Contents - Previous - Next

This is the old United Nations University website. Visit the new site at http://unu.edu

Summary

Some illustrative problems related to nutrient data bases that have the potential to affect the outcomes of epidemiological research have been presented here. This list is by no means exhaustive of all problems that are encountered in this kind of diet related research. In general, additional food composition data will improve the power of many epidemiologic research projects while standardization and careful documentation of data bases will allow more appropriate comparisons between studies. Bioavailability, nutrient (and non-nutrient) interactions, and the influence of environmental factors on food composition all have an impact on the outcome of diet-related epidemiologic investigations. These factors represent new challenges in nutrient data base management.

References

1. S. Gingham, R. R. Williams, J. J. Cole, and W. P. T. James, "Dietary Fibre and Regional Large-bowel Cancer Mortality in Britain," Br. J. Cancer, 40: 456-463 (1979).

2. R. R. Butrum and S. E. Gebhardt, "Nutrient Data Bank: Computer-based Management of Nutrient Values in Foods," J. Am. Oil Chemist' Soc., 53 (1976).

3. J. F. Calder, M. W. Wachira, T. Van Sant, M. S. Malik, and R. N. Bowry, "Diverticular Disease, Carcinoma of the Colon and Diet in Urban and Rural Kenyan Africans," Diagn. Imaging, 49: 23-28 (1980).

4. B. M. Calkins, D. J. Whittaker, P. P. Nair, A. A. Rider, and N. Turjman, "Diet, Nutrition Intake, and Metabolism in Populations at High and Low Risk for Colon Cancer: Nutrient Intake," A. J. C. N., 40: 896 905 (1984).

5. L. B. Dales, G. D. Friedman, H. K. Ury, S. Grossman, and S. R. Williams, "A Case-control Study of Relationships of Diet and Other Traits to Colorectal Cancer in American Blacks," Am. J. Epidemiol., 109: 132-144 (1979).

6. H. N. Englyst, S. A. gingham, H. S. Wiggins, et al., "Nonstarch Polysaccharide Consumption in Four Scandinavian Populations," Nutr. Cancer, 4: 50-59 (1982).

7. J. E. Enstrom, "Colorectal Cancer and Beer Drinking," Br. J. Cancer, 35: 674 (1977).

8. S. Graham, H. Payal, M. Swanson, A. Mittleman, and G. Wilkinson, "Diet in the Epidemiology of Cancer of the Colon and Rectum," J. Natl. Cancer Inst., 61: 709-714 ( 1978).

9. W. Haenszel, H. B. Locke, and M. Segi, "A Case-control Study of Large Bowel Cancer in Japan," J.N.C.I., 64: 17-22 (1980).

10. R. G. Hansen, B. W. Wyse and A. W. Sorenson, Nutritional Quality Index for Food (AVI, Westport, Conn., 1979).

11. M. W. Hinds, L. N. Kolonel, J. Lee, and T. Hiroshata, "Associations between Cancer Incidence and Alcohol/Cigarette Consumption among Five Ethnic Groups in Hawaii," Br. J. Cancer, 41: 929 940 (1980).

12. International Agency for Research on Cancer Intestinal Microecology Group, "Dietary Fiber, Transittime, Fecal Bacteria, Steroids, and Colon Cancer in Two Scandinavia Populations," Lancet, 2: 207-211 (1977).

13. M. Jain, G. M. Cook. F. G. Davish, M. G. Grace, G. R. Howe, and A. B. Miller, "A Case-control Study of Diet and Colo-rectal Cancer," Int. J. Cancer, 26: 757-768 (1980).

14. O. M. Jensen, "Cancer Risk among Danish Male Seventh Day Adventists and Other Temperance Society Members,"J.N.C.I., 70: 1011-1014 (1983).

15. O. M. Jensen and R. MacLennan, "Dietary Factors and Colorectal Cancer in Scandinavia," Isr. J. Med. Sci., 15: 329 334 (1979).

16. O. M. Jensen, R. MacLennan, and J. Wahrendort, "Diet, Bowel Function, Fecal Characteristics, and Large Bowel Cancer in Denmark and Finland." Nutr. Cancer, 4: 5-19 (1982).

17. L. J. Kinlen, "Meat and Fat Consumption and Cancer Mortality: A Study of Strict Religious Orders in Britain," Lancet, 1: 946-949 (1982).

18. L. K. Kolonel, J. H. Hankin, A. M. Nomura, and S. Y. Chu, "Dietary Fat Intake and Cancer Incidence among Five Ethnic Groups in Hawaii," Cancer Res., 41: 3727-3728 (1981).

