Childhood obesity is a major public health challenge. At present there is a lack of convincing evidence about suitable and effective strategies for the prevention of childhood overweight. Recently, an obesity prevention evidence framework has been proposed(Reference Swinburn, Gill and Kumanyika1). Key policies include: (i) building a case for action on obesity; (ii) identifying contributing factors and points of intervention; (iii) defining opportunities for action; (iv) evaluating potential interventions; and (v) selecting a portfolio of specific policies, programmes and actions. Therefore, a systematic analysis of determinants of overweight in the micro- as well as the macro-environment is necessary to provide a sound basis for developing strategies against overweight. The systematic analysis should include an analysis of the determinants of overweight prevalence as well as overweight incidence, separately. Childhood overweight (and not only obesity) is predictive for adult morbidity and mortality(Reference Dietz2). In addition, the life-long persistence and health consequences of overweight and obesity in many children suggest a strong need for the prevention of overweight(Reference Dietz2). Primary prevention strategies address the whole population, in particular normal-weight subjects, and are aimed at preventing the incidence of overweight. Therefore, it is important to analyse determinants of incidence. In addition, determinants of the prevalence of overweight need to be addressed by strategies of secondary or tertiary prevention (i.e. treatment of overweight and/or obesity). To our knowledge, determinants of the incidence and prevalence of childhood overweight have not been compared systematically.
Most of our present knowledge is based on cross-sectional data. These studies have investigated the influence of lifestyle determinants on childhood overweight (e.g. lifestyle factors(Reference Dietz and Gortmaker3–Reference Robinson9)), but only few studies have addressed familial, social and lifestyle factors together(Reference Jouret, Ahluwalia and Cristini10–Reference Vogels, Posthumus and Mariman17). In these cross-sectional studies, parental obesity, low socio-economic status (SES), high weight gain during infancy and television (TV) viewing were found as main determinants of prevalence. Contrary to cross-sectional data, there are only very few longitudinal studies investigating the development of overweight(Reference Maffeis, Talamini and Tato14, Reference Davison and Birch18–Reference Valerio, D’Amico and Adinolfi21). In these studies parental overweight was found as the main determinant. In addition, rapid weight gain in early life was found as a significant predictor in two studies(Reference Dubois and Girard19, Reference Reilly, Armstrong and Dorosty20), as was SES(Reference Dubois and Girard19, Reference Valerio, D’Amico and Adinolfi21). In two studies high TV viewing(Reference Reilly, Armstrong and Dorosty20) and high energy intake(Reference Davison and Birch18) were significantly associated with the development of overweight.
Although the complexity of childhood overweight is generally known, interactions between determinants have been considered in only three cross-sectional studies(Reference Kleiser, Schaffrath Rosario and Mensink11, Reference Singh, Kogan and Van Dyck15, Reference Vandewater and Huang16). Here we present a study where we systematically analysed cross-sectional as well as longitudinal data of the Kiel Obesity Prevention Study (KOPS) to characterise individual and ecological determinants of the prevalence as well as the incidence of overweight in children aged 5 to 16 years. The analysis should provide a sound basis to develop strategies for primary prevention as well as treatment of overweight.
Methods
Study populations
Study design and recruitment procedures of KOPS have been described previously(Reference Plachta-Danielzik, Pust and Asbeck22). Briefly, participants were obtained from three groups participating in KOPS. Group 1 was a representative group of 4997 children aged 5–7 years which was recruited as part of the school entry examination in Kiel, Germany between 1996 and 2001. Group 2 consisted of 4487 children aged 9–11 years who were examined during a school examination between 2000 and 2005. Group 3 consisted of 3237 adolescents aged 13–16 years examined in schools between 2004 and 2006. Participation was voluntary and there were no eligibility criteria except willingness to participate. Signed informed consent was obtained and the study protocol was approved by the local ethical committee.
Questionnaires addressing determinants of overweight (answered by the parents for groups 1 and 2, by the adolescents themselves for group 3) were available for 1837 children aged 5–7 years, 2303 children aged 9–11 years and 2109 adolescents. Thus, the total data of 6249 children and adolescents were used to analyse the determinants of prevalence.
