Fruit & Vegetable Screeners: Validity Results

Performance of Two New Cognitively Enhanced Fruit & Vegetable Short Assessment Forms
Thompson FE, Subar AF, Radimer K, Smith AF, Midthune D, Rosenfeld S, Kipnis V.
National Cancer Institute; Cleveland State University
Presented at the Fourth International Conference on Dietary Assessment Methods, September 2000

Cognitive interviewing about standard fruit and vegetable questions led to refinement and expansion of short diet assessment tools to measure usual individual intake of fruits and vegetables (f&v). The "all-day" instrument includes 12 foods; the "by-meal" instrument includes questions about the same foods, but asks about consumption of two foods (fruit; vegetables other than potatoes, salad, beans) by time of day (morning, midday, evening). Portion size questions are asked on both. Participants of the US National Cancer Institute’s Eating at America’s Table Study who had completed four 24- hour dietary recalls, one per season, over the previous year, were randomly divided into two groups, and mailed one questionnaire per group. Response rates were 38% (n=242) in the "all-day" group and 35% (n=224) in the "by-meal" group. After appropriate transformation and exclusion of outliers, a measurement error model was used to estimate the relationship between each instrument and true intake (estimated by the model). Median intakes of f&v were estimated fairly closely by both screeners when portion size information was included (Men: 5.5, 4.9, 4.8; Women: 4.3, 4.6, 5.5 for recalls, all-day, and by-meal). Median intakes were underestimated when portion size information was not included (Men: 5.5, 3.3, 3.1; Women: 4.3, 3.4, 3.3 for recalls, all-day, and by-meal). Correlations between truth and the complete by-meal screener (Men: 0.68; Women: 0.52) were similar to those between truth and the complete all-day screener (Men: 0.59; Women: 0.56). However, measurement error was present, creating substantial attenuation for both screeners (All-Day: Men: 0.48, Women: 0.33; By-Meal: Men: 0.53; Women: 0.40). We conclude that both screeners perform adequately for estimation of population medians; portion size questions are particularly important for men.