Reproducibility and validity of diet quality scores derived from food-frequency questionnaires.

backgroundDiet quality scores (DQSs) are frequently used to study diet-disease relations but the validity of these scores derived from FFQs has rarely been evaluated.

objectivesTo evaluate the validity and reproducibility of 6 commonly used DQSs derived from the FFQ, including the Alternate Healthy Eating Index-2010, Dietary Approaches to Stop Hypertension Trial score, alternative Mediterranean diet score, and 3 plant-based diet indices (overall, healthful, and unhealthful).

methodsThis analysis included 1394 participants from the Men's Lifestyle Validation Study (N = 652) and the Women's Lifestyle Validation Study (N = 742). Participants completed a 152-item FFQ at the beginning of the study and 1 y later, as well as completed 2 weighed 7-d dietary records (7DDRs) and donated 2 blood samples 6 mo apart between FFQ assessments. The reproducibility of the FFQs was evaluated by rank intraclass correlation coefficients (ICCs). The validity was assessed by comparing FFQ-derived DQSs with those from the average of two 7DDRs using Spearman rank correlation coefficients deattenuated for random measurement error in the 7DDRs (rs). Furthermore, we calculated the correlations between DQSs and plasma biomarkers of diet, including fatty acids, folate, carotenoids, retinol, and α- and γ-tocopherol.

resultsSix FFQ-derived DQSs demonstrated moderate to high reproducibility (energy-adjusted ICCs: 0.61-0.84) and validity (energy-adjusted, deattenuated rs = 0.56-0.80) in both men and women. We consistently observed expected correlations between FFQ-derived DQSs with plasma fatty acids, including long-chain n-3 (ω-3) and trans fatty acids, most carotenoids, and γ-tocopherol (rs > 0.2).

conclusionsOur study demonstrates the validity of the FFQ to evaluate overall diet quality using 6 commonly used DQSs. In addition, these DQSs have qualitatively demonstrated biological relevance, as indicated by their correlations with circulating biomarkers.