Since all three groups belonged to the same total population (=all children participating in the school entry examination between 1996 and 2001 in Kiel), a subgroup of children was identified who had been examined twice within a 4-year follow-up period: (i) subgroup A comprising 1683 children examined at age 5–7 as well as 9–11 years (n 1683); and (ii) subgroup B comprising 9- to 11-year-old children re-examined at age 13–16 years, n 918). For the analysis of incidence, only persistent normal-weight and incident overweight children were considered; 183 and 103 persistent overweight as well as forty-three and fifty-six remitted (e.g. who normalised weight status) children of subgroup A and B, respectively, were excluded from analysis. In our longitudinal analysis complete data sets were available for 1087 children and adolescents (687 and 400 of subgroup A and B, respectively). For analysis of cross-sectional data all children who were investigated twice were considered at one age only. Data of the first examination were used unless the questionnaire of lifestyle habits was missing at the first measuring time but available at the second. Then data of the second measurement were used.
Tanner stages (pubic hair stages for both sexes; breast stages for girls, genitalia stages for boys) were self-estimated by the adolescents using standard pictures(Reference Tanner23) on scales from 2 to 5. This procedure has been validated by Duke et al.(Reference Duke, Litt and Gross24) in forty-three females aged 9–17 years and twenty-three males aged 11–18 years.
Definition of overweight
Height and weight were measured and BMI was calculated(Reference Plachta-Danielzik, Landsberg and Johannsen25). International BMI cut-offs for child overweight (including obesity) were applied using the International Obesity Taskforce standards(Reference Cole, Bellizzi and Flegal26). In addition, waist circumference was measured midway between the lowest rib and the top of the iliac crest at the end of gentle expiration. Fat mass was calculated from tetrapolar bioelectrical impedance analysis measurements using a population-specific algorithm(Reference Plachta-Danielzik, Landsberg and Johannsen25). Children were characterised as ‘overwaist’ and ‘overfat’ according to British reference values(Reference McCarthy, Cole and Fry27, Reference McCarthy, Jarrett and Crawley28) due to missing international and German standards.
Determinants of overweight
Potential risk factors for overweight were assessed using a questionnaire that addressed the following determinants.
Family factors
Parental weight and height were self-reported and parents were classified as ‘normal weight’ (BMI < 25 kg/m2), ‘overweight’ (BMI ≥ 25 kg/m2) or ‘obese’ (BMI ≥ 30 kg/m2). Weight and height of siblings were also self-reported by parents and classified in categories according to international BMI reference percentiles(Reference Cole, Bellizzi and Flegal26, Reference Cole, Flegal and Nicholls29). Occurrence of nutrition-related diseases (hypertension, diabetes mellitus, hypercholesterolaemia, stroke, myocardial infarction) was asked and classified in categories of ‘no’, ‘in grandparents only’ or ‘already in parents’. Parental smoking habits were classified in categories of 0 (‘no’), 1–15 (‘middle’) and >15 cigarettes/d (‘heavy’).
Social factors
SES was determined according to parental education, i.e. highest level attained by either parent: ‘low’ = 9 school years, ‘middle’ = 10 school years, ‘high’ = 12 school years and more. Single parenthood (‘yes’, ‘no’) as well as nationality (‘German’ and ‘non-German’) were dichotomised.
Early life determinant
Birth weight was adopted from the well-baby check-up book and classified into categories (‘low’, ‘middle’, ‘high’) using German reference percentiles(Reference Kromeyer-Hauschild, Wabitsch and Kunze30) taking into account gender and duration of pregnancy.
Lifestyle factors
Physical activity and media time were categorised using age- and sex-specific cut-offs (determined from distribution and recommendations). Regular physical activity was assessed as membership in a sports club and training hours per week (4-week test–retest correlation in 14-year-old adolescents was r = 0·50, P < 0·01 for duration of physical activity(Reference Landsberg, Plachta-Danielzik and Much31)). Physical activity was categorised as ‘very low’ (0 h/week for all age groups), ‘low’ (5–7-year-olds: >0–≤1 h/week; 9–11-year-olds: >0–≤2 h/week; 13–16-year-old boys: >0–≤3·5 h/week; 13–16-year-old girls: >0–≤2·5 h/week), ‘middle’ (5–7-year-olds: >1–≤2 h/week; 9–11-year-olds: >2–≤4 h/week; 13–16-year-old boys: >3·5–≤6 h/week; 13–16-year-old girls: >2·5–≤4·5 h/week) and ‘high’ (5–7-year-olds: >2 h/week; 9–11-year-olds: >4 h/week; 13–16-year-old boys: >6 h/week; 13–16-year-old girls: >4·5 h/week).
