COFA when UK or US data are used, while energy levels are 150 per cent and 145 per cent, respectively. The retinol activity levels are all lower using data sources other than COFA; being 77 per cent (UK) and 38 per cent (US) of the level obtained using COFA.

When the nutrients available for consumption from all foods, with thiamin, niacin and vitamin C adjusted as described in the methods section, levels are higher when data other than COFA are used. The exceptions are magnesium, retinol activity and vitamin C for both UK and US data, and thiamin, riboflavin and niacin equivalents when the UK data are used. The higher contribution of meat fat to total available fat suggested by the use of UK and US data has the effect of reducing the relative importance of the added fats and oils as a source of fat in the national diet. Using COFA data, added fats contribute 60 per cent more fat in the national diet than the meats; using the other data sources suggests that the contributions of meats and added fats are about equivalent. Consequently, the ratio of animal fats to vegetable fats obtained using UK and US data is also higher.

Alcohol content varies with the data source, being lower when UK data are used (93 per cent) and slightly higher when US data are used (102 per cent). The retinol activity data vary from 67 per cent (US) to 93 per cent (UK) of those of COFA. Vitamin C is also 16 per cent lower using data sources other than COFA.

Table IV. Effect of different sources of EPF and nutrient composition on nutrients available for consumption per capita per day

a = rounded from more detailed individual data items.
Thiamin and vitamin C adjusted for losses with processing and cooking. Niacin equivalents calculated

b = niacin equivalents

c = without fortification of flour, total calcium = 926 mg; iron = 13.1 mg; thiamin = 1.14 mg and niacin equivalent = 37.5 mg; without fortification of skim milk powder, retinol activity = 2672 mg

d = without fortification of flour and rice, iron = 12.1 mg; thiamin = 1.20 mg, riboflavin = 2.18 mg and niacin equivalent = 40.5 mg

Table V. Macronutrient contribution to total energy ^a (per cent)

^a adjusted to ignore minor contributions to total energy from minor sources such as organic acids

The use of the UK data indicates that grain products are a major source of calcium, providing 25 per cent of the total, as compared with only 5 per cent obtained using the other food composition sources, while the use of the US data suggests that grains contribute twice as much iron and thiamin and nearly 50 per cent more niacin and riboflavin. Both results are due to fortification of wheat flour, with calcium in the UK, and wheat flour and rice with iron, thiamin, niacin and riboflavin in the US. The fortification of wheat flour with iron, thiamin and niacin in the UK is not as apparent in these results.

Table V presents the macronutrient data using the three data sources, as per cent contribution to total energy. Data sources other than COFA result in a higher contribution from fat and a consequent lower contribution from carbohydrate, alcohol and protein.

Nutritional Adequacy of the Food Supply

Table VI shows that the use of the three data sources give values for protein, retinol activity (even with contribution of offal discounted), thiamin, riboflavin, niacin equivalents, and vitamin C that are at least 50 per cent in excess of recommended intakes. For calcium, COFA and the US data suggest that there is little excess available in the food supply relative to the requirements of the population, while the use of the UK data suggests there is a comfortable excess of 41 per cent of this nutrient. With the exception of COFA, the available level of zinc is at least 29 per cent in excess of requirements. The use of COFA data also gives lower levels of iron in excess of the WPRDI than the other data sources. The adjustment of the US and UK databases used to “remove” the fortifying nutrients from wheat flour or rice reduces these differences, with the excess of WPRDI for iron, thiamin and niacin equivalents then being lower (compared to COFA) when US or UK data are used.

• Discussion

Quantity of Food Consumed

EPFs are highly variable and have the potential to markedly affect the estimates of nutrient intakes. There are many possible reasons for the variation reported in the different databases. It may be cultivar related, or due to local preference for a particular unit size or stage of maturity; or it may reflect the degree of pre-market trimming of inedible or unattractive components. For example, mature carrots used to be marketed in Australia with their green leaves. These are now removed prior to sale. Alternatively, differences in EPF may reflect differences in the parts of the food that are considered edible in the local community. For example, in some food cultures, spinach stalks are discarded, in others they are consumed. Further, in Australia, in response to the demand for lower fat meats, there have been changes in developing animals with different characteristics, in preslaughter feeding and handling practices, in butchering techniques and in retail fat trimming practices (20). These affect both EPF and nutrient composition.

Table VI. Effect of differences in source of EPF and nutrient composition on the assessment of the nutritional adequacy of the food supply (per cent in excess of WPRDI^a)

a = WPRDI (ABS, 1993); Mg & Zn calculated for this paper

b = values in parentheses are for comparisons without contribution from offal meats

c = values in parentheses for calcium, thiamin, riboflavin and niacin equivalents are for comparisons without fortification of wheat flour or rice.

Nutrients Available for Consumption

Users have a number of sources of food composition databases available to them. This is particularly apparent in countries such as Australia where a national food analysis program is only of recent origin. Local food availability, food regulations, food preferences and preparation practices all influence the actual gross and nutrient composition of foods. The food composition database selected for use, unless specific to the local food supply, can have a marked effect on the outcome of a study, both on nutrients, and on foods as sources of nutrients.

These and other factors influence the actual composition of a food, and the relevance of the food composition database used. There are, however, other factors that influence the interpretation of food composition data, and the comparability of data from different sources. For example, the analytical methods used to determine nutrient levels and the mode of expression of these nutrient data may vary between food composition tables.

Methods of Analysis. These can have a large effect on the reported value of a nutrient in a food. This effect can be so striking that data from two different tables cannot necessarily be combined and be expected to provide a meaningful assessment of the dietary intake for that nutrient.

