Introduction

Lifestyle behaviours such as diet, physical activity, and sedentary behaviour have a significant impact on the health and wellbeing of children and young people [1, 2]. Previous studies suggest that the transition from childhood to adolescence is a critical period for behaviour formation [3, 4]. Adolescence is often accompanied by increases in screen time, decreases in physical activity, and poorer dietary behaviours, characterised by low consumption of fruit and vegetables, higher consumption of snacks and sugar-sweetened beverages, and breakfast skipping [3, 5,6,7,8]. The acquisition of poor lifestyle behaviours during adolescence can result in higher adiposity and cardiometabolic risk [9,10,11], lower cardio-respiratory fitness [9, 11], poorer self-rated health [12], poorer mental and emotional health [9, 11, 13], and lower quality of life [10].

Much of the evidence on health behaviours of adolescents comes from studies looking at individual health behaviours in isolation. However, lifestyle behaviours are likely to co-occur within groups of young people [9, 14]. For example, it has been shown that high screen time and high energy-dense snack consumption tend to cluster in young people, as does low fruit consumption and low physical activity [9, 14]. The effect of clusters of behaviours on health is likely to be greater than the effect of individual behaviours [15, 16]. A better understanding of how, and in whom lifestyle behaviours cluster, and how clusters of behaviours evolve over critical life periods of childhood is essential for the development of targeted interventions. However, to date, much of the evidence on clustering of health behaviours has come from cross-sectional studies [9, 17] or research that has not examined changes in clusters from childhood to adolescence [9]. Previous research has shown that a high proportion of 5–6 years old children belonged to an unhealthy behavioural cluster (high sedentary and low physical activity, or high TV and high energy-dense snack consumption) and tended to remain in the same cluster or move to the other unhealthy cluster by age 9 years [18]. Furthermore, in a study of adolescents aged 14–16 years at baseline, there was a high stability in clusters of lifestyle behaviours (healthy and unhealthy) over a 2-year period [19]. However, it is unknown how clusters of lifestyle behaviours change as children become adolescents. Given the significant social, environmental, and interpersonal changes that go alongside this transition [3, 20], it is imperative to understand in whom and how clusters of behaviours change during this period to help inform public health interventions.

An understanding of the determinants of behavioural clusters is required to underpin public health interventions aimed at promoting positive changes in health behaviours [17]. Given that behaviour change interventions that are theoretically based have been shown to be more effective than those without a theoretical underpinning [21, 22], it is useful to utilise behavioural theories to provide a framework for studying determinants of behavioural clusters. There is support for the use of socio-ecological models for understanding key determinants of health behaviours [23]. The socio-ecological model acknowledges that children are embedded within larger social systems and that characteristics of children interact with environments that underlie health behaviours [24].

Previous research has identified that indicators of socio-economic position, factors from a most pertinent level of the socio-ecological model, are important influences on children’s behavioural clusters, partly because they influence children’s attitudes, experiences, behaviours, and exposures to health risk factors [25]. For example, cross-sectional studies have shown that higher parental education is associated with behavioural clusters including high physical activity in schoolchildren [26], while girls and children from lower-educated parents have a higher prevalence of unhealthier clusters (for example, clusters including high consumption of sugar-sweetened beverages, high screen time, and low sleep duration) [26, 27]. However, wider socio-economic determinants of changes in clusters of lifestyle behaviours are poorly understood.

The results from this study provide support for intervening on multiple behaviours during childhood and adolescence, and to focus on gender differences and parental characteristics related to socio-economic disparities. More longitudinal analyses in different countries that help to better understand how young boys and girls from different socio-economic positions change their lifestyle behaviours over time may guide research to develop more specific and targeted interventions that better consider the needs of different subgroups of young people.

Methods

Study design and participants

Data from the UK Millennium Cohort Study (MCS) (https://cls.ucl.ac.uk/cls-studies/millennium-cohort-study/) were used for the present analyses. Details of the design and sampling of MCS have been published elsewhere [28]. In brief, the MCS is an observational nationally representative longitudinal birth cohort study, which examines socio-economic, and health-related circumstances of children born in the UK between September 2000 and January 2002. At baseline (Wave 1), 18,818 children were first assessed when they were 9 months of age. For the present analyses, data when participants were aged 7 years (MCS4: January–December 2008; n = 14,043) and aged 14 years (MCS6: January 2015–March 2016; n = 11,872) were used (Fig. 1).

Fig. 1
figure 1

Sample selection flow-chart

Full size image

Measures

All assessments of behaviours (physical activity, screen time, and dietary behaviours) were collected via parent report when participants were aged 7 years and via self-report when participants were aged 14 years. For the present analyses, variables were dichotomized considering the values which were closer to the guideline recommendations [29, 30] and the sample distribution.

