点击显示 收起
1 From the Lipid Metabolism Laboratory, Jean Mayer US Department of Agriculture Human Nutrition Research Center on Aging at Tufts University, Boston, and the Frances Stern Nutrition Center, New England Medical Center, Boston.
2 Supported by grant HL39326 from the National Heart, Lung, and Blood Institute, National Institutes of Health, and contract 53-3K06-5-10 from the US Department of Agriculture Agricultural Research Service. 3 Address reprint requests to EJ Schaefer, Lipid Metabolism Laboratory, Jean Mayer USDA Human Nutrition Research Center on Aging at Tufts University, Boston, MA 02111.
ABSTRACT
Background: We compared the validity of a semiquantitative food-frequency questionnaire in assessing intakes of macronutrients (absolute amounts and percentages of energy) by 19 subjects fed natural-food diets of known composition. In small subsets (n = 5 or 6), we also tested 3-d diet records.
Objective: The objective of this study was to investigate the efficacy of food-frequency questionnaires and diet records in subjects fed natural-food diets of known composition.
Design: Each subject consumed 3 different diets for 6 wk and self-reported his or her food intake by using a food-frequency questionnaire and a diet record. The diets varied in their chemically analyzed contents of fat (15–35% energy), saturated fat (5–14%), monounsaturated fat (5–14.5%), polyunsaturated fat (2.5–10.5%), carbohydrate (49–68%), and cholesterol (108–348 mg/d).
Results: The food-frequency questionnaire significantly underestimated fat, saturated fat, monounsaturated fat, and protein intakes and significantly overestimated carbohydrate intake with the high-fat diet. The percentage of energy from fat was significantly underestimated for the high-fat diet and significantly overestimated for the very-low-fat diet. Estimates from the food-frequency questionnaire differed significantly from actual intakes for fat (absolute and percentage), saturated fat (absolute and percentage), monounsaturated fat (absolute and percentage), and protein (percentage) in the high-fat diet and for polyunsaturated fat (absolute and percentage), saturated fat (percentage), fiber (absolute), and cholesterol (daily absolute; in mg/d) in the lower-fat diet. Estimates from the diet records better agreed with actual intakes than did estimates from the food-frequency questionnaire except for monounsaturated fat (absolute and percentage) in the high-fat diet and polyunsaturated fat (percentage) in the lower-fat diet and the very-low-fat diet.
Conclusion: Our data indicated that the food-frequency questionnaire did not provide reliable estimates of absolute amounts of dietary fats or cholesterol.
Key Words: Food-frequency questionnaire • dietary assessment • diet records • dietary assessment • macronutrient intake • protein • polyunsaturated fat • saturated fat • monounsaturated fat • cholesterol
INTRODUCTION
Dietary assessment involves the use of diet records, diet-history questionnaires, 24-h recalls, or food-frequency questionnaires. Each method has its strengths and weaknesses (1). The semiquantitative food-frequency questionnaire is especially useful for ranking individual intakes to describe relations between diet and disease in large epidemiologic studies (2). Because of the convenience of using a self-administered, computer-scannable questionnaire that can be completed in 20–30 min, the food-frequency questionnaire also has become popular as a means of measuring absolute dietary intakes and for determining dietary adherence throughout clinical diet trials (3–6). Caution is indicated when the food-frequency questionnaire is used for dietary assessment purposes, however, because the validity of this instrument for characterizing absolute nutrient intakes has not been assessed (1, 7).
The characteristics of dietary assessment methods were well described previously (2, 7–9). One shortcoming of most methods involving diet recall is that they often lack validation against some objective reference method that eliminates reporting bias (10). To date, there have been no true validation studies in which a food-frequency questionnaire was compared with an observed diet with a chemically defined composition. Concurrent validity, which is often established by comparison with some other method, such as 3–7-d weighed-food records, involves systematic respondent-related reporting bias and therefore is inadequate.
During the initial concurrent validation study for the food-frequency questionnaire used in the present study, dietary intake estimates from the questionnaire were compared with four 7-d weighed-food records. Intake estimates from the food-frequency questionnaire were adjusted for total energy intake by using the residuals from regression models; energy intake was the independent variable and nutrient intake was the dependent variable. Correlations between adjusted energy and nutrient intakes ranged from 0.4 to 0.7 when the original food-frequency questionnaire was used (11). An expanded version of this food-frequency questionnaire showed similar correlations for energy- and nutrient-adjusted intakes (12, 13).
