Long-Term Change in BMI for Children with Obesity Treated in Family-Centered Lifestyle Interventions

Childhood obesity is increasing worldwide [1, 2], often leading to obesity in adulthood [3]. Children and adolescents living with obesity are at risk of impaired mental health, i.e., loneliness, reduced self-esteem, and impaired quality of life [4‒6] as well as an increased risk of developing prediabetes, type 2 diabetes, and hypertension compared to children never classified as having obesity [7, 8]. However, weight reduction can improve mental health as well as reduce risk of obesity-related comorbidity [7, 9, 10]. In Denmark, a plateau in the prevalence of childhood obesity has been observed during the past two decades, but for children from families with lower socioeconomic status (SES) an increase in body mass index (BMI) z score has persisted [11‒14] resulting in increased differences in the prevalence between socioeconomic strata [15‒17].

Initiating lifestyle interventions early in life (age 6–9 years) seem to augment adherence and thus increase the possibility to improve overall outcome compared to initiating treatment later in childhood [18‒21], which emphasizes the importance of early intervention. Lifestyle interventions of at least 2–3 years duration have been reported to increase the likelihood of a sustained reduction in BMI [18, 22‒24]. However, limited knowledge exists regarding weight change after completion of an intervention for childhood obesity. In an extensive review, Kumar et al. [23] recommended a staged approach when initiating treatment of obesity, taking the age of the child, the severity of obesity and presence of obesity-related comorbidities into consideration.

Danish registries provide an ideal setting for investigating the long-term effects of interventions for childhood obesity. In all Danish schools, mandatory health check-ups including measurement of height and weight are performed by community health nurses at fixed time-points. When combining these standard procedures with data from interventions and national registries (i.e., on SES), it is possible to evaluate the long-term effect of interventions after they are terminated.

Our aim was to investigate the long-term development in BMI z score in children with obesity participating in one of two community-based family-centered lifestyle interventions by merging the data from the interventions, the mandatory health check-ups at school and the national registries. The interventions were compared to a no-intervention group consisting of children with obesity, who were not invited to participate in an intervention. In addition, we investigated whether sex, age of the child, family structure, SES, or immigration status modified the effect of the interventions as well as subgroup comparisons.

This pragmatic observational study included Danish children with obesity treated in either the Aarhus-intervention (between January 1st 2010 and December 31st 2020), the Randers-intervention (between January 1st 2014 and December 31st 2020), or children never invited into the interventions (recruited from Aarhus Municipality, between January 1st 2010 and December 31st 2020). Both interventions were real-life obesity treatments for children and originally not planned to include a scientific evaluation. During childhood, all Danish children have mandatory health check-ups, performed by general practitioners and/or community health nurses, including assessment of weight and height. The children in the current study were all identified and followed using health check-ups at the schools. Additionally, child and family characteristics were obtained using the Danish national registers at Statistics Denmark.

We included children aged 4–17 years with obesity, as defined by the International Obesity Task Force (IOTF) guidelines (iso-BMI ≥30 kg/m² for age and gender) [25]. The inclusion visit was the time-point with a measured weight and height, closest to the day of enrollment into one of the two interventions.

For the no-intervention group, the inclusion visit was the time-point, at which obesity was first identified. Exclusion criteria were to have declined participation in the interventions (31 children excluded) or a time gap of more than 6 months between the inclusion visit and an observation containing anthropometrical data (20 children excluded). To exclude potential typing errors identified as extreme changes (i.e., outliers) in consecutive BMI z-scores, data from 73 children were reviewed individually in a blinded manner by the two chief physicians JMB and ETV. In the exploratory subgroup comparison of duration of the interventions only, children treated less than 1 year and still enrolled when data were extracted were excluded (21 children), as it was uncertain if these children were to complete a minimum of 1 year in one of the interventions.

