Food composition and empirical weight methods in predicting nutrient intakes from food frequency questionnaire

Although the empirical weight (regression-based) method has theoretical advantages over the traditional food composition method in predicting nutrient levels from food frequency questionnaires (FFQs), the empirical assessments have been limited.

We compared the validity of the two approaches for a 44-item questionnaire used in a population-based prospective study in Japan. Based on four 7-day diet records and questionnaire responses collected from a subsample of the prospective study (94 men and 107 women), we developed a food composition table and stepwise regression models to predict intakes of energy and 14 nutrients from the questionnaire.

When we applied the two methods to an independent population (207 men and 166 women) providing a 3-day diet record and responding to a 36-item dietary questionnaire, energy-adjusted and deattenuated correlation coefficients between the questionnaire and the diet records were not higher for the empirical weight method than for the food composition method; the median (range) was 0.22 (0.07-0.57) for men and 0.23 (-0.09-0.62) for women in the former method, and 0.26 (-0.04-0.58) for men and 0.38 (0.18-0.67) for women in the latter method.

We did not find the improved validity of FFQ by empirical weight method in predicting nutrient intakes.