The most obvious example is carbohydrate. The carbohydrate data in COFA and the UK tables represent a direct analysis of the sugars and starch content of foods. The carbohydrate data in the US do not represent direct measures, but rather are calculated “by difference”, a method which includes dietary fibre in the carbohydrate data. The US tables add to the confusion by reporting other measures of fibre components, namely crude fibre and pectin.

The method of determining carbohydrate in foods also affects the associated energy calculations. When carbohydrate is determined “by difference”, the carbohydrate energy conversions factors used are food type specific, and allow for the potential fibre component. This is not the case when carbohydrate is determined by analysis.

Other common examples of the effect of different methods of analysis are vitamin C, vitamin A and dietary fibre. The analytical methods used in the COFA and UK tables for vitamin C, for example, include ascorbic acid and dehydroascorbic acid. The US data are measures of reduced ascorbic acid only. Data for total ascorbic acid including the dehydroascorbic acid form is given in footnotes where available.

Modes of Expression. Nutrients may be expressed differently in different tables. The term dietary fibre may include different components dependent upon the method used. Carbohydrate components (starch, sugars and dietary fibre) are expressed as monosaccharide equivalents in the UK tables (2), but as the direct measure in the Australian tables (4,8–12). Vitamin A is expressed as retinol and β-carotene equivalents in the UK; retinol equivalents, retinol and β-carotene equivalents in COFA and the measures are direct weights, but in the US tables the term used is vitamin A, and the values expressed as retinol equivalents or International Units. Energy may be expressed as kilojoules or kilocalories, and the factors used to calculate energy vary as described above. Total energy may include other energy-contributing components such as organic acids, as in COFA.

Missing Data. In the printed version of the UK tables (2), for example, the fact that there are no measures of zinc for a variety of foods, particularly fish, is clearly indicated. In computer based tables, zeros may be inserted with obvious problems for the user who is unaware that “O” may represent either no nutrient detected at the level of delectability of the analytical method used or, no data available on the level of this nutrient in this food. This can result in an incorrect perception of a food such as fish as a food source of zinc, an inappropriately low value for the total dietary intake of this nutrient, and an incorrect interpretation of the results obtained. This is an obvious example of the value of checking the data on the computer version against the official published copy. A related problem for users of computer databases is that of national differences in nutrient fortification regulations and practices. The impact of this on study outcomes and appropriate interpretation of data is clearly shown in this paper. Specific information is needed to both identify and adjust for these effects and even then the effects may be masked by variable voluntary nutrient additions (e.g. in breakfast cereals).

Nutritional Adequacy of the Food Supply

Assessment of the amounts of a nutrient available per capita against the WPRDI is the basis for monitoring the trends in the nutritional adequacy of the Australian food supply. The level of nutrient in excess of the WPRDI, is used as an indication of the “safety margin” for that nutrient. In recent years, the National Health and Medical Research Council has expressed concern about several nutrients in the Australian food supply — thiamin, calcium and iron (21,22).

Thiamin. The data in Table VI data indicate thiamin at 70 per cent in excess of the WPRDI using COFA, whereas if the standard US data were the basis of the assessment, the 172 per cent excess would be grounds for complacency. Grains are the major source of thiamin when all three data sources are used, the absolute level of thiamin contributed by grains when US data are used is much greater compared to the other two sources. This reflects the level of thiamin fortification of wheat flour in the US. Conversely, “removing” this added thiamin from the US or UK wheat flour suggests that there is a lower excess of 35 per cent of WPRDI compared to COFA at 70 per cent. The reasons for this are the naturally higher level of thiamin in Australian flour due to higher extraction rates, and, at that time, a segment of the flour supply contained voluntarily added thiamin.

Iron. The use of COFA shows that iron levels are 45 per cent in excess of WPRDI (considerably lower than obtained using the previous Australian food tables at 93 per cent of the WPRDI (25). This information coupled with a national survey of schoolchildren in 1985 indicating that 9 per cent of 15 year old girls had compromised iron status based on biochemical assessment (23) led directly to a recommendation to “increase the consumption of iron containing foods” in the national dietary guidelines (21). By contrast, the use of the UK or US food composition tables would not have caused such a degree of concern, at 53 per cent and 115 per cent in excess of WPRDI, respectively, the US results reflecting the iron fortification of wheat flour and rice. Differences in breakfast cereal fortification with iron also have an effect, but varies with product as well as country.

Calcium. Calcium intakes are of considerable concern in Australia due to the prevalence of osteoporosis and data indicating a decrease in consumption of milk and milk products in adolescence (24). In 1992 a recommendation “to increase the intake of calcium containing foods” was added to the revised dietary guidelines for Australians (21). The use of COFA or US data to assess the adequacy of this nutrient in the food supply, indicate only a small safety margin. The level of calcium suggested by the standard UK data, however, would not raise the same degree of concern because of the calcium fortification of flours in the UK.

Country Specific Food Composition data

All the data sources are derived from national food composition databases developed to best represent the local food supply. COFA is based primarily on a national food analysis program. The UK database is underpinned by a national analytical program, however, the US database is primarily compiled from analytical data produced by independent, mainly US based researchers.

From the results in this paper, the use of UK and US food tables gives nutrient estimates that are closer to those obtained using TCAF, the “old” Australian tables, than using COFA (25) probably reflecting the previous reliance on the data from these two sources in the compilation of the “Australian” food composition tables. Even with access to the more recent US and UK data used in this paper, the Australian food analysis program has shown that there are real differences in the composition of locally produced and consumed foods.

• Conclusion

The analysis in recent years of foods currently available and consumed in Australia has provided the first opportunity to assess the effect of using data from other countries on perceptions of foods and nutrients available in Australia.

