Childhood obesity remains a significant health problem in Canada [1]. Evidence shows that family-based multicomponent interventions that integrate self-regulatory strategies (ie, goal setting, graded tasks, and self-monitoring) and support changes at the familial and individual levels are necessary to significantly affect child weight outcomes (eg, BMI, waist to hip ratio, and total fat mass [2-9]). However, a 2018 meta-analysis [4] found that family-based multicomponent behavioral interventions had a small effect in reducing children’s BMI in efficacy trials versus standard-of-care controls (β=−.16, 95% CI −0.24 to −0.07).

Mobile health (mHealth) interventions offer a promising adjunct or alternative to in-person treatments to support lifestyle behavior change [10,11]. Several reviews [12-15] and meta-analyses [16,17] have suggested that mHealth interventions offer multiple advantages to in-person interventions (eg, real-time data collection, intervention in natural environments, lower costs, health behavior tracking with feedback, and incorporation of gamified elements), which may appeal to children and youth [12]. Data on the efficacy of mHealth interventions for the prevention and management of childhood obesity are promising but limited as this is still a rapidly evolving field of research [12,14,18].

mHealth interventions for children living with obesity are most often evaluated using randomized controlled trials and, in some cases, evaluate the dose of the intervention received to provide a better understanding of their effects [19-24]. Dose-response analyses are often measured in terms of total minutes or percentage of content examined; however, this approach does not provide a nuanced picture of how users may benefit from different mHealth intervention components (ie, what design elements of the app may be more successful in engaging participants and promoting health behavior change) [25,26]. Studies examining how intervention exposure affects behavior change cannot solely focus on the quantity of the intervention received, but must also consider how participants engage with the active ingredients of the intervention—namely, the features that support behavior change.

mHealth interventions are particularly well-suited to examine in greater detail which components of the intervention participants engage with through app analytics data. Recently, there have been calls to develop analytical methods to process the vast amounts of data that are available when using mHealth technologies [27] and identify digital phenotypes (ie, user typologies derived from individuals’ patterns of interactions with specific app features) [28,29]. Although digital phenotypes have been used in other areas of health research (eg, diabetes [30], sleep [31], mental health [32]) and dietary and physical activity behaviors in a nonclinical sample [33], little attention has been paid to the treatment of obesity in childhood. Some studies have investigated which app features participants use [23,33] and individual characteristics associated with partial or total use of an intervention [33-36]. However, most studies evaluated usability derived from self-reported measures (eg, asking participants about their preferences and use of app features), total app use, or the use of individual features instead of focusing on patterns of app use [23,34,37].

To our knowledge, no study targeting childhood obesity has identified user typologies based on participants’ engagement with objectively measured components of an mHealth intervention. To address this gap, this study aimed to (1) identify digital phenotypes of Canadian children with overweight or obesity and their parents who used an mHealth app (the Aim2Be app [25]) over a 3-month period, (2) explore whether participants’ characteristics differed by digital phenotype, and (3) evaluate 3-month changes in children’s BMI z scores (zBMI) and dietary, physical activity, and screen time behaviors across digital phenotypes.

This study was a secondary analysis of data collected from a randomized controlled trial evaluating the efficacy of the Aim2Be app (version 2) to improve lifestyle behaviors and adiposity among children with overweight or obesity [25,38]. The trial was registered on ClinicalTrials.gov (NCT03651284) on August 29, 2018 [25]. The CONSORT-EHEALTH (Consolidated Standards of Reporting Trials of Electronic and Mobile Health Applications and Online Telehealth) checklist [39] is available in Multimedia Appendix 1. Data analyzed in this study were collected from March 2019 to June 2020.

The evaluation protocol was approved by the Children’s and Women’s Research Ethics Board at the University of British Columbia (H16-03090/H17-02032), the Health Research Ethics Board at the University of Alberta (Pro00076869), the Hospital for Sick Children Research Ethics Board (REB1000059362), the Hamilton Integrated Research Ethics Board (project 4250), and the Children’s Hospital of Eastern Ontario Research Ethics Board (18/01E). All the participants provided web-based consent before participating in the study.

The detailed study protocol has been published elsewhere [25]. The participating families (N=214) were recruited from 6 weight management clinic sites across Canada, as well as through Facebook. Children were eligible to participate if they were aged 10 to 17 years and were overweight or obese, as defined by the age- and sex-specific World Health Organization cutoffs [40]. After providing consent, eligible participants completed a web-based survey, three 24-hour dietary recalls, and received assessment tools for height (measuring tape) weight (digital scale), and physical activity (Fitbit Flex 2, Fitbit Inc) to complete baseline measurements. Participants completed follow-up assessments at the 3- and 6-month follow-ups. Families randomized to the experimental group (107/214, 50%) had access to the app after completing baseline measures. Waitlisted control families (107/214, 50%) were given access to the app after completing their assessment at the 3-month follow-up. This study combined data collected from baseline to 3 months in the intervention group, and from 3 to 6 months in the waitlisted control group (Figure 1). Randomization was successful, and participants’ characteristics did not differ between the intervention and the waitlisted control group; however, our analyses were not based on the randomization group but dependent on users’ engagement.

