Study design and study population
The nationally representative German Health Interview and Examination Survey for Children and Adolescents (KiGGS) was performed between 2003 and 2006 by the Robert Koch Institute. The aim of the KiGGS survey was to collect comprehensive data on the health status of children and adolescents aged 0 to 17 years. Participants were enrolled in two steps, in the first step, 167 sample points were randomly drawn stratified by federal state and community size. In the second step, participants were randomly selected from local population registries stratified by age. Children and adolescents with a migration background were also included. The final net sample included 17,641 participants, who lived in Germany . The survey was approved by the German federal data protection office and by the ethics committee of Charité - University Medicine Berlin. Participants were informed in detail about the study objectives, interview and examination procedures as well as the handling of data records and analysis under pseudonymous conditions, and gave their written consent. Design and methods are described in detail elsewhere .
Data collection and adaption of study variables
KiGGS includes several physical examinations, conducted by trained staff, among them blood pressure, body height and body weight measurements. Systolic and diastolic blood pressure (BP) were measured using an automated oscillometric device at an interval of two minutes. The arithmetic mean of two consecutive measurements was used for the analysis. Body height was measured according to a standardized protocol to the nearest 0.1 cm using a portable stadiometer. Body weight was measured in underwear to the nearest 0.1 kg with an electronic scale . Since the body mass index (BMI) is dependent on growth related changes in body composition, a relative measure of BMI is more appropriate for adolescents. Therefore, BMI z-scores of the body mass index percentiles were calculated according to Schaffrath Rosario et al. . For the analysis of biomarkers, blood samples were collected, separated into aliquots, frozen and stored at −40°C . The fasting period was documented for every participant. To assess the usual intake of selected foods, a Food Frequency Questionnaire (FFQ) was used . An energy intake index was calculated, summarising the multiplied amounts and mean energy contents of the FFQ items. Health-related behaviour like alcohol consumption, physical activity and smoking status of the participants were assessed with a self-administered questionnaire. Medication and supplement use of adolescents during the last seven days were determined with a standardised interview conducted by a physician . Additionally, parents were asked about their income, occupational status and education. With this information a family socio-economic status index was calculated as described previously .
In this study associations of dietary indices with biomarkers of long term nutrition were analysed. These biomarkers were selected from the biomarkers available for the study population. Since dietary indices are used to evaluate the accordance of an individual’s diet with nutritional recommendations, a positive association with biomarkers of dietary exposure (ferritin, HbA1c, folate, vitamin B12) was expected. Biomarkers of cardiovascular status (homocysteine, uric acid, C - reactive protein, total cholesterol, high-density lipoprotein cholesterol) were selected, which are predictive for levels in adulthood . Homocysteine was determined with fluorescent particle immunoassay (Axsym; Abbott, Wiesbaden, Germany). Uric acid was determined by the uricase-PAP method (Hitachi 917; Roche, Mannheim, Germany). Total cholesterol was analysed using an enzymatic assay (cholesterol oxidase-PAP method) produced by Roche. High-density lipoprotein cholesterol (HDL-C) was determined directly with a homogenous enzymatic colorimetric assay (Roche). Serum C-reactive protein (CRP) was measured by Immunoturbidimetry (Hitachi 917). During the course of the survey the reagent produced by SCIL (Martinsried, Germany) was replaced by Roche. Due to parallel measurement, data derived with the SCIL reagent could be converted into a CRP value that corresponds with the new method. Serum vitamin B12, serum ferritin, and serum folate were measured by electrochemi luminescence immunoassay (Elecsys E2010; Roche). During the survey, the method for determination of folate was changed by the manufacturer. In the present analysis data obtained with both methods were analysed separately (first/second period), since conversion of the data was not feasible. HbA1c was analysed using high-performance liquid chromatography (Diastad; Biorad, Munich, Germany). Biochemical measures in the KiGGS study were described in detail elsewhere .