19. P. A. Kurup, N. Jayakumari, M. Indira, et al., "Diet, Nutrition Intake, and Metabolism in Population at High and Low Risk for Colon Cancer: Composition, Intake, and Excretion of Fiber Constituents," A. J. C.N., 40: 942-946 (1984).

20. K. Liu, J. Stamler, D. Moss, D. Garside, V. Persky, and L. Soltero, "Dietary Cholesterol, Fat, and Fiber, and Colon-cancer Mortality," Lancet, 2: 782-785 (1979).

21. J. L. Lyon and A. W. Sorenson, ' Colon Cancer in a Low-risk Population," A.J.C.N., 31: s227-s230 (1978).

22. R. MacLennan, O. M. Jensen, J. Mosbech, and H. Vuori, "Diet, Transit Time, Stool Weight, and Colon Cancer in Two Scandinavian Populations, A.l.C.N., 31: s239-s242 (1978).

23. S. L. Malhotra, "Dietary Factors in a Study of Colon Cancer from Cancer Registry," Med. Hypotheses, 3: 122-126 (1977).

24. O. Manousos, N. E. Day, D. Trichopoulos, F. Gerovassilis, A. Tzonou, and A. Polychronopoulai, "Diet and Colorectal Cancer: A Case-control Study in Greece," Int. J. Cancer, 32: 1-5(1983).

25. R. A. McCance and E. M. Widdowson, eds., The Composition of Foods, 4th ed. (HMSO, London, 1978).

26. A. J. McMichael, M. B. McCall, J. M. Hartshorne, and T. L. Wooding, "Patterns of Gastrointestinal Cancer in European Migrants to Australia: The Role of Dietary Change," Int. J. Cancer, 25: 431-437 (1980).

27. A. J. McMichael, J. D. Potter, and B. S. Hetzel, "Time Trends in Colo-rectal Cancer Mortality in relation to Food and Alcohol Consumption: US. United Kingdom, Australia, and New Zealand," Int. J. Epidemiol., 8: 295-303 (1979).

28. A. B. Miller, G. R. Howe, M. Jain, K. L. P. Craib, and L. Harrison, Food Items and Food Groups as Risk Factors in a Case-control Study of Diet and Colo-rectal Cancer," Int. J. Cancer, 32: 155-161 (1983).

29. B. Modan, "Patterns of Gastrointestinal Neoplasms in Israel," ls. J. Med. Sci., 15: 301-304 (1979).

30. L. W. Pickle, M. H. Greene, R. G. Ziegler, A. Toledo, R. Hoover, H. T. Lynch, and R. F. Fraumeni, Jr., "Colorectal Cancer in Rural Nebraska," Cancer Res., 44: 363 369 (1984)

31. J. W. Powles and D. R. R. Williams, "Trends in Bowel Cancer in Selected Countries in relation to Wartime Changes in Flour Milling," Nutr. Cancer, 6: 40-48 (1984).

32. R. W. Rawson, "The Total Environment in the Epidemiology of Neoplastic Disease: The Obvious "Ain't Necessarily So," Cancer Incidence in Defined Populations, Banbury Report, no. 4 (Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y., 1980), pp. 109-119.

33. B. S. Reddy, G. Ekelund, M. Bohe, A. Engle, and L. Domellot, "Metabolic Epidemiology of Colon Cancer: Dietary Pattern and Fecal Sterol Concentrations of Three Populations," Nutr. Cancer, 5: 3440 (1983).

34. R. L. Rizek, B. P. Perloff, and L. P. Posati. "USDA's Nutrient Data Bank," Food Tech. in Australia, 33: 3 (1981).

35. P. Rozen, S. M. Hellerstein, and C. Horwitz, "The Low Incidence of Colorectal Cancer in a High-risk Population," Cancer, 48: 2692-2695 (1981).

36. G. N. Stemmermann, A. M. Y. Nomura, and L. K. Heilbrum, 'Dietary Fat and the Risk of Colorectal Cancer," Cancer Res., 44: 4633 4637 (1984).

37. A. R. P. Walker, "The Relationship between Bowel Cancer and Fiber Content in the Diet," A.J.C.N., 31: s248-s251 (1978).

