American Journal of Epidemiology Copyright© 1991 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved
Vol. 134, No 6 Printed in U.S A.
An Evaluation of a Food Frequency Questionnaire for Assessing Dietary Intake of Specific Carotenoids and Vitamin E among Low-Income Black Women
Ralph J. Coates,1-3 J. William Eley,1-3 Gladys Block,4 Elaine W. Gunter,5 Anne L. Sowell,5 Clifford Grossman,1 and Raymond S. Greenberg13
The National Cancer Institute diet questionnaire was evaluated for use in a lowincome black population. Data were collected from 91 women aged 30-69 years who were hospital outpatients in Atlanta, Georgia, June through August, 1988. Six ethnic and regional foods added to the questionnaire were found to be important contributors to intakes of several nutrients. Although 17 records were identified as containing probable recording or reporting errors, intakes of carotenes, alpha-carotene, betacarotene, cryptoxanthin, and vitamin E were significantly and positively associated with serum levels of their referent nutrients. Among nonsmokers, correlation coefficients ranged from 0.32 to 0.45, adjusted for age, body mass index, alcohol and calorie intakes, medications and vitamin supplement use, and serum cholesterol and triglycerides. When questionnaires containing identified errors were omitted, correlations ranged from 0.30 to 0.54. There were no correlations between dietary intakes of lycopene and lutein and blood levels (-0.06 to 0.09). Among smokers, diet-serum correlations were reduced (0.00 to 0.32). These correlations are similar to those reported in research on vitamin E and carotenoids in other populations. These results suggest that the questionnaire is as valid for use in this population as it is in other populations. Am J Epidemiol 1991; 134:658-71. blacks; carotene; carotenoids; diet; nutritional status; questionnaires; smoking; vitamin E
Recent epidemiologic research on the relations between dietary intakes and risk of cancer of a number of specific sites has yielded inconsistent results among studies and study populations (1). Such inconsistencies raise questions not only about rela-
tions between diet and cancer but also about the validity of methods used to assess dietary intake in different populations. The technique epidemiologists most commonly use to assess usual diet has been the food frequency questionnaire (2, 3). Because the
Received for publication December 31, 1990, and in final form April 25, 1991. 1 Division of Epidemiology, Emory University School of Public Health, Atlanta, GA. 2 Nutrition Division, Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control, Atlanta, GA. 3 Winship Cancer Center, Emory University School of Medicine, Atlanta, GA. 4 Division of Cancer Prevention and Control, National Cancer Institute, National Institutes of Health, Bethesda, MD. 5 HANES Laboratory, Center for Environmental Health and Injury Control, Centers for Disease Control,
Atlanta, GA. Reprint requests to Dr. Ralph J. Coates, Division of Epidemiology, Emory University School of Public Health, 1599 Clifton Road, N.E., Atlanta, GA 30329. This work was supported in part by cancer education grant CA17998-15 from the National Institutes of Health. J. W. E. is a recipient of an American Cancer Society Career Development Award. The authors thank Lone Click, Amanda Gearing, Mark Dorminy, Dr. David Simon, Dr. Melvin Moore, Theresa Gillespie, Nayana Ambardekar, Hiawatha Martin, Bonnie Boles, Eric Giddens, and Joel Webber for their assistance in conducting this study.
Food Questionnaire Evaluation in a Black Population
validity of the food frequency questionnaire may be affected by educational or cultural factors that influence recall (3, 4), the validity of the technique may vary across populations. In a review of methodological issues to be considered in assessing the relation between diet and health, the National Research Council's Committee on Diet and Health (5) noted that neither of the commonly used questionnaires developed by Willett et al. (6) and Block et al. (7) had been evaluated in a sufficiently diverse set of populations. Although there is no "gold standard" to use in validating food frequency questionnaire estimates of intake, these instruments have been evaluated by correlating intake estimates with serum levels of specific nutrients (8-10). Serum vitamin E and carotenoid levels do vary with intake (8, 11) and, therefore, may be used as indicators of intake. As standards, serum nutrient levels offer the added advantage of being independent of the subject's recall (8). In a number of different populations, food frequency questionnaire estimates of usual intake of carotenoids have been correlated with serum levels of carotenoids. The populations studied include Boston area health care researchers and staff (8), volunteers in clinical trials in several US cities (9, 12, 13), and hospital clinic patients in Boston (14). In two of these studies (8, 14), vitamin E intakes have been found to be correlated with serum alphatocopherol levels. To our knowledge, no research has been published on the validity of food frequency questionnaires for use in assessing intake in low-income black populations. We therefore evaluated the widely used Health Habits and History Questionnaire developed by Block and colleagues at the National Cancer Institute (10). We examined the distribution of responses to dietary intake questions, evaluated the contribution of specific ethnic and regional foods to nutrient intakes, and examined the association between Health Habits and History Questionnaire estimates of diet intakes and serum levels of a number of carotenoids and vitamin E.
