In the present study, the validity of the KiGGS FFQ was evaluated in comparison to a diet history method instrument. Due to measurement errors and limitations within every dietary assessment method, only relative validity can be determined. The FFQ showed a fair to moderate agreement in ranking participants towards their intake for most food groups compared to the DISHES interview . Only white bread and pasta/rice showed slight agreement. The correlation coefficients varied between 0.22 for pasta/rice and 0.69 for margarine. A reasonable to good correlation between the two instruments was found for 67% of the food groups . The average of the observed correlation coefficients was higher or equal to other FFQ validation studies for adolescents [6–9, 29–32]. The observed correlation coefficients are also similar to results from FFQ validation studies for adults [33–36]. Individual, higher coefficients for adults may be caused by an established meal structure and therefore a better memory on portion size and frequency. By contrast, the food frequency and portion sizes of adolescents are not constant . Agreement of mean intake is rather low in most food groups. Some food groups - like milk, mineral water, eggs, meat fish, fruits and potato products - show small average differences. However, on the individual level there is a wide range of differences for every food group. The FFQ should therefore perhaps not be used to estimate absolute intakes. Other youth validation studies on food group level came to similar results [9, 29].
The validation was performed using food consumption data from the EsKiMo module. This offered the advantage of a large validation sample that is representative of German adolescents, which also made it possible to evaluate the validity in subgroups. However, there may have been a tendency to select participants who were especially interested in their health and nutrition, since the EsKiMo participants agreed to participate for a second time. Calculation of correlation coefficients is a common method in validation studies . One main reason may be that it facilitates comparisons with other study results . However, correlation coefficients only measure the strength of the association between two methods, not the agreement, and can be a misleading indicator of validity [40, 41]. Nevertheless, calculating correlation coefficients was included in this study since small correlation coefficients can be indicators of potential error sources . Additional analyses, like Bland Altman analysis or ranking classification, can avoid misleading conclusions. For the Bland Altman analysis it is assumed that the differences between two measurements are normally distributed . Since in our study the differences were not normally distributed, and this could not be improved by log-transformation, the differences between the two instruments were calculated on the basis of untransformed data. Furthermore, we included an adapted analysis, which approximates the analysis of limits of agreement. Percentiles (2.5/97.5) of differences between the methods were calculated, which also represents 95% of differences. There are some limitations to be considered in relation to this validation study. For the assessment of validity, the reference method should have independent error sources . Contrary to this, the reference instrument DISHES also relies on the memories of the participants and their perceptions of portion sizes, like the FFQ. This may result in unrealistically higher estimates of validity. Since the EsKiMo study was not primarily designed as a validation study, the choice of another reference method was not applicable. However, the DISHES interviews were conducted by trained nutritionists and supported by standardized software, while the FFQ was self-administered. Dietary intake information was more detailed and assessed in a meal-based structure. The DISHES interview used a variety of tableware models, standard portions and a picture book for estimating portion sizes. Furthermore, the list of food items assessed by the FFQ was fixed, while the DISHES interview was open-ended. The DISHES interview therefore seems an acceptable method of comparison. The DISHES method was previously validated for adults, but not for adolescents. It has also been used in some large nutrition surveys. Nevertheless, a pre-test was conducted to test feasibility among adolescents and the food-code database was adopted for younger persons. The FFQ was filled in by respondents several days before the DISHES interview was conducted. The sequence of instruments is relevant, since one measurement may affect a later response . However, the reverse sequence would probably have a larger effect, since the diet history is a more comprehensive instrument which may have a larger impact on a person's memory and awareness of the actual diet. In addition, the items in DISHES are more detailed and asked in a face-to-face setting. We therefore think the influence on recall of the applied sequence is minor.
The FFQ seems to be suitable for all considered subpopulations, since most food groups showed reasonable to acceptable correlation coefficients. Only the groups pasta/rice, white/brown bread, and cakes/pastries showed correlation coefficients of below 0.3; these should be interpreted with caution. Despite these results, certain differences between the BMI and socio-economic groups were found. As expected, older participants (aged 16 to 17 years) showed a tendency towards higher correlation coefficients than younger ones (12 to 13 years), because their cognitive abilities were better developed . Furthermore, older adolescents choose their food themselves more often; they are also more conscious of what they eat. Correlation coefficients were lower for overweight adolescent girls compared to normal-weight girls. This finding might be expected, since thinness and body image have an important influence on female adolescents' dietary reporting . Boys are less likely to be concerned about their body image. This relationship is in line with results from other studies [46, 47]. In addition, participants who live in families with a low socio-economic status showed lower correlations more often than participants in families with high socio-economic status. To our knowledge, similar studies in such subgroups have not been performed among adolescents. However, some studies among adults found an inverse association between socio-economic status and underreporting [48, 49], which is a potential source of bias in nutritional epidemiology and could be one reason for lower reporting validity. Nevertheless, the difference between subgroups is marginal and most food groups showed acceptable to good correlations. The KiGGS-FFQ is thus also suitable for groups with lower socio-economic status and higher body mass index.
Some differences in ranking and mean estimates between the instruments may be caused by differences in the measurement of portion sizes. While the DISHES interview assesses food intake with a variety of standard portions, tableware models and a picture book, the FFQ uses predefined, simple categories. The variability of values measured by the FFQ is therefore rather low. The relatively weak ranking agreement in the case of vegetables is discussed in other studies among adolescents [9, 29], and also in adult populations [33, 35, 50]. One possible explanation is again related to difficulties in estimating portion size in some food groups. Some broad food items like "cooked", "frozen", "tinned", and "raw vegetables" may complicate the estimation of these predefined portion sizes. For instance, lettuce and tomatoes both belong to the raw vegetables group, even though one portion of each may have very different weights. In addition, adolescents in particular may have problems defining the origin of their foods, because they normally do not prepare meals themselves. Accordingly, dividing vegetables into the groups frozen and tinned seems difficult for this age group. This difference was not assessed in the DISHES interview. These items were therefore grouped. The food group pasta/rice showed slight agreement among ranking participants in terms of food intake. One possible explanation is the different use of the two products. Pasta is often the main component of a meal like spaghetti bolognaise, while rice is eaten as a side dish. It is therefore difficult to predefine a portion size for both products together. The DISHES interview assesses the amounts separately for every food and in as much detail as possible. The group white bread also showed only slight agreement in ranking. This may be due to a lack of experience among adolescents regarding the classification of bread.