Use of OR to design food frequency questionnaires in nutritional epidemiology

Nutritional epidemiology, investigating the relationship between diet and disease, often uses food frequency questionnaires (FFQs) to assess a population’s habitual dietary intake. An FFQ should include enough food items (i.e. questions) to capture sufficient information on all nutrients of interest. However, it should not be too long in order to avoid the fatigue of respondents. Although the procedure of selecting questions is done by an expert, it is neither standardized nor transparent, and very time consuming. Moreover, it is hard to select questions in such a way that all nutrients of interest are sufficiently covered within a relatively short questionnaire. The resulting questionnaire is probably not optimal, e.g. with the same number of questions more information might be obtained. We have developed a 0–1 knapsack model to optimize the selection of questions for FFQs with interest in multiple nutrients. With this FFQ model we generated FFQs with interest in energy and 9 nutrients. We found that the FFQ model can be a valuable tool to optimize FFQs. With the FFQ model the selection of questions is less time-consuming and more standardized and transparent than in a manual procedure, and the resulting food lists of FFQs are either shorter or provide more information.