This pragmatic, real-life study followed children treated in one of two community-based lifestyle interventions located in two different cities in the eastern part of Jutland, Denmark. The children were included into the interventions based on their place of residence. The treatment for both interventions consisted of consultations with the children and their families and the interventions were managed by the parents in-between consultations. Both interventions were pragmatic and adapted toward the needs of the individual child. A more detailed description of the interventions can be found in the appendix. The differences between the two interventions are outlined below.

The Aarhus-intervention enrolled children 5–8 years of age with obesity. Children were identified with obesity at a mandatory health check-up and after identification the child’s parents were contacted by the community health nurse, who invited the child and family to participate in the intervention. At that time, an informed consent for participation was obtained. The intervention (i.e., treatment) was managed by specialized community health nurses employed at the municipality. During the intervention, the nurse conducted 3–4 consultations in the child’s home focusing on eating habits, mental health, screen time, daily physical activity, and sleep duration. In addition, voluntary participation in weekly supervised physical activity was offered free of charge. The intended duration of the intervention was 1 year during which height and weight was assessed at time of inclusion, and after 2 months, 6 months, and after 1 year. Children with at least three documented consultations within the first year were categorized as having 1 year of treatment.

The Randers-intervention enrolled children and adolescents 5–17 years of age with overweight or obesity but following our inclusion criteria, we excluded children not classified as having obesity. Participants were referred to the intervention from various entities such as general practitioners, hospital departments, or through self-referral; however, most of the children were referred from the community health nurse performing the health check-up at the schools. Informed consent from the parents was obtained before initiating treatment. The intervention was conducted by specialized community health nurses employed by the municipality. The families were seen at a local healthcare center with consultations focusing on eating habits, mental health, screen time, daily physical activity, and sleep duration. In contrast to the Aarhus-intervention, the Randers-intervention was more dynamic and did not include a maximum duration of the treatment nor a fixed number of consultations. The normal treatment period was up to 3 years and the consultations could be as frequent as every 8 weeks, depending on the need of the individual child and family.

The no-intervention group consisted of all identified and remaining children with obesity living in Aarhus Municipality, who did neither participate in nor declined participation in the intervention. Why these children were not invited into the intervention is unknown and can only partly be explained by the fact that Aarhus-intervention only enrolled children aged 5–8. The demographic characteristics of this group and differences with the intervention groups are presented as part of the descriptive results. Unexposed follow-up for children in the Aarhus-intervention (observations with obesity prior to the inclusion visit in the interventions) were also included in the no-intervention group.

BMI z-score (adjusted for age and sex) was defined by using a validated Danish reference material and was calculated for each observation [26]. The outcome was defined as change in BMI z-score per year (SD/year) with a 95% confidence interval (CI). The IOTF cut-off for obesity in boys at age 8 is BMI (and corresponding BMI z-score) equal to or greater than 19.8 kg/m² (z-score 2.87), at age 14: 27.6 kg/m² (z-score 2.79), and at age 18: 30 kg/m² (z-score 2.66) [25, 26]. Children were seen at health check-ups with height and weight measurements as a minimum at age 5–6, age 9–10 (not mandatory, but often performed regardless), and at age 14–15 years. Children were considered lost to follow-up, if they did not participate in a health check-up or moved away from the municipality. However, due to the study design it was not possible to identify the number of children lost to follow-up.

Data from children participating in or declining participation were extracted from data-capturing tools used by community healthcare nurses (TM-Sund in the Aarhus-intervention, NOVAX in the Randers-intervention). For all children included in this study, observations (containing height and weight) recorded at the mandatory health check-ups were also extracted from TM-Sund or NOVAX.

Data from children included in this study were merged with the Danish national registers at Statistics Denmark (DST) and were linked to their parents by the using unique family ID. The children were categorized as living with two adults or not living with two-adults by the variable family type recorded in the Danish Population register. As marker for SES, we used the variables Highest Completed Education, Danish Education Register, and the Equalized Household Income, Danish Household Register, respectively. To define the highest completed household education, the variable was recoded into duration of education (years) and stratified into: primary education (<10 years), high school or vocational education (10–12 years), short- and medium-cycle higher education (12–15 years) and university degree or equivalent (>15 years) for the parent with the longest education [27‒29]. To define household income, the Equalized Household Income (adjusted for tax payed, alimony and number of family members) was categorized into low, medium, and high income defined by tertiles [30, 31]. To account for inflation, the categorization was updated annually at time of inclusion. Immigration status was defined for each child as either being first, second generation immigrant or of Danish origin by the Immigration Register.

