Background: Food frequency questionnaires (FFQs) are important tools for dietary assessment in large epidemiological studies, playing a crucial role in evaluating the relationship between diet and health. However, adapting the food lists in FFQs to align with specific study objectives or target populations presents a considerable challenge. Methods: The present study develops a framework using mixed-integer linear programming (MILP) to minimize the food list of an FFQ using a vegetarian population in Germany as a proof of concept. Constraints of the optimization ensured that the selected food items have a certain nutrient coverage and variance coverage, as well as an appropriate aggregation level. Nutrient intake for three scenarios for FFQs was compared with 24 h recalls (24HR) using R2, calculated through linear regression. The three scenarios were: 1. FFQ reflecting the effect of categorizing portion sizes, 2. FFQ reflecting the effect of selecting food items, 3. FFQ reflecting the effect of categorizing portion sizes and selecting food items. Results: Length of minimized FFQs increased with a higher proportion of nutrient coverage and variance coverage. Including aggregation of food items produced shorter FFQs than FFQs that only contain food items at a lower aggregation level. R2 values across the three scenarios showed that the FFQ captured most of the between-person variation in nutrient intake that was observed in the 24HR. Conclusions: MILP offers a reliable and data-driven framework for compiling optimized FFQs.
Blaurock et al. (Sat,) studied this question.