Validation and calibration of food-frequency questionnaire measurements in the Northern Sweden Health and Disease cohort

Abstract

AbstractObjectives:To evaluate the reproducibility of, and to compare and calibrate, diet measures by the Northern Sweden 84-item food-frequency questionnaire (FFQ) with measures from 24-hour diet recalls (24-HDR).Design:Randomly selected respondents (n=246) from the EPIC (diet-cancer) and MONICA (diet-cardiovascular disease) study cohort in Northern Sweden were invited to answer the FFQ twice over a one-year interval (FFQ1 and FFQ2), and to complete ten 24-hour recalls (reference method) in the months between. Plasma β-carotene concentrations were determined from a subset of 47 participants.Setting:Vasterbotten and Norrbotten, Northern Sweden.Participants:Ninety-six men and 99 women, who completed the study.Results:The reproducibility of the FFQ was high in terms of both mean energy and nutrient intakes and relative ranking of participants by intake levels (median Pearson correlation of 0.68). Moderately higher food intake frequencies were recorded by FFQ1 compared with 24-hour recalls for dairy products, bread/cereals, vegetables, fruits and potato/rice/pasta, whereas meat, fish, sweet snacks and alcoholic beverage intakes were lower. The median Spearman coefficient of correlation between FFQ1 and the average of ten 24-HDR measurements was 0.50. Daily energy and nutrient intakes were similar for FFQ1 and 24-HDR measurements, except for fibre, vitamin C, β-carotene and retinol (FFQ1<24-HDR) and sucrose and cholesterol (FFQ1>4-HDR). Pearson coefficients of correlation between FFQ1 and 24-HDR corrected for attenuation due to residual day-to-day variation in the 24-HDR measurements ranged from 0.36 to 0.79 (median 0.54). Adjustment for energy had only very moderate effects on the correlation estimates. Calibration coefficients estimated by linear regression of the 24-HDR on the FFQ1 measurements varied between 0.30 and 0.59 for all nutrients except alcohol, which had calibration coefficients close to 1.0. These low calibration coefficients indicate that relative risk estimates corresponding to an absolute difference in dietary intake levels measured by the FFQ will generally be biased towards 1.0. Plasma β-carotene levels had a Pearson coefficient of correlation of 0.47 with the 24-HDR measurements, and of 0.23 with FFQ1 measurements.Conclusions:The Northern Sweden FFQ measurements have good reproducibility and an estimated level of validity similar to that of FFQ measurements in other prospective cohort studies. The results from this study will form the basis for the correction of attenuation and regression dilution biases in relative risk estimates, in future studies relating FFQ measurements to disease outcomes.