The data presented in this paper show that using overseas data sources to estimate nutrient availability can produce significant errors in the assessment of the nutrient adequacy of the food supply, and of the relative significance of foods as sources of nutrients. The implications for the development of nutrition programs, goals and targets are obvious.

This paper makes a strong case for ongoing support for the local food composition program, and for the use of the Australian food composition database in all Australian nutrition programs and research studies.

While the value of good food composition data which are relevant to the local food supply has been demonstrated in this study, high standards in use of the data will not occur unless users are adequately trained. Such education should include the need to know about local food determinants, as well as how to use and interpret information such as different analytical methods, modes of data expression, and the rates for sampling, method choice and analytical quality assurance (26,27).

• Acknowledgments

The Australian Bureau of Statistics kindly made available unpublished details of the estimates of foods available for consumption. We thank Michael de Looper, Australian Institute of Health and Welfare, for assistance with the WPRDI calculations.

• References

(1)   English, R. (1981) Food Tech. Aust. 33, 103–106.

(2)   Paul, A.A., & Southgate, D.A.T. (1978) The Composition of Foods, HMSO, London

(3)   US Department of Agriculture (1976- ) Composition of Foods: Raw, Processed, Prepared, Agric. Handbook No. 8 series, USDA, Washington, DC

(4)   Cashel, K., English, R., & Lewis, J. (1989) Composition of Foods, Australia, Australian Government Publishing Service, Canberra

(5)   English, R. (1986) Trans. Menzies Found. 11, 25–34

(6)   Thomas, S. & Corden, M. (1970) Tables of Composition of Australian Foods, Australian Government Printer, Canberra

(7)   Thomas, S., & Corden, M. (1977) Metric Tables of Composition of Australian Foods, Australian Government Printer, Canberra

(8)   English, R., Lewis, J., & Cashel, K. (1990) Composition of Foods, Australia, Volume 2, Cereals and Cereal Products, Australian Government Publishing Service, Canberra

(9)   Lewis, J., & English, R. (1990) Composition of Foods, Australia, Volume 3, Dairy Products, Eggs and Fish, Australian Government Publishing Service, Canberra

(10) English, R., & Lewis, J. (1990) Composition of Foods, Australia, Volume 4, Fats and Oils, Processed Meats, Processed Fruit and Vegetables, Australian Government Publishing Service, Canberra

(11) Lewis, J. & English, R. (1990) Composition of Foods, Australia, Volume 5, Nuts and Legumes, Beverages and Miscellaneous Foods, Australian Government Publishing Service, Canberra

(12) Lewis, J., Holt, R., & English, R. (1992) Composition of Foods Australia, Volume 6, Infant Foods, Australian Government Publishing Service, Canberra

(13) Australian Bureau of Statistics (1993) Apparent Consumption of Foodstuffs and Nutrients, Australia, 1990–91, ABS, Canberra

(14) English, R. (1987) Towards Better Nutrition for Australians, Australian Government Publishing Service, Canberra

(15) Health Targets and Implementation Committee (HTIC) (1988) Health for All Australians, Australian Government Publishing Service, Canberra

(16) Holland, B., Unwin, I.D., & Buss, D. (1988) Third Supplement to McCance and Widdowson's The Composition of Foods, 4th edition: Cereals and Cereal Products, Royal Society of Chemistry, Nottingham

(17) Holland, B., Unwin, I.D., & Buss, D. (1988) Fourth Supplement to McCance and Widdowson's The Composition of Foods, 4th edition: Milk Products and Eggs, Royal Society of Chemistry, Nottingham

(18) Cashel, K., & Greenfield, H. (1994) Br. J. Nutr. 71, 753–773.

(19) National Health and Medical Research Council (1992) Recommended Dietary Intakes For Use in Australia, Australian Government Publishing Service, Canberra

(20) Warren, B., & Channon, H. (c1990) Lamb Cutting Notes 1. More Fat Means Less Saleable Meat, Rutherglen Research Institute, Victoria

(21) National Health and Medical Research Council (1992) Dietary Guidelines For Australians, Australian Government Publishing Service, Canberra

(22) National Health and Medical Research Council (1989) Report of the 108th Session of the Council, Australian Government Publishing Service, Canberra.

(23) English, R., & Bennett (1990) Med. J. Aust. 152, 582–586.

(24) English, R., Cashel, K., Lewis, J., Bennett, S., Berzins, J., Waters, A., & Magnus, P. (1988) National Dietary Survey of Schoolchildren Aged 10–15 Years, No. 1, Foods Consumed, Australian Government Publishing Service, Canberra

(25) Cashel, K., & Greenfield, H. (1995) J. Food Comp. Anal. (in press)

(26) Greenfield, H. (1990) Food Aust. 42, S1–S44

(27) Greenfield, H., & Southgate, D.A.T, Food Composition Data: Production, Management and Use, Elsevier Applied Science, London

Quality Control in the Use of Food and Nutrient Databases for Epidemiologic Studies

I. Marilyn Buzzard, Sally F. Schakel, Janet Ditter-Johnson

Nutrition Coordinating Center, Division of Epidemiology, School of Public Health, University of Minnesota, 1300 South Second Street, Minneapolis, MN 55454-1015, USA

This paper describes procedures used by the Nutrition Coordinating Center (NCC) at the University of Minnesota for maintaining food and nutrient databases. NCC's databases are designed to support an automated system for dietary data collection and nutrient calculation for clinical trials and other nutrition research and large population-based studies. The three major databases include the Nutrient Database, the Food Database, and the Brand Database. The Nutrient Database consists primarily of food composition data for “core” (non-recipe) foods. The Food Database drives the system's interactive prompting for detailed food descriptions and specification of amounts consumed. This database also contains all of the non-nutrient data required to link food descriptions with one or more entries in the Nutrient Database and convert amounts consumed to gram weights for nutrient calculation. The Brand Database, which contains food and nutrient information for commercial products, is used to update the other two databases. Each of these databases and the quality control procedures used for maintaining them are described. Because many of the studies using NCC databases are long term projects, time-related database maintenance and quality control procedures are required. These procedures permit routine updating to reflect the changing marketplace and the availability of new or improved data, while also ensuring comparability of dietary data collected at any point in time.

