In epidemiologic studies of diet, investigators often want to be as comprehensive as possible in the data collection of food intake. Investigators often worry, however, that even a full-length dietary questionnaire will fail to capture foods that may be unique to an individual's or a subgroup's dietary habits. One way to address this concern has been to add an “open-ended” question that asks, for example, “Are there any other foods that you eat at least once a week?” This article examines the extent to which such a question contributes to nutrient estimates and rankings.
Multiethnic studies present additional difficulties. If a study includes both Chinese Americans and whites, for example, then Chinese foods may be added. Investigators may then be concerned that questions about these foods should also be asked of the whites, for comparability between the groups. Asking about such additional foods to the whites, however, adds to respondent burden, and may add logistic burden as well. Therefore, this article also examines to what extent asking ethnic-specific food items affects nutrient estimates and rankings among both whites and the relevant ethnic group.
The Study of Women's Health Across the Nation (SWAN) is a multicenter, multiethnic cohort recruited through community- and population-based sampling in 7 geographic areas.1 The 3302 participants were women age 42 to 52 years. Four of the centers (Boston, MA; Chicago, IL; Detroit, MI; and Pittsburgh, PA) enrolled only non-Hispanic white or black women. Three centers enrolled whites plus one other ethnic group: Hispanics in Newark, NJ, Chinese Americans in Oakland, CA, and Japanese Americans in Los Angeles, CA. Interviews were conducted in English or the relevant language, at the choice of the participant.
The study was begun in 1994, and the data presented here were collected in 1996—1997. The study was approved by the Institutional Review Boards at each site, and informed consent was obtained from study participants.
The development of the SWAN dietary questionnaire, a modification of the 1995 version of the Block food-frequency questionnaire (FFQ),2,3 has been described in detail elsewhere.1,4 The questionnaires were administered by in-person interview using 3-dimensional models for portion size quantitation. Major U.S. sources of phytoestrogens (including tofu, soy milk, and soy-based meat substitutes) are standard on Block FFQs. Some questions on the Block95 FFQ (ie, the “any other fruit” and “any other vegetable” line items) were omitted from the SWAN questionnaires to reduce the time required. The resulting core FFQ consisted of 103 food items asked of all 3302 respondents. In addition, all respondents were asked an open-ended “other foods page” (“Are there any other foods you eat at least once a week that we haven't mentioned?”).
For Hispanic respondents, 8 foods were added after examining nutrient contributors among Hispanics in national data.5,6 We identified 12 additional foods for Chinese respondents based on focus groups, and 16 additional foods for Japanese respondents were identified based on 24-hour recall data (Table 1). Women in those ethnic subgroups were asked about the core food list plus the additional ethnic foods. An “ethnic foods page” containing the additional ethnic-specific foods was also administered to the white respondents at the relevant sites (Hispanic foods at the Newark site, Chinese foods at the Oakland site, and Japanese foods at the Los Angeles site.) For those whites, the ethnic foods page was administered after the other foods page.
Thus, the total dietary assessment consisted of 3 components: the full core FFQ of 103 items, the open-ended other foods page, and the ethnic foods administered both to the relevant ethnic group and to whites at the same site. Analyses for this report are limited to the 3039 women whose questionnaires were determined to be reliable and who had completed the FFQ, the other foods page and, when applicable, an ethnic foods page (Table 2).
In this article, we investigate whether collecting the other foods page and the ethnic foods page data has an impact on nutrient estimates or ranking. We calculated Spearman correlations, quintile classification, and weighted kappa statistics to describe the relationship between nutrient estimates with and without the inclusion of the other foods page or ethnic foods page data. The absolute nutrient amounts con-tributed by the other foods page and ethnic foods page are described in terms of mean nutrient amount and percent of the full FFQ.
Other Foods Page
Only 510 women (17%) reported any additional foods on the other foods page. Among those who reported any additional foods, the median number of foods reported was 1.0 (data not shown). The 2 largest categories of added foods were fruits and vegetables (which would have been coded on the main FFQ if the “any other fruit” and “any other vegetables” line items had been included in the questionnaire). No soy-based foods were volunteered in response to this “other foods” question. The Hispanic, Chinese, and Japanese respondents were asked about their ethnic foods before the other foods page was administered. In contrast, for whites and blacks, the other foods page preceded the ethnic foods page, and thus these results for other foods are not influenced by the ethnic foods page.
