Culture-specific food frequency questionaires: Development for use in a cardiovascular study

Abstract/Resume

The object of this work was to develop self-administered, culture-specific food frequency questionnaires (FFQs) for use in a prevalence study of cardiovascular disease. The cultures included the general Canadian population, south Asian Canadians and Chinese Canadians and the FFQs were based on the structure of our previously reported FFQ used in the Canadian Diet, Lifestyle, and Health Study. Food record data for each of the culture groups were available from previous studies. A database was used to identify food items and serving sizes to be included in each questionnaire. The foods contributing most of the 19 food components of the general Canadian sample were calculated. This article describes the methodology for the initial development of the FFQs and discusses the initial evaluation of the general Canadian questionnaire. Validation of all three questionnaires against seven-day food records is in progress.

(Can J Diet Prac Res 1999; 60:27-36)

Cette recherche avait pour but d'elaborer des questionnaires de frequence de consommation (QFC) a remplir par le repondant et tenant compte des aspects culturels. Les questionnaires seront utilises dans une etude de prevalence des maladies cardiovasculaires. Les groupes culturels representes etaient la population canadienne en general, les Canadiens d'origine sud-asiatique et les Chinois canadiens; les QFC etaient bases sur la structure des QFC utilises ant'rieurement dans l'etude sur l'alimentation, le mode de vie et la sante des Canadiens. Les donnees des releves alimentaires de chacun des groupes culturels ont ete tirees d'etudes anterieures. Une base de donnees a ete utilisee pour repertorier les aliments et les grosseurs de portions a inclure dans chaque questionnaire. Les aliments qui contribuent le plus aux 19 composantes alimentaires de l'echantillon general de Canadiens ont ete calcules. Cet article decrit la methodologie d'elaboration initiale du QFC et traite de l'evaluation initiale du questionnaire destine a la population canadienne en general. La validation des trois questionnaires par rapport a des releves alimentaires de sept jours est en cours. (Rev can prat rech dietet 1999; 60:27-36)

INTRODUCTION

The culturally diverse population of Canada provides a unique opportunity to study the role of dietary factors in dietdisease relationships. Despite living in the same environment, multiethnic populations have demonstrated a wide range in incidence of diseases like cancer and heart disease; variations in dietary practices offer a likely explanation for this diversity (1). The development of appropriate methodology for culture-sensitive dietary assessment is essential for identifying the role of diet in the etiology of chronic diseases (2, 3). The primary objective of dietary assessment in most epidemiologic studies is to estimate the individual's usual intake of foods and dietary components over a long period. Whether in a large population study or a small clinical setting, assessment and evaluation of usual dietary intake is a challenge. A food frequency questionnaire (FFQ) is generally preferred over food records and other diet assessment methods, because of its ability to describe usual long-term diet and for its ease of administration (4). However, the FFQs developed for one population are not necessarily suitable for another population, since the types of foods and the portions vary from population to population (3). This paper describes the principles and procedures applied in developing three ethnic-specific FFQs for a prevalence study of cardiovascular disease (SHARE Study of Heart Assessment and Risk in Ethnic Groups) that may be used to study other populations. The three selfadministered, population-specific FFQs developed are: the general Canadian, the south Asian, and the Chinese FFQ.

METHODS AND RESULTS

The foundation for these culture-specific FFQs was the structure of our previously developed FFQ, reported by Jain et al. (5) in a Toronto Diet Validation Study (TDV Study), and further expanded for use in the Canadian Diet, Lifestyle, and Health Study (DLH Study) cohort. The FFQs for each culture group were then developed from food record data in the TDV Study and from the pilot studies with the different culture groups.