The Nutrition Coordinating Center (NCC) at the University of Minnesota maintains food and nutrient databases to support a system for the collection and nutrient calculation of dietary data. The system is designed primarily for clinical trials and other medical research and epidemiologic studies investigating relationships between diet and health (1–3). The Minnesota nutrition data system has been used for hundreds of research studies over the past two decades. The majority of these studies have been funded by the US National Institutes of Health. A brief overview of the requirements for the system will provide a basis for understanding the functionality of the databases that drive the system.

• Overview of the Minnesota Nutrition Data System

The Minnesota nutrition data system was designed to meet the needs of its usersprimarily large, population-based nutrition research studies. These needs include the following: standardized procedures for collecting food intake data, especially for multi-centered studies collecting data at many different centers by many different individuals; a high level of specificity for describing foods, including methods of food preparation and brand identification; food and nutrient databases that are frequently updated and which contain no missing values of nutritional significance; a food composition database that continues to expand to include the nutrients of current research interest; rapid data processing; and database maintenance procedures that permit accurate comparison of food and nutrient intakes over time.

To avoid the need to maintain different versions of the system for different users, the most stringent requirement demanded by any one user is provided to all users. For example, if only a few studies need specificity for sodium, this level of specificity is provided to everyone; those who are not interested in sodium intake may opt not to ask questions about salt use in food preparation, and the system will automatically assign the default amounts. If one study needs foods added to the database for a special study population, these foods become available to all users.

The three major components of the current version of the Minnesota nutrition data system are shown schematically in Figure 1. They include interactive data collection, automated coding, and nutrient calculation. Over the years the system has evolved to take advantage of technological advances and more accurate and efficient procedures for collecting and processing the data (4). When the system was first developed about 20 years ago, the first two steps were not automated (1,2). Food intake data were collected on paper, and the quality of the data depended largely on the skills of the interviewer or on how well subjects were trained to keep food records. The data collection process has now been automated so that the computer provides all of the prompts required to describe foods at the appropriate level of detail (5).

Figure 1. Major components of the Minnesota nutrition data system

Figure 2. The three databases that drive the Minnesota nutrition data system

Similarly, coding was initially done on paper by trained food coders, and the coded data were then entered into the computer by data entry operators. Despite the use of duplicate data entry and subsequent computerized edit checks, there was potential for transcription and other errors. About ten years ago, an on-line coding system was developed which allowed coders to enter the data directly into the computer (6). Edit checks could then be invoked at the point of data entry, which greatly enhanced the accuracy and efficiency of the coding process.

The final step of completely automating the coding process was not possible until we had completed the development of the interactive data collection component. Only when all of the detail required for coding is captured by the computer can the coding be totally automated. This enhancement, which was implemented about five years ago, resulted in substantial improvements in accuracy and standardization, as well as savings in time and effort, since coding has traditionally been the most labor intensive part of processing dietary data.

The three major components of the automated system have now been incorporated into a software package that is currently being used at approximately 150 research institutions in the US and Canada (5,7). A customized version of the system was developed for collecting 24-hour dietary recalls for the Third National Health and Nutrition Examination Survey (NHANES III) which is now in its fifth year of data collection (8).

The three databases required to maintain the automated nutrition data system are shown schematically in Figure 2. They include the Nutrient Database, the Food Database, and the Brand Database. Each of these databases will be described in greater detail below.

• The Nutrient Database

The NCC Nutrient Database is the smallest of the three databases. It consists of the following data fields:

• food codes for approximately 1700 “core” (i.e., non-recipe) foods

• food descriptions for each food code

• amounts per 100 g for each of 94 food components for each food

• reference codes for each nutrient value

• serving sizes designated by the US Food and Drug Administration and the US Department of Agriculture (USDA) for labeling purposes

• food group designations for several different food grouping classification schemes.

Our philosophy is to keep the number of foods in the Nutrient Database as small as possible to minimize maintenance efforts and facilitate rapid updating (6). Foods are included only in their “as eaten” state; for example, foods that are never eaten raw are not included in this database. The majority of the foods in the Nutrient Database are single ingredient foods, but there are also a number of commonly consumed multi-ingredient processed foods such as cheese, bread, and sausage. Each food entry is described in detail in a text field; the Latin or scientific name is also included if applicable.

The 94 food components in the current NCC Nutrient Database include: energy; the proximate nutrients (protein, fat, carbohydrate, and alcohol, plus water and ash); animal and vegetable protein plus 18 amino acids; 23 individual fatty acids; cholesterol; starch; six simple sugars; total dietary fiber and three fiber fractions; nine minerals; 17 vitamins, including two vitamin A fractions and four fractions of vitamin E; plus caffeine, aspartame, and saccharin. Every nutrient value in the database is associated with a reference code documenting the source of the data and/or the method used to impute the data, if applicable (9). The database also includes fields for three different food grouping schemes to facilitate analysis by food groups and to accommodate different research objectives.