Physical activity

Physical activity was reported as the number of days per week that the cohort member (child participant) took part in moderate-to-vigorous physical activity (MVPA). For the present analysis, MVPA was classified as ‘ < 5 days/week’ and ‘ ≥ 5 days/week’.

Screen time

The number of hours spent watching television (TV) and playing electronic games on a typical weekday was reported. Time spent TV viewing was dichotomized into ‘ < 3 h/day’ and ‘ ≥ 3 h/day’, and time spent using electronic games was dichotomized into ‘ < 1 h/day’ and ‘ ≥ 1 h/day’.

Dietary behaviours

Participants reported the servings of fruit eaten per day and the number of days per week breakfast was eaten. Fruit consumption was dichotomized into ‘ < 2 servings/day’ and ‘ ≥ 2 servings/day’, and the frequency of eating breakfast was dichotomized into ‘not every day’ and ‘everyday’.

Socio-economic variables at age 7 years

When participants were aged 7 years, parents reported their child’s sex, ethnicity, and their weekly household income according to the Organisation for Economic Co-operation and Development (OECD), participants whose OECD was below 60% of the median were considered in situations of poverty. Parents also reported their highest National Vocational Qualification (NVQ) level achieved and who the child lived with at home.

Statistical analyses

Analyses were conducted using Stata software version 14 (Stata, College Station, TX). Baseline sample characteristics were computed by sex and summarized as median and interquartile range (IQR) for continuous variables and frequencies and percentages for categorical variables. Kolmogorov–Smirnov test and histograms were used to assess the variable distribution. Mann–Whitney U tests were used for sex differences in continuous non-parametrical variables, and chi-square tests were used for sex differences in proportions.

At both time points, Joint Correspondence Analysis (JCA), which is an extension of the Multiple Correspondence Analysis (MCA), was conducted to explore the patterns in the categorical lifestyle variables. The analysis included five variables (physical activity, TV viewing, electronic games playing, breakfast eating, and fruit consumption). Analyses were carried out for boys and girls separately, as significant sex differences were shown in the lifestyle behaviours at baseline (Table 1). The dimensions revealed from the JCA that contained most of the data variability were used to inform the cluster analysis. The Calinski-Harabasz rule [31] was used to determine the optimal number of cluster solutions. The final number of clusters was decided based on interpretability and Calinski-Harabasz indexes at both time points and for both sexes. So, the number of clusters less or equal to five, with a higher Calinski-Harabasz index at both time points and for both sexes were selected. K-means cluster analysis was used to create three clusters of behaviours using the coordinates obtained from the JCA at both waves of assessment independently and for boys and girls separately. The stability (tracking) of cluster membership between ages 7 years and 14 years was examined using Kappa (K) statistics. The degree of tracking was evaluated using the cut-offs suggested by Landis and Koch [32]: < 0.2, poor; 0.2–0.4, fair; 0.41–0.6, moderate; 0.61–0.8, good; > 0.81, very good. The K value for the stability of cluster membership between 7 and 14 years was 0.254 for girls and 0.265 for boys, which shows fair agreement between baseline and follow-up cluster membership.

Table 1 Characteristics of the sample at age 7 by sex expressed as median (IQR) or n (%)
Full size table

As the tracking of clusters was fair, we determined the profile of changes in behavioural clusters. Participants were classified depending on their continuity in the same cluster over time or by their change from one cluster to another. The categories to represent change between age 7 and 14 years were ‘remained in healthy cluster’, ‘remained in mixed cluster’, ‘remained in unhealthy cluster’, ‘changed to an unhealthier cluster (i.e. moved from healthy at age 7 to mixed or unhealthy at age 14; or from mixed at age 7 to unhealthy at age 14)’, and ‘changed to a healthier cluster (i.e. moved from unhealthy at age 7 to mixed or healthy at age 14; or from mixed at age 7 to healthy at age 14)’. Multinomial logistic regressions and the calculation of the corresponding odds ratios (OR) were used to examine the associations between socio-economic variables at baseline and the changes in cluster membership between ages 7 and 14 years. These models were performed for boys and girls separately and were adjusted for weight status and body fat percentage. Anthropometric covariates were collected by trained researchers during home visits when children were aged 7 years (MCS 4).

Results

From the initial MCS sample of 18,818 children, a final sample analytical sample of 9339 children (4660 girls and 4679 boys; 49.6% of the initial sample) had complete data on all variables of interest and were included in the present analyses (Fig. 1).

Characteristics of participants at age 7 years

Parents of boys had a high education level (NVQ above 3) (40.27% vs. 38.65%; p = 0.035) and had higher weekly family income (£405.27 ± 232.47 vs. £397.99 ± 234.50; p = 0.038) compared to girls (Table 1).

