Effects of a novel mobile health intervention compared to a multi-component behaviour changing program on body mass index, physical capacities and stress parameters in adolescents with obesity: a randomized controlled trial

A simple randomized (1:1) controlled not blinded trial was conducted in the Children’s Hospital of Eastern Switzerland, which is a specialised childhood obesity management center (trial registration number: NCT03270423, date of registration: 26/07/2017, www.​clinicaltrials.​gov). Between September 2016 and September 2017, we recruited study participants among adolescents with overweight or obesity aged 10 to 18 years, requesting treatment in our center. Subjects were either referred by their physician or contacted us after seeing the study advertisement (flyers, media release). Participants were eligible if they had a BMI above the 90th percentile (>P.90) [14] with at least one risk factor or co-morbidity or a BMI above the 97th percentile (>P.97). Exclusion criteria were: major somatic or psychiatric diseases, assessed as described elsewhere [15], without adequate treatment; taking medication with the potential to affect weight (e.g. antiepileptic drugs or methylphenidate); participation in another obesity treatment program in the past year or already using a mobile phone application which promotes weight loss. The study took place between January 2017 and November 2018 according to study protocol. It included an intensive intervention phase of 5.5 months and a maintenance phase of 6 months. Study physicians recruited 41 study participants who, under the surveillance of the independent hospital pharmacist, were randomized to either the PM group or the CON group using the “randList” software (datinf gmbh Tübingen). To motivate children and adolescents to participate in the study, we offered five smartphones per group to the five most “successful” participants of the PM and the CON group according to a score of maximum 15 points, taking into account participation (missing appointments resulted in fewer points) and changes in physical performance and BMI-SDS.

Baseline investigations were performed, on average, 5 weeks before the start of the intervention. After baseline investigations and randomization, the intensive intervention phase started, lasting 5.5. months and followed by two visits (at nine and 12 months after intervention start) to ensure maintenance of behavioural changes (maintenance phase, see below).

During the intensive intervention phase, the PM group had an individual multi-component BCI following Swiss guidelines [4], including handouts on nutritional education and physical activity [16], and was equipped with a smartphone with the PathMate2 (PM) app [17]. The PM app was developed with the MobileCoach open-source software for health interventions [18, 19], aiming to support behavioural changes in accordance with state-of-the-art multi-component interventions. The effects of earlier versions of the PM app have been evaluated in longitudinal studies [20, 21]. The gamified PM app included two chat channels (Fig. 1): in the first channel, a conversational agent (virtual coach: Anna or Lukas) chatted with patients; in the other channel, HCP (human coaches) and patients were able to chat with each other [17]. The conversational agent chatted daily with the PM participants, encouraging them to achieve challenges like a number of steps per day, performing relaxing breathing exercises [for more details [22]], taking photos of their meals, or answering quality of life questions in order to earn virtual rewards. Most of the time and to allow efficient and reproducible conversational turns, patients were able to respond to Anna/Lukas with predefined answer options. In the app dashboard, the patients could see their progress during the game. Furthermore, a dashboard overview alerted HCP, in case of lack of chat interaction during more than 2 days. Usage time of all mobile apps, and self-reported comfort with and carrying of the smartphone on the body were also recorded.

On the other hand, during this intensive intervention phase, the CON group had an individual multi-component BCI following Swiss guidelines [4] without the PM app. This BCI uses behavioural change techniques such as goal setting, self-monitoring, stimulus control and behavioural contracting to support a healthy lifestyle. The CON received homework exercises based on worksheets. They had to bring their expanding handbook at each hospital visit. During on-site visits, topics like physical activity, diet, mental health and well-being, motivation for healthy choices in everyday life as well as family eating activities and communication habits were discussed.

During the intensive intervention phase of 5.5 months, the PM group had daily interactions with the conversational agent, four on-site visits and two remote counselling sessions via telephone as well as a maximum of 10 min chat time per participant with the human coach, resulting in a total contact time with HCP of approximately 6 h per participant. During the same period the CON group had seven on-site visits with total contact time with HCP of approximately 8 h per participant.

At the end of the intensive intervention phase the use of PM app was ended and the PM group returned the smartphone with the PM App to the HCP.

The maintenance phase served as the self-responsible implementation of a healthy lifestyle according to what had been learned in the intensive intervention phase. It started after the end of the intensive phase and lasted 6 months, including two BCI reflecting sessions, identical in both groups at nine and 12 months after the start of the study. In the 9-month-visit, both groups had a physical examination, a biofeedback session and a BCI dialogue on goal setting and emotions. The final 12-month-visit consisted of a physical examination, a biofeedback session with completion of stress questionnaires and a discussion about maintaining a sustainable healthy lifestyle.

