The (ab)use of food frequency questionnaire data in substitution modelling in nutritional epidemiology: a critique

Why this matters for everyday food advice

Many headlines claim that swapping one food for another can lengthen life or prevent disease. Behind those headlines are complex studies that rely on long checklists called food frequency questionnaires, where people report how often they eat different foods. This review asks a simple but important question: are those questionnaires good enough to support confident advice about what to replace on your plate?

How scientists try to study food swaps

Because it is difficult and often unethical to assign people to strict diets for many years, nutrition research usually follows people in daily life instead. A common approach is substitution modelling, which estimates what might happen to health if, for example, a serving of red meat is replaced with fish or plant protein. To do this, researchers need reasonably accurate numbers for both the food being reduced and the food taking its place. Food frequency questionnaires are attractive because they are cheap and easy to use in very large groups, and they aim to capture usual eating habits over long periods.

What this review set out to check

The authors examined 100 studies from 21 countries, published between 2018 and 2024, that used only food frequency questionnaires in substitution models. They asked whether the specific food or nutrient measures used in those models had been tested against better reference methods, such as detailed food records or repeated 24 hour recalls. They also looked at how clearly the studies reported those tests and how closely questionnaire results matched the comparison methods.

What the review discovered

More than half of the studies used food questionnaire measures that had not been properly checked for accuracy, even though many appeared in high impact journals. In 62 percent of studies, the description of how well the questionnaires performed was minimal or missing. When validation data were available, the match between questionnaire estimates and reference methods was often only fair to moderate. For some nutrients and food groups, average intakes were off by as much as several hundred percent, and accuracy varied widely from one item to another.

Why small errors become big problems

Food frequency questionnaires are known to work better for ranking people from low to high intake than for measuring exact amounts. They tend to have systematic errors, such as consistent underestimation of total energy, that differ by nutrient and food group. In substitution models, those errors affect at least two things at once: the food being reduced and the food being increased. Instead of cancelling out, the errors can add together and bend the estimated effect in either direction. This means that the neat numbers often reported for swapping one daily serving of a food for another may not reflect real world changes as reliably as they appear.

What needs to change

The authors argue that relying on untested questionnaire estimates for detailed food swap calculations is not good enough, especially when results help shape dietary guidelines. They call for tools designed specifically for substitution questions, better use of more accurate short term diet records and biomarkers where possible, and clearer reporting of how well each food or nutrient measure performs. Until such improvements become routine, results from substitution models built mainly on food frequency questionnaires should be viewed with caution, and not taken as precise instructions for how a single food swap will affect individual health.

Citation: Louie, J.C.Y., Bhowmik, J. The (ab)use of food frequency questionnaire data in substitution modelling in nutritional epidemiology: a critique. Eur J Clin Nutr 80, 458–468 (2026). https://doi.org/10.1038/s41430-026-01712-7

Keywords: food frequency questionnaire, substitution modelling, nutritional epidemiology, dietary measurement error, dietary guidelines