Dietary assessment and construction of dietary indices
Usual consumption of several food groups during the “last few weeks” was assessed using a self-administered, semi quantitative FFQ. The questionnaire was developed by the Robert Koch Institute and includes 45 food items. The frequency of consumption was assessed within ten categories, similar for all food items: never, once a month, two to three times a month, one to two times a week, three to four times a week, five to six times a week, one time per day, two to three times a day, four to five times a day, more than five times a day. In addition, participants had to estimate the usual portion size of the food item, which was given in five item specific categories. Several pictures were used to illustrate the portion sizes. The FFQ and a covering letter were sent by postal mail to the participants, several weeks prior the examination visit. The first page of the FFQ provides instructions about the completion of the questionnaire. Additionally, a telephone hotline was offered for any support in completing the questionnaire. At the examination visit the questionnaire was checked for completeness, and further support was offered. The FFQ was validated in comparison to the dietary history method DISHES and showed fair to moderate ranking validity for most food items (Spearman correlation coefficients from .35 to .69 with most values above .5), except for pasta/rice (.22) and white bread (.31) . The validity of the FFQ is comparable to other FFQs for adolescents .
The food based dietary guidelines “Optimized Mixed Diet” (OMD) were developed to facilitate the adoption of a healthy diet to children and adolescents. The concept was described in detail elsewhere . For this study, three dietary indices were selected, since these were developed especially for children and adolescents in Germany, taking into account the OMD recommendations. While the Healthy Nutrition Score for Kids and Youth (HuSKY) and the Indicator Food Index (IFI) were originally developed for KiGGS, the Healthy Food Diversity Index (HFD) was initially developed for adults and then adapted to adolescents. Additionally, for this study, a simple index of fruit and vegetable intake was calculated to compare with the more complex ones, since in the German national health monitoring these food groups are used as a main indicator of a healthy diet. For all dietary indices an increasing score is associated with a healthier diet.
Healthy Nutrition Score for Kids and Youth (HuSKY)
The HuSKY was developed for the KiGGS study to compare eating habits of children and adolescents with the OMD guidelines . To develop the index, 38 FFQ items were aggregated into eleven food groups corresponding to the guidelines . Then, the ratio of food intake to food intake recommendation was calculated for each food group. On base of the sex and age-specific guidelines the ratio was allocated with points. For most food groups, intakes below the recommendation were proportionally allocated up to 100 points. If participants exceed the double recommended amount, points were proportionally subtracted from 100. The points of all food groups were summarized and afterwards, the HuSKY was standardized on a scale from 0 to 100. The HuSKY offers a valuable instrument to evaluate overall eating habits in a population, but is not intended to assess specific aspects of dietary behaviour in detail.
Healthy Food Diversity Index (HFD)
The HFD was originally developed for the German Nutrition Survey (GeNuS) of 1998 among adults to assess the food diversity and the health value of an individual diet . It considers three aspects: the number, distribution, and health value of all consumed foods. The index increases when the variation in food intake becomes healthier. Therefore the Berry-Index , which was applied in economic food diversity studies, was multiplied by a food-specific health factor based on the food consumption guidelines of the German Nutrition Society . For our study, the HFD was adapted to adolescent’s diet. The intake of 41 FFQ items was used to calculate the index score. The food specific health factors were calculated according to the OMD guidelines . Higher values of the HFD reflect a healthier diet. The consideration of both diversity and dietary recommendations seems to be the advantage of the HFD.
Indicator Food Index (IFI)
A further, relatively simple index was developed previously in the research group . Consumption of seven food groups of the KiGGS FFQ (fruits, vegetables, brown bread, soft drinks, fast food, chocolate, and salty snacks) was used as an indicator of a favourable or unfavourable diet. Therefore, frequency of each food group intake was categorized as healthy (2 points), neutral (1 point) and unfavourable (0 points). The points were defined using dietary guidelines and as a consensus of nutrition experts during a dietary indices expert meeting at a KiGGS symposium. By adding the points of all seven indicator food items an index with a scale from 0 to 14 was calculated. A score from 0 to 5 points was rated as an unfavourable, 6 to 10 points as a neutral and 11 to 14 points as a favourable diet. It should be emphasized, that this index covers only a few foods and therefore reflects only a selected proportion of the diet and not an overall dietary pattern. Furthermore, in comparison with HFD and HuSKY the estimation of health values for single food groups is relatively simple.