Data regarding mental disorders were obtained from the Danish National Patient Register [32, 33]. Children were categorized as having a family history of mental disorder if a parent had a psychiatric diagnosis at the time of the inclusion visit.

For descriptive analyses at time of inclusion, normal distributed data were analyzed using a one-way ANOVA (several means) or Student’s t test (2 groups), while non-normal distributed data were analyzed using Wilcoxon Rank-Sum test (2 groups) or a Kruskal–Wallis test (>2 groups). A Fisher’s exact test was used for categorical variables.

To model development in BMI z-score of repeated measurements during follow-up, a mixed effects model with random intercepts for participants and a random-coefficient for individual trends during follow-up was used. We present both unadjusted analyses to provide a crude estimate, and adjusted analyses to facilitate comparisons and to gain precision. By adjusting the model, we expected that developments over time would largely be insensitive to differences in risk factors between children. The development in mean BMI z-score for each group was modelled using linear splines with knots placed at inclusion and at 6, 12, 36, and 120 months. The model was adjusted for measures at inclusion of age, BMI z-score, sex, family type, highest completed household education, equalized household income, immigration status and presence of psychiatric diagnoses for the child and parents. A multiple imputation with chained equations was applied to replace missing values for family type (n = 164 [5.4%]), education (n = 29 [1.0%]), income (n = 278 [9.1%]), and immigration status (n = 154 [5.1%]). We used Rubin’s rule to obtain overall estimates based on the analyses of each of the 100 imputed datasets [34].

We examined the stratified effect of age, family structure, income, parental level of education and immigrations status before we examined their potential for effect modification for the developments in BMI z-score in the combined intervention groups as compared to the no-intervention group. We used the categorized variables of sex (only for effect modification), age (tertiles), family structure (not-two adults vs. two adults family), income (by the median), education (≤12 vs. ≥13 years of education), and immigration status (Danish origin vs. immigrant), respectively. In addition, we only examined the dichotomized effect of the duration of the interventions (less than vs. more than 1 year) because it was not possible to perform an effect modification analysis here.

We report estimates with 95% CIs and use 5% as significance level. Analyses were done in Stata 15 College Station, TX, USA: StataCorp LLC [35].

We included 3,040 children living with obesity for further analyses, 445 in the Aarhus-intervention, 258 in the Randers-intervention and 2,337 in the no-intervention group (shown in online suppl. Fig. S1; for all online suppl. material, see https://doi.org/10.1159/000540389).

Children in the Randers-intervention were younger compared to no-intervention. Children in the Aarhus-intervention were younger with a higher proportion of girls compared to the Randers-intervention and the no-intervention group. Parents in the Randers-intervention had a higher level of education and fewer were immigrants compared to the Aarhus-intervention and no-intervention. The equalized household income was higher in the Randers-intervention compared to the no-intervention group. Lastly, children in the Randers-intervention had a higher proportion of psychiatric diagnoses compared to the Aarhus-intervention (shown in Table 1).

This study included a median and maximum follow-up of 3.3 (Q1; Q3: 1.7; 6.6) and 10.3 years for the Aarhus-intervention, 1.9 (Q1; Q3: 0.9; 3.4) and 6.9 years for the Randers-intervention, and 2.7 (Q1; Q3: 1.3; 4.8) and 10.1 years for the no-intervention group, respectively (not shown in a table).

