The construction of dietary patterns is an alternative to the use of single nutrients or food items in studies of children's food intake. Country-specific dietary patterns have recently been identified for children and adolescents in several European countries 1 2 3 4 5 6 7 8 9 . The dietary pattern approach may illuminate diet-disease associations that are not revealed when single nutrients or food items are used alone 10 11 . Dietary patterns can also be helpful in evaluating adherence to certain diets over time and risk of disease 11 . The most commonly used method for dietary pattern identification is principal component analysis (PCA), which constructs new linear factors by grouping together correlated variables 12 .

In epidemiology, tracking is defined as the stability or maintenance of a given variable over time 13 . Accordingly, dietary tracking may be regarded as maintenance of nutrient intakes, food consumption or dietary habits over a certain time period 4 . Food intake and dietary habits normally change during childhood and adolescence due to individual factors like physiological development, changes in parental influence and social and environmental changes 14 . Dietary habits have mainly been tracked in adults or between adolescence and adulthood, while few studies have tracked diet during childhood 15 and from childhood to adolescence 4 6 14 16 17

It has been difficult to demonstrate a consistent relationship between BMI in children and dietary factors like fat intake or total energy intake in cross-sectional studies and longitudinal observation studies 18 19 20 . While most cross-sectional studies have found negative associations between energy-rich foods and overweight, some longitudinal studies have reported a positive association between weight gain over time and consumption of high-fat foods and sugar-sweetened drinks 21 22 . A comprehensive review of studies examining the relationship between dietary intakes, eating behaviours and childhood obesity concluded that more research is needed, particularly in the form of studies exploring the joint effect of multiple dietary behaviours 23 . Two recently published studies have documented a positive relationship between energy-dense diets and overweight in children 24 25 . Further longitudinal studies exploring the association between complex diets and overweight in children are needed.

We have previously reported the association between PCA-identified eating patterns and overweight in a cross-sectional study in 9- to 10-year-old children in Telemark County, Norway 3 . The present study is a follow-up of the same children three years later, in early adolescence (7^th grade, 12 to 13 years of age). The aim of the study was to a) describe eating patterns in early adolescence and b) to determine the association between eating patterns and overweight from middle childhood to early adolescence.

Four distinct eating patterns were identified at both time points (n = 924 and n = 800 for the 4^th and the 7^th grade, respectively), which were comparable over time (Table 1). The main composition (high-loading items) obtained in the 4^th grade was maintained for all four eating patterns in the 7^th grade (Table 1). However, some modifications due to changes of items between patterns could be observed, mainly related to items with low factor loadings (between -0.30 and +0.30). A separate PCA performed on the follow-up sample (n = 427) showed only minor differences in eating pattern composition and respective factor loadings compared to the whole sample. The changes observed over time were mainly related to the "varied Norwegian" pattern. Meat for dinner and juice during and between meals, which were part of this pattern in the 4^th grade, did not have high loadings for the pattern in the 7^th grade, while several main meals and food items were added to the 7^th-grade pattern: breakfast, lunch, rice, water, jam/honey and margarine/butter on bread (Table 1). Furthermore, low-fat white cheese and yoghurt products, which loaded positively in the "varied Norwegian" pattern in the 4^th grade, loaded positively in the "dieting" pattern in the 7^th grade. In fact, all yoghurt products loaded positively in the "dieting" pattern in the 7^th grade. The main change in the "snacking" pattern was the addition of sugar-sweetened, carbonated soft drinks and "eating between meals" at the 7^th-grade stage (Table 1). Only small changes were observed for the "junk/convenient" pattern over the time period studied (Table 1).

In the 4^th grade, the "snacking" pattern explained the largest amount of variance in the data, followed by the patterns "junk/convenient", "varied Norwegian" and "dieting". In the 7^th grade however, the "junk/convenient" pattern and the "varied Norwegian" pattern accounted for the largest variations in overall diet. The names or 'labels' that were used to describe the patterns in the 4^th grade were highly appropriate to the patterns extracted in the 7^th grade, and seemed to describe the composition of the 7^th-grade patterns even better than the composition of the 4^th grade patterns. The "varied Norwegian" pattern was more prominent at the 7^th-grade stage, as it explained a larger proportion of the variance in intake frequencies and included a larger variety of foods, drinks and main meals.