38. B. K. Watt, "Tables of Food Composition: Uses and Limitations," Contemp. Nutr., 5: 2 ( 1 980).

39. B. K. Watt and A. L. Merrill, Composition of Foods: Raw, Processed, Prepared (USDA, Washington, D.C., 1963).

40. R. Zaldivar, W. H. Wetterstrand, and G. L. Ghai, "Relative Frequency of Mammary, Colonic, Rectal, and Pancreatic Cancer in a Large Autopsy Series," Zb/. Bakt. B, 169: 474-481 (1979)

Epidemiological uses of food composition data in the european context

Introduction
Nutritional epidemiology
The problems
Suggestions for improvement
Ongoing activities
Summary
References

LENORE ARAB
Klinisches Institut fur Herzinfarktforschung, Heidelberg, Federal Republic of Germany

Introduction

Instead of presenting a long list of users and potential users of food composition data in Europe, and an even longer list of the needs that are not yet fulfilled in that field, this paper focuses on one specific user group and their problems - the nutritional epidemiologists. A close look at the difficulties faced by persons conducting international nutritional epidemiology sheds light on many of the European and non European problems in the use of food composition data.

Relevant to the broader context of users and needs of food composition data in Europe is the collection of issues presented at the workshop entitled "Towards Compatibility of Food Nutrient Composition Data Bases in Europe" |22]. At this meeting, held in Wageningen, the Netherlands, EUROFOODS was founded and work committees established to direct their efforts towards solving European food problems in five areas. Responses to a questionnaire circulated in conjunction with this meeting revealed the breadth and depth of food-nutrient data bases in Europe, and their limitations [3]. A recent update on the number of food entries in the systems of various European countries can be found in figure 1.

Nutritional epidemiology

In the attempt to determine and quantify nutritional factors related to the onset of disease in the human population. epidemiologic analyses are being applied in cross sectional, case-control, and cohort studies (see chap. 6). Relationships between the eating behaviour of individuals and the onset or prevalence of risk factors and chronic diseases are being examined and quantified. Conducting such studies internationally, by broadening the range of observation, enhances the ability to identify risk differentials and thereby strengthens the conclusions drawn from population studies.

Nutritional epidemiology incorporates the attempt to collect, systematically and in an unbiased manner, information on the eating behaviour of individuals and groups for use in illuminating or testing correlates and causality of disease.

The problems

The difficulties of nutritional epidemiology as they relate to the nutritional side begin with problems in dietary assessment methodology [9] (see chap. 10). This includes the questions asked, how they are asked, who asks them, built-in validation attempts, the degree of specificity strived for, the forms, booklets, and guidelines used, the conditions surrounding the information exchange, and motivational measures, among other factors. In addition to these difficulties, quantification of amounts [8,13,21], translational difficulties, coding of consumed foods into computer-readable form [2], differences between tables [17] and between foods, and what to do about missing values present trying and complex obstacles. Some of these issues are related directly, others indirectly, to food composition tables.

These problems are more clearly seen in perspective against the background of a typical dietary assessment. Taking, for example, the often-used 24-hour recall method to compare intakes in high-risk individuals and a control group, the following steps, listed with their possible shortcomings, are necessary:

1. Question the individual about what was eaten the previous day. (Memory failure; interviewer-related bias.)

2. Question the individual about the amounts of the foods eaten. (Memory failure; estimation difficulties; non-edible portions - bones, pits, skin. )

3. Code foods into machine-readable form, usually numbers. (Few codes leading to compromised information; errors in reading or writing codes.)

4. Convert portion sizes into gram amounts. (Plate waste, refuse deduction, portion size calculation from recipes.)

5. Enter subject identification, date, meals, foods, and amounts. (Transcription errors. )

6. Check entered data for correctness. (Oversight; difficult forms; no printout of food names. )

7. Correct the errors. (Renewed typing errors.)

8. Recheck the entered corrections (Oversight or elimination of this step altogether.)

9. Merge this information into a common data base with food-nutrient information and calculate the average nutrient intake for the day. (Non-standard algorithms; program errors.)

10. Group foods for comparison of frequency and amounts consumed of basic food groups between individuals. (Double counting; mistakes in grouping.)

11. Compare intake of nutrients between case and control groups. (Missing values in nutrient tables resulting in artifactual differences.)

12. Test for significant differences between groups. (Invalid methods selected.)

Every step in this process is fraught with potential errors. Questioning about previous intake varies from interviewer to interviewer [16] unless, as is done in some cases, the subject is asked to reconstruct on paper his meals of the previous day [1] or a computer program prompts questions [6]. The subject may inadvertently adjust responses to cues from the interviewer. Subjectivity in recording responses is also a danger. Estimation of the amount eaten is subject to memory failure [16], estimation difficulties, and misquantification or incorrect subtraction of non-edible portions [15]. The coding of foods usually involves a compromising of the available information to fit the length and breadth of the coding system used (see chap. 12). The coding process itself, unless computerized [18], generally involves searching for the correct number from code books, transcribing it onto paper, and having these numbers entered into a terminal. This process allows for many possible reading, transcribing, and keyboarding errors. Different individuals may code the same information differently [9].

