Study design and study population
The target population of the German Health Interview and Examination Survey for Children and Adolescents (KiGGS baseline study) consists of all children and adolescents aged 0 to 17 years, with the exception of those in hospitals, state institutions, or foster homes. The survey was conducted by the Robert Koch Institute. The design, sampling strategy, and study protocol have been described elsewhere in detail [32]. Briefly, the sample was drawn using a two-stage clustered and stratified sampling procedure. In the first stage, 167 sample points representative of German communities were selected with regard to community size and federal state. In the second stage, for every age, participants were randomly selected from the local population registries. The survey was approved by the Federal Office for Data Protection and the Charité-Universitätsmedizin Berlin ethics committee. Participants aged 14 years or older and all parents provided written informed consent before the interview and examination procedures. The overall response rate was 66.6% [32].
Data collection
The 167 sample points were covered by four study teams between May 2003 and May 2006. Parents and participants older than 11 years of age were asked to complete different self-administered questionnaires in the study centres [33]. These included questions on socio-demographic characteristics as well as health and health related behaviours. In addition, participants underwent a computer-assisted medical interview and a physical examination (e.g. body weight and height measurement) conducted by trained staff. Lastly, non-fasting blood and urine samples were also obtained [34].
Dietary assessment
Participants aged 12–17 years were further asked to complete a self-administered, semi-quantitative food frequency questionnaire (FFQ) as well. To cover the most relevant food groups for this population group, the semi-quantitative FFQ was developed by the Robert Koch Institute in consultation with several experts in the field of dietary assessment among children and adolescents. The development of the FFQ is described in detail elsewhere [35]. The FFQ was validated against the modified dietary history method DISHES (Dietary Interview Software for Health Examination Studies) and showed fair to moderate ranking validity for food intake amounts for most of the food items (Spearman correlation coefficients from .35 to .69 with most values above .50) [22]. The FFQ included questions on the average food consumption frequency, as well as the average consumed portion size, for 45 food items in the last few weeks. Categories for frequencies were identical for all food items: never; once per month; 2–3 times per month; 1–2 times per week; 3–4 times per week; 5–6 times per week; once per day; 2–3 times per day; 4–5 times per day; more than 5 times per day. Food-specific portion sizes were assessed by five categories and often illustrated with pictures, e.g. using standard household measures (cups, spoons, etc.). Food frequency information was recoded into frequency consumption of these foods per month (1 month was set equal to 4 weeks; for example, once per week = 4, once per day = 28, more than five times per day = 168). For frequency bands such as one or two times per day, the arithmetic mean was used. Portion sizes were converted into equivalent gram amounts using the standard portion sizes provided in the FFQ. The average food intake was then calculated by multiplying the recoded frequencies and portion sizes (average food intake = food frequency (per month) x portion size (g)).
If the frequency of consumption was given, but information on portion size was missing, the middle category of portion size provided in the FFQ was imputed as it represents the most frequently chosen portion size for this age group. If the food frequency was missing, then the food item was considered as not having been consumed (average food intake = zero).
Food items were grouped into 34 food groups, according to a former dietary pattern analysis based on a modified diet history interview (DISHES) in a subgroup of this study population [36].
Total energy intake was calculated by multiplying the intake and mean energy contents of the FFQ food items. The energy content of every food group was calculated based on weighted estimates of consumption frequencies of specific foods within the food groups based on the comprehensive DISHES data (e.g. several different breads on the total amount of bread).
Assessment of biomarkers
Several biomarkers were measured using the blood and urine samples collected in KiGGS. Venous blood samples were obtained from the participants if the parents and the adolescents themselves gave consent. Serum was separated and transported on dry ice to a central laboratory according to a standardised protocol. Samples were kept at -40 °C until analysed. Pre-analytic and analytic standards have been previously described in detail [34].
For the present analysis, the available indicators of nutrient status were selected, such as serum vitamin B12, serum folate and serum ferritin. These were analysed using electrochemiluminescence-immunoassay (Elecsys E 2010, Roche Mannheim, Germany). During the survey, the manufacturer changed the method for measuring folate. A conversion factor could not be applied; therefore, separate analyses were performed for the two measurement methods (serum folate 1, serum folate 2).
Furthermore, for this analysis, biochemical and physiological cardiovascular risk factors including blood pressure, total serum cholesterol, low density and high density lipoprotein cholesterol (LDL-C and HDL-C), homocysteine, uric acid, and HbA1c were also selected. Standardised measurements of systolic and diastolic blood pressure were obtained using an automated oscillometric blood pressure device (Datascope Accutorr Plus) [32, 34, 37]. The means of two independent readings for systolic and diastolic blood pressure were used. Total cholesterol was analysed using an enzymatic assay (cholesterol oxidase-PAP method, Roche Mannheim, Germany). LDL-C and HDL-C were determined directly with a homogenous enzymatic colorimetric assay (Roche Mannheim, Germany). Homocysteine was measured with fluorescent particle immunoassay (Abbot, Wiesbaden, Germany). Uric acid was determined by the uricase-PAP method (Hitachi 917; Roche Mannheim, Germany). HbA1c was analysed using high-performance liquid chromatography (Diastad; Biorad, Munich, Germany) [34].