The Pearson's correlation coefficients for factor scores for the four corresponding eating patterns at the two time points ranged from 0.44 to 0.60 (Table 2). Additionally, the pattern scores for "snacking" and "dieting" in the 4^th grade were inversely related to "varied Norwegian" scores in the 7^th grade (Table 2). The weighted κ-values for being in the same tertile at the two time points were 0.50, 0.49, 0.37, and 0.44 for the "junk/convenient", "varied Norwegian", "snacking" and "dieting" eating patterns, respectively.

At both time points, 50% of the participants were boys and 50% were girls. Weight and height were obtained for 955 (50% boys and 50% girls) of the 1,045 participating children at the 4^th-grade stage (91%), and for 865 (49% boys and 51% girls) of the 1,095 at the 7^th-grade stage (79%). For children with complete weight and height measurements at both time points (n = 540), BMI correlated with r = 0.82 (95% CI: 0.79, 0.84). The distribution between normal weight, overweight and obesity was 80%, 16% and 4% in the 4^th grade and 81%, 15% and 3% in the 7^th grade, respectively. While the incidences of overweight/obesity among boys and girls did not differ in the 4^th grade (21% for boys and 20% for girls (p = 0.851)), a significant difference was seen in the 7^th grade (21% for boys and 15% for girls (p = 0.001)).

In the follow-up sample (n = 427), 345 (80%) of the children were still of normal weight (48% boys, 52% girls), while 20 (5%) changed from overweight to normal weight (55% boys, 45% girls), 21 (5%) changed from normal weight to overweight (71% boys, 29% girls) and 41 (10%) remained overweight (49% boys, 51% girls). Cohen's Kappa for being in the same weight group at both time points was 0.61 (95% CI: 0.50, 0.72).

At the 7^th-grade stage (n = 800), the "varied Norwegian" pattern scores were negatively associated with maternal overweight and positively associated with physical activity and with maternal and paternal educational level. The "snacking pattern" scores were positively associated with sedentary behaviour and negatively associated with family income and maternal and paternal educational level. The dieting pattern scores were positively associated with maternal educational level and were higher for girls than boys, while the "junk/convenient" pattern scores were higher for boys than girls (p < 0.05 for all, data not shown).

Cross-sectional analysis of dietary patterns and overweight in the 7^th grade showed the highest incidence of overweight in the lower tertile of the "varied Norwegian" pattern (23%) and the upper tertile of the "dieting" pattern (24%) (Table 3). The lowest incidence of overweight was observed in the upper tertile of the "varied Norwegian" pattern (12%) and in the two lower tertiles of the "dieting" pattern (13%). Children ranked in the upper tertile of the "varied Norwegian" pattern were less likely to be overweight than those in the lower tertiles (Table 3), in contrast to what was found in the 4^th grade, where high scores for the "varied Norwegian" eating pattern were associated with an increased risk of overweight 3 . Studying the association between single foods and meals and overweight at both time points showed a lower intake of vegetables and less regular breakfast eating among overweight compared to normal weight children at the 7^th grade stage.

Independently of this, children ranked in the upper tertile of the "dieting" pattern were more likely to be overweight than those in the lower tertiles in the 7^th grade. After adjustment for influential confounders in the model, the associations remained significant (Table 3). No significant associations were observed between either the "junk/convenient pattern" or the "snacking pattern" and overweight in the 7^th grade.

Multiple linear regression analysis of the relationship between changes in eating pattern scores and changes in BMI categories over time revealed no significant associations, but weak trends in the data could be observed (Table 4). Compared to children who remained within the "normal weight" category at both time points, children who changed from normal weight to overweight tended to reduce their "junk/convenient" and "snacking" patterns scores and increase their "dieting" pattern scores, while children who changed from overweight to normal weight tended to increase their "snacking" pattern scores. Furthermore, those who were overweight at both time points tended to reduce their scores for the "varied Norwegian" pattern and increase their scores for the "dieting" pattern, compared to the children who stayed normal weight between the 4^th and 7^th grade (Table 4).

Finally, we examined the likelihood of remaining in the overweight category over time relative to unchanged/increased adherence to a pattern (no change or positive change in scores between the 4^th and 7^th grade), or reduced adherence to a pattern (negative change). Children with stable or increased "varied Norwegian" pattern scores showed a significantly lower risk of remaining in the overweight category, compared to children with decreased "varied Norwegian" pattern scores. The association was independent of parental characteristics and child activity/inactivity (Table 5).