Development of the food item list

This is the first step in compiling an FFQ. The questionnaires should measure the individual's habitual consumption of all main food items during a specified time. The estimates of specific types of foods consumed, average frequency of consumption and usual portion sizes should be sufficiently accurate and detailed to allow calculation of total energy intake, minimum nutrient intake and intake from major food groups. To allow for all of the potential dietary risk/preventive factors for various diseases (cancer, heart disease), many different aspects of diet had to be considered. For example, attention had to be given to different sources of fats, given the hypotheses for the importance of types of fats for heart disease and cancer as well as some non-nutritive substances e.g. components of allium family, cruciferous vegetables, phytoestrogens, quercetin etc. Given the importance of various plant food constituents and their possible preventive activity, accurate measurement of consumption of different types of fruits, vegetables, cereals, meats, fish, and refined carbohydrates was felt to be important. The inferences would focus on the relation between diet and disease, after adjustment for energy intake.

Instead of an ad hoc list of food items, we created a database (6) using food records from three studies:

seven-day food records from 208 participants in the TDV study (1989-91) for the general Canadian population (5),

four-day food records from the 51 participants of south Asian origin in a pilot study (July-August, 1995) and

four-day food records from the 26 participants of Chinese origin in a pilot study (SHARE) in southern Ontario (October-November, 1996).

One global FFQ may not serve for the various ethnic populations in the study. The choice of a single multi-ethnic FFQ versus separate FFQs depends on the degree of overlap in food items between the communities. Communities adhering to traditional food patterns, especially older people and first generation immigrants, exhibit less overlap, and this pattern was observed for the population in this study (data shown later). Since the development principles applied were similar for all three populations, the FFQ for the general Canadian population is described in this paper in greater detail than are the other two FFQs.

Food list for the general Canadian population

The main contributing food items for the general Canadian FFQ were selected from the seven-day food records from the TDV Study conducted on a random, population-based sample of 208 participants (98 men and 110 women) in Toronto, Canada, between May 1989 and July 1991 (5). Although this FFQ was essentially structured on the FFQ we previously described (5) and used for our DLH study, we used this opportunity to make changes where necessary in the existing list and to enhance our confidence in the new food list. We have used the term 'Canadian' for the general Canadian FFQ to distinguish it from other culture groups that we studied; however, it is at best representative of the Toronto population only. The majority of this group was of British or European descent, and less than 3 % were of south Asian or Chinese origin. Of the 633 potentially eligible subjects, 216 (34%) were initially interviewed for the TDV Study, when they were given detailed instructions in their homes on how to keep the seven-day food records that were later collected by interviewers. Participants were not required to weigh foods but were asked to measure the volume of serving dishes to help assess portion sizes.

The information on food records was coded and the nutrients were calculated using a nutrient database of over 3,000 foods at the Epidemiology Unit of the National Cancer Institute of Canada (NCIC). This database was developed from Agriculture Handbook No. 8 (AH-8) food composition tables (7) expanded for Canadian foods to code the TDV Study (5). AH-8 was used for this analysis since the Canadian Nutrient File (8) was under revision during the TDV Study in 1989, and we were using the NCIC database for most of our diet-disease studies in Canada. The impact of differences between the Canadian Nutrient File and the AH-8 is difficult to assess (9) as there is no published report on it. For the purpose of this work, it may not be very substantial since the NCIC nutrient database more closely reflects the Canadian values than those in the AH-8. Nutrient calculations were performed with a personally developed software package.

A frequency tabulation was computed using Statistical Analysis Software to identify the most commonly occurring items. For the general Canadian group, the data were handled as a single data set (seven days for 208 subjects = 1,434 days of records). In that set, 956 uniquely coded food items were identified (total frequency = 11,206). Intakes of energy and 18 nutrients (protein, total fat, carbohydrates, saturated fat, oleic acid, linoleic acid, cholesterol, dietary fibre, calcium, iron, simple sugars, thiamin, riboflavin, niacin, vitamin A, vitamin C, vitamin E, and beta-carotene) from each of these items were calculated and expressed as percentages of the respective total intakes for the group. The nutrient contribution from each unique food code (frequency times amount for all subjects combined) for each of the 18 nutrients and the energy intake was ranked to detect the importance of each food for its inclusion as a separate food or a grouped item. (For example, of all the energy consumed by all subjects, 4% was from 2% milk, so it was listed separately). Of the 956 foods, we were able to collapse 833 foods into 158 food groups according to the following criteria:

conceptual similarity (for example,14 citrus fruit juices were combined as one item, "orange, grapefruit juice"),

respondent's ability to make the necessary connection (for example, `ricotta cheese' as a cottage cheese item),

similarity in nutrient content per usual serving (10), and

importance of a particular food or nutrient to the dietdisease hypotheses.

The 122 remaining items (and water) resulted in a loss of 592 frequencies (14 of the 158 foods or groups were initially computed as part of another food item/group and later separated for cognitive reasons). These 833 foods accounted for 96.7% of total energy, 95.6% of total protein, and over 94% of daily intake of all other 17 nutrients for the population. Appendix 1 presents the food list for the general Canadian group including the 158 items and the energy contribution (percent) of each of the food or food groups to the population's daily intake. It also gives the rank of that particular item in its contribution to the total population's intake (data on other nutrients are available from the author on request). These may serve as guidelines for others interested in developing their own instruments. A similar list is also available for the other 19 nutrients and each of the 956 items, but its inclusion was beyond the scope of this paper.

Evaluation of the adequacy of the food list

To check the internal validity of the food list, a nutrient composition database was developed for each of the 158 food items or groups, adjusting for both the frequency and the amount in grams of each item within a group. Nutrient values were computed per 100 grams of the food. For example, if orange juice was reported 24 times, giving a total amount of 13,942 grams, and grapefruit juice was reported 16 times, for a total amount of 9,564 grams, the weight of the two foods was used to calculate the weighted nutrient composition per 100 grams of the grouped item "orange, grapefruit juice". The food records of 203 people (95 men and 108 women, mean age: 60.29.7, range: 28 to 75 years) from the original sample of 208 were recalculated using this nutrient composition database. Five subjects were deleted because of inadequate data.

The average intakes of energy and 18 selected nutrients were compared to the intakes calculated from food records (Table 1). The two sets of intake information were compared using paired t-tests and Pearson's correlation coefficient. There were no significant differences in the two tabulations, suggesting no significant loss of information by pooling selected items. The correlations were excellent (all 0.94 or greater) and the only nutrient that differed between the original food records and the recalculated records was saturated fat. However, categorical analysis by quartiles showed a 66% exact agreement for saturated fat and no misclassification into extreme categories. It appears that the recalculated FFQ is able to capture a greater proportion of total fat unaccounted for by saturated fat, oleic and linoleic acid.

Since the intent of this exercise was to make the shorter-list FFQ representative of the food records, the high correlations obtained were not unexpected. This computation did not take into account the portion sizes as listed on the FFQ. The amounts originally reported on the food records were used and it is therefore impossible to verify the portion size attributions for each food item. Some of the dietary components of interest in relation to the study, e.g. food items and food groups, allium family, phytoestrogens, have not been reported here and will be calculated from foods and food groups later.

Food items list for the south Asian and the Chinese FFQ

The initial lists of foods were derived from four-day food records obtained in separate studies from 51 Canadians of south Asian origin (from India, Pakistan, Sri Lanka and Bangla Desh) and 26 Canadians of Chinese origin (from mainland China, Hong Kong, or Taiwan). The subjects were randomly selected from a Hamilton area list of surnames of south Asian or Chinese origin. They had lived in Canada for at least five years. They were contacted by mail and then telephone; subjects who agreed to take part were invited to a clinic for various tests, including a 24hour diet recall. Of the contacted subjects, approximately 25% were recruited for the studies. They were given a food record at this time to take home, to be returned to the centre by mail. The south Asian subjects were interviewed by researchers of the same community, the Chinese by a non-Chinese nutritionist. The food records were coded, giving a unique code to every new item or recipe not existing on the FFQ in use for the DLH Study (5). A total of 221 unique foods (2,083 frequencies) were reported on the south Asian FFQ and 284 foods (2,176 frequencies) on the Chinese FFQ (131 and 145 with frequencies over two, respectively). As far as possible, foods similar to those in the general Canadian FFQ list were combined and retained on the list.

No nutrient calculations were performed on these data at this stage because no suitable foodbank was available and the number of subjects per study was small. Almost all foods could be grouped into a short list with no appreciable loss of frequently reported foods. Comprehensive lists were compiled after consultations with people of the community and dietitians of the culture group. A cut-off point of frequency greater than two, together with regrouping and addition of items considered missing by experts, resulted in 163 items on the south Asian FFQ and 169 items on the Chinese FFQ. Items that occurred less frequently than twice were either combined with other similar foods or deleted. Although the ultimate aim was also to identify the major contributors of various nutrients, it was felt that items occurring less often than twice for the whole sample were not the major contributors for any particular nutrient. An open-ended section at the end of the list prompts the respondent to indicate other frequently eaten foods, their frequency of consumption and amount. These items will be given unique food codes and handled as the main study may determine.

The number of food items common on the general Canadian list and the south Asian FFQ was 109; for the Chinese FFQ, it was 119. Another 20 foods items from the Canadian FFQ were grouped as 10 food items on the south Asian FFQ, and 21 items from the Canadian FFQ were grouped into 10 Chinese FFQ items. Thus the south Asian FFQ had 44 food items/groups unique to that group, and the Chinese FFQ had 40 such items. Table 2 lists unique foods that were incorporated with minor modifications in the two community-specific FFQs. In general, beverages, fruits and desserts were common on all three FFQs. Major differences were noted in the cooking practices of vegetables and meats, the frequency of consumption of various foods by different groups, and the associations necessary for cognitive recognition.

Frequency of food consumption

There is some controversy about the need to estimate actual frequency of intake vs. simple categorization of intakes (e.g. once a day, four or five times a week, etc.) (11,12). However, studies have shown that the main determinant of variation in measured dietary intakes is frequency of consumption of the individual food items (13). Therefore, the three questionnaires developed in the current study asked respondents to estimate frequency of food consumption.

Portion size

It was decided to use `semi-quantitative' FFQs, with questions not only about the frequency of consumption of different food items but also the habitual portion sizes. Many traditional FFQs ask only about frequency of consumption. Recent studies, however, have shown that accuracy could be slightly improved by also asking questions about the habitual portion sizes (14). For the Canadian FFQ, based on the seven-day food record, the average and range of food serving sizes were computed for each of the groups to identify a standard reference serving size for each food on the FFQ. Examination of the average amount per food item and its range reported on the food records for this work was used only as a guide. The computed averages were unworkable for a number of items (for example, six bottles of a 360-ml beer were coded as 2160 ml for one subject). All reported portions were therefore manually examined to obtain frequently reported portions and the most frequently reported for that mode of distribution was used.

In the culture-specific FFQs, the 'average' portion size was specified and respondents were asked to indicate whether their usual portion was smaller than average, average, or greater than average.

To increase the accuracy of portion size estimates, photographs of some foods in average (medium), less than average (small), and greater than average (large) amounts were printed on the questionnaires. The dishes were cooked, measured (for volume) and weighed before being photographed. The portions were then discussed with members of the various ethnic groups for their representativeness.

Precoding and scanner-readable format

The FFQs can be used with any ordinary data entry software. To save costs and time in the long run, a precoded and scanner-readable format was also developed The scanner format can be read by a Datafax system that creates an ASCII file.

Pilot testing of the FFQs

Colleagues and the general population were asked to complete the south Asian and Chinese FFQs (n=5 for each) and comment, mentioning any commonly occurring items that were not listed. This occurred for only two items for the Chinese population. Weaknesses in questionnaires were evaluated by simple descriptive comments from narticinants about ambiguity of questions, omission of frequently consumed foods, portion sizes, etc. For the General Canadian FFQ, this testing was done with an FFQ similar to that used in the DLH study.

Validity and calibration studies

The principle aim of developing the FFQs was that they should provide the best possible ranking of individuals by their habitual intake level of foods and nutrients. This ranking capacity is being evaluated by comparing the FFQs with seven-day food records and will be reported in the future. The seven-day food records may not be an ideal `gold standard' since intakes may be altered during the record period, and they may not represent intakes over the entire year. The food banks for the south Asian and Chinese nutrient calculations are being compiled based on various sources, including reported recipes and the literature.

DISCUSSION

The purpose of this paper has been to describe the methodological approach, which may be adapted to develop population-specific dietary questionnaires, whether for population studies or for clinical research. These procedures have been used in various forms by other investigators (6,10,15). Use of these methods resulted in FFQs that might be more suitable for the general Canadian population as well as Canadians of south Asian and Chinese origin. The results presented here are a preliminary evaluation of the FFQ; a more formal assessment of calibration, validity and reliability is currently underway. The results of nutrient intakes computed from the 158-item general Canadian FFQ (Table 1) support the adequacy of the food list and the nutrient content assumptions. The nutrient intake values in our study were generally comparable to intakes reported for participants in the Ontario Health Survey (Table 1) (16-18). The approximately 10% lower values in our study may be due to an actual difference in intake, since the mean age of our study population was much higher than that of the Ontario Health Survey population (61 years vs. approximately 41 years). The general Canadian FFQ described here is limited in its generalizability since it was based on a sample of a Toronto population only, aged 28-75 years.

Participation rates of 33% for the Toronto population and approximately 25% for other groups included in the study are likely inadequate for the data to be sufficiently representative of the populations studied. The main reasons for refusing to participate were a lack of time or interest. One of the limitations of FFQs developed on the basis of food records could be the seasonal effect of the food record collection period. It was not a limitation for the general Canadian FFQ described here, since the food records were collected over two years. It could, however, be a limitation for the other two community-specific FFQs although addition of food items perceived to be important after consultations with community members has likely corrected for these problems.

The process undertaken here suggests that it is feasible to compute culture-specific FFQs for population, health or clinical studies desiring to capture population-specific intakes. A U.S. study found that correlations between food records and a self-administered FFQ were lower among blacks than among whites and lower among women with fewer years of education (19). These differences between ethnic groups may have been due to inadequate representation of food items for blacks on their FFQs. It will be important to determine the necessity of these culture-specific questionnaires when the nutrient data are available from the south Asian and the Chinese questionnaires, i.e., how much information would have been lost by calculating nutrients from only those items common to the 'general' questionnaire. Work over the past decades indicates that there is an upper limit to the degree with which individuals' habitual dietary intakes can be measured, most studies showing correlations of 0.4 - 0.7 (4) in dietary validation studies of FFQs. This implies that calibration and validation substudies will generally be needed to correct for measurement errors in studies estimating relative risks of disease. In clinical situations or surveillance studies, however, where absolute intakes are important, calibration sub-studies may be used to achieve standardisation between clinics and populations, i.e. the questionnaire measurements can be adjusted for betweencentre differences in systematic over- or underestimation at the group level (20).

RELEVANCE TO PRACTICE

Whether in a large population study or a small clinical setting, assessment and evaluation of usual dietary intake is a challenging subject. This work describes the various steps that may be adapted by dietitians in research and practice to develop dietary assessment tools for their study population.

Acknowledgements This research was partly supported by funds from the National Cancer Institute of Canada. The author wishes to thank Malcolm Koo, Department of Public Health Sciences, University of Toronto, and several members of the community for their advice during this process.