Assessment of anthropometric markers
Body height was measured without shoes, with an accuracy of 0.1 cm, by trained staff using a portable Harpenden stadiometer (Holtain Ltd.; Crymych, UK). Body weight was measured while participants were only wearing underwear and no shoes, with an accuracy of 0.1 kg, using a calibrated electronic scale (SECA, Birmingham, UK). Body mass index (BMI) was calculated as body weight (in kilograms) divided by body height squared (in meters) [32].
Assessment of other variables
Within KiGGS, health related information was assessed through self-administered questionnaires. Regular alcohol consumption was defined as drinking at least one glass of beer, wine, or liquor per week. Smoking habits were assessed with the following question: ‘Do you currently smoke?’ ‘daily’, ‘several times a week’, ‘once a week’, ‘more seldom’ or ‘no’ [38]. This variable was categorized into ‘yes’ or ‘no’ (with only those adolescents who never smoke being categorized into “no”). Regarding physical activity, adolescents were asked: ‘In your leisure time, how often are you physically active in such a way that you start to sweat or become slightly out of breath?’. The subsequent question: ‘How many hours per week?’ was used in this analysis.
Medication and supplement use during the last 7 days prior to the interview was determined with a standardised computer-assisted interview conducted by trained physicians. Adolescents were present at the standardised computer-assisted interview where primarily the parents were asked: ‘Has your child taken any medications in the last 7 days? Please also mention the use of any ointments, liniments, contraceptive pills, vitamin and mineral supplements, medicinal teas, herbal medicines or homoeopathic medicines’. Parents were asked in advance to bring prescriptions or original containers to the study centre for the purpose of verification [39, 40].
Study population
For the present analysis, all adolescents between 12 and 17 years were selected from the KiGGS survey sample (a total of 2953 boys and 2801 girls). Of these, 292 participants were excluded because they did not provide a blood sample. 263 participants were further excluded because they had no, incomplete (more than twenty missing values), or implausibly high total food intake data. That is, if the estimated total food intake exceeded 10 kg/day, the total beverage intake exceeded 15 l/day, or food intake exceeded 4 kg/day, combined with beverage intake above 6 l/day. Lastly, two girls who were pregnant were also excluded because pregnancy influences the biomarker profile. Therefore, the final analysis is based on a total sample of 5197 adolescents (2646 boys and 2551 girls).
For specific analyses, persons with missing values (for specific serum variables, for instance) were excluded. Moreover, persons with diabetes mellitus or taking diabetes medication were excluded from the analysis of HbA1c (9 participants); persons taking cardiovascular medication from the analysis of blood pressure (179 participants); persons with vitamin supplement use from the analysis of folate and vitamin B12 (296 participants); and persons with mineral supplement use from the analysis of ferritin (115 participants). Because hormonal contraceptives influence blood lipids [41], girls with hormonal contraceptive use were excluded from the analysis of blood lipids (397 participants).
Statistical analyses
Statistical analyses were conducted for boys and girls separately using the SAS version 9.4 (SAS Institute Inc., Cary, USA).
To correct for non-response and disproportional sampling, a weighting factor was used for all analyses. Since the sample is based on a clustered and stratified design, all analyses were performed with complex survey procedures. Differences with p-values <0.05 were considered statistically significant.
Selected study characteristics were calculated using PROC SURVEYMEANS and PROC SURVEYFREQ. Linear regression models with PROC SURVEYREG were used to examine mean values (with 95% CI) of biomarkers according to dietary pattern scores. For the analysis, dietary patterns were divided into quintiles and acted as independent variables. The first model (Model 1) was only adjusted for age. The second model (Model 2) was adjusted for age, physical activity (hours per week, continuous), smoking status (yes/no), regular alcohol consumption (yes/no), and mean caloric intake in kcal/day (continuous). The analyses concerning blood lipids and blood pressure were additionally adjusted for BMI in kg/m2 in Model 2. Socioeconomic status influences eating behaviour and, consequently, biomarker status. Therefore, socioeconomic status is part of the chain of causation and was, thus, not considered as a confounder. To examine age differences, additional analyses were conducted stratified by two age groups (12–14 years, 15–17 years) and adjusted according to Model 2. Trend tests were conducted by including the mean score of each pattern quintile as a continuous variable into the models.
Dietary patterns were identified separately for boys and girls using principal component analysis (in SAS: PROC FACTOR method = prin) on 34 food groups. For each food group, the mean amount of grams per day was standardised to a mean of 0 and standard deviation of 1 (z-transformation).
The resulting components were linear combinations of the included variables and explained as much of the variation in the original variables as possible. The components were rotated by an orthogonal transformation (resulting in uncorrelated components) to achieve a simpler structure with greater interpretability. To identify the number of principal components to be retained, the following criteria were used: the criterion of eigenvalues exceeding 1 (the interpretation of this criterion being that each component should explain a larger amount of variance than a single standardised variable in order to be retained), the scree plot (a graphical presentation of eigenvalues), and the interpretability of each component (dietary pattern) [42]. For good interpretability of each component, an adequate number of food groups with high loadings within a component were necessary. According to Hatcher 2007, components with at least 3 ‘significant’ loadings, which were loadings greater than or equal to |0.4|, were selected [42]. Each obtained component represents a linear combination of all food groups, which were weighted by their factor loadings. Higher factor loadings indicate that the food variable contributes more to the development of the component. Each participant had a score for all identified dietary patterns, which were standardised to a mean of 0 and a standard deviation of 1. These scores rank individuals according to the degree to which they conform to each food consumption pattern. The pattern scores were labelled according to the food groups with high loadings.