Both interventions demonstrated similar reductions in BMI z-score during the first 6 months compared to the no-intervention group (Aarhus-intervention: −0.15 SD/year vs. Randers-intervention: −0.24 SD/year, p = 0.23). A rebound in BMI z-score occurred during 6–12 months of follow-up for the Aarhus-intervention as compared to both the no-intervention group (0.18 SD/year) (shown in Table 2) and to the Randers-intervention (0.18 SD/year) (data not shown in a table). Considering the other time-intervals, i.e., 1–3 years and 3–10 years, we did not observe any statistical differences between the three groups (shown in Table 2 (the adjusted model) and Figure 1). The estimates of BMI z-scores development over time were nearly identical with and without adjustment, although precision improved marginally after adjustment (shown in Table 2).

By including all available observations during follow-up, we observed that only children in the Randers-intervention obtained a significant reduction in the annual change of BMI z-score as compared to the reference (no-intervention group) (−0.05 SD/year, CI: −0.08; −0.02, p < 0.001), while no change was observed for children in the Aarhus-intervention compared to the reference (0.01 SD/year, CI: −0.01; 0.04). The results were similar in the corresponding unadjusted model (shown in Table 2). By comparing the two interventions, we observed a significant reduction in the annual change of BMI z-score (−0.07 SD/year, CI: −0.10; −0.03) for children treated in the Randers-intervention as compared the Aarhus-intervention (not shown in a table).

By including all available observations for children in the interventions, we observed that income above the 50th percentile was associated with a reduction in BMI z-scores as compared to income below the 50th percentile (−0.05 per year, CI: −0.02; −0.09, p = 0.003). Immigrant status (first or second generation) was associated with an additional increase in BMI z-score (0.04 per year, CI: 0.00; 0.07, p = 0.03) as compared to children of Danish origin (shown in Table 3).

Children participating in the interventions for at least 1 year were slightly younger (7.1 years, Q1; Q3: 6.5; 8.9 vs. 7.4 years, Q1; Q3: 6.7; 9.3, p = 0.02) and having a higher BMI z-score (3.1 SD [0.7 SD] vs. 2.9 SD [0.7 SD], p = 0.02) as compared to children participating less than 1 year (data not shown). An annual reduction in BMI z-score were observed for both groups during follow-up, (durations for at least 1 year: −0.08 SD/year, CI: −0.10; −0.06, p < 0.001, less than 1 year: −0.04 SD/year, CI: −0.08; 0.00, p = 0.022). However, when comparing the two, children with the duration of less than 1 year increased BMI z-score by 0.04 SD/year, CI: 0.00; 0.08, p = 0.037 (shown in Table 3).

As illustrated in Figure 2, a comparable change in BMI z-score was observed during the first 6 months of follow-up regardless of the duration of the interventions (p = 0.65). However, during the following 6 months the children treated less than 1 year increased their BMI z-score compared to children with longer duration of the interventions (0.24 SD/year, 95% CI: 0.06; 0.42, p = 0.01) (not shown in a table). Despite the difference in the annual change of BMI z-score observed between the groups “duration less than a year” or “duration for at least 1 year,” we did not observe any differences when comparing these groups to the no-reference group (shown in Table 3 [7. Duration of the interventions]).

No effect modification for development in BMI z-score was observed for the combined interventions as compared to no-intervention with respect to sex (p = 0.37), age (middle vs. lowest tertiles [p = 0.36], highest vs. lowest tertiles [p = 0.30], highest vs. medium tertiles [p = 0.45]), family type (p = 0.90), income (p = 0.85), education (p = 0.17), and immigration status (p = 0.74) (data not shown).

In this study, we report long-term results from two community-based lifestyle interventions treating children with obesity compared to an untreated reference group. The two interventions had significant and comparable short-term (6 months) reductions in BMI z-score when compared to children with obesity not invited into any intervention; however, only the Randers-intervention with its more frequent clincal consultations and longer lasting intervention was able to maintain this reduction throughout follow-up. Nonetheless, the overall reduction in BMI z-score in the Randers-intervention was only modest and smaller than what is usually considered clinical relevant (a reduction of BMI z-score of 0.25 SD) [36, 37] and the overall clinical impact could therefore be questioned. Importantly by using effect modifcation analyses, we observed comparable effects of the interventions regardless of SES or immigrant background. However, for children in both interventions, lower family income, immigrant background, or shorter duration of the interventions (<1 year) were associated with an unfavorable increase in BMI z-score.