[Reference]

References

[Reference]

1. American Institute for Cancer Research. 1.1 Patterns of Diet and 1.2 Patterns of Cancer. In: Food, Nutrition and the Prevention of Cancer: a global perspective. American Institute of Cancer Research, Washington, 1997: 22-52.

[Reference]

2. Cassidy CM. Walk a mile in my shoes: culturally sensitive food-habit research. Am J Clin Nutr 1994;59(suppl):190s-197s. 3. Buzzard IM, Sievert YA. Research priorities and recommendations for dietary assessment methodology. Am J Clin Nutr 1994;59 (suppl) :275s-280s.

[Reference]

4. Willett WC. Future directions in the development of food-frequency questionnaires. Am J Clin Nutr 1994;59(suppl):171s-174s.

5. Jain M, Howe GR, Rohan T. Dietary assessment in epidemiology: Comparison of a food frequency and a diet history questionnaire with a seven-day food record. Am J Epidemiol 1996;143:953-960.

6. Howe GR, Harrison L, Jain M. A short diet history for assessing dietary exposure to N-nitrosamines in epidemiologic studies. Am J Epidemiol 1986;124:595-602.

7. Watt BK, Merrill AAL Composition of foods: Raw, Prepared. Washington DC: Agriculture Research Service, US Department of Agriculture, 1963, expansion March 1972. (USDA Handbook No. 8)

8. Canadian Nutrient File. Health Protection Branch, Health Canada, Ottawa, Ontario, Canada, 1997.

[Reference]

9. Verdier P. The Canadian Nutrient File: How Canadian are the data? J Can Diet Assoc 1987;48:21-23.

10. Block G, Hartman AM, Dresser CM, et al. A databased approach to diet questionnaire design and testing. Am J Epidemiol 1986;124:453-469.

11. Tylavsky FA, Sharp GB. Misclassification of nutrient and energy intake from use of closed-ended questions in epidemiologic research. Am J Epidemiol 1995;142:342-352.

12. Hunter DJ, Sampson Lt Stampfer MJ, et al. Variability in portion sizes of commonly consumed foods among a population of women in the United States. Am J Epidemiol 1988;127:1240-1249.

13. Willett W. Food frequency methods. In: Willett W. Nutritional Epidemiology. New York: Oxford University press, 1990: 69-91.

14. Haraldsdottir J, Tjonneland A, Overvad K. Validity of individual portion size estimates in a food frequency questionnaire. Int J Epidemiol 1994;23:787-796.

[Reference]

15. Bohlscheid-Thomas S, Hoting I, Boeing H, et al. Reproducibility and relative validity of food group intake in a food frequency questionnaire developed for the German part of the EPIC project. Int J Epidemiol 1997;26:s59-s70.

16. Bright-See E. Assessment of relative validity of the Ontario Health Survey Food Frequency Questionnaire. J Can Diet Assoc 1994;55:33-38.

17. Pomerleau J, Ostbye T, Bright-See E. Place of birth and dietary intake in Ontario. 1. Energy, cholesterol, carbohydrate, fiber, and alcohol. Prev Med 1998;27:32-40.

18. Pomerleau J, Ostbye T, Bright-See E. Place of birth and dietary intake in Ontario. II. Protein and selected micronutrients. Prev Med 1998; 27:32-40.

[Reference]

19. Kristal AR, Feng Z, Coates RJ, Oberman A, George V. Associations of race/ethnicity, education, and dietary intervention with the validity and reliability of a food frequency questionnaire. Am J Epidemiol 1997;146:856-869.

20. Kaaks R, Plummer M, Riboli E, et aL Adjustment for bias due to errors in exposure assessments in multi-center cohort studies on diet and cancer a calibration approach. Am J Clin Nutr 1994;49:254s-250s.

[Author Affiliation]

MEERA JAIN, PHD, Department of Public Health Sciences, Faculty of Medicine, University of Toronto