A recurring problem with dietary assessment instruments is underreporting of intakes by subjects. Educational level, race, sex, obesity status, and age influence eating habits and the reporting of dietary intakes (14–18). There is concern as to which macronutrients are underreported because the underreporting of intakes of high-fat foods could result in lower estimates of fat and energy intakes. Recently, the validity of using absolute dietary estimates based on food-frequency data for assessing fat intakes was questioned with respect to relations between dietary fat and breast cancer (19, 20). The purpose of the present study was to assess the agreement of macronutrient and dietary cholesterol intakes with intake estimates from the food-frequency questionnaire, 3-d diet records, and actual intakes on the basis of chemically analyzed diets of identifiable foods.
SUBJECTS AND METHODS
Thirty-two subjects were enrolled in a 3-phase study designed to assess the effects of various diets on plasma lipoprotein concentrations (21, 22). A total of 19 of these subjects participated in the study. Each subject was asked to complete a food-frequency questionnaire and a 3-d diet record during each of the 3 phases of the study, allowing for 57 possible sets of food-frequency questionnaires and diet records during the study. The mean (±SD) age of the subjects was 60 ± 11 y and the mean body mass index (in kg/m2) was 27 ± 3. The mean age of the women in the study was higher than that of the men (66 ± 8 and 54 ± 11 y, respectively). The study was conducted according to the ethical guidelines of the Human Investigation Review Board of the New England Medical Center and Tufts University School of Medicine.
Each of 3 diets was fed for 6 wk. The phase-1 diet was similar to the typical American diet, the phase-2 diet was similar to the National Cholesterol Education Program (NCEP) Step II diet, and the phase-3 diet was a very-low-fat diet . During all phases, dietary energy intakes were controlled in all subjects to maintain weight within 1 kg. Body weight was assessed thrice weekly. Energy intake was adjusted for weight maintenance by means of proportional increments or decrements of all foods so that the quantity and composition of foods eaten remained consistent. All meals were prepared by the staff of the Metabolic Research Unit Nutrition Services Laboratory at the US Department of Agriculture Jean Mayer Human Nutrition Research Center on Aging at Tufts University. The subjects were fed meals from a 3-d rotating menu that consisted of breakfast, lunch, dinner, and 1 or 2 snacks. One meal/d was consumed at the Metabolic Research Unit, 5 d/wk. Two full days' worth of meals and all other meals were packaged and provided to the subjects, who were asked to report any food they were unable to eat. As an independent measure of adherence to the diets, serum lipids were monitored on 3 occasions during weeks 4, 5, and 6 of each diet phase. Blood was collected after a 12-h fast and samples were processed according to the method of the Lipid Research Clinics (23).
Food-frequency questionnaires and diet records were self-administered and were provided to the subjects in random order during weeks 3, 4, or 5 of each phase. The subjects were instructed to complete the questionnaire by reporting what they ate during each phase. As part of informed-consent procedures at the outset of the study, subjects were informed as to the nutrient contents of the diets and were provided with a list of foods and mixed dishes available on each 3-d menu. Therefore, conditions were optimal for accurate dietary recall. However, the subjects were middle aged and elderly, so their memories may not have been optimal.
Diet records and food-frequency questionnaires were given to the subjects by registered nurses when the subjects visited the center. The subjects completed these dietary assessment instruments on their own at home and returned them during subsequent visits to the center. Questionnaires were considered to be incomplete if entire sections were left blank or if questions involving the type and amount of fat were left unanswered. If the portions or foods listed were not specific or if data for <3 d were recorded, we excluded the incomplete instruments from the analysis rather than requesting subjects to complete another questionnaire or diet record. Out of a total of 114 possible records from 19 subjects, 45 records were used in our analyses. Six subjects filled out only 1 instrument, 7 subjects completed 2 instruments, 1 subject completed 3 instruments, 4 subjects completed 4 instruments, and only 1 subject completed all 6 instruments.
We used the expanded version of the Willett food-frequency questionnaire (12). Completed food-frequency questionnaires were sent away to be scanned by computer and the results were provided in an unadjusted format for energy, percentage of energy from macronutrients, absolute amounts of macronutrients and dietary fiber (in g), and absolute amount of dietary cholesterol (in mg).
To obtain a standard against which to measure efficacy, triplicate preparations of each complete 3-d menu for each diet were chemically analyzed for protein, carbohydrate, fat, fatty acids, and dietary cholesterol contents by Hazelton Labs America, Inc (Madison, WI). Dietary fiber was estimated by using food-composition tables from the GRAND database (release 867; US Department of Agriculture Human Nutrition Research Center, Grand Forks, ND).
Diet records were analyzed by a registered dietitian using FOOD PROCESSOR II (ESHA Research, Salem, OR). Recipes for mixed dishes were provided by the Metabolic Research Unit to enable accurate analysis of the entries on the diet records. For statistical analysis, each chemical value was subtracted from the values obtained from the food-frequency questionnaire and the diet record. The resulting differences were analyzed by repeated-measures analysis of variance using SAS PROC MIXED, version 6.12 (SAS System for Mixed Models; SAS Institute, Inc, Cary, NC).
RESULTS
Our data were based on self-administered, complete food-frequency questionnaires filled out by 19 subjects on 28 occasions and diet records filled out by 12 subjects on 17 occasions. The nutrients provided by the chemically analyzed (actual) diets and the mean dietary intakes as reported by the food-frequency questionnaire during the 3 diet phases are shown in Tables 1 and 2. Energy intakes reported on the food-frequency questionnaire were lower than actual intakes in all 3 phases. The study diets varied in their chemically analyzed contents of fat (15–35%), saturated fat (5–14%), monounsaturated fat (5–14.5%), polyunsaturated fat (2.5–10.5%), carbohydrate (49–68%), fiber (24–50 g/d), and cholesterol (108–348 mg/d) (Table 1). Furthermore, mean intakes as reported on the food-frequency questionnaire during phase 1 differed significantly from actual intakes in their contents of fat, monounsaturated fat, and saturated fat. Phase-2 intakes as assessed by the food-frequency questionnaire differed significantly from actual intakes in their contents of polyunsaturated fat and dietary fiber and cholesterol. There were no significant differences between actual intakes and estimates from the food-frequency questionnaire during phase 3.
View this table:
TABLE 1.. Actual absolute intakes and intake estimates from the semiquantitative food-frequency questionnaire (SQFFQ) during the 3 diet phases1
View this table:
TABLE 2.. Energy compositions of actual intakes and intake estimates from the semiquantitative food-frequency questionnaire (SQFFQ) during the 3 diet phases1
There were significant differences in percentages of energy from fat, saturated fat, monounsaturated fat, and protein between actual intakes and estimates from the food-frequency questionnaire during phase 1 (Table 2). During phase 2, the percentages of saturated fat and polyunsaturated fat as reported on the food-frequency questionnaire were significantly different from actual intakes. During phase 3, the percentages of fat and monounsaturated fat estimated from the food-frequency questionnaire were significantly different from actual intakes. No other dietary components reported on the food-frequency questionnaire differed significantly from actual intakes in phases 1, 2, or 3.
Estimates from 3-d diet records were significantly different from actual intakes of monounsaturated fat (absolute and percentage of energy) during phase 1 and of polyunsaturated fat (percentage of energy) during phases 2 and 3 (Tables 3 and 4). There were no other significant differences between the estimates from the 3-d diet records and actual intakes during the 3 phases.
View this table:
TABLE 3.. Actual absolute intakes and intake estimates from 3-d diet records during the 3 diet phases1
View this table:
TABLE 4.. Energy compositions for actual intakes and intake estimates from 3-d diet records during the 3 diet phases1
The percentage overestimation and underestimation of various dietary components when the food-frequency questionnaire was used is shown in Table 5. Energy intakes with all 3 diets were underestimated when the food-frequency questionnaire was used. During phase 1, intakes of fat (absolute and percentage of energy), saturated fat (absolute and percentage of energy), monounsaturated fat (absolute and percentage of energy), and protein (percentage of energy) were significantly underestimated and carbohydrate intake (percentage of energy) was significantly overestimated. During phase 2, underestimation of polyunsaturated fat intake (absolute and percentage of energy) and overestimation of saturated fat intake (percentage of energy) were significant. During phase 3, the food-frequency questionnaire significantly overestimated intakes of fat and polyunsaturated fat (percentage of energy). Percentage of energy from fat was significantly underestimated by the food-frequency questionnaire during phase 1 and significantly overestimated by the foodfrequency questionnaire during phase 3.
View this table:
TABLE 5.. Percentage under- and overestimations of mean dietary components with the semiquantitative food-frequency questionnaire1
Total blood cholesterol concentrations, presented as the mean of 3 measurements obtained during weeks 4, 5, and 6 of each diet phase for the 19 subjects, were 6.8 ± 0.8, 5.7 ± 0.8, and 5.7 ± 0.8 mmol/L in phases 1, 2, and 3, respectively. LDL-cholesterol concentrations were 4.6 ± 0.7, 3.8 ± 0.7, and 3.8 ± 0.7 mmol/L; HDL-cholesterol concentrations were 1.2 ± 0.3, 1.1 ± 0.2, and 1.0 ± 0.2 mmol/L; and triacylglycerol concentrations were 1.4 ± 0.6, 1.5 ± 0.7, and 1.8 ± 0.9 mmol/L in phases 1, 2, and 3, respectively (21, 22).