As part of the continuous national health monitoring, the Robert Koch Institute regularly conducts telephone health interview surveys in representative samples of the German adult population (GEDA) . Since the number of questions in a telephone interview is limited, only questions concerning the consumption of fruits, vegetables and fruit/vegetable beverages were included. These items can be used to build a simple indicator for a healthy diet but not to represent the general diet . For the present study, a similar fruit/vegetable index was calculated to compare it with the more complex dietary indices. Standardised portions per day were calculated for six FFQ items (cooked, raw, frozen and tinned vegetables; fresh and tinned fruits). Subsequently, the portions were summarised. According to the nutritional recommendations of the German Nutrition Society up to one portion juice per day was added to fruit and vegetable consumption .
For the present analyses we excluded participants without blood samples (N=292) and those, who did not completed the FFQ (N=263). Furthermore, pregnant participants were excluded from the analyses (N=2). Overall, out of 5,755 KiGGS participants our analyses included 5,198 participants. To avoid bias by medication use we excluded participants with diabetes and antidiabetic medication in the analyses for glycohaemoglobin (HbA1c; N=77). Furthermore, we excluded participants who used oral contraceptives in the analyses of serum lipids (N=432) and those who used oral contraceptives and antihypertensive medication in the analyses of blood pressure (N=468).
The sample of the KiGGS
study was drawn by a clustered and stratified design, therefore all analyses were performed with complex-samples procedures of SPSS version 18.0 (SPSS Inc., Chicago, Illinois, USA). Since sex differences in dietary habits and pubertal status may be expected in this age group, we conducted separate analyses for boys and girls. To enhance representativeness for the German population structure, statistical analyses were weighted. For the comparison of the different dietary indices, scores were grouped into quintiles. Consequently the interpretation of the scales was similar for the indices, with higher quintiles indicating a healthier diet. The biomarkers values were generally not normally distributed. After a log transformation a normal distribution was also not achieved for all biomarkers but the results of the models did not change. Therefore the untransformed data are presented. Mean values of biomarkers with 95% confidence intervals were calculated according to quintiles of dietary indices, using complex-samples general linear model (CSGLM), and tested for trends. Additionally, regression coefficients for the association between biomarkers and indices were calculated by including the ranked index score as a continuous variable. All analyses were stratified for sex and adjusted for age (continuous), energy intake (continuous), BMI
z-scores (continuous), alcohol consumption (yes, no), season of data collection (spring, summer, autumn, winter), physical activity (every day, 3–5 times/week, 1–2 times/week, 1–2 times/month, never), smoking status (yes, no), and family socio-economic index (low, medium, high status). The prevalence of obesity [35
] was calculated for each quintile of the dietary indices. A trend test was conducted by logistic regression analysis, including the ranked index score as a continuous variable. Cronbach’s alpha [36
], which is a function of the correlation and the number of items in a scale [37
], was calculated to estimate the internal consistency of the dietary indices. The degree of agreement in ranking classification of the dietary indices was evaluated with calculation of the weighted kappa coefficient (κw
) using the formula [38
A cross table (5x5) of frequencies was calculated to derive the observed proportion of agreement (Ow) and the expected proportion of agreement by chance (Cw). The weighting factors were 1 for complete agreement, .75 for people differing one category, .5 for people differing two categories, .25 for people differing three categories, and 0 for complete disagreement. P-values less than .05 and non-overlapping 95% confidence intervals were considered statistical significant.