With principal component analyses, we identified three dietary patterns among boys and two among girls in a population-based sample of German adolescents. Dietary patterns showed significant associations with nutrient intake. Because of the higher densities of vitamins, minerals and fibre, the 'healthy' patterns are more favourable compared to the 'western' and 'traditional and western' patterns which were associated with higher energy density, higher percent of energy from unsaturated fatty acids, lower percent of energy from carbohydrates and lower nutrient densities of several vitamins and minerals. The 'traditional' pattern was characterised by favourable as well as less favourable aspects.
Most of the dietary patterns were associated with health related lifestyle characteristics e. g. physical activity and frequency of family meals. Furthermore, the 'western' and 'healthy' dietary patterns among boys and the 'traditional and western' pattern among girls were correlated to socioeconomic status.
The identification of dietary patterns in a representative sample of the adolescent population and their relation to socioeconomic status, nutritional behaviour and nutrient intake has been rarely examined. Empirical evaluated dietary patterns are specific for the examined study population and reflect culturally influenced eating habits. Furthermore, studies used different methods e. g. for food grouping. Therefore, deviations between dietary pattern compositions are obvious. Nevertheless, the patterns found among German adolescents were, to some extent, similar to those found in previous studies in this age group in other countries. Comparison of dietary patterns among adolescents between different countries can give useful information on similarities in food consumption behaviour between different populations. For such comparison however cultural and economic conditions should be similar so that it is likely that the food supply is similar. If similar dietary patterns are found in such countries, it could be interesting to obtain further information concerning the associations between patterns and health in this age group.
Our 'western' pattern was partly comparable to the 'western' patterns found among adolescents in other countries e. g. in the Western Australian Pregnancy Cohort Study (Raine Study)  and in the Korean Nutrition Health and Nutrition Examination Survey (KNHANES)  and also to the 'high fat and sugar' pattern found in a representative sample of adolescents in the 1995 Australian National Nutrition Survey . All these patterns were characterized by pizza, hamburger, soft drinks and meat or meat products. In contradiction to our study, alcohol consumption was not analysed in those studies. In Germany, we observed that adolescents already consume relevant amounts of alcoholic drinks .
Our 'healthy' patterns were to some extent comparable to other 'healthy' patterns found among adolescents e. g. in a population based sample of Spanish adolescents, and in samples of adolescents in Finland (Cardiovascular Risk in Young Finns Study), Japan (Japanese female dietetic course students), and Australia (Raine Study) [4, 18, 23, 40] and to the 'fruit, salad, cereals, and fish pattern' found in the 1995 Australian National Nutrition Survey .
Our findings on the association between dietary pattern and nutrient intake were especially similar to results of the 1995 Australian National Nutrition Survey . Adolescents in Australia with a higher score of the 'fruit, salad, cereals, and fish' pattern had a higher energy-adjusted intake of dietary fibre, beta carotene, folate, vitamin C and protein similar as German adolescents with higher scores of the 'healthy' patterns. The 'high fat and sugar' pattern was associated with a lower energy adjusted intake of dietary fibre, folate and iron .
Among German adolescents, despite the high intakes of meat in the 'western', 'traditional' and 'traditional and western' patterns, iron density decreased with increasing scores of these patterns, whereas iron density increased with increasing scores of the 'healthy' patterns. This can be explained by the fact that meat and processed meats are less important sources of iron than bread, juices, vegetables, fruits and breakfast cereals among German adolescents . However, this does not consider differences in bioavailability of iron between animal and vegetarian foods. Furthermore, white bread, a characteristic component of the 'traditional' pattern, has lower iron content than whole grain bread.
Most former studies [23, 27, 40] determined dietary patterns for boys and girls together. Among German adolescents, we found different dietary patterns between boys and girls. Even between the 'healthy' dietary patterns, differences concerning the food group loadings are notable. Further differences between boys and girls were found. Firstly, for boys the most predominant dietary pattern was the 'western' pattern, among girls the 'healthy'. Second, among boys many associations between the 'healthy' dietary pattern and socioeconomic or behavioural factors (e. g. physical activity, frequency of joined family dinner) were found. Among girls there was no association between socioeconomic or behavioural factors and the 'healthy' dietary pattern. Thirdly, the two predominant dietary patterns found among German girls explained only 13.2% of variance in food group intake, while among German boys, the three predominant dietary patterns explained 18.1% of variance. This may indicate that there is more individual diversity within the diet of girls or that boys have more often similar food preferences. Fourthly, particularly notable among German boys was the increasing age from the lowest to the highest quartile of the 'western' pattern. Among girls, with increasing scores of the 'traditional and western' pattern age was decreasing. The present study used a cross sectional design. Thus, we are not able to evaluate long term changes. But it may be assumed, that boys change their eating behaviours during adolescence toward a less favourable dietary pattern, whereas girls do not.
Among U.S. middle and high school students in Minnesota (Project EAT)  and students in Vyronas region, Athens  high scores of 'take away food' or 'fast food' patterns were associated with increasing age among both sexes.
Previous studies investigated the relationship between dietary patterns and socioeconomic parameters. In a representative sample of adolescents in Australia and among Greece students, no association was found [21, 27]. In a German study, there was no association between clusters of fat intake patterns and mother's educational background . In Australia (Raine Study), Finland (Cardiovascular Risk in Young Finns Study), and Scotland (The Survey of Sugar Intake among Children in Scotland), the 'healthy' or 'vegetables' patterns were associated with higher education of the mother , both parents  or the parent who mainly provides the food . The 'western' pattern in the Australian Rain study and the 'traditional' pattern in Finland (Cardiovascular Risk in Young Finns Study) were associated with lower educational background. Furthermore, in Scotland (The Survey of Sugar Intake among Children in Scotland), the 'vegetables' pattern was associated with higher socioeconomic status and higher household income  and among adolescents of the Balearic Islands higher scores of the Mediterranean pattern were associated with higher parental socio-economic status among girls .
