Combining self‐report dietary assessment instruments to reduce the effects of measurement error

Abstract

With the advent of a self‐administered automated 24 hour dietary recall instrument (24HR), developed at the National Cancer Institute (NCI), it is now possible to envision its use in cohort studies investigating the relationships between diet and disease. Understanding that all dietary assessment instruments are subject to measurement error, we compare 3 possible strategies for estimating intake: 1) administering 1–12 24HRs per subject; 2) administering a single food frequency questionnaire (FFQ); and 3) administering a combination of 24HRs and a single FFQ. We used data from the National Cancer Institute's Eating at America's Table Study, which collected 4 24HR, one per season, and an FFQ in a nationally representative sample of 965 U.S. adults. We compared the precision, power and required sample size for the different strategies under the assumption that the 24HR is unbiased for true usual intake. We found that the strategy of 4–6 administrations of the 24HR is sufficient for most nutrients and food groups, and that the combined use of multiple 24HR and FFQ data often provides estimates superior to either alone. In most cases, adding an FFQ to one or more 24HRs improved precision more than adding an additional 24HR. If self‐administered automated 24HRs are to be used in cohort studies, one should consider administering 4–6 24HRs and an FFQ.