Fig. 1. Number of foods on which food composition data is available in Europe (it should be noted that nutrient information is often borrowed; the sum is therefore not representative of the total amount of original data).

The calculation of food intakes into nutrients depends on the availability of information on the foods as consumed (cooked, fried, etc.) for the nutrients of interest. Missing values in food tables are often handled interchangeably as zero values for these nutrients, which can result in false or artifactual results [4]. Inaccuracies in the calculation of nutrients from foods can also result from faulty programming, although this is seldom a major difficulty. Differences in estimating nutrient intakes have been documented between systems with identical sources of nutrient information, for a number of reasons [11].

Regrouping the tens of thousands of food items on the market in a particular country is also generally non-standardized. Comparisons are therefore subject to differences in the systems used [16]. An example of this is butter being included in either the dairy group or the fats and oils group, and egg-rich products such as quiches and egg noodles being grouped together with cereal and grain products.

Mention should also be made of the misuse of methods for the testing of hypotheses as a major problem in nutritional epidemiology. For example, the results of a single 24-hour recall or a frequency questionnaire of food consumption are often used to determine interrelationships and the interpretation made that no underlying relationship exists. Strong interactions may be statistically insignificant due to low subject number or days of observation and great intraindividual day-to-day variability [5, 7].

Suggestions for improvement

Since the use of food composition tables in nutritional epidemiology concerns mainly estimation of the composition of food consumed by individuals, it makes sense to regard the entire process as closely interrelated- the dietary assessment, food coding, and referencing of tables. Some of the problems beginning with the assessment of dietary behaviour are: (1) non-standardized methodology; (2) portion size estimation; (3) translation; (4) mixed dishes; (5) different food codes; (6) different food grouping systems; (7) differences in food tables; (8) food compositional differences; (9) missing values; and (10) definitions of edible parts.

In many cases only minimal effort would be required to improve and simplify nutritional analysis in international epidemiological studies. The components of food-nutrient data-base systems (data files and programs) are compartmentalized and presented diagramatically in figure 2, emphasizing the individual areas needing enhanced development as well as their interdependence. The numbers in the diagram link relevant system components to numbered suggestions in the following text.

1. Standardized Dietary Assessment Methods

Although basically common terminology is used in dietary assessment, the procedures followed are non-standardized and seldom described or documented. A 24-hour recall for one person may be a computer-based decision tree [6], for another it might be an empty piece of paper, a pencil, and a simple question on the subject "Please write down what you ate yesterday." Handbooks on standardized procedures for different dietary assessment methods would improve the situation.

2. Standardized Portion Sizes

All subjects have difficulties in estimating the exact portion sizes of foods consumed (particularly retrospectively), requiring estimations and assumptions to be made by the interviewer or coder. The difference between weight of purchased product (often known) and that of actual item consumed (without bone, pits, skin, with water, etc.) are all part of this issue. The use of standardized portion sizes for individual foods within countries could help. The application of standard weights and density measures where applicable, as well as common models for small, medium, and large portions, cuts of meat, portions of grains, noodles, and baked products, would be useful. A standard set of food models for use in dietary assessment could be a step in the right direction.

Fig. 2. Components of an extensive dietary assessment system.

Table 1. Native-language food-table entries for apples, butter, beer, and beef

3. Names and Translations

The difficulty of translating foods from one language to another cannot be underestimated. Even within a country, colloquial food names are inexact and can be misleading. Table 1 gives a comparison of the differences between countries in the names of simple foods in food tables. Here apples, butter, beer, and beef, as entered in the food tables of IO countries, are compared. Apples and butter between countries are somewhat comparable. Beer, however, brings in the complication of different products with different formulations and alcohol content. And beef is even more troublesome, as different countries have different standards for cutting and names for the various cuts, different fat contents and extent of trimming. A considerable part of an international coding workshop held in Heidelberg in February 1985 was devoted to the description of food items. (Unpublished proceedings available from the author.) Table 2 presents a list of many of the published food nutrient tables currently available in Europe. To use these tables one needs not only a good translation of the food items and measured parameters, but also an understanding of the language to read the introduction in which critical information on using the tables is found. A common pitfall in translating foods is in situations where an identical or similar food name is used in two countries but an entirely different product is meant. Another problem is that of foods whose names convey little meaning as to content, such as those found in table 3.

