Towards transparent development of food frequency questionnaires : scientific basis of the Dutch FFQ-TOOL tm: a computer system to generate, apply and process FFQs

Summary

Introduction
A well-designed food frequency questionnaire (FFQ) aims to assess habitual intake
of foods or nutrients of interest in the target population. Therefore, the food list of
an FFQ has to be adapted to the prevailing dietary habits of the target population
and updated when re-used some time after initial development. However, due to
lack of expertise, time, or finances, FFQs are often re-used without adaptations. To
simplify and standardize the development of new FFQs, we developed a computer
system, the Dutch FFQ-TOOLTM, to generate, apply, and process FFQs. The aim of
this thesis was to provide the scientific basis for development of this system.
Methods
We first characterized FFQ design by a systematic review of validation studies.
For selecting food items in the system, we evaluated a simple procedure, called
MOM2, which selected food items on the basis of explained variance in nutrient
intake without taking covariance from other food items into account. To improve
questions in FFQs, we conducted focus group discussions to investigate problems
encountered by adults when filling out FFQs. Using the information from these
studies; we developed a prototype of the system, which used the Dutch National
Food Consumption Survey 1998 for the selection of food items. This prototype was
used to generate an FFQ, which was subsequently validated against biomarkers and
3-day food records.
Results
In the systematic review, we observed that FFQs with more items (>200 items)
were better able to rank people according to their intake of most nutrients than
shorter FFQs (<100 items). MOM2 appeared suitable to select food items that
contributed importantly to variance in nutrient intake, leading to only a few
more food items than regression analysis. Focus group discussions showed that
36 out of 40 respondents were confused by examples that were meant to clarify
questions and 31 out of 40 respondents had difficulties in identifying consumed
foods, because categorization of foods was not logical to them. This information
was used to develop standard questions and answering categories which were
included in the Dutch FFQ-TOOLTM. The validation of a first FFQ generated by the
system showed that validity coefficients between FFQ and true intake were 0.50
for protein, 0.36 for potassium and 0.50 for PUFA, as assessed by the methods of
triads applied to the FFQ, food records, and biomarkers of intake. These results
are similar to those from validation studies of other FFQs.
Conclusion
We have developed a flexible data-based computer system that can generate FFQs
using standardized procedures for multiple nutrients of interest, which is suitable
to assess habitual food and nutrient intake of adults with sufficient validity.