All participants performed biofeedback training exercises during hospital visits, under the guidance of HCP, at baseline, and three (data not shown), 5.5 and 12 months after the start of the intervention. Adolescents did a two-minute relaxation exercise: they had to perform abdominal breathing while trying to decrease the amplitude of the skin conductance response, which they could observe in real-time on the computer screen. Skin conductance finger sensors measured the sweat gland activity, which is a sensitive indicator of arousal [23] (NeXus-10 Mark II, Mind Media, Netherlands). Plasma cortisol level, as stress marker, was measured before and after each biofeedback relaxation exercise.

Outcomes were assessed at baseline (on average 5 weeks before the start of the Intervention), at the end of the 5.5-month intensive intervention phase (T1), and after the six-month maintenance phase (T2, at 12 months). All measurements were carried out in a non-fasting state and at the same time of day for each subject.

The primary outcome was BMI-SDS. BMI is calculated as weight in kilograms divided by height in meters squared. Since BMI normally increases with age in children and adolescents and taking into account the age- and sex-related skewness of BMI distribution, BMI was adjusted according to the LMS method for sex, age as well as the skewness of BMI distribution, and indicated as standard deviation score (BMI-SDS), identical to BMI z-score, as referred to on the WHO web site (https://​www.​who.​int/​tools/​growth-reference-data-for-5to19-years/​indicators/​bmi-for-age) and in Braegger et al. [14]).

Secondary outcomes included body composition, waist-to-height ratio, physical capacities, blood pressure, heart rate and stress assessment. Muscle and fat mass were evaluated by bioelectrical impedance analysis (InBody720) [24, 25]. Waist circumference (WC) was measured with a flexible non-elastic band, and defined as the smallest abdominal circumference between the lowest rib and the upper anterior iliac spine [26]. The intra- and inter-evaluator coefficient of variation of waist circumference measurement in our HCP team was 1.3 and 6.6% respectively (means ±SD: 83.3 ± 1.1 cm and 78.9 ± 5.2 cm, respectively). The measurements of each patient were always performed by the same HCP. Waist-to-height ratio (WHtR), an indicator of obesity-related cardiovascular risk, was calculated using the formula: waist circumference in centimetres divided by body height in centimetres [27]. Puberty was estimated by Tanner stages [28, 29]. Blood pressure (at the right upper arm with an appropriate cuff) and heart rate were measured sitting quietly for 5 min, following standardized procedures [30]. Physical capacities (strength, agility, flexibility, endurance, balance) were assessed by a modified Dordel-Koch-Test (Dordel- Koch-Test [31, 32] plus the plate tapping selected from the Eurofit Test [33]). The modified Dordel-Koch-Test was always conducted by the same physical education teacher for all participants throughout the study and included eight tests:

Power and sample size were calculated on the basis of the changes observed in our childhood obesity management centre [7], which showed a mean BMI-SDS change (± SD) of − 0.33 ± 0.44 after 1 year of individual multi-component BCI. We expected to achieve similar results with our two study groups. Data simulation from the above distribution showed that 17 patients per group would provide 80% power to detect a significant BMI-SDS change in one of the treatment groups using Wilcoxon signed rank tests. To account for a few dropouts, we aimed to recruit 20 patients per group.

Statistical analyses were performed using rank tests because some outcomes were not normally distributed. The distribution of outcomes across all patients and within each of the treatment groups was described by medians with ranges. P-values for the significance of changes from baseline to T1 (D1) or T2 (D2) were calculated using Wilcoxon signed-rank tests. P-values for differences between the two groups (at individual time points and for change from baseline to T1 or T2) were obtained with Wilcoxon rank sum tests. Associations between numeric variables were described with Spearman rank correlations. Confidence intervals for these coefficients were derived from the normal approximation obtained with Fisher’s transformation. Associations between categorical variables were described by frequencies and tested with Fisher’s exact tests. A significance level of 0.05 was applied for all tests without correction for multiple testing given the exploratory nature of the analyses performed for the secondary outcomes. Data analysis was performed using the statistical software R, version 3.5.1 (2018-07-02). For the primary outcome we also conducted a last observation carried forward (LOCF) analysis and a responder analysis with Fisher’s exact test (responders defined as participants showing BMI-SDS change < 0). Additional analyses of group differences with generalized estimating equations (GEE) as well as calculations of intra- and intergroup effect sizes were performed for the primary outcome and for physical capacities.