Self-reported media time was assessed as hours per day spent in TV viewing and computer use on a typical weekday (4-week test–retest correlation in 14-year-old adolescents was r = 0·68, P < 0·01(Reference Landsberg, Plachta-Danielzik and Much31)). In a previous study on 5- to 11-year-old children(Reference Grund, Krause and Siewers32), TV viewing had been compared with (i) energy expenditure as assessed by the combined use of indirect calorimetry and 24 h heart-rate monitoring (time > FLEX heart rate) and (ii) aerobic fitness (submaximal oxygen consumption, O2-pulse). However, there were no significant differences in either energy expenditure or fitness between groups of children watching TV for ≤1 h/d v. >1 h/d. Daily time spent for media use was categorised as ‘low’ (5–7-year-olds: 0 h/d; 9–11-year-olds: 0–<1 h/d; 13–16-year-old boys: 0–<2 h/d; 13–16-year-old girls: 0–<1·5 h/d), ‘middle’ (5–7-year-olds: >0–≤1 h/d; 9–11-year-olds: ≥1–<2 h/d; 13–16-year-old boys: ≥2–<2·5 h/d; 13–16-year-old girls: ≥1·5–<2 h/d), ‘high’ (5–7-year-olds: >1–≤2 h/d; 9–11-year-olds: ≥2–<3 h/d; 13–16-year-old boys: ≥2·5–<3·5 h/d; 13–16-year-old girls: ≥2–<3 h/d) and ‘very high’ (5–7-year-olds: ≥2 h/d; 9–11-year-olds: ≥3 h/d; 13–16-year-old boys: ≥3·5 h/d; 13–16-year-old girls: ≥3 h/d).
Nutrition was assessed using a twenty-six-item FFQ based on the WHO MONICA FFQ adapted to children(Reference Mast, Körtzinger and Müller33). An index of dietary pattern was calculated(Reference Landsberg, Plachta-Danielzik and Much31). Consumption of ≥3 ‘healthy’ foods (wholemeal bread, fruit, vegetables, fish, cheese) and <3 ‘risk-related’ foods (white bread, sausage, soft drinks, fast food, sweets/chips) at least 3–5 times/week were summarized to a ‘healthy dietary pattern’. Consumption of ≥3 ‘risk-related’ foods and <3 ‘healthy’ foods at least 3–5 times/week corresponded to a ‘risk-related dietary pattern’. Other combinations were mentioned as ‘mixed dietary pattern’. The FFQ was validated against a 7 d diet record in children aged 5–7 years (n 24) and 9–11 years (n 61)(Reference Pust34). Additionally, differences in the dietary pattern index were analysed when either parents or children completed the FFQ. There were non-systematic differences in several food items when compared with parental reports, i.e. healthy as well as unhealthy foods were over- and underestimated by children. Four-week test–retest percentage agreement (reliability) of dietary pattern in 14-year-old adolescents was 67·6 %.
Statistics
The statistical analyses were performed with the SPSS 15·0 for Windows (SPSS Inc., Chicago, IL, USA) and STATA 11 (Stata Corp., College Station, TX, USA) statistical software packages. Results are presented as median and interquartile range.
Multilevel logistic regression analyses were performed to identify independent risk factors for prevalence and incidence of overweight. A multilevel approach was used to account for the hierarchical data structure (level 1: students; level 2: schools) and thus to control for clustering of participants in schools. It was performed with STATA 11 (XTMELOGIT command). Schools were used as random effect, risk factors of overweight were considered as fixed effects. Categorical determinants were converted in dichotomous dummy variables. Reference categories are marked in Table 2. In the first model all potential determinants were considered. In a second model interaction terms between lifestyle factors and age, parental weight status and parental education were considered additionally. Level of significance was set at P < 0·05 (two-sided). Missing values were considered as separate covariates but their estimated values are not presented. Due to small selection biases with respect to the total population (data on BMI provided by school physicians), data were weighted on the distribution of the total population with regard to weight status of the children (in cross-sectional data analyses) and SES (in longitudinal data analyses)(Reference Danielzik, Pust and Landsberg35, Reference Plachta-Danielzik, Bartel and Raspe36). Students who were under-represented in the study population get a higher weight factor for data analysis and vice versa. All analyses were stratified for sex. Age and pubertal stages were considered as confounders.
Additional analyses
Since some studies have found associations between several food items and overweight(Reference James and Kerr5, Reference Pereira, Kartashov and Ebbeling8), we tested the influence of soft drinks, fast food, sweets, fruit and vegetables instead of the dietary index within the logistic regression analysis.
In our previous analysis of determinants of overweight in 5- to 7-year-old children, different determinants were observed between overweight and obesity(Reference Danielzik, Czerwinski-Mast and Langnase37). Therefore, we stratified the analysis by overweight and obesity (according to international cut-offs for BMI(Reference Cole, Bellizzi and Flegal26)).
BMI is widely used as a measure of fat mass. However, BMI is only an indirect parameter of total body fat and does not reflect body fat distribution(Reference Lobstein, Baur and Uauy38). Thus, ‘overwaist’ and ‘overfat’ were used as dependent variables in logistic regression analyses instead of overweight.
In the analyses of determinants of incidence, 4-year changes in determinants were considered. Therefore new categories were created with consistent values as well as inconsistent values (categories of change; with the exception of parental education, birth weight and nationality which were unchangeable variables).
Results
Characterisation of the study populations
The study populations are characterised in Table 1. Overweight prevalence was 18·3 % in boys and 19·2 % in girls. Four-year cumulative incidence rates were 10·1 % in boys and 8·2 % in girls. The distributions of all potential determinants of overweight are shown in Table 2 for the cross-sectional as well as the longitudinal study group at baseline. Within the longitudinal cohort family members more often were normal weight, parents had a better education and lived together more often, compared with the cross-sectional cohort. In addition, the children of the longitudinal cohort were more often German and had more favourable lifestyle behaviours. These differences were due to the fact that the longitudinal cohort consisted of normal-weight children only. In addition a selection bias was obvious and was corrected in multivariate analyses by using weight factors (see Statistics).
Table 1 Characteristics and weight status of children and adolescents of the cross-sectional as well as the longitudinal cohort, Kiel Obesity Prevention Study
IQR, interquartile range; SDS, standard deviation score; WC, waist circumference; FM, fat mass; NW, normal weight; OW, overweight; OB, obese.
*Incident overweight and persistent normal-weight children only.
†According to bioelectrical impedance analysis.
‡According to international BMI reference percentiles(Reference Cole, Bellizzi and Flegal26).
Table 2 Characterization and distribution of potential determinants of overweight stratified by sex, Kiel Obesity Prevention Study
NW, normal weight; REF, category which is used as reference in multivariate analyses (Tables 4–7); OW, overweight; OB, obese; UW, underweight.
*Incident overweight and persistent normal-weight children only.
†Hypertension, diabetes mellitus, hypercholesterolaemia, stroke and/or myocardial infarction.
‡Calculated from FFQ concerning frequency of consumption of healthy and risk-related foods(Reference Landsberg, Plachta-Danielzik and Much31).
Four-year changes in determinants
Within the 4-year follow-up period, 9·9 % and 19·8 % of mothers and fathers became overweight while 13·8 % and 10·9 % of mothers and fathers who were overweight at baseline re-normalised their weight. Among siblings, 16·4 % gained weight and 8·7 % ameliorated their weight status. Some 3·6 % of parents started smoking and 19·9 % of former smokers became non-smokers. Moreover, 17·8 % of children who lived with one parent only at baseline lived with two parents at follow-up and 7·9 % changed from a two- to a one-parent household. Four-year changes in lifestyle variables are presented in Table 3. Overall, 58 %, 68 % and 32 % of the children remained within the same category of physical activity, media time and nutrition, respectively. Children who changed a category more often improved their physical activity and nutrition level but they increased media time consumption.
Table 3 Four-year changes in lifestyle variables of children of the longitudinal cohort stratified by sex, Kiel Obesity Prevention Study
T0, baseline; T1, 4-year follow-up.