At age 7, girls participated in MVPA less frequently and skipped breakfast more frequently than boys. Boys consumed less servings of fruit and spent more time watching TV and playing electronic games than girls (Table 1).

Characteristics of behavioural clusters

For boys, the two dimensions (x and y) revealed by the JCA analyses explained 76.1% and 26.1% of the variability in the data at baseline, and the 89.4% and 10.4% variability in the data at follow-up. For girls, the two dimensions revealed by the JCA analyses explained 82.4% and 17.5% of the variability in the data at baseline, and the 86.5% and 13.4% variability in the data at follow-up (SF 1).

Three behavioural cluster solutions were produced: a healthy cluster (HC), a mixed cluster (MC), and an unhealthy cluster (UC) (Table 2). At baseline, 41.6% of boys and 52.6% of girls were in the HC, 36.5% of boys and 31.2% of girls were in the MC, and 21.9% of boys and 16.2% of girls were in the UC.

Table 2 Cluster characteristics at baseline and follow-up represented as n (%)
Full size table

Among boys and girls at baseline, the HC was characterised by no breakfast skipping, lower TV viewing, higher MVPA, higher fruit consumption, and lower e-games playing than in the MC and UC (Table 2). The UC was characterised by lower MVPA, higher breakfast skipping, lower fruit consumption (especially in boys), higher TV viewing, and higher e-games playing than HC and MC. The MC was characterised by lower MVPA, low breakfast skipping, low TV watching higher, high electronic games playing, and fruit consumption lower than HC, but higher than UC (Table 2).

At follow up, there were differences in the behaviours that characterised each cluster, and clusters differed more between boys and girls than at baseline (Table 2). In boys, the three clusters had a similar percentage of high e-games playing (between 28.8 and 37.4%), while 76% of the girls with high e-games playing were in the UC. In girls, the MC was characterised by high fruit consumption (similar to HC), while in boys the MC was characterised by low fruit consumption (Table 2).

Changes in behavioural clusters between ages 7 and 14 years

Almost a fifth of boys (19.1%) who were in the HC at baseline remained in the HC at age 14 compared with 13.1% of girls (Table 3). A similar proportion of boys and girls who were in the MC at age 7 remained in the MC at age 14 years (11.2% and 9.1%), and a higher proportion of boys (7%) who were in the UC at age 7 remained in the UC at age 14 years compared to girls (3.4%). A high percentage of boys (26.4%) became healthier compared to girls (9.4%). A higher proportion of girls became unhealthier compared to boys (57.2% vs. 33.9%).

Table 3 Changes in behavioural clusters from baseline to follow-up represented as n (%)
Full size table

Associations between indicators of socio-economic and change in behavioural cluster profiles

After controlling for socio-economic indicators and anthropometric factors, compared to the boys who remained in HC, boys from low-income households were more likely to change to a healthier cluster than those from a higher-income household (OR (95% CI) = 1.43 (1.10, 1.84)) (Table 4). Boys with parents of higher education were less likely to remain in the MC, remain in the UC, and change to an unhealthier cluster, but also less likely to change to a healthier cluster than those of less educated parents (NVQ below 3 or other qualifications). Boys living with a single parent or other guardian were more likely to remain in MC, remain in UC, and change to an unhealthier cluster, but also less likely to change to a healthier cluster compared to those living with both parents.

Table 4 Boys’ odds ratios (ORs) and 95% confidence intervals (CI) for being in the different longitudinal cluster profiles according to socio economic status and at age 7 (n = 4679)
Full size table

Compared to the girls who remained in HC, girls from lower-income households were more likely to remain in UC and change to an unhealthier cluster (Table 5). Girls with higher educated parents were less likely to remain in MC, remain in UC, and change to an unhealthier cluster, but also less likely to change to a healthier cluster than those with lower parental education. Girls whose parents had other qualifications were two times more likely to remain in UC (OR (95% CI) = 2.25 (1.40, 3.62)), and two times more likely to change to a healthier cluster (OR (95% CI) = 2.14 (1.30, 3.52)) compared to girls who parents have lower education levels. Girls from White ethnic groups were less likely to remain in MC and less likely to change to a healthier cluster compared to girls from other ethnic groups. Girls living with a single parent or other guardian were more likely to remain in UC and change to an unhealthier cluster compared to those living with both parents.

Table 5 Girls’ odds ratios (ORs) and 95% confidence intervals (CI) for being in the different longitudinal cluster profiles according to socio economic status at age 7 (n = 4660)
Full size table

Discussion

The aims of this study were to identify the changes in behavioural clusters from childhood to adolescence and to examine the associations between indicators of socio-economic status and changes in behavioural clusters over time. To the best of our knowledge, these are the first analyses to have examined changes in clusters of health behaviours over 7 years, to have considered the transition from childhood to adolescence, and to associate these changes to socio-economic indicators. Findings show low consistency in behavioural clusters over the 7 years ages 7 and 14 years with clear sex differences, as girls tend to change to unhealthier clusters, while boys tend to remain in healthier clusters. Indicators of socio-economic status during childhood appear to be important determinants of changes in behavioural clusters over the 7-year period.