Sources of data for the NCC Nutrient Database are described in detail elsewhere (9). The primary sources are USDA data tapes and publications, information from manufacturers of brand name products, and the scientific literature. The USDA Nutrient Data Base for Standard Reference is the major USDA data set used by NCC (10). The USDA Survey Database (11) provides many imputed values that are not included in the Data Base for Standard Reference. USDA Handbook No. 8 (12) provides additional information not included in the Standard Reference data sets such as specific factors for calculating energy values, standards for enrichment of grain products, and values for the amounts of separable lean and fat of retail beef cuts. Other USDA data sets used by NCC include various provisional tables and bulletins (9).

Nutrient data for brand name products are becoming increasingly important as the consumption of processed foods continues to increase in the US. Values for commercial products are obtained from the NCC Brand Database (described below). The scientific literature is another important source of nutrient data, especially for those nutrients included in the NCC database that are not currently provided by USDA. Food composition tables from other countries are occasionally used to obtain values for foreign foods not included in the USDA data sets.

• The Food Database

The NCC Food Database exceeds the size of the Nutrient Database by approximately ten-fold. It includes the hierarchy of food descriptions that drives the interactive prompting for detailed food identification (5, 7). The hierarchy consists of about 17,000 food descriptions for foods consumed in North America. This includes brand name descriptions as well as generic descriptions, in addition to a large number of ethnic and regional foods, dietary supplements, and medications containing caffeine and sodium. The hierarchy is organized in a manner that reflects the way people think about foods, rather than according to any scientific classification. The hierarchical organization facilitates the prompting for food description detail by presenting a series of menu selections that become progressively more detailed until the food is adequately described.

For each food in the hierarchy of food descriptions, the Food Database provides all of the data required to link the food with one or more entries in the Nutrient Database and to convert amounts expressed in various common units to gram weights for nutrient calculation (5,7). Examples of other data fields in the Food Database include codes that designate the type of food preparation method; ingredient listings and amounts for recipes and formulations; designation of ingredients that require further description (such as the type of fat used in a recipe); default assignments, based on nationally representative market research or food consumption data, which designate the most common of the available options when a subject cannot provide the level of detail requested; any geometric shapes (e.g., cube, sphere, wedge, or cylinder) in which the food might be described; one or more density conversion factors, depending on the various forms in which a food can be measured (e.g., solid, chopped or grated); and other amount conversion factors, such as the weights of food-specific portions (e.g., small, medium, or large piece; slice; or package), raw to cooked yields, and edible portion conversions. Also included is a maximum serving size for each food in the database to serve as a quality control check for unusually large amounts.

Sources of data for the Food Database are documented by reference codes. The primary sources are the coding manual section of the USDA survey database (11) and information from manufacturers. Several other USDA publications related to food weights, yields, and portion sizes are also used for the Food Database. These publications are referenced by Schakel et al. (9).

NCC currently maintains two separate versions of the Food Database. Food descriptions in the one version are linked to the NCC Nutrient Database, whereas in the other version, the foods are linked to the USDA Survey Database (11). The latter version, which has been customized for collecting 24-hour dietary recalls for NHANES III (8), includes some additional modifications to enhance comparability of nutrient calculations with calculations from the USDA surveys.

• The Brand Database

The NCC Brand Database contains food and nutrient information for selected categories of commercial products. This database continues to grow as the food marketplace expands and changes. In the US there are approximately 1000 new products introduced every month (13). So trying to keep up with even the most popular foods is a never ending process. We currently maintain data for about 7000 products in the Brand Database. Information from this database is used routinely to update both the Nutrient Database and the Food Database.

Brand name information is used for several different purposes. In some cases it is needed to adequately identify the food that is consumed. For example, subjects may describe a food by its brand name, such as “Coke,” rather than by a more generic description, such as “car bonated cola beverage.” Brand name data may also be needed to differentiate between similar products that differ significantly in composition. For example, if different brands of carbonated cola beverages differ with respect to caffeine content, knowing the brand facilitates identification of the appropriate caffeine level. Another very important use of brand information is to help determine the amount of the food consumed. For example, a subject may report consuming a small container of low fat yogurt. Since low fat yogurt is available in several different “small” sizes, knowledge of the brand will often permit accurate determination of the amount consumed.

Figure 3. Example of a screen from the Food Database Maintenance System used for the NHANES III survey

Although our goal is to update each food category at least annually, changes in the marketplace often determine priorities for updating. For example, we might update frozen entrees and ready-to-eat cereal several times during the year due to the influx of many new products and product reformulations in these categories, whereas other categories that are not changing so rapidly might not be completely updated for several years, other than obtaining manufacturers' information on a few new products that appear on food intake records obtained from research subjects.

Not all brand name food categories are included in the Brand Database. If there are no significant differences among brands within a food category with respect to nutrient content or serving amount, that category is not included in the database. For example, brand name canned vegetables are not included in the database. The current version of the Brand Database includes 25 food categories.

Examples of the types of data fields in the Brand Database include: the product code, an arbitrary number assigned by NCC; the product name; a detailed description of the product; the name of the manufacturer and the product's Universal Product Code (UPC); the product category designation; density information, if available; package size; serving size; servings per package; ingredient listing; and preparation instructions. Nutrient values provided by the manufacturer, including label data as well as analytical or calculated data, are included in the database. Although analytical data are preferred for nutrient calculation, studies sometimes prefer label values for developing educational materials. Nutrient values obtained from other sources, such as from the literature, are also included. Sources of all information are indicated by reference codes, and the dates of receipt of the data at NCC are noted.