DISCUSSION
Previous investigations of the accuracy of dietary assessment methods focused on estimates of energy intakes because objective measures of energy intake that are free of recall bias, such as total energy expenditure by the doubly labeled water technique, are available. Most subjective reports of energy intake are underestimates of actual intakes (18, 24, 25). When 4 dietary assessment methods were evaluated with the doubly labeled water reference method, younger women came closest to their actual energy intakes when 24-h recall was used, whereas older women came closest when the food-frequency questionnaire was used (15). No significant correlations between energy expenditure and reported energy intakes were evident, regardless of the dietary assessment method used. Underreporting of energy intake ranged from 20% in nonobese adults to 50% in obese subjects (9). In our study, we found that energy intake was underestimated by 20% on average.
To normalize the results of studies that use food-frequency questionnaires, it is common to energy-adjust raw data (12, 13, 16, 26). Adjustments for energy require the assumption that portion size and fat, carbohydrate, and protein intakes are all reported with the same degree of accuracy. Other assumptions are that the omissions or biases in the reporting of any foods or food groups are minimal and that reporting errors are proportional to the amounts of food eaten. In fact, reporting bias appears to exist for certain foods, especially those high in fat, sugar, or alcohol (24). If fat intake is underreported because of bias, the percentage of energy from fat will be falsely low. Results from our study showed underreporting of energy and fats with the food-frequency questionnaire. It is doubtful that reporting bias contributed to these findings because all food and drink was provided to the subjects.
In studies in which the food-frequency questionnaire was compared with 7-d weighed-food records, correlation coefficients between the 2 reporting methods ranged from 0.4 to 0.7 (11, 12). Such analyses merely show the concurrent validity of one method compared with another and do not compare reported intakes with a response bias–free standard of intakes. If food-frequency questionnaires are to be used to assess absolute nutrient intakes, the degree to which these estimates agree with actual intakes from diets fed under metabolic ward conditions needs to be evaluated. In our study, we found significant underestimation of actual intakes of fat, saturated fat, monounsaturated fat, and polyunsaturated fat when the food-frequency questionnaire was used. Because the food-frequency questionnaire underestimated intakes of fat and fatty acids during the higher-fat diet and overestimated intakes during the lower-fat diet, this instrument appears unable to differentiate a high from a low fat intake and is thus of questionable use for assessing a subject's baseline dietary intake or dietary adherence during a clinical trial in which fat intake is important.
Diet records may not provide valid estimates of actual intakes either. For example, when 2 groups with restrained and unrestrained eating habits were examined, both groups altered their intakes when they were asked to keep food records (26). The restrained eaters decreased their consumption of fatty foods by 19% and increased their consumption of fruit and vegetables by 43% during the recording period (27). Dietary intakes were reported with better accuracy with 3-d food records than with the food-frequency questionnaire in our study. One possible explanation for the higher correlation between actual intakes and estimates from diet records is the ability of the researcher to use almost-exact recipes or more-specific foods in the analysis. Food-frequency questionnaires provide a limited list of foods and do not allow recipes or specific ingredients to be inputed for analysis.
In our study, conditions for dietary recall were improved by allowing the subjects access to menus and food lists throughout the study. This should have increased, rather than decreased, the accuracy of recall with both dietary assessment methods. The subjects were given common foods and most of these foods are listed on the food-frequency questionnaire. It is doubtful that limitations in food choices on the questionnaire were responsible for the lack of agreement; rather, it is likely that the food-frequency questionnaire did not allow for adequate quantification of fat or energy intakes. It is also possible that subjects were less likely to underreport intakes while eating a weight-maintenance diet when all meals and snacks were provided by the researcher and when body weights were frequently measured, as was the case in this study.
We knew that the subjects in our study were complying with the diets because significant and predictable reductions in plasma LDL-cholesterol concentrations were achieved with the therapeutic diets (21, 22). Furthermore, the subjects maintained their weight during the 3 phases of the study. Limitations of our study included the small sample size. Additionally, adherence to completing the food-frequency questionnaire or the 3-d diet records was imperfect, as was commonly reported in other studies. A review of reasons for nonadherence did not reflect any systematic bias by body mass index or other known characteristics.
Our data indicate that the food-frequency questionnaire may be unreliable and inadequate for assessing absolute and relative macronutrient intakes. Use of the questionnaire also needs to be critically examined to determine whether the questionnaire is adequate for assessing adherence in diet-related clinical trials, especially when fat intakes are of interest. Three-day diet records provided significantly better estimates of actual intakes than did the food-frequency questionnaire in a controlled metabolic study in which all foods and drinks were provided. These results indicate that a larger study on the applicability of food for dietary characterization of absolute and relative nutrient intakes should be undertaken. Although the expense of such a study may be a deterrent, the information that it could obtain is of significant importance to the field of dietary assessment.
REFERENCES