Foods listed on the other foods page contributed less than 2% to the mean nutrient estimates for all nutrients and in all ethnic groups (data not shown). As an illustration, only 13 kcal were added to the mean nutrient intake among blacks and only 16.7 kcal among whites. In addition, Spearman correlation coefficients (rS) between the nutrient estimates with and without the foods added on the other foods page were all at least 0.98, indicating that the other foods page also did not alter the relative ranking of respondents in this study. For example, despite the addition of fruits and vegetables on the other foods page, the correlation between nutrient estimates with and without the other foods page was 0.99 for dietary fiber, vitamin C, and folate in all ethnic groups. There was strong agreement using classification into quintiles with and without the other foods page (95% exact agreement; weighted kappa = 0.97 for dietary fiber, for example). This was also true when the analysis was restricted to those who added any foods on the other foods page (weighted kappa = 0.94 among whites and 0.92 among blacks for dietary fiber). Quintile classifications for other nutrients were equally high.
Ethnic Foods Page Among Whites
We administered the ethnic foods page to whites at 3 sites. The majority of these respondents reported consuming some of the ethnic foods shown in Table 1; 58% of whites at the Newark site ate Hispanic items, 84% of whites at the Oakland site consumed some Chinese items, and 53% of whites at the Los Angeles site reported consumption of Japanese items. However, as Table 3 indicates, these ethnic foods added very little to their intake of any nutrient. With the exception of phytoestrogens (discussed further later in this article), Japanese foods contributed no more than 1% to the mean intake of any nutrient, Hispanic foods contributed no more than 2%, and Chinese foods contributed no more than 4% to any nutrient estimate among whites at those sites. In addition, almost all of the correlations between the nutrient estimates calculated with and without the ethnic foods page were 0.99, indicating that those foods were eaten with insufficient frequency or quantity to alter respondents’ ranking relative to each other with regard to nutrient intake. The weighted kappa statistic for agreement by quintile was over 0.95 for most nutrients at all 3 ethnic sites.
The median amounts of the phytoestrogens daidzein and genistein that were added by the ethnic foods page among whites at both the Oakland and Los Angeles sites were zero, because 88% of whites reported no additional soy-based foods above the amount contributed by the foods on the core food list. Among the 12% of whites who reported eating any of the soybean foods on the ethnic foods page, the median amount added was 767 μg of genistein and 460 μg of daidzein at the Oakland site, and 1768 and 1422 μg of those nutrients at the Los Angeles site. These contributed 10% to 12% of the mean genistein and daidzein at Oakland, and 31% to 41% of the mean of those nutrients at Los Angeles. Despite the increases in the absolute value of the mean, the Spearman correlations between the nutrient estimate with and without the ethnic foods page were 0.95 for both these phytoestrogens, indicating that respondent ranking was altered very little. For daidzein, the weighted kappa statistic was 0.92 for whites at Los Angeles and 0.95 for whites at the Oakland site. Thus, although some whites occasionally ate some ethnic-specific foods, that did not materially alter their ranking or categorization.
Ethnic Foods Page Among the Relevant Ethnic Group
Among Hispanic, Chinese, and Japanese Americans, inclusion of the ethnic-specific foods increased the mean (Table 4). These foods contributed as little as 2% and as much as 59% to the nutrient averages in these ethnic groups. Despite these increases in absolute amounts, correlations and categorizations between nutrient estimates with and without those foods were high. Excluding phytoestrogens, the median correlation (rS) for the other 12 nutrients shown was 0.975 among Hispanics, 0.94 among Chinese, and 0.955 for Japanese. Correlations were somewhat lower among Japanese American respondents for the phytoestrogens, daidzein (rs = 0.67) and genistein (rs = 0.77). The phytoestrogens among the Chinese Americans were less affected by the Chinese ethnic foods (rS = 0.97 for daidzein).
This analysis has shown that, in the context of a full-length questionnaire of approximately 100 food items, the additional foods acquired as a result of an open-ended question regarding “any other foods you eat at least once a week” add little to point estimates. Similarly, these additions have virtually no effect on rankings or categorizations of respondents with regard to their nutrient intake. Spearman correlation coefficients were 0.99 and weighted kappa statistics were greater than 0.9 between nutrient estimates with and without the other foods page for each of the ethnic groups examined. These results on the other foods page, which was administered by an interviewer, are consistent with other data using self-administration. Willett et al.9 found that only 7% of Boston-area nurses volunteered any additional foods on their 116-item self-administered FFQ. In that study, correlations with reference data did not differ at all for 13 nutrients and differed by only 0.01 for 5 nutrients. Block et al.10 found similar results for a 100-item self-administered FFQ.