Table I. A partial list of NCC edit limits for entering nutrient values

• Quality Control Procedures

Quality control procedures are critical for maintaining food and nutrient databases because the potential for error in dealing with hundreds of thousands of data elements is very great. Quality control procedures are designed both to reduce the potential for error and to increase the likelihood of identifying errors that occur inadvertently. The quality control procedures used by NCC for database maintenance are similar for the three databases described above. These procedures include the following: data evaluation based on established criteria; automated data entry whenever possible; comparison of new with pre-existing data; wellorganized data entry screens; edit checks at the point of data entry; review of all manually entered data; checks for consistency among related data fields; and review of data fields within food groups for consistency with expected ranges of values.

When data are available from multiple sources, criteria are used to select the most appropriate values (9). Analytical data are generally preferred over calculated data; however, the quality of the analytic procedures used, the extent to which the data represent nationwide sampling and eating habits, and the currency of the data are also important considerations. USDA data are generally preferred over other data sources, and refereed publications are preferred to other publications, such as meeting proceedings or text books. If no published data are available, we occasionally use unpublished data, such as those provided by a reputable laboratory. Because missing nutrient values are calculated as zeros, they can result in significant underestimations of nutrient intakes (14, 15). Therefore, when values are not available from any sources, we calculate or impute them using established procedures (3, 15–17). A great deal of nutritionist effort is involved in imputing data to ensure that there are no missing values of nutritional significance in the database. Imputed values are replaced with analytical data when they become available.

Whenever possible, data entry is automated to enhance efficiency and reduce the potential for data entry errors. Most USDA data are now available in electronic form, and we are able to link many of our Nutrient Database entries with USDA food descriptions via the reference codes. We hope that manufacturers will eventually provide product information electronically, which would greatly reduce the potential for error. When database changes are required for an entire classification of foods, rather than for a single food item, a computer program is written to implement these global changes. For example, when the default for “margarine, regular stick, brand unknown” was changed to reflect more recent market research data, the new default code number was globally inserted into all recipes that included the default margarine.

Well-designed data entry screens can reduce the potential for error. Such screens may be formatted in a manner that simulates the format of the input documents, thus decreasing the amount of eye movement required by the data entry operator. An example of a well designed data entry screen is shown in Figure 3. Note that the different types of data are separated by labeled boxes, so specific data are easy to locate. The screen is not too crowded, which makes it easy to view.

To further reduce the potential for error, data are entered into the database in the format in which they are received. For example, nutrient values provided by the manufacturer may be entered into the Brand Database as amount per serving, amount per Reference Daily Intake (RDI), or amount per 100 g. Edit checks at the point of data entry permit immediate correction of keying errors. These checks flag data that are out of a given range or do not conform to other field specific restrictions. A partial list of edit limits used for entering nutrient values is presented in Table I. Since nutrient composition may vary substantially among food groups, the limits are usually food group specific. Another type of edit check is the flagging of incomplete data sets; for example, some data fields require that the reference codes be designated before the computer will accept the data.

All data entered manually into the database are cross-checked by a second database nutritionist. Data that are manually entered into the Nutrient Database are stored in a temporary file. Each database nutritionist is assigned a color that identifies the values entered by that individual; if changes are recommended by the second nutritionist, these changes must be verified by the nutritionist who originally entered the data. Only after verification of any changes are the new data accepted for posting to the permanent data file.

Reports are routinely generated comparing new data with previous data. For example, when a new version of the USDA Nutrient Database for Standard Reference is installed, a report of differences between the new and the pre-existing values is generated. Any unusually large differences are verified through communication with USDA staff. Similarly, reports are generated to compare new data from manufacturers with any pre-existing data for the same product. If differences are noted, the manufacturer is contacted to determine if the differences are due to a reformulation of the product or to improved composition data. This information allows us to determine whether a new entry needs to be created or an existing entry updated. For this reason, we routinely request information for all food products marketed by a manufacturer, not just for the new products.

Table II. Examples of relational edits for the Nutrient Database

^a proximates include protein, carbohydrate, fat, alcohol, water and ash
^b pro=protein; carbo=carbohydrate; alc=alcohol
^c acceptable difference depends on type of food

New versions of the Food and Nutrient Databases are generally released concurrently. During the four-week period prior to the release of the new versions, all modifications to the databases cease while the efforts of the database nutritionists are devoted to the various quality control procedures. Three types of quality control reports are generated prior to the release of the new versions. One type, the relational edits, are reports that examine consistency among different fields in the database (3). For example, appropriate relationships among nutrients in the Nutrient Database are verified by comparison of calculated values with expected values. The calculated values must fall within an acceptable range of the expected result. Table II lists a few examples of the 28 relational edits currently generated for verification of the Nutrient Database. Examples of relational edits for the Food Database include verification of rules such as: every recipe ingredient reported by volume must have a density; and every food that can be described in terms of a geometric shape must have a solid density. There are currently 26 of these relational edits for the Food Database. Although many of these edits are invoked at the point of data entry, others must be verified before a new version of the Food Database is released.

Another type of quality control check conducted before the release of a new version of one of the databases is a review of computer listings of selected fields by food category. For example, vitamin A values for all processed cheeses in the Brand Database are scanned for any outliers which must then be verified from the original data source. This type of review is conducted for all nutrients in the Nutrient Database that are not included in the relational edit type of consistency checks shown in Table II. Many of the non-nutrient data fields are also subjected to this type of review. For example, the options for various food shapes in the Food Database are compared within food groups, and any differences must be justified.

A final quality control report that is generated before the release of a new version of the Food and Nutrient Databases is the calculation of nutrients for a test set of menus specifically designed to include a wide variety of foods, as well as all of the functionalities of the calculation software. These calculations are compared with calculations from the previous versions of the databases; any differences must be verified as the result of intended modifications to the databases or to the calculation software.