Although both interventions were overall comparable, focusing on family involvement and guidance related to eating habits, mental health, sedentary behavior and sleep duration, a possible explanation for the modest, however significant, reduction in BMI z-score observed in children in the Randers-intervention might relate to a more dynamic approach in the treatment of childhood obesity. The Randers-intervention had an explicit focus on the needs of the individual family including ability to offer a higher number of consultations as well as a longer duration of the intervention (up to 3 years) as compared to the Aarhus-intervention (∼1 year). Therefore, we wanted to verify if earlier described associations between longer duration of lifestyle interventions (2–3 years) and more favorable weight development for children with obesity could be rediscovered for this study [18, 22].

We found evidence for positive effects on BMI z-score development among children participating at least 1 year compared to children participating less than 1 year, essentially confirming earlier findings and emphasizing the importance of flexibility and the possibility for long-term engagement in lifestyle interventions when treating childhood obesity. In addition, we observed a comparable development in BMI z-score during the first 6 months of follow-up for children with long and short duration, which underlines the importance of continuous treatment beyond the first year. For children with obesity, early dropout from a lifestyle intervention is a major concern and has been reported to be associated with older age [18, 20, 21, 38], higher degree of obesity [39], lower parental education [18, 39], and symptoms of depression at the mother [21, 39]. In line with these earlier findings [18, 19, 21, 38], we observed higher age in children with the shortest duration of the interventions, which can possibly be explained by higher degree of autonomy. In contrast to a previous finding [39], a higher degree of obesity at inclusion was, in our study, associated with longer a duration of the interventions. However, this association has not been consistently observed in other publications [18, 21, 38].

Lower SES or having immigrant background have both been reported to be associated with an increased risk of developing obesity, the metabolic syndrome [16, 40, 41] and a decrease in compliance toward lifestyle interventions [42]. In our study, we did not observe any effect modifications regarding changes in BMI z-score comparing the combined interventions to an untreated reference group, indicating that lifestyle interventions have comparable effects on childhood obesity regardless of sex, age, SES, and immigration status. In addition, we did not observe associations between adherence and parental education or psychiatric diagnoses, which has earlier been reported [18, 21, 39]. This could be explained by the relatively small differences across the societal strata in Denmark, perhaps due to the free access to education and healthcare for all citizens.

Although no effect modification was observed in the interaction analyses, our explorative comparisons revealed that children in the interventions with immigrant background or a low family income (below the 50th percentile) showed an increased BMI-score as compared to Danish origin and high family income, respectively. Our results support previous reports [40‒43] and indicate that children with these characteristics are at a higher risk of an unfavorable weight development, despite comparable effect of the actual interventions. This may suggest that in order to increase the effectiveness of such family-centered interventions, these children may benefit from a more specialized and individualized approach.

The strength of the present study lies in the inclusion of a substantial number of children with multiple observations, as well as the identification of an untreated no-intervention group. This unique combination of clinical data and Danish registries made it possible to follow the children beyond the completion of the interventions and into a real-life setting. This combination of data reduced the risk of differential loss to follow-up as well as selection and recall bias [44].