The most important finding in the present study was that children adhering to a "varied Norwegian" eating pattern over time had lower risk of remaining overweight than children with declining adherence to this pattern. Cross-sectional analysis of data obtained at the 7^th-grade stage evidenced a significantly lower risk of being overweight for children with high "varied Norwegian" eating pattern scores, and an increased risk of being overweight for children with high "dieting pattern" scores. Neither cross-sectional nor longitudinal data analysis results indicated an increased risk of overweight for children with high scores for eating patterns characterised by the frequent intake of unhealthy food items.

Tracking was observed for all eating patterns between the two time points. Strong correlation coefficients were observed between factor scores for similar patterns at the two time points, suggesting reasonable stability in individual eating habits over the time period studied. However, significant inverse correlations were observed between two of the 4^th-grade patterns and "varied Norwegian" in the 7^th grade, possibly indicating that children with high 4^th-grade scores for the "snacking" or "dieting" patterns had lower 7^th-grade scores for the "varied Norwegian" pattern, and consequently less frequent consumption of the food items included in the dietary guidelines. Likewise, the positive correlation between the "varied Norwegian" pattern in the 4^th grade and the "dieting" pattern in the 7^th grade could indicate a shift toward more frequent use of fat- and sugar-reduced food and drink items (light products) by many children over the time studied. The weighted κ-values confirmed moderate individual tracking 31 of the "junk/convenient", "varied Norwegian" and "dieting" patterns, and fair tracking of the "snacking" pattern, during the three-year follow-up period. A combination of stability and changes in eating patterns of children and adolescents has also been reported in previous studies 4 6 14 15 32 . The present study supports the proposition that eating habits are tracked from middle childhood into early adolescence, but also emphasise the important changes in children's food habits that occur during this particular period of life.

The correlation between BMI measurements at the two time points was high. In the follow-up sample, the majority of children were still of normal weight (80%), 5% of the children moved from overweight to normal weight, 5% moved from normal weight to overweight and 10% remained overweight. A number of studies have investigated BMI tracking during childhood and adolescence 33 34 and from childhood and adolescence into adulthood 35 36 . The results consistently show a high degree of tracking from childhood to adolescence, a finding supported by the present study.

An important finding in the present study was that children with a high adherence to the 'varied Norwegian' pattern in the 7^th grade were less likely to be overweight than children with a low adherence to this pattern, and that children with no change or a positive change in this pattern over time were less likely to remain overweight. It is of particular interest to note that while high scores for the "varied Norwegian" eating pattern were associated with an increased risk of overweight in the 4^th grade 3 , high scores for the same pattern in the 7^th grade were associated with a decreased risk. This change in direction may partly be explained by the changes observed in the composition of the "varied Norwegian" pattern over time, but it is more likely to be explained by individual changes in eating habits by age. Our results showed lower intake of vegetables and less regular breakfast eating among overweight compared to normal weight children at the 7^th grade stage. A decreased intake of vegetables and more irregular meal patterns by age have been observed in a nationwide study of Norwegian adolescents 37 . In our study, this trend was only observed among overweight children. The positive association between the "varied Norwegian" pattern and BMI status was evident independent of parental characteristics, gender and physical activity of the child, and corroborates current dietary guidelines which recommend eating a variety of foods including vegetables, fruit, unrefined cereal products and fish and having regular main meals. Regular meals and, especially, regular breakfast intake, have been associated with a reduced risk of overweight in several studies 7 38 39 40 41 42 . A study of food and activity patterns in a longitudinal study of Greek children confirmed that eating frequency, breakfast consumption and adherence to a healthy diet characterised as Mediterranean were negatively associated with overweight 7 . In a study of Norwegian adolescents, in which Telemark was one of two participating counties, eating four (rather than fewer) meals per day was significantly negatively related to being overweight 40 .