The three NCC databases are currently maintained separately, but a project is underway to integrate them into a single database management system. This will enhance our quality control by permitting us to automate more of the data entry than is now possible. For example, selected information from the Brand Database will be able to be automatically transferred to the Food Database rather than having to be manually entered.

• Time-Related Database Maintenance and Quality Control Procedures

The need for comparability of dietary data over time requires use of time-related procedures for maintaining the Food and Nutrient Databases (18). Time-related database maintenance procedures permit recalculation of previously collected food intake data at any subsequent time to take advantage of new or improved data, including updates to both the Food Database and the Nutrient Database. Use of these procedures ensures comparability within long-term studies and among studies by eliminating the confounding due to using different databases and coding practices at different time periods.

Database changes that reflect real changes in the foods people are eating or in food preparation methods, must be differentiated from those changes that represent new or improved data for foods that have not changed. Time-related changes include most marketplace changes, such as new products, new serving sizes of existing products, product reformulations, and discontinued products. Changes in food preparation methods, such as increased trimming of fat from meats or use of less salt and fat in recipes, may also represent time-related changes. Non-time-related database changes include such changes as new or improved analytic data, or better data for calculation of nutrient retentions in cooked or processed foods.

Procedures for time-related database maintenance require quality controls to ensure consistency among subsequent database versions. For every change made to the Food Database, the nutritionist must indicate whether or not the change is retroactive to previous versions of the database. The computer will not accept a change unless this information is entered. For example, improved data, such as more accurate values for raw to cooked conversion factors, are always retroactive to all previous versions of the Food Database. Edit checks at the point of data entry prevent making changes that would compromise consistency with previous versions of the database. For example, deletion of a food in the Food Database cannot be retroactive to previous versions because studies must be able to edit food intake records collected on previous versions.

New versions of the Food and Nutrient Database are released approximately every six months. Time-related changes are handled differently for maintaining the Food Database, which is used for data collection and coding, than they are for maintaining the Nutrient Database, which is used for nutrient calculation. The current version of the Food Database must always reflect the current marketplace. Products no longer on the market must be deleted from the database to make sure that they are not selected when they are no longer available. For the Nutrient Database, however, foods must never be deleted, since a new version of this database may be used for calculating nutrients for dietary data collected at any time in the past. Discontinued foods continue to be updated to reflect improved nutrient data and the addition of new nutrients to the database.

Thus, for a long term study, many versions of the Food Database must be used for collecting and coding the data, whereas a single version of the Nutrient Database is used for calculating nutrients for the entire study. Whenever a new version of the Nutrient Database is released, dietary intake data for the entire study may be recalculated on that version. This will ensure that the most current data are used for nutrient calculations and that the calculations are comparable throughout the study.

To facilitate ongoing editing of dietary data for long-term studies, NCC has developed a Multi-Version Food Database that collapses all existing versions of the Food Database into a single database. Each subsequent release of the Multi-Version Food Database incorporates only those data which have changed since the previous release of the database. This eliminates the redundancy that exists among individual versions of the database. Use of the Multi-Version Food Database permits the editing of food intake data collected at any point in time. For example, NHANES III will have used 12 different versions of the Food Database for collecting and coding dietary data over the six years of the survey. Because all editing of dietary recalls must be done using the version on which the data were collected, the editing would be very cumbersome without the functionality of the Multi-Version Food Database.

All of the quality control procedures previously mentioned have been incorporated into the maintenance software for the Multi-Version Food Database. Careful adherence to time-related procedures for maintaining databases allows investigators to recalculate their dietary data automatically at any future time to take advantage not only of new nutrients and other food components that have been added to the Nutrient Database, but also of improved nutrient and non-nutrient data that have become available. These procedures make it possible to monitor trends in foods and nutrient intakes over time, which is especially important for meeting the objectives of ongoing dietary surveys and long term research studies.

• Acknowledgments

Funding to support this work has come primarily from the National Heart, Lung, and Blood Institute and the National Cancer Institute of the National Institutes of Health, Bethesda, MD, and from the National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, MD. Specific grants and contracts include the following: NIH/N01HV-12903, NIH/RO1-HL-42165, NIH/R01-CA-36522, and CDC/200-89-7014.

• References

(1)   Dennis, B., Ernst, N., Hjortland, M., Tillotson, J., & Grambsch, V. (1980) J. Am. Diet. Assoc. 77, 641– 647

(2)   Tillotson, J.L., Gorder, D.D., & Kassim, N. (1981) J. Am. Diet. Assoc. 78, 235–240

(3)   Sievert, Y.A., Schakel, S.F., & Buzzard, I.M. (1989) Contr. Clin. Trials 10, 416–425

(4)   Buzzard, I.M., Price, K.S., & Feskanich, D. (1991) in The Diet History Method, L. Kohlmeier (Ed.), Smith-Gordon and Company, London, pp. 39–51

(5)   Feskanich, D., Buzzard, I.M., Welch, B.T., Asp, E.H., Dieleman, L.C., Chon, K.R., & Bartsch, G.E. (1988) J. Am. Diet. Assoc. 88, 1263–1267

(6)   Buzzard, I.M. (1989) in Nutritional Status Assessment of the Individual, The Food and Nutrition Press, Inc., Trumbell, pp. 87–98

(7)   Feskanich, D., Sielaff, B.H., Chon, K., & Buzzard, I.M. (1989) Comp. Meth. Prog. Biomed. 30, 47–57

(8)   McDowell, M., Briefel, R.R., Warren, R.A., Buzzard, I.M., Feskanich, D., & Gardner, S.H. (1990) in Proceedings of the 14th National Nutrient Databank Conference, The CBORD Group, Inc., Ithaca, NY, pp. 125–131

(9)   Schakel, S.F., Sievert, Y.A., & Buzzard, I.M. (1988) J. Am. Diet. Assoc. 88, 1268–1271.

