Lifestyle factors such as high television (TV) viewing(Reference Hancox, Milne and Poulton1–Reference Kleiser, Schaffrath Rosario and Mensink3), sugar-sweetened beverage consumption(Reference Ludwig, Peterson and Gortmaker4–Reference Malik, Schulze and Hu6) and excess snacking(Reference Kubik, Lytle and Story7) are associated with increased obesity risk while high physical activity (PA)(Reference Molnar and Livingstone8–Reference Janssen, Katzmarzyk and Boyce10) and longer sleep duration(Reference Kleiser, Schaffrath Rosario and Mensink3, Reference Chen, Beydoun and Wang11, Reference Hitze, Bosy-Westphal and Bielfeldt12) appear to be protective. However, in most studies these behaviours account for a minor proportion of the variance of overweight and obesity. Moreover, usually only one or selected lifestyle factors have been explored. However, individual lifestyle factors tend to cluster among themselves. For example, snack frequency and sugar-sweetened beverage consumption increase with the hours of TV viewing(Reference Gorely, Marshall and Biddle13, Reference van den Bulck14) while sedentary behaviours and physical activities co-occur(Reference Nelson, Gordon-Larsen and Adair15–Reference te Velde, De Bourdeaudhuij and Thorsdottir17). A combination of health behaviours may have synergistic effects on overweight. Only a few studies have addressed the clustering of lifestyle factors. Understanding the interrelationships among lifestyle factors will provide insights for health research and for development of public health programmes that can impact on the obesity epidemic.
Within the Kiel Obesity Prevention Study (KOPS)(Reference Müller, Asbeck and Mast18, Reference Danielzik, Pust and Landsberg19), we conducted a cross-sectional study in German eighth-grade students to identify lifestyle clusters composed of PA, commuting mode, media time, nutrition, alcohol consumption and smoking. These clusters were related to obesity indices. In addition, a longitudinal study was performed to investigate the impact of these clusters on the incidence of overweight and obesity.
Methods
Study population
Data were collected as part of KOPS, which is an ongoing study on the determinants and on the effect of the prevention of childhood overweight started in 1996 in Kiel, Germany(Reference Müller, Asbeck and Mast18, Reference Danielzik, Pust and Landsberg19). In KOPS, three groups of children and adolescents were enrolled: (i) 6-year-old children (n 4997; as part of a school entry examination in Kiel between 1996 and 2001); (ii) 10-year-old children (n 4487; within a school examination between 2000 and 2005); and (iii) 14-year-old adolescents (n 3237; as part of secondary schools in Kiel between 2004 and 2006). The heads of all thirty-five secondary schools in Kiel were contacted and informed, of which three refused to participate. All eighth-grade students (n 6783) of the remaining schools were invited to attend the study. Participation was on a voluntary basis and there were no eligibility criteria except the willingness to participate. In total, 48 % (n 3237) of the students agreed to take part in the investigation. No information is available regarding the 3546 students who did not participate in the study. Signed informed consent was obtained and the study protocol was approved by the local ethics committee.
Of the 3237 adolescents, 1343 had missing data on lifestyle factors (n 1320) or obesity indices (n 23). Thus, the ‘study population’ of the present study consists of 1894 14-year-old adolescents of the third KOPS group. Since all three study groups of KOPS belonged to the same total population, a ‘subsample’ of 389 adolescents was identified whose obesity indices had been measured twice within a 4-year follow-up period (at the age of 10 (T0) and 14 (T1) years).
Lifestyle factors
Physical activity and media time
PA, commuting mode to school and media time were assessed with a questionnaire. Weekly duration of regular structured PA outside school as well as weekly duration of unstructured PA (e.g. jogging and skating) was provided by the adolescents. Four-week test–retest correlation (r) was 0·50 (P < 0·01) and 0·35 (P < 0·01) for the duration of structured and unstructured PA, respectively(Reference Landsberg, Plachta-Danielzik and Much20). Mode of commuting to school was given and categorized as active (walking and cycling) and inactive (driving by car or bus) commuting. Test–retest percentage agreement (reliability) was 91·6 %. Self-reported media time was assessed as hours per day spent in TV viewing and computer use on a typical weekday. The 4-week test–retest correlation (r) was 0·68 (P < 0·01). In a previous study on 5–11-year-old children, TV viewing had been compared with (i) energy expenditure as assessed by the combined use of indirect calorimetry together with 24 h heart rate monitoring (time more than Flex heart rate); and (ii) aerobic fitness (submaximal oxygen consumption, O2 pulse)(Reference Grund, Krause and Siewers21). However, there were no significant differences in either energy expenditure or fitness between groups of children watching TV ≤1 h v. >1 h/d.
Nutrition and risk behaviour
A twenty-six-item validated FFQ was used to describe dietary intake(Reference Mast, Kortzinger and Müller22). Consumption of three or more ‘healthy’ foods (whole meal bread, fruits, vegetables, fish, dairy products) and fewer than three ‘risk-related’ foods (white bread, sausage, soft drinks, fast food, sweets/chips) at least 3–5 times a week was summarized as a ‘healthy dietary pattern’(Reference Landsberg, Plachta-Danielzik and Much20). Consumption of three or more ‘risk-related’ foods and fewer than three ‘healthy’ foods at least 3–5 times a week corresponded to a ‘risk-related dietary pattern’. Other combinations indicated a ‘mixed dietary pattern’. The 4-week test–retest percentage agreement (reliability) of dietary pattern was 67·6 %.
Risk behaviour was defined as alcohol consumption and smoking behaviour. Habitual alcohol consumption was assessed according to: never, once a month, 2–3 times a month and at least once a week. For the analysis, the latter three categories were combined. Adolescents were also asked whether they smoked or not. The 4-week test–retest percentage agreement (reliability) was 78·3 % for alcohol consumption and 95·6 % for smoking.
Obesity indices and blood pressure
Obesity indices were measured anthropometrically (height, weight, waist circumference (WC)) and bioelectric impedance analysis (fat mass (FM) calculated with a population-specific algorithm)(Reference Müller, Asbeck and Mast18, Reference Danielzik, Pust and Landsberg19, Reference Plachta-Danielzik, Landsberg and Johannsen23) was performed at the age of 10 and 14 years. International age- and gender-specific cut-offs were used to define overweight and obesity by BMI(Reference Cole, Bellizzi and Flegal24). German references were used to calculate the sd score of BMI(Reference Kromeyer-Hauschild, Wabitsch and Kunze25). Blood pressure was measured using a sphygmomanometer on the right arm after a 5-min rest.
Social characteristics
Educational level of the adolescents was defined using the attended school type. Germany’s school system is divided into three different types of school: ‘Hauptschule’ (nine school-years), ‘Realschule’ (ten school-years) and ‘Gymnasium’ (at least twelve school-years) depending on the level of academic performance of the students. In the present study, educational level was expected to be ‘low’ in the ‘Hauptschule’, ‘middle’ in the ‘Realschule’ and ‘high’ in the ‘Gymnasium’. The nationality of the adolescents and their families was acquired and transformed into a dichotomous variable (German v. non-German).
Statistical analysis
Results were presented as median and interquartile range. A non-parametric Kruskal–Wallis test was used to determine between-group differences with respect to continuous variables. The χ 2 test compared prevalence of categorical variables among groups.
Lifestyle clusters
Two-step cluster analysis was used (the Statistical Package for the Social Sciences statistical software package version 16·0 (SPSS Inc., Chicago, IL, USA)) as an explanatory tool to identify specific lifestyle patterns. This method is designed to handle large data sets and enables the input of both categorical and continuous variables(Reference Everitt, Landau and Leese26). The categorical variables included commuting mode to school, dietary pattern, alcohol consumption and smoking behaviour. The continuous variables included duration of structured and unstructured PA and media time. The choice of similarity measure and the determination of the number of clusters were based on the log-likelihood distance and Schwarz’s Bayesian criterion. To demonstrate the robustness of the cluster solution, the solution was re-examined on an internal random sample of 50 % of the study population. The different clusters are given labels that characterize the behavioural pattern. The identified clusters were then compared regarding the actual difference on the lifestyle behaviours by means of ANOVA and a Bonferroni post hoc test. Multivariate logistic regression analyses were performed to estimate OR for overweight and obesity.
Longitudinal data
Obesity indices of the subsample of 389 adolescents were measured twice assuming an identical lifestyle pattern during the follow-up period (T1–T0). The follow-up period was 4·5 (sd 0·4) years. T1 data were corrected to 4 years by linear extrapolation using the following equation: BMIcorrected (T1) = [(BMI (T1)−BMI (T0)]/[(age (T1)–age (T0) × 4]+BMI (T0).
All analyses were performed with SPSS version 16·0 and the level of significance was set at P < 0·05 (two-sided).
Results
Characteristics of the study population and the subsample
The study population consisted of 1894 14-year-old adolescents (48·5 % boys). Most of the adolescents had high and middle educational levels (41·6 % and 38·9 %, respectively). Non-German adolescents made up 14 % of the total. When compared with the study population, the percentage of boys was higher (52·4 %) and the percentage of non-Germans was lower (7·5 %) in the subsample of 389 subjects. Obesity indices and lifestyle factors in the study population and the subsample are shown in Table 1.
Obesity indices and lifestyle factors in the study population of 14-year-old adolescents (n 1894) and in the subsample of 389 subjects at the age of 10 (T0) and 14 (T1) years: Kiel Obesity Prevention Study, 2000–2006
IQR, interquartile range; BMI-SDS, sd score of BMI; FM, fat mass; WC, waist circumference; PA, physical activity.
Cross-sectional data
Characterization of lifestyle clusters
The present study identified three lifestyle clusters. Based on the cluster criteria, cluster 1 was characterized by ‘low activity and low-risk behaviour’, cluster 2 by ‘high media time and high-risk behaviour’ and cluster 3 by ‘high activity and medium-risk behaviour’ (Table 2). A generally healthy lifestyle pattern could not be identified in our study population.
Cluster centres for the lifestyle factors included in the cluster analyses: cross-sectional data, Kiel Obesity Prevention Study, 2004–2006
PA, physical activity.
a,b,cMean values within a column with unlike superscript letters were significantly different between the clusters (post hoc Bonferroni testing). Significant differences between the clusters: †P < 0·01 (χ 2 test).
There were 39.1 % of adolescents who were affiliated to cluster 1. This cluster was characterized by low-structured and unstructured PA and moderate media time. Of the adolescents, one-fifth had a risk-related dietary pattern. All the adolescents in this cluster reported never drinking alcohol, and only 0·5 % smoked regularly.
Cluster 2 was observed in 26·3 % of adolescents. Subjects of this cluster showed moderately structured and unstructured PA and high media time. In one-fourth of the adolescents, the dietary pattern was risk-related; more than one-third smoked and almost all consumed alcohol at least once a month.
Of the adolescents, 34.6 % were in cluster 3. This cluster was characterized by high Z-scores for structured and unstructured PA combined with low Z-scores for media time. A risk-related dietary pattern was found in 2·9 % of the adolescents of this cluster. There were 8.0 % who smoked and 43·0 % who consumed alcohol at least once a month.
Table 3 shows the sociodemographic characteristics of the clusters. Clusters 1 and 2 had a higher proportion of boys, whereas cluster 3 comprised more girls. The age of the adolescents was highest in cluster 2. The clusters differed with regard to the distribution of the educational level. In cluster 1, the proportion of subjects with middle and high educational levels was similar. Cluster 2 contained the highest proportion of adolescents with a middle educational level, whereas most adolescents in cluster 3 had a high educational level.
Sociodemographic characteristics of the study population within lifestyle cluster: cross-sectional data, Kiel Obesity Prevention Study, 2004–2006
IQR, interquartile range.
Significant differences between the clusters: *P < 0·05 (Kruskal–Wallis test); †P < 0·05 (χ 2 test).
Associations with overweight
Table 4 summarizes the associations between the lifestyle clusters and obesity indices as well as blood pressure. Weight was lowest among adolescents in cluster 3. Adolescents of cluster 2 were taller than adolescents of the other clusters. BMI was lowest among adolescents in cluster 1, whereas FM was lowest among adolescents in cluster 2. There was no significant difference in the prevalence of overweight and obesity between the clusters with regard to the total study population. However, prevalence tended to be lowest in cluster 3. Regarding boys and girls separately, girls in cluster 3 had a significantly lower prevalence of overweight (11·0 %) than girls in clusters 1 (17·0 %) and 2 (14·3 %; P = 0·017). Considering the educational level of the adolescents, no differences in the prevalence of overweight and obesity existed between the clusters.
Associations between lifestyle clusters and obesity indices and BP of the study population: cross-sectional data, Kiel Obesity Prevention Study, 2004–2006
IQR, interquartile range; BMI-SDS, sd score of BMI; FM, fat mass; WC, waist circumference; BP, blood pressure; BPsys, systolic BP; BPdias, diastolic BP.
Significant differences between the clusters: *P < 0·05 (χ 2 test); †P < 0·01 (Kruskal–Wallis test).
Blood pressure did not differ between the clusters.
Longitudinal data
Associations with the development of overweight
Table 5 shows the longitudinal data of obesity indices in the subsample by lifestyle cluster identified at the age of 14 years. At T0, the lowest weight and BMI were found among children in cluster 1. By contrast, WC was lowest in cluster 3. At T1, compared with the other clusters, weight and WC were lowest in cluster 3.
Obesity indices and prevalence of overweight and obesity at T0 and 4 years later at T1 and 4-year incidence of overweight and obesity in the subsample (n 389) by lifestyle cluster: longitudinal data, Kiel Obesity Prevention Study, 2000–2006
T0, 10-year-old children; T1, 14-year-old adolescents; IQR, interquartile range; BMI-SDS, sd score of BMI; FM, fat mass; WC, waist circumference.
‡Four-year changes of weight, height, BMI, FM and WC were adjusted for baseline values.
Significant differences between the clusters: *P < 0·05 (Kruskal–Wallis test); †P < 0·05 (χ 2 test).
Adjusted 4-year changes of FM were found to be lowest in cluster 2. Stratifying for gender, these results were confirmed in girls (data not shown).
There was no difference in the prevalence of overweight and obesity between the clusters either at T0 or at T1. However, the 4-year incidence rates of obesity were significantly lower in cluster 3. Sex and educational level had no confounding effect on prevalence and incidence rates.
Discussion
The aim of the present study was to characterize lifestyle clusters in German adolescents and to analyse their potential association with overweight. Cluster analyses identified three groups of adolescents sharing similar lifestyle behaviour patterns. These clusters varied by sociodemographics and tended to differ in their associations with overweight and obesity. Compared to cluster 3, clusters 1 and 2 tended to have a higher prevalence of overweight and obesity. In addition, the incidence rate of obesity was lowest in cluster 3.
Lifestyle factors
Today, participating in PA as well as the use of electronic media has a great role in adolescents’ recreational behaviour. German adolescent boys and girls spend 3·5-fold and 2·5-fold time in using electronic media than in exercising and playing sports, respectively(Reference Lampert, Sygusch and Schlack27, Reference Lampert, Mensink and Romahn28). In addition, 20 % of 11–17-year-old adolescents in Germany smoke, one-third of the boys and one-quarter of the girls consume alcohol regularly(Reference Lampert and Thamm29). Several studies have shown that modifiable lifestyle factors are interrelated.
The increasing impact of using electronic media is often stated to be one of the main reasons for the supposed lack of PA in adolescents. The so-called ‘displacement hypothesis’ proposes that sedentary behaviours displace PA(Reference Mutz, Roberts and van Vuuren30). However, controversial findings exist concerning these associations(Reference Marshall, Biddle and Gorely2, Reference Biddle, Gorely and Marshall16, Reference Sallis, Prochaska and Taylor31). Correlations between PA and inactivity (high media times) are often small or do not exist. However, in the present study, PA and media time were inversely but weakly correlated (r = −0·08 and −0·14 in boys and girls, respectively). Adolescents with the highest PA (cluster 3) were characterized by the shortest media times. Conversely, adolescents in cluster 1 did not have the longest media time. These findings are in line with the results of te Velde et al.(Reference te Velde, De Bourdeaudhuij and Thorsdottir17) suggesting that certain sedentary behaviours, such as TV viewing and computer use, are not necessarily barriers for PA. One reason for that could be that PA is executed typically in the afternoon while sedentary behaviours are more prevalent in the evening.
Lifestyle clusters
In our study population, we identified three lifestyle patterns. Adolescents with low PA levels (cluster 1) did not show any risk behaviour in terms of alcohol consumption and smoking. By contrast, high media times were combined with poor nutrition and high-risk behaviour in terms of alcohol consumption and smoking (cluster 2). In addition, high activity was correlated with low media time, healthy nutrition together with high alcohol consumption and smoking (cluster 3). Strictly speaking, none of these clusters was considered to be absolute healthy.
te Velde et al.(Reference te Velde, De Bourdeaudhuij and Thorsdottir17) characterized a so-called healthy behaviour pattern in 8–13-year-old children. However, the authors did not consider risk behaviour in the analysis and labelling of the cluster was based on low time spent on TV viewing during dinner and during the day and low computer use combined with high PA. Associations with risk behaviour stayed unclear. Boone-Heinonen et al.(Reference Boone-Heinonen, Gordon-Larsen and Adair32) performed a cluster analysis including risk behaviour in 11–21-year-old US adolescents. They found that high smoking prevalence was associated with junk food consumption in boys and food restriction in girls. In contrast to our findings, the sedentary behaviour cluster was characterized by low alcohol consumption and low smoking prevalence.
The clusters identified in our study differed by sociodemographic characteristics. Adolescents of the ‘high media time and high-risk behaviour’ cluster were older than adolescents of the other clusters. Since alcohol and cigarette consumption increase with age in adolescents(Reference Lampert and Thamm29), this could explain the high consumption in this cluster. In contrast to te Velde et al.(Reference te Velde, De Bourdeaudhuij and Thorsdottir17), different behavioural patterns for boys and girls could not be found in the present study. Identified clusters had identical characteristics for both sexes.
Associations with overweight
We found no clear associations between lifestyle clusters and overweight and obesity in our study population. Prevalence did not significantly differ between the clusters but tended to be lowest in cluster 3. However, 4-year incidence of obesity was significantly lowest in cluster 3 and highest in cluster 1, suggesting that low activity plays a major role in the development of childhood obesity.
While previous studies used cluster analyses to explore patterns of these behaviours(Reference Nelson, Gordon-Larsen and Adair15, Reference Gorely, Marshall and Biddle33, Reference Hulshof, Wedel and Lowik34), few analyses examined the association of lifestyle clusters with overweight(Reference te Velde, De Bourdeaudhuij and Thorsdottir17, Reference Boone-Heinonen, Gordon-Larsen and Adair32). te Velde et al.(Reference te Velde, De Bourdeaudhuij and Thorsdottir17) found the lowest prevalence of overweight in the healthy behaviour cluster (characterized by high PA and low TV viewing and computer use) supporting our findings. In the study of Boone-Heinonen et al.(Reference Boone-Heinonen, Gordon-Larsen and Adair32), the cluster characterized by participation in many school clubs and sports had the lowest risk of overweight.
Study strength and limitations
A strength of the present study is cluster analysis instead of characterization of isolated individual behaviours. In addition, the study provides longitudinal data to test associations between lifestyle clusters and weight change.
However, some limitations should be addressed. Since we assumed that adolescents of the subsample had retained their lifestyle behaviour pattern over a 4-year time period from 10–14 years of age, we cannot rule out some changes in lifestyle behaviours over that time. It would have been more precise to repeat cluster analysis, at the age of 10 and 14 years, and to account for changes in lifestyle behaviours. However, previous studies have shown that PA and sedentary behaviour track from childhood to adolescence(Reference Janz, Dawson and Mahoney35, Reference Kristensen, Moller and Korsholm36).
While the weakness of self-reported measures of lifestyle behaviours of PA and nutrition is well known(Reference Sirard and Pate37, Reference Willett38), our activity and nutrition questions were based on questionnaires demonstrated to be valid(Reference Grund, Krause and Siewers21, Reference Mast, Kortzinger and Müller22) and reliable(Reference Landsberg, Plachta-Danielzik and Much20) and have been shown to be important predictors of obesity(Reference Plachta-Danielzik, Landsberg and Lange39). Only one typical sedentary behaviour, i.e. time spent in TV viewing and computer use, was assessed. However, adolescents engage in a wide range of activities that have not often been assessed, e.g. doing homework, reading, personal care and social interactions(Reference Jago, Anderson and Baranowski40). Using electronic media is the most popular sedentary recreational behaviour among German adolescents(Reference Lampert, Sygusch and Schlack27).
For a large proportion of respondents, data in terms of lifestyle behaviour were missing (40·8 %), because on the one hand, the questionnaire was enlarged during the examination. Thus, items required for this analysis, such as commuting mode to school and unstructured PA, were not available for 76·2 % of adolescents measured in 2004. On the other hand, incomplete questionnaires were returned. In comparison with the excluded adolescents, the proportion of adolescents with a high educational level was lower and the frequency of non-Germans was higher in the study population, but the distribution of gender, prevalence of overweight and obesity and participation in structured PA were similar. Since cluster solution was reconstructed in a random sample, our results and conclusions would have been comparable when including a larger sample. Nevertheless, dropout bias cannot be completely ruled out.
Conclusions
Explicit healthy lifestyles do not exist among adolescents of the KOPS. An active lifestyle pattern reduces the incidence of obesity. Our findings provide strong support for targeted interventions that account for diversity of lifestyles and address-specific lifestyle clusters. This strategy seems to be more promising than one-for-all programmes.
Acknowledgements
Sources of funding: KOPS is supported by grants from Deutsche Forschungsgemeinschaft (DFG Mü 5-1, 5-2, 5-3, 5-5), ‘Kompetenznetz Adipositas (Competence Network on Obesity)’ funded by the Federal Ministry of Education and Research (FKZ: 01GI0821), Wirtschaftliche Vereinigung Zucker, Precon and WCRF. The present study has been facilitated by the EU-funded HOPE project: Health Promotion through Obesity Prevention across Europe (the Commission of the European Communities, SP5A-CT-2006-044128). The study does not necessarily reflect the Commission’s views and in no way anticipates the Commission’s future policy in this area. Conflict of interest declaration: The authors have no conflict of interest. Authorship responsibilities: B.L. had the original idea, conducted the statistical analyses, did the interpretation of the data and wrote the manuscript. B.L., D.L., M.J. and J.S. acquired the data. S.P.-D. and 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.