It has been difficult to demonstrate an increased risk of overweight and obesity for children with a high adherence to unhealthy dietary patterns 5 8 19 23 . However, the British ALSPAC study reported a positive relationship between high-energy diets at ages five and seven and overweight at age nine 24 . In the European DONALD study 25 , a small but positive association between consumption of high-energy convenience foods and bodyweight was found among boys. In the present study, we were unable to demonstrate any significant associations with overweight for the eating patterns characterised by frequent intake of unhealthy food items through either cross-sectional or longitudinal data analyses. This may be due to misreporting, including over-reporting of healthy food items and/or under-reporting of unhealthy food items. It is known that food and drink items perceived as 'unhealthy' are under-reported more often than other foods, especially if the children are overweight 43 44 45 . Another explanation could be that 12- to 13-year-old children eat unhealthy food items when their parents are not aware of it. On the other hand, children that are lean and physically active may tolerate a higher intake of high-energy foods, and are not as likely to under-report these items. Any misreporting is likely to have attenuated the association between eating behaviour and overweight.

Dieting behaviour has been associated with overweight in adolescence 46 47 . In the present study, children with high adherence to a "dieting" pattern were more likely to be overweight than those with low adherence to this pattern in 7^th grade as well as in 4^th grade. Many so-called "light" products or typical dieting products are intended as substitutes for food items with a high fat or sugar content. Our results did not show any benefit with regard to weight reduction in children having stable or increased scores on the dieting pattern over time. Rather than encouraging weight control mainly through the increased use of fat- and sugar-reduced food products, parents should strive to include more unrefined plant foods, fish, water, and regular meals in their family diet.

The objective measurement of the weight and height of the children at both time points is a major strength of the present study. Self-reported data have been shown to underestimate overweight prevalence in adolescents 48 . Other strengths of the present study include identical data collection at both time points and inclusion of a broad range of background variables, which are likely to capture a significant proportion of the variability in socioeconomic status and health behavior. The longitudinal design and the simultaneous assessment of BMI, diet and physical activity are important in order to evaluate the influence of dietary behavior and physical activity over time. However, our study also has limitations. A problem common to papers using PCA is that the factors are identified based on scree plots, eigenvalues and the interpretability of the components, which involves a good deal of subjectivity. This does not mean that it is not a good method, but the reader must be aware of these limitations. The FFQ did not include portion sizes, and calculating energy and nutrient intakes was not feasible. Consequently, we could not evaluate the validity of the reported intakes based on total energy intake. However, the reproducibility and validity of PCA-derived dietary patterns assessed using FFQs have previously been found to be satisfactory 49 . The FFQ was completed by the parents, and it is likely that the dietary data reflect the parents' "dietary image" rather than the true habitual diet of the children 50 . Parental reporting was chosen to avoid different data collection conditions at the two time points and to reduce under-reporting, which is thought to be common in adolescents 51 . Even if the number of children participating at both time points was reasonably large, there was a loss to follow-up, which was mainly due to incomplete data. The incidence of overweight was lower in the 7^th grade than in the 4^th grade. The parents did not receive any feedback regarding the child's weight status after the 4^th grade study in order to reduce the risk of bias in how parents of overweight children reported the dietary behaviour in the follow-up study. However, we cannot exclude the possibility that the decrease in overweight could be due to underrepresentation of overweight and obese children in the follow-up study, especially among girls. Greater focus on body image and weight control 51 in adolescence may be a reason for lower participation by overweight girls in the 7^th grade.

To provide a better understanding of the longitudinal relationship between diet and overweight in adolescents, additional research is needed. We intend to repeat the study when the children reach the age of 15 to 16 years.

We observed slight to moderate stability of eating patterns between middle childhood and early adolescence, but important changes in individual eating habits were also observed. Children adhering to a "varied Norwegian" eating pattern over time were less likely to remain overweight than children with declining adherence to this pattern. The association was evident independent of parental characteristics, gender and the physical activity levels of the children. Our results indicate that in order to reach normal weight, overweight children should be encouraged to eat regular main meals and retain a diverse diet including unrefined plant foods, water and fish, rather than fat- and sugar-reduced foods and drinks.

BMI: Body Mass Index; FFQ: Food Frequency Questionnaire; PCA: Principal Component Analysis; IOTF: International Obesity Task Force; CI: Confidence intervals.

The authors declare that they have no competing interests.

IMO and MVS were responsible for the study design and data collection. MVS was responsible for the statistical analyses. ALB and IMO assisted in the statistical analyses, and ALB contributed to the interpretation of results. IMO drafted the manuscript. All authors read and approved the final manuscript.