(10) US Department of Agriculture (1993) Nutrient Database for Standard Reference, Release 10, USDA, Washington, DC

(11) US Department of Agriculture (1993) Nutrient Data Base for Individual Food Intake Surveys, Release 6, National Technical Information Service, Springfield, VA

(12) US Department of Agriculture (1976- ) Composition of Foods: Raw, Processed, Prepared, USDA Agric. Handbook No. 8 series, USDA, Washington, DC

(13) Gorman, B. (1990) Prep. Foods New Prod. Ann. 159, 16–18, 47–52

(14) Buzzard, I.M., Price, K.S., & Warren, R.A. (1991) Am. J. Clin. Nutr. 54, 7–9

(15) Posati, L. (1985) in Proceedings of the Tenth National Nutrient Databank Conference, National Technical Information Services, Springfield, VA, pp. 124–133

(16) Schakel, S.F., Warren, R.A., & Buzzard, I.M. (1990) in Proceedings of the 14th National Nutrient Databank Conference, The CBORD Group, Inc., Ithaca, NY, pp. 155–165

(17) Westrich, B.J., Buzzard, I.M., Schakel, S.F., & McGovern, P.G. (1993) in Proceedings of 18th National Nutrient Databank Conference, Washington, ILSI Press, p.276

(18) Buzzard, I.M. (1991) in Proceedings of the 16th National Nutrient Databank Conference, The CBORD Group, Inc., Ithaca, NY, pp. 73–77

Construction of a Database of Inherent Bioactive Compounds in Food Plants

Andrew D. Walker, Roger Preece

Institute of Food Research, Computing Group, Norwich Research Park, Colney Lane, Norwich, NR4 7UA, UK

Jenny A. Plumb, Roger Fenwick, Bob K. Heaney

Institute of Food Research, Food and Molecular Biochemistry Department, Norwich Research Park, Colney Lane, Norwich, NR4 7UA, UK

Numerous incidences have been recorded where naturally-occurring dietary components have contributed to chronic and acute illness and occasionally to human fatalities. The causative agents, termed “natural (or inherent) toxicants”, are commonly present in food plants in order to provide protection against fungal, insect and herbivore attack. Perhaps the best known example of the effects of such a toxicant is the severe gastrointestinal and neurological disturbances observed following consumption of damaged, green or sprouted potatoes containing high levels of glycoalkaloids. Other classes of compounds with well-defined physiological effects include glucosinolates (in brassica vegetables and condiments), lectins (in legumes), isoflavones (in soya), cyanogenic glycosides (in cassava and legumes) and psoralens (in parsnip and celery).

In order to study the varying biological effects of these natural toxicants (or of naturally occurring non-nutritional compounds offering protection against heart disease or cancers), it is essential that the available information on the content of these bioactive compounds in foods is readily accessible to workers in the plant science, food science, nutrition and clinical areas. Although databases exist which contain data on the nutrient compositions of foods, readily accessible/critically evaluated data on the occurrence and levels of inherent biologically active compounds in foods is not yet available. Researchers at the Institute of Food Research (IFR) have designed and are currently compiling a database to include information on occurrence, levels and factors affecting levels of natural toxicants, anti-nutrients and protective factors in foods.

The database, constructed using the ORACLE Relational Database Management System (RDBMS) (1), provides information on the levels of specific compounds in food plants, on the country of origin, the plant part analyzed, any preparation methods used; information on varieties and sample numbers is included together with full citations for all references used. The relational data model used can be used to construct the database using other RDBMS. The system was originally developed on a DEC VAX minicomputer and has been successfully installed on an IBM PC 486 compatible.

Primary literature searches on specific natural toxicant occurrence and factors affecting levels have been carried out using CD ROM databases (e.g. Agricola and Food Science and Technology Abstracts) and are routinely updated using Current Contents on disk. Collected references are read and any relevant references cited therein are additionally obtained for use as a secondary source of data. Reprints describing occurrence data for a number of naturally-occurring toxicants are critically assessed for data quality according to defined criteria. These criteria, developed at IFR after wide consultation, contain guidelines on acceptability of analytical method, in sampling, unequivocally-identified plant species, etc. Only data which satisfy these criteria are entered into the database.

Currently, eight compound classes (covering 65 compounds) including glycoalkaloids, glucosinolates, psoralens (furocoumarins), alkenylbenzenes, saponins and hydrazines have been entered. One hundred and twenty foods have been coded giving a total number of records of about 1100.

Together with the records from individual references, further fields have been entered including a thesaurus of alternative compound names, a thesaurus of food names, and a textual “comments” screen containing information on the way in which levels may change as a result of processing, storage, cooking, agronomic and environmental conditions etc. All references relevant to this field have been entered. Output screens have been developed linking food name, food part, preparation method and compound name. The mean levels of each compound together with maximum and minimum levels are calculated from all the data in the database assigned a satisfactory quality code, and are presented with supporting information on the number of records and references used to obtain these data.

Planned developments include expansion of the number of foods and compounds within the database, inclusion of data relating to toxicological effects, and the setting up of a European database on non-nutrient composition of foods.

• Acknowledgment

This project has been funded by the Ministry of Agriculture, Fisheries and Food.

• Reference

(1) ORACLE 6 The ORACLE Corporation UK Ltd, Bracknell, UK