Determinants of prevalence (cross-sectional data)
Significant determinants of prevalence of overweight were family, social, early life and lifestyle factors (Table 4). The main determinant with the highest odds ratio was parental obesity (boys: OR = 2·1; 95 % CI 1·5, 3·0; girls: OR = 3·7; 95 % CI 2·7, 5·1). Low physical activity increased the risk of overweight in boys (OR = 1·5; 95 % CI 1·1, 2·0) while high media time was a significant determinant in girls (OR = 1·7; 95 % CI 1·2, 2·4). High birth weight (OR = 1·5; 95 % CI 1·1, 1·9) as well as increasing age (OR = 1·1; 95 % CI 1·1, 1·2) were risk factors of overweight in boys only. Girls of low SES had an increased risk of overweight when compared with girls of high SES (OR = 1·6; 95 % CI 1·2, 2·1). When the model was extended by interaction terms, family and lifestyle factors lost significance while the interaction term between media time and weight status of mothers became significant in boys (OR = 1·2; 95 % CI 1·1, 1·3) as did interaction terms between media time and age (OR = 1·0; 95 % CI 0·9, 1·0) and parental education (OR = 1·2; 95 % CI 1·0, 1·3) in girls (Table 5). Figure 1 illustrates the significant interactions. An increased risk for overweight with increasing media time consumption was obvious for boys of obese mothers (Fig. 1(a)), girls at the age of 5–11 years (Fig. 1(b)) and girls from families of middle and high SES (Fig. 1(c)).
Table 4 Determinants of prevalence of overweightFootnote * stratified by sex derived from multilevelFootnote † logistic regression analysis (model 1), Kiel Obesity Prevention Study
OW, overweight; OB, obese; UW, underweight.
* According to international BMI reference percentiles(Reference Cole, Bellizzi and Flegal26).
† Adjusting for clustering effect in schools; reference categories are given in Table 2. Significance indicated by P < 0·05.
Table 5 Determinants of prevalence of overweightFootnote * stratified by sex derived from multilevelFootnote † logistic regression analysis (model 2), Kiel Obesity Prevention Study
OW, overweight; OB, obese; UW, underweight.
* According to international BMI reference percentiles(Reference Cole, Bellizzi and Flegal26).
† Adjusting for clustering effect in schools; reference categories are given in Table 2. Significance indicated by P < 0·05.
Fig. 1 Prevalence of overweight according to media consumption and stratified by: (a) maternal weight status (
, normal weight; 
, overweight; 
, obese) in boys; (b) child’s age (
, 5–7 years; 
, 9–11 years; 
, 13–16 years) in girls; and (c) socio-economic status according to parental education (
, low; 
, middle; 
, high) in girls, Kiel Obesity Prevention Study
Determinants of incidence (longitudinal data)
Parental obesity (OR = 4·4; 95 % CI 1·5, 13·1), parental smoking habits (OR = 2·5; 95 % CI 1·1, 5·5) as well as low physical activity (OR = 4·1; 95 % CI 1·2, 14·4) were the significant determinants of incidence of overweight in boys (Table 6). In addition, incidence of overweight decreased with increasing age of the boys (OR = 0·8; 95 % CI 0·6, 1·0). Taking into account interaction terms, low physical activity (OR = 27·7; 95 % CI 1·2, 618) remained a significant determinant of incidence (Table 7). In girls, obesity of the father (OR = 6·8; 95 % CI 1·7, 27·9) was the only significant determinant of incidence of overweight (Table 6).
Table 6 Determinants of incidence of overweightFootnote * stratified by sex derived from multilevelFootnote † logistic regression analysis (model 1), Kiel Obesity Prevention Study
OW, overweight; OB, obese; UW, underweight.
* According to international BMI reference percentiles(Reference Cole, Bellizzi and Flegal26).
† Adjusting for clustering effect in schools; reference categories are given in Table 2. Significance indicated by P < 0·05.
Table 7 Determinants of incidence of overweightFootnote * stratified by sex derived from multilevelFootnote † logistic regression analysis (model 2), Kiel Obesity Prevention Study
OW, overweight; OB, obese; UW, underweight.
* According to international BMI reference percentiles(Reference Cole, Bellizzi and Flegal26).
† Adjusting for clustering effect in schools; reference categories are given in Table 2. Significance indicated by P < 0·05.