4. Mixed Dishes

People often eat dishes that are prepared elsewhere and the identity of whose ingredients remains unknown to them. Salads, stews, soups, cakes, casseroles - the list of foods in which one cannot determine what has been added just by looking is practically endless. Mixed dishes in general present problems, as hundreds of different recipe variations are involved for each dish of a given name. They are such an important part of European diets, however, that they cannot be ignored. A simple example is pizza, which is known everywhere by this name but is prepared in a myriad of ways - in some areas it is made without tomatoes, in others without cheese, and when cheese is used sometimes it is mozarella, sometimes Gouda, sometimes emmentaler. Standard recipes would at least prevent arbitrary differences and allow for calculations on a common basis when exact information is not available.

Table 2. Major European food tables

Table 3. Some mixed dishes translated into English

Despite the fact that the large proportion of the European diet comes from mixed dishes, relatively few prepared dishes have been analysed and presented in composition tables [3]. There are different methods for imputing the nutrient values of food that have undergone various preservation and preparation processes, none of which has been shown to be the most accurate [14]. Recommended procedures for imputing values would at least enhance the comparability of results.

5. A Standard Code

Different coding systems with different orientations, different groupings of food, and different levels of specificity [2] stand in the way of data exchange, comparison of results, and recalculation of surveys based on foreign tables. A common coding system would allow comparisons between studies of the frequencies and amounts of individual foods consumed. Also, it would help bypass the translational problems in identification of the same and similar foods. It would also simplify the application of common grouping schemes for foods.

6. Basic Food Groups

The problem of grouping foods and products has been addressed earlier. Since tens of thousands of food items are consumed in Western populations, it is necessary to aggregate individual foods into groups for further analyses. Brand-name items usually fall together, but egg dishes, vegetable casseroles, and stews are foods that typically differ from one grouping scheme to another. Different schemes for grouping foods can lead to antifactual differences or false conclusions regarding epidemiologic associations. Agreement on at least the major food groups and guidelines for alternate grouping schemes would be useful.

7. Food-table Format

Differences between different food tables are apparent as soon as one opens two different tables. The format with which foods and nutrients are listed differs, some tables have one food per page, others lists of foods and lists of nutrients. The ordering of the nutrients differs, as do the units of measure, the exactness, and the number of significant digits. There are also differences in the factors used for calculation [4]. Again, the need for a standardized format is demonstrated here (see chap. 10).

8. Food-nutrient Differences

The nutrient composition of an apple or potato in the FRG is not likely to be the same as in Holland, and is even less likely to be the same as in Sweden or in Spain. First of all, different species are used; second, different amounts of sunshine influence the nutrient level; third, the trace elements are strongly influenced by the soil. The breadth of variation may be as great within a country as between countries. The fat content of similar foods differs tremendously from country to country. Little can be done here in general, but this problem discourages the exchange of data, particularly important when large amounts of the products are being exchanged within the European Common Market or eaten on vacation outside the country. Data exchange requires some means for the user to find the proper food in a language he can understand (source and sampling information) and a presentation of the values whose analytical method and accuracy are available to him. Since variability of values within a table can be considerable, information is needed to enable the end-user to know the laboratory intra-sample variation and variation between species, as well as the origin of the food sampled and the time of analysis.

9. Missing Values

There are many types of missing values in nutrient tables. As Greenfield and Southgate have pointed out, there are imputed values, borrowed values, values calculated from a measured factor, values that are available for many foods but not available for others (absent), and nutrients that are not included at all [19]. For the epidemiologist the values which are labelled missing bring further analyses to a halt. Zeros in the analyses make the determination of a gradient of risk impossible - an estimation is almost always preferable. Greater exchange of nutrient information between countries would allow for the eradication of many of the missing values in tables. Efforts to correlate information on nutrient losses and gains in preparation of foods would help improve the quality of estimation of nutrient contents in preferred foods which have not been analysed.

10. Edible Parts

So-called "edible parts" of foods can differ from culture to culture. The French for example consume only the soft inner portion of Camembert cheese. Germans eat the crust as well, and would not expect this to be missing in the nutrient analyses. Descriptions of the food sample analysed would help prevent misuse of foreign data.

This list of possible problems with food composition data is not exhaustive. Additional suggestions for improvement would include increased availability of data [3] in machinereadable form and a solution for updating data more easily and less expensively.

Contents - Previous - Next