MATERIALS AND METHODS Study population
From June 6 through August 8, 1988, black women aged 30-69 years attending outpatient clinics of Grady Memorial Hospital in Atlanta, Georgia, were asked to participate in a nutrition study. Clinics were selected to include patients visiting the hospital for a variety of conditions that were unlikely to affect nutritional status. Medical students consulted daily with hospital clinic staff to identify patients of the appropriate age and sex from those who were waiting in clinics. Men and nonblack patients were excluded. In addition, 84 women were excluded who were pregnant or lactating; were too ill to be interviewed; had a history of diabetes or intestinal digestive disorders; or were taking anticonvulsants, anticoagulants, broad spectrum antibiotics, lipid-lowering medications, or systemic steroids. Of 132 eligible patients asked to participate, 28 (21.2 percent) refused, and 10 (7.6 percent) were unable or declined to complete the interview or blood collection. For three patients (2.2 percent), insufficient serum was available for determination of the complete set of serum nutrients, and they were excluded from this analysis. The final sample consisted of 91 patients (68.9 percent of those eligible). Of these 91 patients, 64 were from a walk-in clinic, 14 were from an eye clinic, and 13 were from the orthopedic; general medical; or ear, nose, and throat clinics. Patients had a wide range of health conditions: aches and pains (41 percent), eye problems (13 percent), and injuries, chronic illnesses, and other conditions. Interview and dietary data
During a private interview lasting approximately 45 minutes, medical students used a modified version of the Health Habits and History Questionnaire (10) to collect dietary, medical history, demographic, and health habits information. Students had been trained by study investigators and coordinators with experience in collecting dietary
Coates et al.
data for epidemiologic research. Interviewers met weekly with investigators to review the interview process. The Health Habits and History Questionnaire and its development have been described in detail elsewhere (7, 10). Briefly, respondents were asked to recall for the year before their interview the frequency of their consumption and portion sizes of 98 specified foods and food groups; the frequency of eating at different types of restaurants; and the frequency, types, and amounts of fats used in cooking and added to vegetables. In open-ended questions, respondents also were asked to recall their intakes of any other frequently consumed foods not on the 98-item food list. In addition, respondents were asked whether they ate visible fat on meat and ate skin on chicken. Additional questions assessed frequency and quantity of specific types of vitamin supplements. Interview responses were recorded on a form that was edited for completeness and coded. Forms were key entered twice using data entry programs with edit checks, and discrepancies in values were resolved during the second entry. Dietary intakes of specific nutrients were calculated by the existing Health Habits and History Questionnaire programs that sum the products of frequency times portion size times the nutrient content of each food item. The nutrient values and portion sizes were developed by Block et al. (7, 10), using data from the Second National Health and Nutrition Examination Survey, US Department of Agriculture Handbook 8, revised, and other sources. The calculated dietary intakes were adjusted for frequency of eating in restaurants, eating chicken skin, trimming fat from meat, and adding fats to vegetables. Because participants were interviewed during the summer when consumption of seasonal fresh fruits and vegetables occurs, "inseason" rather than yearly Health Habits and History Questionnaire estimates of nutrient intakes were used when available. The estimate of Handbook 8 carotenoids from the Health Habits and History Questionnaire programs was derived from the US Department of Agriculture Handbook 8
vitamin A values; however, for tomato products, the Handbook 8 values have been adjusted downward because these foods contain primarily lycopene rather than betacarotene (10). The specific carotenoids (alphaand beta-carotene, cryptoxanthin, lycopene, lutein, and other xanthins) were estimated on the basis of published analyses and unpublished data provided by Dr. Gary Beecher of the US Department of Agriculture. When published assays were unavailable for a particular food, estimates of specific carotenoids were based on the carotenoid distributions of similar foods. Estimates for the specific carotenoids involved more recent assay methods, required a number of assumptions, and are not directly related to the carotenoid estimates represented by the Handbook 8 carotenoids. Therefore, it is not possible to add alpha-carotene, betacarotene, and cryptoxanthin to obtain the Handbook 8 carotenoid value. For this study, we added ethnic and regional foods to the 98-item Health Habits and History Questionnaire list. These included such items as chicken or turkey pot pie, Kool-Aid (General Foods, White Plains, New York), plums, grapes, grits, okra, blackeyed peas, and specific organ meats and pork foods considered by Atlanta nutritionists to be consumed frequently by the patient population. Nutrient content and portion sizes for okra and black-eyed peas were available from the Health Habits and History Questionnaire nutrient data base. We obtained nutrient values for the remaining foods from the US Department of Agriculture Handbook 8, revised (15), and estimated age- and sex-specific portion sizes using similar foods from the 98-item listing. We estimated carotenoid values for these foods using Handbook 8 provitamin A values and the carotenoid distributions of similar foods. When running the Health Habits and History Questionnaire dietary analysis programs, we used the default options with two exceptions; options were selected to include information on specific cold cereals and the 19 ethnic and regional foods we added to the questionnaire.
