Flegal K M, Larkin F A, Metzner H L, Thompson F E, Guire K E
Department of Biostatistics, School of Public Health, University of Michigan, Ann Arbor 48109.
Am J Epidemiol. 1988 Oct;128(4):749-60. doi: 10.1093/oxfordjournals.aje.a115028.
Differences in energy estimates between a food frequency questionnaire and a multi-day dietary record can be partitioned into distinct components due to differences in reported frequency of consumption and in reported serving size, and to differences in nutrient composition between the questionnaire standards and the foods reported on the records. The effect of each component on the relative validity of the questionnaire can be assessed by examining its contribution to the differences between the two methods in estimated group intake and in the relative ranking of individual respondents. This methodology was used for the 1984-1985 University of Michigan Food Frequency Study, in which the estimated energy intake from a quantitative food frequency questionnaire was compared with that from 16 days of food records collected over the course of a year from 228 white and black men and women aged 24-51 years. For all race-sex subgroups, mean energy intake estimated from the questionnaire was significantly greater than mean intake estimated from the record. Within race-sex subgroups, the correlations between estimates from the two methods were low, and agreement in classification by tertiles was poor. The differences in group mean energy intake between the methods were due to the effects of discrepancies in both serving size and frequency of consumption. However, the low correlations and poor agreement in classification for individual respondents were due principally to the effect of discrepancies in frequency. These results suggest that improving the accuracy of frequency estimation is a key element in increasing the relative validity of food frequency questionnaires used for epidemiologic research.