Additional analyses
When including individual food items (soft drinks, fast food, sweets, fruit and vegetables) instead of the nutrition index none of these items reached significance (data not shown).
When stratifying the analyses according to overweight and obesity the same determinants reached significance, whereas the odds ratios were higher for obesity but also had higher 95 % confidence intervals (data not shown).
Explained variance (Nagelkerke’s R 2) was 14·3 % for determinants of prevalence (for both sexes combined). Data were re-analysed with overwaist and overfat as dependent variable. Explained variance was 11·3 % and 19·2 % for overwaist and overfat, respectively.
Within the analysis of incidence 4-year changes in determinants did not reach significance.
Discussion
Determinants of prevalence
In KOPS parental overweight and obesity were found as main determinants of overweight risk in German children and adolescents (Table 4), as in other studies(Reference Jouret, Ahluwalia and Cristini10–Reference Maffeis, Talamini and Tato14). By contrast, in the literature the impact of lifestyle factors was not uniform. A high media time increased the risk of overweight(Reference Janssen, Katzmarzyk and Boyce6, Reference Robinson9, Reference Kleiser, Schaffrath Rosario and Mensink11). Nutrition and physical activity were not strongly associated with the risk of overweight in multivariate analyses(Reference Janssen, Katzmarzyk and Boyce6, Reference Kleiser, Schaffrath Rosario and Mensink11, Reference Lasserre, Chiolero and Cachat12, Reference Maffeis, Talamini and Tato14).
We found in KOPS that low physical activity as well as high media time increased the risk for overweight (Table 4). However, poor nutrition habits reached no significance as a risk factor but surprisingly entered our analysis as a protective factor. Our finding might suggest a bias in overweight children due to the assessment instrument. However, our FFQ was validated against 7 d food records and a sufficient agreement was found (r = 0·3–0·4 for several food items)(Reference Mast, Körtzinger and Müller33, Reference Pust34). Under-reporting may affect data quality in overweight children. However, two validation studies could not show that under-reporting was common in overweight children only(Reference Champagne, Baker and DeLany39, Reference Sichert-Hellert, Kersting and Schoch40). The inverse effect of nutrition disappeared when interaction terms were taken into account (Table 5). We take this as evidence for a minor effect of nutrition on prevalence of childhood overweight. We found sex differences in determinants of overweight. Low physical activity was significantly associated with overweight in boys whereas high media time increased overweight risk of girls. This is in contrast to the study of Jouret et al.(Reference Jouret, Ahluwalia and Cristini10) in which no sex differences were found in media time consumption but in physical activity: structured physical activity was associated with overweight in girls only. A recent study of Perez-Pastor et al.(Reference Perez-Pastor, Metcalf and Hosking41) showed that mother’s obesity may affect only daughter’s obesity whereas father’s obesity affected son’s obesity only. In KOPS this sex-specific influence could not be confirmed; obesity of both mothers and fathers had an influence on overweight in boys as well as girls (Table 4).
Considering interaction terms (Table 5; Fig. 1) showed that a more complex understanding of childhood obesity is needed. As in the study of Vandewater and Huang(Reference Vandewater and Huang16), we found that TV viewing and weight status of the children was moderated by parental weight status and age of the children. The risk of overweight increased with TV viewing in children with at least one obese parent but not in children with normal-weight parents(Reference Vandewater and Huang16). In addition, there was a further interaction between media time and parental education. We found that high media time increased the prevalence of overweight in children with higher parental education. Thus, a high parental education did not protect against the negative impact of high media consumption. This finding is in line with the study of Singh et al.(Reference Singh, Kogan and Van Dyck15) in which the association between obesity and TV viewing and physical activity was more pronounced in children of higher SES groups.
Thus, reduction of media time should be a target of obesity treatment programmes in children and adolescents of obese mothers and of families from middle and high social status.
Determinants of incidence
In KOPS parental obesity, parental smoking habits and low physical activity were significant risk factors for incidence of overweight. By contrast, parental overweight had no significant effect on incidence (Table 6). There is evidence that genetic and environmental factors, which are related to parental obesity, have a greater effect before the age of 6 years(Reference Sorensen, Holst and Stunkard42). An effect is therefore more clearly seen in the cross-sectional analyses of children than in the analysis of longitudinal data.