Food Questionnaire Evaluation in a Black Population
Serum nutrient data
After the respondent was interviewed, a nonfasting venous blood sample was obtained. A single serum specimen was collected using an evacuated 15-ml collection tube without anticoagulant. Each tube was covered to protect it from light and allowed to stand upright for 0.5 hour. The tube was centrifuged for 15 minutes at 2,500 rpm, and the serum was separated. The serum specimen was frozen at —70°C for up to 2 months before shipment on dry ice to the Centers for Disease Control's HANES Laboratory for nutrient analysis. Specimens were analyzed by highperformance liquid chromatography (16) for levels of alpha-tocopherol, alpha-carotene, beta-carotene, cryptoxanthin, lycopene, and lutein/zeaxanthin. Serum total cholesterol and serum triglycerides were measured by discrete analysis using the Dupont automatic clinical analyzer (Dupont, Wilmington, Delaware) (17). Control sera were prepared at the HANES Laboratory from pooled human serum. Interrun and intrarun coefficients of variation ranged from 1.4 percent to 11.4 percent, except for alphacarotene which ranged up to 34.4 percent (interrun) because of the extremely low levels found in most sera. Data analysis
For descriptive purposes, we calculated means and medians of the nutrient intake and serum nutrient levels using untransformed variables. A body mass index was calculated from self-reported height and weight (weight (kg)/height (m)2) (18). We classified respondents as taking a medication that increases vitamin A or E if they were taking cholinergic or adrenergic agents, betablockers, acetylcholinesterase inhibitors, alpha-methyldopa, clonidine, hydralazine, diuretics, replacement hormones, or calcium channel blockers and as taking a medication that reduces vitamin A or E if they took nitrates, potassium chloride, or antacids (19-25). We used Health Habits and History Ques-
tionnaire edit programs to identify potential errors in the diet section of respondent records. The edit programs identified records for which values were outside the valid range for each variable, frequencies of consumption of each food were unusual, or total frequencies were unusual. The unreasonable totals for which possibly serious errors were identified included the number of portion sizes coded "medium," "small," or "large" (>99 percent), the number of frequencies for which "once per" was used (>75 percent), and the total number of food servings eaten by a respondent per day (30). To evaluate the need to include any of the 19 additional food items in a food frequency questionnaire developed for use in this population, we used the Health Habits and History Questionnaire nutrient sources program to estimate for the entire study population the percentage of each nutrient contributed by each food item. The nutrients used for this evaluation were vitamin E, specific carotenoids, and other nutrients commonly analyzed in cancer epidemiologic studies (calories, protein, total fat, saturated fat, cholesterol, carbohydrate, dietary fiber, vitamins A and C, and calcium). We considered the foods to be important contributors if, when foods were ranked by percentage contribution to the nutrient, they ranked in the upper 90 percent of the cumulative percentage distribution. To examine the associations of nutrient intakes with serum nutrient levels, we calculated Pearson product-moment correlation coefficients (26, 27). Because techniques to measure serum lutein levels had not been developed for this study, we used serum lutein/zeaxanthin as a referent for the Health Habits and History Questionnaire dietary lutein measure. A serum referent for the Health Habits and History Questionnaire Handbook 8 carotenoids was created by summing the provitamin A serum carotenoids for which we had laboratory measures, i.e., alpha-carotene, beta-carotene, and cryptoxanthin. In addition to the Handbook 8 carotenoids measure of provitamin A carotenoids, the nutrient analysis pro-
grams produce three other measures of provitamin A: alpha-carotene, beta-carotene, and cryptoxanthin. These three intake measures were combined into a single measure of provitamin A intake and were correlated with the serum provitamin A index. We used plasma alpha-tocopherol as the referent for the vitamin E estimate. In addition, because some investigators (14) have combined both food and supplemental sources of vitamin E into a single dietary vitamin E intake estimate, we combined dietary vitamin E and vitamin E from supplements to form a "total vitamin E" intake measure which was correlated with serum alpha-tocopherol. Because a number of other characteristics have been reported to affect serum levels of vitamin E and carotenoids, we used multiple regression analysis (26, 27) to calculate partial correlations adjusted for these effects. Serum carotenoid levels have been found to be associated with vitamin A supplementation (12, 28), total calories (12), alcohol (9, 12, 14, 29, 30), body mass index (8, 9, 13, 14, 29, 30), serum cholesterol (8, 9, 12, 14, 30, 