Figure 2 provides a flowchart for enrolment, randomization, intervention and follow-up (maintenance phase) of study participants. Tables 1 and 2 display baseline characteristics. 87% of the participants (27 out of 31) were referred to us from their physician for obesity therapy and 13% contacted us directly wanting to participate in our study. After randomisation and the start of intervention, two patients dropped out due to intercurrent diagnosis of psychiatric disease and thus were excluded from all analyses. Three further patients, who dropped out after the first 5 months of the study, were included in the analyses of outcomes at T1 but excluded in the analysis of outcomes at T2. For all variables except BMI-SDS, the number of patients in the analysis was further reduced by non-compliance with the study plan or physical incapacity due to illness.

The increase of BMI-SDS (median 2.61, range 1.7–3.5) from baseline to the start of the intervention was not statistically significant (p-value 0.09). Subjects’ characteristics were similar among groups (Tables 1 and 2), though mental health problems assessed by SDQ total score displayed different distributions. 51.6% of mothers and 45.2% of fathers were of Swiss origin.

Physical performance and capacities were similar in both groups, except CON boys achieving better results in jump distance and 6-min-run compared to PM boys (Additional File 1). With few exceptions, subjects had lower values than the age- and gender-specific references [31, 37].

Results of the primary and secondary outcomes are presented in Figs. 3, 4, 5, 6, 7 as well as in additional files 2 to 6 and 11.

After half a year, 51% of the PM participants were still using the App and 54% were active on achieving challenges. The average app usage rate was 71.5% and the average adherence rate was 57.2% during the first 5.5 months of the study. During this period, all CON kept their hospital appointments, 95% brought, as asked, their handbook to each visit and 68% did the assigned homework.

According to self-reports 3 months after study start, PM subjects carried, on weekdays, 10 h the smartphone with them (median, range 0–21.5), whereas the time the smartphone stayed outside their bedroom at night was 8 h (range 0–13); these results were similar for weekends and did not change during the course of the 5.5 month-app intervention. The average time per day that adolescents spent with their smartphone was 16.2 min, with YouTube and the PM app being the two most frequently used apps (for details, see Additional file 7, Tables a&b). Moreover, the PM app was used, on average, for 48 s per day, which reflects approximately 16% of the overall smartphone usage during the intervention period. Thus, no serious adverse events with respect to smartphone addiction were evident.

Therapy success (reduction in BMI-SDS) was associated with neither baseline age, sex, BMI-SDS, psychological or school problems of the participants, nor BMI, foreign nationality or psychological problems of the parents.

From T0 to T1, patients with a greater reduction in BMI-SDS also had a greater reduction in waist circumference and in fat percentage, with little or no increase in muscle mass (analysis performed for the whole cohort, Additional file 8, see most right column in scatter plots). However, changes in BMI-SDS, waist circumference, muscle and fat mass from T0 to T1 were not significantly associated with changes in physical capacities, or with the levels achieved at T0 or T1 (Additional file 8, see first three columns from left to right). Thereby, even if most patients improved their physical fitness and capacities, and many also reduced their BMI-SDS and body fat, the magnitude of these changes varied individually, and these individual variations were not related to each other.

We found a weak positive correlation between changes in chronic stress and changes in waist circumference and BMI-SDS after 12 months (Additional file 9). Otherwise, no significant associations were found between BMI-SDS and stress parameters (TICS scores, cortisol before biofeedback, cortisol change after biofeedback) or changes in BMI-SDS, WC, WtHR and stress changes during the study (data not shown).

Attrition rates observed in childhood obesity management programs, mainly in the USA, are reported to reach 27 to 73% [38], while the dropout rates of multi-professional BCI in group setting for children and adolescents in Switzerland range between 10 and 20% [6]. Despite an extensive preliminary assessment and a signed informed consent, but in accordance with the above numbers, 24% of our subjects withdraw between the recruitment and the start of the study. However, 94% of the eligible PM patients who started the study also completed it. Moreover, half of the PM participants were still using the App daily after half a year, in contrast to the high dropout rates reported in digital health interventions with adherence dropping after a few weeks [39, 40]. The high adherence rate seen in the PM group may be explained by the daily encouragement through a peer-appealing conversational agent with real-time feedback, the rewarding game system, the competition with peers and the low threshold of communication with HCP. During the first 5.5 months of the study the CON group also showed a high compliance with 100% attendance at hospital appointments and 85% of the eligible CON patients completing the study. This high adherence to therapy could be attributed to the structured BCI and to the personal counseling from childhood obesity specialists.