In our study low physical activity was the only significant lifestyle determinant of incidence of overweight. The effect remained even after controlling for interactions with parental weight status and SES (Table 7). In contrast to the present study, Maffeis et al.(Reference Maffeis, Talamini and Tato14) did not find lifestyle variables to significantly affect the change in relative BMI over a 4-year period when parental obesity was taken into account. Davison and Birch(Reference Davison and Birch18), who analysed predictors of change in girls’ BMI from age 5 to 7 years, showed that girl’s BMI at age 5 years, family risk of overweight, mother’s increase in BMI, father’s enjoyment of activity, energy intake and girl’s percentage fat intake reached significance. Gortmaker et al.(Reference Gortmaker, Must and Sobol4) showed that watching TV for more than 5 h/d increased the 4-year incidence of overweight in US children.
Comparison of determinants of prevalence and incidence of overweight
Parental obesity and smoking habits as well as low physical activity were significant determinants of prevalence as well as incidence, whereas social factors influenced overweight prevalence only. These data may be taken as evidence for the idea that a societal approach is more important in the treatment of childhood overweight than in primary prevention. In addition, the impact of lifestyle factors may also differ: while high media time added to increased prevalence, low physical activity was the major determinant of incidence. Thus primary prevention programmes should involve the family and focus on increasing physical activity. By contrast, in treatment programmes, family involvement as well as a societal approach is important in combination with a lifestyle approach addressing physical activity and media consumption in children and adolescents of obese mothers and from families of middle and high social status.
Age is differently added to prevalence and incidence. The risk of being overweight increased with age while the risk of becoming overweight decreased. Both findings indicate that the older the children are, the more likely they are to be already overweight. Our data thus argue in favour of early treatment and prevention of overweight.
Limitations
Although many individual and ecological factors were considered in the present study, only 14 % of the variance of overweight could be explained (Tables 4–7). Our definition of overweight was based on BMI which might be a poor indicator of fat mass. Therefore, analyses were repeated using waist circumference and percentage body fat mass. However, this did not increase explained variance (see Results). Additional variables which were significant determinants of overweight in other studies like sleep duration(Reference Reilly, Armstrong and Dorosty20), infant weight gain(Reference Dubois and Girard19, Reference Reilly, Armstrong and Dorosty20), mother’s weight gain(Reference Davison and Birch18) and smoking habits during pregnancy(Reference Dubois and Girard19) were not included in our analyses. Since we have analysed this in a subgroup of the KOPS population we do not assume that they would increase explained variance. Genetic influences were not directly considered but were included in weight status of parents and siblings. If all these determinants explain only less than one-fifth of the variance of overweight, one may question if the approach proposed by Swinburn et al.(Reference Swinburn, Gill and Kumanyika1) is sufficient to combat the obesity epidemic. Recent studies from our group have shown that weight gain is due to a relatively small positive energy balance(Reference Plachta-Danielzik, Landsberg and Bosy-Westphal43). Thus differences in lifestyle factors between overweight and normal-weight subjects are too small to be detected with conventional epidemiological methods. This idea is in line with an alternative strategy to combat the obesity epidemic which was recently published by Hill(Reference Hill44). Hill promoted small changes in diet and physical activity to prevent further weight gain. However, it could be questioned if this ‘easy option’ approach is sufficient to solve such a complex phenomenon like overweight.
Conclusions
Treatment of overweight should involve family and social environment and should mainly address high physical activity as well as low media consumption. Measures of primary prevention should also involve family and should preferentially address high physical activity. Beyond these conventional measures, alternative approaches like the small-changes approach should be tested.
Acknowledgements
This work was supported by grants from Deutsche Forschungsgemeinschaft (DFG Mü 5-1, 5-2, 5-3, 5-5), ‘Competence Network on Obesity’ funded by the Federal ministry of Education and Research (FKZ: 01GI0821), Wirtschaftliche Vereinigung Zucker, Precon and WCRF. The sponsors of the study had no role in study design, data collection, data analyses, data interpretation or writing of the paper. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication. The authors declare no conflict of interest. S.P.-D. had the original idea, did the statistical analyses and interpretation of the data, and wrote the manuscript. B.L., M.J. and D.L. acquired data. M.J.M. supervised the study, did the interpretation of the data, and wrote the paper. All authors discussed the data and approved the final version of the paper.