31), serum triglycerides (8, 14), cigarette smoking (9, 13, 14, 32), and a number of medications, including lipid-lowering agents and exogenous estrogens (13, 30, 32). Serum vitamin E levels have been associated with vitamin E supplementation (30, 32), calories (8, 14), alcohol (14), age (14), serum cholesterol (8, 31), serum triglycerides (1, 8, 14, 30, 31), and cigarette smoking (12, 14, 30). Correlations between nutrient intakes from foods and serum referents were adjusted for intakes from supplements. The correlation of total vitamin E with serum alphatocopherol was not adjusted for supplement intake. The statistical significance of the diet-smoking interaction was evaluated by including in the regression models interaction terms of cigarette smoking status (1 = yes/0 = no) times diet intake. To determine if observed correlations between carotenoid and vitamin E intakes and serum levels of the nutrients varied by interviewer or by respondent characteristics, we subdivided the study population by sex and race of the interviewer and by age, education, income, and dietary practices of re-
spondents. We conducted additional analyses to determine if correlation coefficients improved when we excluded records identified by the Health Habits and History Questionnaire edit programs to be in error. We also excluded the 19 additional ethnic and regional foods from the nutrient analysis, recalculated intakes, and recalculated correlation coefficient based on the unmodified Health Habit and History Questionnaire. Because the nutrient intake, serum nutrient, body mass index, vitamin supplement dose, and alcohol intake measures were skewed toward higher values, we used natural logs of these variables in the correlation and regression analysis. If the untransformed value was zero, the transformed value was set to ln(0.1). These transformations decreased the skewness of all variables, but several variables were not normalized, including dietary intakes of alcohol and lycopene; serum levels of alpha-carotene, cryptoxanthin, lycopene, and triglycerides; and intakes of vitamin A and vitamin E from supplements. However, for each multivariable regression model, plots of residuals between predicted and observed serum nutrient levels indicated that normality, linearity, and homoscedasticity assumptions were met in each regression model, and Cook's D statistics showed no excessive influence of any observations (26, 27, 33). RESULTS
This study population of black women was largely middle-aged with less than high school levels of education and very low levels of education (table 1). They were, on average, overweight with body mass indexes above the 27.3 value commonly used to define overweight (18). A larger proportion of those women smoked than do black women nationally (36 percent) (34). Although we excluded individuals taking a number of medications, approximately a third took medications that increase levels of vitamin A or E, and a smaller number took medications that decrease those vitamins. Health Habits and History Questionnaire
Food Questionnaire Evaluation in a Black Population
TABLE 1. Demographic characteristics and nutrition-related health habits of study participants (n = 91): Atlanta, Georgia, 1988 Characteristic
Age (years) Weight (kg) Height (cm) Body mass index (weight (kg)/height (m)2) Household income ($/month)* Education (years) Current smokers (%) Medications which may Increase vitamin A or E (%) Decrease vitamin A or E (%)
Mean ± standard error
46.2 ±1.15 45 74.7 ± 1.80 73.9 163 ±0.76 163 30.3 ± 0.71 29.5 700 700 10.6 ±0.28 11 45.1 36.3 13.2
•Study participants were asked their monthly household income category: s$200, $201 -$400, $401 -$600, $601 -$800, $801-$1,000, >$1,000. The mean and median incomes were in the $601-$800/month category
edit programs identified possible respondent errors in 17 (18.7 percent) of the dietary records. Four respondents gave the same time unit (day, week, or month) for more than 75 percent of the foods they reported eating. Three subjects reported the same for portion size (medium or small) for 100 percent of the food items. Ten women reported eating four or fewer food items each day. Of the 19 ethnic and regional foods added to the 98 items in the questionnaire, chicken or turkey pot pie, ham hocks, grits, okra, Kool-Aid black-eyed peas, grapes, and plums were in the top 90 percent of food contributors to calories or one of the nutrients (table 2). Chicken pot pie was a primary contributor to intake of the Handbook 8 carotenoids; Kool-Aid was a primary contributor to calories, carbohydrate, and vitamin C intakes; and black-eyed peas were a top contributor to dietary fiber intake. Nutrient intakes and serum nutrient levels are presented in table 3. More than half of the study population used vitamin supplements. For each nutrient, median values were less than the means, and distributions were skewed to higher values. In the regression models that included interaction terms between cigarette smoking and dietary intake, the p-value of the term
varied with the nutrient: