Accuracy of food-frequency questionnaires

Department of Nutrition Harvard School of Public Health 665 Huntington Avenue Boston, MA 02115

Dear Sir:

In their recent article in the Journal, Schaefer et al (1) concluded that a food-frequency questionnaire developed by our group did not provide reliable estimates of absolute intakes of dietary fat or cholesterol. They based this conclusion on an extremely small evaluation of 19 subjects who were fed 3 different diets for 6-wk periods in a tightly controlled metabolic study. Although I agree that food-frequency questionnaires are not an optimal method for assessing group means in intervention trials (2), the conclusions reached by Schaefer et al are based on a highly artificial context distinctly different from that for which food-frequency questionnaires were designed. Because measurement of long-term dietary intake is the objective, this and most food-frequency questionnaires are based on average intakes of foods over the previous year. In this particular application, subjects were cycled through 3 different diets in addition to their customary diets for 6-wk periods. Thus, it is not surprising that there would be a blurring of dietary intakes assessed by the food-frequency questionnaire.

Second, the subjects were given all the foods that were to be eaten and thus did not participate in the selection, purchase, or preparation of foods, which would normally be the case when food-frequency questionnaires are used in epidemiologic studies. Furthermore, to obtain a contrast in diets in metabolic studies, recipes are typically altered from what participants might normally use. Allowing participants to see the menus does not at all replicate the usual involvement of subjects in determining, and thus developing a knowledge about, their long-term dietary patterns.

Schaefer et al suggested that the diet records provided more accurate information in their study, but they acknowledged that the results were artificial because the investigators knew the exact recipes to use for dishes that were reported (because they provided the food themselves). However, their experience illustrates one of the serious problems with diet records in that fewer than one-third of the subjects actually completed the records. In this contrived example, this made no difference because everyone was fed the same food. However, in a more realistic context, persons completing diet records are likely to have diets different from those who do not complete diet records, which could lead to a serious bias. Additionally, it was shown in many intervention trials using objective measures of dietary intake that individuals overreport compliance when they use diet records (2). This is likely to be the result of better compliance during recording of the diets than on other occasions. There appear to be similar biases for 24-h dietary recalls when individuals know that they will be interviewed about their previous day's diet (3). Thus, the most unbiased measure of average group compliance in intervention studies appears to be neither food-frequency questionnaires nor diet records but rather "surprise" 24-h dietary recalls conducted on random days by telephone interview (2) or the use of biomarkers when possible.

Additional data on the validity of various dietary assessment methods would be useful, but the study design used by Schaefer et al could be highly misleading.

REFERENCES

1. Schaefer EJ, Augustin JL, Schaefer MM, et al. Lack of efficacy of a food-frequency questionnaire in assessing dietary macronutrient intakes in subjects consuming diets of known composition. Am J Clin Nutr 2000;71:746–51.
2. Willett WC. Nutritional epidemiology. 2nd ed. New York: Oxford University Press, 1998.
3. Buzzard IM, Faucett CL, Jeffery RW, et al. Monitoring dietary change in a low-fat diet intervention study: advantages of using 24-hour dietary recalls vs food records. J Am Diet Assoc 1996;96:574–9.