Comparison of a food frequency questionnaire with a diet record

The associations between levels of nutrients derived from a 24-hour diet record and a food frequency questionnaire were assessed in a sample of 433 men and women. The food frequency questionnaire was administered three years after the completion of the diet record. Spearman correlations were all statistically significant; they varied from 0.36 for energy to 0.15 for vitamin A. Comparison of distributions into fifths showed few people grossly misclassified. A model was developed to assess the correlations which would be expected under various conditions of within and between subject variance for each nutrient, errors in measurements and drift in intake over time. The best possible correlation that could be obtained using the within and between subject variations in intake previously published, and with no measurement error or drift over time, was 0.60 for energy and 0.34 for vitamin A. Using a realistic measure of measurement error (standard deviation on log scale of 0.12 for diet record method, 0.06 for drift over time and 0.18 for food frequency method) the correlations obtained in the modelling were very similar to that observed. This study shows that it may be appropriate to use a food frequency questionnaire instead of a diet record to estimate intakes in population based epidemiological studies.