Asking white respondents about ethnic-specific Hispanic, Chinese, or Japanese foods also had minor effects on point estimates or rankings, although the majority of whites reported occasional consumption of some items on each of the ethnic food lists. With the exception of the 2 phytoestrogens, correlations between nutrient estimates with and without the ethnic foods were 0.99 and weighted kappa statistics were higher than 0.9. Even for the phytoestrogens, the correlations with and without the ethnic foods page were 0.95 and the weighted kappa statistics were higher than 0.9, presumably because the major U.S. contributors of phytoestrogens (tofu, soy milk, and soy-based meat substitutes) are standard on Block core FFQs. Whether presentation of these foods to the white respondents actually improved the nutrient estimates (as opposed to simply increasing them) cannot be addressed by these data, but the correlations suggest that any improvement would be trivial.
Inclusion of ethnic-specific questions for persons in those ethnic groups is consistent with a sound questionnaire design, in which the goal is to include foods that may be important contributors to nutrient intake in that population. In those ethnic groups, such foods contributed substantially to group mean nutrient estimates. Even then, however, most correlations between estimates with and without the added ethnic foods were 0.9 or higher. Only phytoestrogens among the Japanese Americans (rs = 0.67 and 0.77) were more substantially affected, which may have in part been a reflection of the fact that 6 of the 16 added foods were soy products.
Other studies have shown that inclusion of additional culturally specific foods in questionnaires for certain minority groups increases the absolute value of the nutrient estimates (eg, Tucker et al.7 for Puerto Ricans and Coates et al.8 for blacks). Whether such additional foods improve correlations with reference data is less clear. Coates et al.8 did not find improvement in correlations, consistent with our present results, whereas Tucker et al. did.7 In the latter study, however, the correlations were derived by recoding 24-hour recall data rather than by actually administering a questionnaire, and thus may overestimate the correlations that could be achieved.
In summary, these analyses indicate that inclusion of an “other foods” section in a full-length FFQ does not result in sufficient improvement of nutrient estimates to justify the additional effort. Among whites, ranking and categorization are also not importantly altered by Hispanic, Chinese, or Japanese ethnic foods, even for highly concentrated food components such as the phytoestrogens, in the context of a well-designed core food list. For Hispanic, Chinese-, or Japanese Americans, however, addition of some ethnic-specific foods does increase nutrient estimates and has a modest effect on rankings for some nutrients, notably on phytoestrogen rankings among Japanese Americans.
We thank the study staff at each site and all of the women who participated in SWAN. Clinical Centers: University of Michigan, Ann Arbor (U01 NR04061, MaryFran Sowers, PI); Massachusetts General Hospital, Boston, MA (U01 AG12531, Robert Neer, PI 1994–1999, Joel Finkelstein, current PI); Rush University, Rush–Presbyterian–St. Luke's Medical Center, Chicago, IL (U01 AG12505, Lynda Powell, PI); University of California, Davis/Kaiser (U01 AG12554, Ellen Gold, PI); University of California, Los Angeles (U01 A12539, Gail Greendale, PI); University of Medicine and Dentistry–New Jersey Medical School, Newark (U01 AG12535, Gerson Weiss, PI); and the University of Pittsburgh, Pittsburgh, PA (U01 AG12546, Karen Matthews, PI).
1. Huang M-H, Schocken M, Block G, et al. Variation in nutrient intakes by ethnicity: results from the Study of Women's Health Across the Nation (SWAN). Menopause
2. Mares-Perlman JA, Klein BEK, Klein R, et al. A diet history questionnaire ranks nutrient intakes in middle-aged and older men and women similarly to multiple food records. J Nutr
3. Subar AF, Thompson FE, Kipnis V, et al. Comparative validation of the Block, Willett, and National Cancer Institute food frequency questionnaires. Am J Epidemiol
4. Huang M-H, Harrison GG, Mohamed MM, et al. Assessing the accuracy of a food frequency questionnaire for estimating usual intake of phytoestrogens. Nutr Cancer
5. Block G, Norris JC, Mandel RM, et al. Sources of energy and six nutrients in diets of low-income Hispanic-American women and children: quantitative data from the HHANES survey, 1982–1984. J Am Diet Assoc
6. Dietary intake source data: United States, 1976–80. Hyattsville, MD: US Department of Health and Human Services. Vital Health Stat 11
7. Tucker KL, Bianchi LA, Maras J, et al. Adaptation of a food frequency questionnaire to assess diets of Puerto Rican and non-Hispanic adults. Am J Epidemiol
8. Coates RJ, Eley JW, Block G, et al. An evaluation of a food frequency questionnaire for assessing dietary intakes of specific carotenoids and vitamin E among low income black women. Am J Epidemiol
9. Willett WC, Sampson L, Browne ML, et al. The use of a self-administered questionnaire to assess diet four years in the past. Am J Epidemiol
10. Block G, Hartman AM, Naughton D. A reduced dietary questionnaire: development and validation. Epidemiology
“The joy of research must be found in doing, since every other harvest is uncertain.”
THEOBALD SMITH,J. BACT.1934;27:19.