The present study found that behavioural clusters are susceptible to changes over a 7-year period. At age 7 years, boys and girls had similar cluster prevalences, being in the HC was the most prevalent and in the UC was the least prevalent. This changed in boys to an almost equal prevalence across the three clusters, whereas among girls, the UC became the most prevalent and the HC the least prevalent at age 14 years. Previous studies have also shown that clusters of health behaviours have low to moderate tracking rates from ages 5–11 to 9–14 years [18]. Within the clusters in the present study, the number of boys with high e-games playing increased over time and was similarly distributed among the three clusters, and the number of boys and girls with high TV watching increased, especially in the UC. These results are in line with other studies reporting an increase in screen time from childhood to early adolescence, with differences in media used among boys and girls [33], and such increases in screen time could be driving the changes seen in certain clusters.

Results also showed that girls tended to change to unhealthier clusters, while boys tended to remain in HC or change to healthier clusters during the transition from childhood to adolescence. Previous literature has shown evidence of sex differences in lifestyle behavioural patterns over time. Research has shown that girls are more likely to become unhealthier in terms of lifestyle patterns of physical activity and sedentary behaviour [34, 35], especially when they came from lower socio-economic backgrounds [34]. Furthermore, a study by Daking et al. [19] found that boys were more likely to show unfavourable changes in lifestyle behavioural clusters (including diet, physical activity, and sedentary behaviour) across the transition from middle to late adolescence compared to girls. These findings show that variability in behaviours included in cluster analyses may influence the direction of sex differences regarding lifestyle patterns. Further research is needed examining the same behavioural clusters over critical time periods to be able to compare and unpack the sex differences in changes over time to inform interventions.

In addition to the gender differences seen in changes in clusters over time, indicators of socio-economic status appeared to impact changes in cluster membership. Among both boys and girls, living with a single parent (or another person different from the parents) was associated with a higher chance of remaining in UC or changing to unhealthier clusters, while higher parental education was associated with a higher likelihood of remaining in HC. For girls, lower family income was also associated with a higher probability of remaining in UC or changing to unhealthier clusters. Lower socio-economic status has been previously examined in association with lifestyle clusters in young people in cross-sectional studies [9]. For example, low parental education [36,37,38], low household income [36], and living in a deprived neighbourhood [39] have been associated with unhealthier lifestyle clusters, which may be related to a higher food insecurity/access to healthy foods [40], and/or lower access to a safe built environment that facilitate the physical activity practice. [41, 42]. Contrastingly, we also found that children with low socio-economic status had higher probabilities of transitioning to healthier clusters. These results may be partially explained by our reference group in the multinomial logistic regression analyses which was ‘remaining in HC’. Most of the children from high-socio-economic environments were in the healthier cluster at baseline (SF 2); therefore, these children had less probability of changing to healthier clusters than children from lower socio-economic groups. The results from this study provide support for intervening on multiple behaviours during childhood and adolescence, and to focus on gender differences and parental characteristics related to socio-economic disparities. More longitudinal analyses in different countries that help to better understand how young boys and girls from different socio-economic positions change their lifestyle behaviours over time may guide research to develop more specific and targeted interventions that better consider the needs of different subgroups of young people.

Strengths and limitations

Strengths of the study include the longitudinal study design, which includes a large nationally representative sample of British children, the inclusion of multiple health behaviours, and a thorough analysis of clustering. However, conclusions must be made in light of the limitations. Participants (12%) were excluded from the original sample due to missing data (SF 3), potentially limiting the generalisability of our findings. The outcome variables (physical activity, screen time, and dietetic variables) and socio-demographic variables were all assessed via questionnaires, which could lead to errors in the measurement due to socially desirable responses and/or inaccurate recall [43,44,45]. Furthermore, parents reported during the first wave of data collection and adolescents during the second wave which could change reporting bias [46].

Conclusions

Lifestyle behaviours cluster in children and are susceptible to change over a 7-year period with a high proportion of boys changing to healthier clusters compared to girls, and a higher proportion of girls changing to unhealthier clusters compared to boys. Children from lower socio-economic status backgrounds including low household income, low parental education, or not living with both parents were not only more likely to change to unhealthier clusters over this period but also more likely to change to healthier clusters. More longitudinal analyses in different countries are needed for a better understanding of the key drivers of change in clusters among subgroups of children, which is essential for the development of interventions targeting multiple health behaviours.