The total smartphone app usage time (including PM app), which was 16.2 min per day during the 5.5-month app intervention, remains (even with max. value of 88.6 min per day) clearly below the 2 h and 42 min reported among 13- to 18-year-olds in the U.S. [41]. Additionally, as the most frequently used app was YouTube, we can argue that our app did not promote mobile gaming addictions.

According to international data [42] only 2% of adolescents exhibit healthy lifestyle behaviours; hence, in pediatric obesity treatment, learning to adopt a sustainable healthy lifestyle is a key goal. In this sense, our study investigated whether a novel smartphone application can support adolescents in adopting long-lasting healthier habits while avoiding yo-yo effects [43]. Although body mass index can only indirectly indicate a healthy behaviour, we have chosen it, in form of BMI-SDS corrected for sex and age, as our primary outcome, as it is the most widely used parameter of measuring obesity.

In the whole cohort, we observed a small but significant improvement in BMI-SDS (− 0.09) after 5.5 months of intervention, and changes were maintained up to one year (− 0.12). These findings suggest that the individual multi-component BCI was effective, similarly to previous studies in various settings and age groups (BMI-SDS change from − 0.06 to − 0.13SD [3]). Interestingly, the number of contact hours in our study was much lower than the 26 h found to be effective through comprehensive and intensive BCI [44]. As a significant decrease in BMI-SDS and waist-to-height ratio were found in the control group only, we conclude, in line with prior work [44], that an intensive personalized behavioural intervention can result in a greater short-term reduction in weight status than a mobile health app program. However, in our study, the effects on BMI-SDS were not maintained at one-year in controls, which may be explained by the wide variations of the CON results and/or the dropout of some patients showing large changes in BMI-SDS from T0 to T1 (two control patients, who dropped out before T2, had a BMI-SDS decrease of − 0.74 at T1, which was larger than that of the entire control group of − 0.37). Nevertheless, the last observation carried forward analysis did not present different results with regard to BMI-SDS (only applied for the primary outcome, Additional file 10). Another explanation for the non-significant BMI-SDS change after one year could be that only two contact hours with HCP between 5.5 and 12 months after study start (maintenance phase) were probably insufficient in order to maintain behavioural changes.

During the intervention, changes in physical capacities and body composition were similar to previous studies [6, 45]. In both groups, we found similar improvements after 5.5 and 12 months in strength measures and agility. The significant improvement of muscular strength, as seen in our study, is of critical importance in improving overall health of adolescents with obesity as well as their comfort in daily life, thus increasing their autonomy [46]. The 6-min-run-test (aerobic fitness) is generally highly and negatively correlated with BMI-SDS [47]; this may explain, why in our study only controls (n = 12) showed an improvement in 6-min-run-test at T1 but not at T2. Although balance in obese children has previously been described as impaired [48], our patients showed optimal balance scores already at baseline. The PM intervention resulted in increased muscle mass at 5.5 months and a reduced percentage of fat mass throughout the study. As no correlations between changes in physical capacities and muscle mass were found in our study, it could be argued that the significant increase in muscle mass in both groups could also be a consequence of a synchronized pubertal growth. However, more than two thirds of the participants had a Tanner stage of 4 or 5, and thus had already passed the pubertal stage, when a maximum increase in muscle mass can be expected [49, 50].

In prior Swiss research, and in consistence with our results, systolic hypertension was present in 47.6% of children with obesity [51]. In our study, we did not observe any change in 6-min run test at 12 months, which may explain the absence of change in blood pressure or pulse rate [52].

The effectiveness of IT-mediated obesity interventions in improving health status in this age group remains uncertain, probably due to insufficient intervention duration, small sample size and short-term follow-up. In the 2018 meta-analysis of Ho et al. [53], six RCTs of internet-based self-monitoring interventions revealed a small reduction of BMI or BMI z-score; however, the quality of evidence was low due to the risk of bias and imprecision. On the other hand, in the review of Chen and Wilkosz [54] ten out of 14 studies evaluating the effectiveness of technology-based interventions for obese adolescents showed no effects on BMI or BMI z-score.

Only a few IT-supported interventions have, in line with our results, reported significant positive effects on the percentage of body fat [55, 56], skeletal muscle mass [55], as well as physical activity [57, 58] and muscular fitness [59].

