Validation of dietary patterns assessed with a food-frequency questionnaire

In their recent article, Hu et al (1) identified dietary patterns using factor-analyzed data from food-frequency questionnaires (FFQs) and they assessed the validity of these patterns in part by examining their association with patterns identified by using factor-analyzed dietary-record data. Their use of dietary-record data to assess the validity of dietary patterns requires comment. Specifically, although dietary records are an acceptable gold standard for validating FFQs for measuring food or nutrient intake, the same strategy does not provide an equivalent validation of dietary patterns. To the extent that dietary records provide more accurate estimates of food intake, they also allow more precise estimation of factor scores than do FFQs. By conducting a factor analysis to identify dietary patterns, however, Hu et al essentially created two factor–analytically derived scales to measure intakes from Western and prudent dietary patterns, with each food or food group representing one (differentially weighted) item in each scale. Validation of FFQ-based dietary patterns against dietary-record-based patterns with use of scales derived from factor analysis based on the same food items is comparable with validation of a scale against the same scale with individual items measured more accurately. In essence, the validation strategy presumes that the item-level data are valid and uses these data rather than an independent indicator of each food pattern. The ability to assess the validity of dietary patterns measured by factor analysis is limited by our understanding of what dietary patterns actually represent. Nutritional anthropologists have researched numerous dimensions of intake patterns—how foods are organized into dishes and dishes into meals, which foods are integral to the meal, and even the time, place, and context in which meals are eaten (2). Measuring patterns by using factor-analyzed FFQ responses assumes that patterns can be characterized adequately by food-intake frequencies and their intercorrelations. Although this method may capture enough variation in eating habits to render measurement of other dimensions unnecessary, examining dimensions of dietary patterns other than with the use of food-frequency data may provide valuable additional information in some instances. For example, effects on iron bioavailability of concurrent consumption of meats as absorption enhancers or phytates as absorption inhibitors (3) illustrate the potential importance of considering the organization of foods into meals. Whether scales derived from factor analysis based on food frequencies alone are acceptably valid measures of actual dietary patterns, therefore, remains to be evaluated. Identifying a more appropriate gold standard for validation will require a more complete conception of what the Western and prudent dietary patterns actually are. Indeed, the greater challenge may be to gain a more complete a priori understanding of dietary patterns before trying to measure them, thus raising the possibility of measuring dietary patterns directly rather than relying on ad hoc interpretations of dietary data. Hu et al’s analysis does provide evidence of food groupings that might have been anticipated a priori. The finding of similar patterns across methods also provides evidence of the reproducibility of their approach. As such, their evaluation showed that FFQs can be a useful and convenient source of dietary data for measuring dietary patterns, even though they were not originally intended for dietary-pattern measurement.