Repeatability of food frequency assessment tools in relation to the number of items and response categories included

Accuracy of measurement is a cornerstone of research in order to make robust conclusions about the research hypothesis.

To examine whether the number of items (questions) and the number of consumption responses (the coding used to measure the frequency of consumption) included in nutritional assessment tools influence their repeatability.

During 2009, 400 participants (250 from Greece, mean age 37 +/- 13 years, 34% males, and 150 from Spain, mean age 39 +/- 17 years, 41% males) completed a diet index with 11 items and binary (yes/ no) responses, a diet index with 11 items and 6-scale responses, and 36-item and 76-item food frequency questionnaires (FFQs) with 6-scale responses. The participants completed these tools twice, with 15 days between the two administrations of the tools. The Spearman-Brown coefficient (r(sb)), Kendall's tau coefficients, and the Bland-Altman method were applied to answer the research hypothesis.

The highest repeatability coefficient was observed for the diet index with 11 items and binary (yes/no) responses (r(sb) = 0.948, p < .001), followed by the diet index with 11 items and 6-scale responses (r(sb) = 0.943, p < .001), the 36-item FFQ with 6-scale responses (r(sb) = 0.936, p < .001), and the 76-item FFQ with 6-scale responses (r(sb) = 0.878, p < .001). Statistical comparisons revealed no significant differences between repeatability coefficients of the first three tools (p > .23), whereas these three tools had significantly higher repeatability coefficients than the 76-item FFQ (p = .002). Subgroup analyses by sex, education, smoking, and clinical status confirmed these results.

Repeatability was found for all food frequency assessment tools used, irrespective of the number of items or the number of responses included.