We believe that using a validated Danish reference material would be the best choice defining BMI z-score the our Danish cohort [26], although we acknowledge that this can complicate interpretation and comparison with international BMI z-score measures. The obesity cut-off is not consistent for all ages when using BMI z-score in children, and this could to same extent explain the reduction observed in the no-intervention group, as we did not suspect this group to reduce their degree of obesity. Due to pragmatic follow-up, we observed a large individual variation in the number of observations, which is why we chose the mixed effect models as the best way to address this challenge. It was not possible to obtain height and weight or determine the exact day of leaving the intervention, and thus, it was not possible to restrict the follow-up to the post-intervention period. Due to the different durations of the two interventions, a higher proportion of the children could be receiving treatment in Randers-intervention compared to the Aarhus-intervention, when data were extracted and this could potentially induce bias. Some children could have received alternative treatment during the follow-up period and this could potentially have attenuated the associations. We cannot explain why a large proportion of the children were not invited to participate in the intervention – this could be investigated in future research projects. Another limitation may have been that children in the no-intervention group all were residents in the city of Aarhus, and this could increase the risk of bias due to differences between cohorts. However, the two cities are located close to each other (40 km) with similar prevalence of overweight and obesity in children leaving school (2012–18: 18% vs. 19%) [45]. The combination of the interventions could be a limitation. If there is a strong different effect depending on the type of the interventions, the interventions may not be considered together. Due to the study design and to the differences between the groups, there is a risk of confounding, despite adjustments.

While similar short-term reductions in BMI z-score were observed for both interventions, only the more dynamic Randers-intervention with longer duration was able to maintain this reduction throughout follow-up. Although an overall minor reduction in BMI z-score (less than 0.25 SD) was observed for the Randers-intervention, the clinical impact may only be modest. It was also particularly critical that the effect of no-intervention group was comparable to Aarhus-intervention after just 1 year of follow-up. Sex, age, SES and immigration status did not modify the effects of the interventions.

We hereby conclude that treating children with obesity in a longer and more dynamic family-centered lifestyle intervention can yield an overall effect. However, this effect may at the best be only modest and still not effective enough to induce a long-term beneficial development in BMI. That said, we still encourage other researchers to continuously evaluate existing obesity programs, thereby increasing our understanding and in addition, improve treatments modalities for these children.

Louise Lindholdt helped identify relevant variable for SES in the register of Statistic Denmark. Jonas Frey Rosborg Schaarup introduced and explained how to use mixed effect models. Kristiane B. Beciher proofread the manuscript before the final submission. We would like to thank the community healthcare workers responsible for measuring anthropometrics and in particular Sara Hyldig, Randers Municipality. We would also like to acknowledge the Danish Regions and the Danish National Center for Overweight for supporting the project.

This study was performed in line with the principles of the Declaration of Helsinki and approved by the Central Jutland Regional Committee on Health Research Ethics with record No. 1-45-70-27-20. The Danish Data Protection Agency approved use of registers from Statistics Denmark. This study has been reported to clinicaltrials.gov (protocol id: 2596) and to the local directory at Aarhus University (record No. 2016051-000001). Consent to participate: The need for informed consent was waived by the Central Jutland Regional Committee on Health Research Ethics. This study was register based, without any contact to the included families and in addition, the authors ensured that no data were reported on an individual level.

R.M.J., H.S., J.N.Ø., E.T.V., and J.M.B. are employed at Steno Diabetes Center Aarhus, a public hospital and research institution situated in the Central Denmark Region, which is partly funded by an unrestricted grant from the Novo Nordisk Foundation. The remaining authors have no conflicts of interest to disclose. The Danish Regions and the Novo Nordisk Foundation had no role in the design and conduct of the study.

R.M.J. is supported by a public grant by the Danish Regions, Denmark “The Joint Grant for Prevention.” The article processing charge is covered by the Danish National Center for Obesity.

R.M.J., H.S., J.N.Ø., S.H., K.S., E.T.V., and J.M.B. conceptualized and designed the study. H.S., J.N.Ø., and J.M.B. supervised R.M.J. and the study process. R.M.J. collected data, designed and carried out the analyses. E.T.V. and J.M.B. reviewed data on outliers. R.M.J. drafted the initial manuscript. All authors critically reviewed, revised and approved the final manuscript as submitted and agree to be accountable for all aspects of the work.

The data that support the findings of this study are not publicly available due to the legislation of the Danish General Data Protection Regulation. For requests regarding data access, contact the corresponding author, R.M.J.