With increasing 'healthy' pattern scores, the percentages of boys taking dietary supplements increased. Among girls, this increase was not significant. Previous studies in Japan (Japanese female dietetic course students)  and Finland (Cardiovascular Risk in Young Finns Study)  found an increasing percentage of supplement users with increasing scores of healthy dietary patterns among both sexes.
In Finland (Cardiovascular Risk in Young Finns Study) and Japan (Japanese female dietetic course students), 'traditional' dietary patterns were found more often in rural areas [4, 18], whereas a 'healthy' pattern was observed more often in urban areas , which follows our results. Dietary patterns are very likely influenced by local food supply. In rural areas of Germany there are less fast food and non-traditional restaurants compared to urban areas. Maybe therefore, boys living in rural areas more often follow a traditional German diet.
Other studies have described an association between a higher level of physical activity and higher adherence to healthier dietary patterns [21, 23, 44], like among boys in our study. Conversely, with increasing hours of watching television, a higher adherence to less healthy dietary patterns (e. g. 'snacky' or 'western'), similar to the association found among German girls with higher scores of the 'traditional and western' pattern in our study, was also previously observed [21, 24, 25, 40].
We also used energy-adjusted dietary pattern scores  (data not presented). This did not essentially change the results concerning the association between dietary pattern scores and socioeconomic and lifestyle characteristics.
In our study, no association between dietary pattern scores and overweight was found. Only few studies have previously investigated the association between dietary patterns and BMI among adolescents. These studies came to different results. In a German study, BMI showed no association to fat intake patterns within a longitudinal analysis . A study in Scotland also found no association between dietary patterns and BMI . Another study observed a lower BMI among those with a healthier dietary pattern and a higher BMI among those with a western dietary pattern . One study even found a higher BMI among persons with a healthier nutrition .
One reason for the different results may be that overweight persons who attempt to lose weight tend to confound the relationship between nutrition and BMI . Furthermore, obese people tend to underreport fatty foods and foods rich in carbohydrates . Besides, it is difficult to detect minor overconsumption of energy in epidemiological studies that track the development of obesity over long periods  and overweight is primarily controlled by energy balance. Another source of error may be the bias of self-reported body weight and height, which often differ from measured figures . In a subsample of KiGGS, validity of self-reported height and weight among 11- to 17-year-old adolescents was assessed. BMI values calculated from self-reported weight and height were lower than those calculated from measured values among both sexes . Thus, we probably underestimate BMI in our study. However, standardized anthropometric measurements were not feasible due to the nature of household visits and interviews taken place in study vans.
Further analysis concerning dietary patterns and BMI were conducted while excluding low energy reporters (about 10% of girls and 7% of boys) . This did not change the results essentially.
Strengths of this study include the utilization of representative population-based data and the comprehensive nutrition interview, which documented food consumption during a period of four weeks, as well as the inclusion of data on nutrition behaviour, lifestyle habits and socioeconomic background.
A limitation of our study is the cross-sectional design, which allows no statements concerning causation.
Potential limitations of the interview-based dietary assessment methods in general are underreporting of usual intake and invalid reporting due to memory gaps and social desirability. DISHES was only validated for adults but in a recent study among adolescents aged 12 to 17 years it showed fair to moderate ranking validity with food group intake assessed with a FFQ . In addition, a pretest on feasibility of the use of DISHES in the age group of 12 to 17 years was conducted, which indicated no particular problems for conducting DISHES in this age group. Furthermore, most interviews where carried out at home of the participants, which gave the opportunity to get additional information from the parents (e. g. concerning name of meals, kind of meat).
Another limitation of the study was the self-reported physical activity which may be affected by misreporting, e. g. overreporting influenced by social desirability . However, within the setting of this survey, more objective methods to determine physical activity would be too time-consuming and expensive and where therefore not feasible.
For children and adolescents with a migration background, a higher proportion of unreachable addresses and non-respondents were expected. Thus an oversampling of this population group was performed in KiGGS. Furthermore, invitation letters, information material and questionnaires were translated into six languages . However, for the dietary assessment in EsKiMo, certain basic German language skills were required. Thus, the proportion of participants with a migration background in EsKiMo was somewhat lower than in KiGGS. However, 6.8% of the EsKiMo participants had a one-side and 11.3% a two-side migration background (unweighted percent). Nevertheless, separate analysis concerning migration background was not reasonable in EsKiMo because of a relatively small sample size which included only 65 boys and 79 girls with a migration background which came from several countries.
Principal component analysis is generally exposed to the limitation of some subjectivity, particularly when grouping the food items, selecting the method of factor rotation, defining the number of patterns to be retained and labelling of the factors . To enable comparability with other studies, we used criteria similar to those reported by other dietary pattern analyses [14, 21]. In addition, we tested a further version of food grouping with 32 food groups instead of 48. This led to very similar results concerning the number of principal components to be retained and the food groups with high loadings within patterns. Correlation coefficients between pattern scores determined with 48 food groups and those determined with 32 food groups ranged between 0.93 and 0.97. McCann also noted that food classification method affected neither the number nor character of the identified patterns .
Dietary patterns derived using principal component analysis generally tend to account for only a small amount of the total variance of diet . The variance explained in the present study was similar to a study among 12 to 17 year-old Scottish adolescents, in which 3 dietary patterns based on 141 food groups accounted for 14.4% of variance among boys and 15.1% among girls . A slightly higher variance than in our study was observed among 12 to 18 year-old Australian adolescents, in which 3 dietary patterns based on 86 food groups accounted for 21.7% of the variance in the data .