In consideration of the study results, when access to a specialized childhood obesity management center is limited and the adolescent’s primary goal is the improvement of physical status and body composition, PathMate2 app could be a feasible option, if combined with individual counselling sessions. The PathMate2 app could also potentially be implemented in a follow-up program after an intensive obesity intervention. However, if the main focus of an obesity intervention is the reduction of BMI-SDS, the PathMate2 app intervention in its current form would probably be inadequate; in this case more contact hours with HCP and longer or repeated intermittent periods of app usage (longer than 6 months) would probably be needed in order to achieve a satisfactory BMI-SDS reduction.

With regard to acute stress response, subjects in our study exhibited significantly lower cortisol levels after the biofeedback relaxation exercises. These findings support the conclusions of prior research [12, 13], that psychophysiological interventions could have a clinical utility in obesity therapy in lowering emotional stress connected with eating disorders.

The present study is, to our knowledge, the only one that has investigated the long-term effect of biofeedback training on obesity related stress response in adolescents. Our hypothesis that the adolescents with obesity who would be most successful in reducing the maladaptive response to chronic stress would also exhibit a decrease in BMI-SDS was not confirmed; however, we found a weak positive correlation between changes in chronic stress (TICS) and changes in waist circumference and BMI-SDS after 12 months. Similarly, the RCT of Emmanouil et al. [60] found no changes in BMI-SDS after an eight-week stress management intervention program in Greek children and adolescents with obesity. Interestingly, baseline markers of acute and chronic stress in our patients, namely cortisol levels and TICS scale, were in the normal range and did not indicate a maladaptive stress response, at least not with the methods used. Although biochemical changes of stress parameters as well as a dysregulation of the cortisol metabolism [61‐64] can be found in patients with obesity, obesity is not always related to a hyperresponsive HPA axis and high cortisol levels [61], and the large interindividual variation in glucocorticoid sensitivity [65] could explain the normal cortisol levels in our patients with obesity.

Further research in a larger sample is therefore needed to prove the long-term effectiveness of biofeedback exercises on distress and obesity and identify the best candidates for psychophysiological intervention.

In our study, we did not find any associations between therapy success and baseline characteristics. However, the two-year Swiss childhood obesity study (KIDSSTEP) [6] as well as 2- to 5-year-follow-up German studies [66, 67] in children with obesity after one-year-interventions showed that the decrease in BMI-z-score was the greatest in children before pubertal age and the smallest in adolescents above 14 years. This association could explain why in our participants, who had a median age of 13.6 years, the BMI-SDS decrease was not significant in the 12-month follow-up.

In the present study, changes in physical capacities were not associated with changes in BMI-SDS, waist circumference, muscle mass or fat mass. Likewise, Reichert et al. [68] concluded that literature data offers only limited support for a causal link between physical activity and adiposity in adolescents; in children, however, the population-based longitudinal observation conducted by Metcalf et al. [69] supported a reverse causality where physical inactivity appeared to be the result of fatness rather than its cause.

The greatest strength of the present study is its randomized controlled design.

Conducting this low-intensity intervention (< 25 contact hours [44]) in a real-world setting with the well-known obesity pitfalls [6] and with adolescents with obesity suffering from challenging somatic and mental comorbidities, makes the results of this study more applicable in clinical practice. The attrition rate after the start of intervention was low and the adherence was high in both groups.

It should be pointed out that data privacy during the patients’ interaction with the PM app was guaranteed using a MobileCoach server located in Switzerland and compliant with Swiss data protection regulations [17, 18].

A limitation of this study is the small sample size due to the well-known recruitment difficulties related to this age group [6, 70, 71] and potential obesity comorbidities (for example psychiatric diseases), the limited time and resources of the families, as shown in the swiss KIDSSTEP study [6], and probably the lack of motivation in participating in a scientific study since the expenses of multi-component BCI in Switzerland are covered by health insurances. Therefore, and because adolescence is characterised by an increased reward-seeking behaviour [72], we decided to reward the best five participants of each group. A potential selection bias due to the attractivity of the smartphone could explain the drop out of five patients after being randomized in the control group and signing the consent forms (Fig. 2, hypothesis: the “less motivated” CON dropped out while the “more motivated” CON chose to continue the study despite not getting into “the smartphone group”). The selection of motivated adolescents might be associated with a better compliance of the control group. Lastly, although co-operating psychologists recommend the use of Trierer Stress Inventar questionnaire to evaluate stress in adolescents from the age of 13 years, TICS is validated on a population above 16 years of age. Otherwise, all other outcome measures were based on objective and validated methods for the age group investigated.