The theoretical foundations of the Aim2Be app have been described elsewhere [25]. Briefly, the app, which was cocreated by Ayogo Health Inc [41] and the Childhood Obesity Foundation with expert input, aimed to promote healthy lifestyle behaviors among children and their parents by targeting dietary, physical activity, screen time, and sleep behaviors while emphasizing healthy body image, strong self-esteem, and living green [25]. The behavior change techniques incorporated in the app are grounded in social cognitive theory [42], the Player Experience and Need Satisfaction Model—an extension of the self-determination theory [43,44]—and the Agency, Challenge, Uncertainty, Discovery, and Outcomes framework [45]. The content within different features of the app was also informed by the Canadian 24-Hour Movement Guidelines [46] and the Canadian dietary guidelines in place at the time of the study (Canada’s Food Guide 2007) [47].

The Aim2Be app features fall under 3 broad domains: behavioral, gamified, and social. The behavioral domain draws on self-regulatory strategies such as goal setting, self-monitoring, and graded tasks to facilitate behavior change by strengthening self-regulatory skills [4,8]. The gamified domain focuses on increasing participants’ enjoyment, engagement, and motivation through various gamification elements (eg, personalization, challenges, uncertainty). The social domain facilitates peer support, behavior modeling, and interaction with other app users or with a coach through different interactive features (eg, answering poll questions, viewing poll results, posting on the social wall, and responding to others’ posts). Social support is also provided to children through a companion app for parents, which aims to facilitate behavioral changes through a positive familial environment, reinforcement strategies, and environmental and stimulus control. Figures 2 and 3 illustrate screenshots of the child and parent interventions, respectively.

In addition to the parent companion app, 2 very similar versions of Aim2Be were developed for preteens (aged 10-13 years) and teenagers (aged 13-17 years), with 3 app features (ie, stages, posting on a social wall, and responding to others’ posts) available only to teenagers. As this study combined data from both teenagers and preteens, features only available to both groups were included in the analyses.

From the 214 parent-child dyads, the mean age of the children was 13 (SD 2.2) years, and the sample was evenly split among boys (104/214, 48.6%) and girls (110/214, 51.4%). Approximately 92.5% (198/214) of the participating parents were mothers, and 71% (152/214) were married or living with a partner. The mean age of parents was 44 (SD 6.2) years. Just over half of the parents (120/214, 56.1%) had not completed a university degree. Approximately 60.3% (129/214) of parents self-identified as having a White or European descent, 16.8% (36/214) reported mixed ethnicity, 5.6% (12/214) reported an East or Southeast Asian descent, 4.2% (9/214) reported a South Asian descent, and 3.3% (7/214) reported an indigenous descent. Household income ranged from <CAD $50,000 (US $37,500; 36/214, 16.8%) to >CAD $150,000 (US $112,500; 35/214, 16.4%). Approximately 30.8% (66/214) of parents reported incomes between CAD $50,000 (US $37,500) and CAD $100,000 (US $75,000), and 25.7% (55/214) reported incomes between CAD $100,000 (US $75,000) and CAD $150,000 (US $112,500).

Table 1 summarizes the results from the LCA and Figure 4 provides plots for selecting LCA indices for both the child and parent models. Fit indices and interpretability of the classes supported a 4-class solution among children. By contrast, the parent LCA fit indices pointed to a 5- or 6-class solution; however, further evaluation of the potential solutions led to the retention of the 5-class solution. Although 6 of the 11 indices showed the 6-class solution as the best option, the relatively small sample for 2 of the classes suggested an overextraction. Therefore, the 5-class model was retained as the final solution for the parents. In addition, the 5-class solution made more substantive sense. Moreover, the average posterior probability for both child- and parent-selected models ranged between 91% and 99%, indicating well-differentiated classes for the 5-class solution. Thus, our results suggest excellent differentiation between the classes.

Figure 5 shows children’s and parents’ digital phenotypes (A and B, respectively). Figure 5A shows 4 children’s digital phenotypes (N=214): unengaged, dabblers, partially engaged, and fully engaged. Unengaged (79/214, 36.9% of users) included children who did not interact with most of the app features with exception of check-ins. Dabblers (42/214, 19.6% of users) regrouped children who did not use most behavioral features of the app (eg, completing tasks and reading or reflecting on articles) but predominantly interacted with gamified and social features, including collections and the digital coach. Partially engaged (61/214, 28.5% of users) included children who were low users of the behavioral features, particularly regarding task completion and reading and reflecting on articles but had greater use of the check-in feature. Partially engaged children had mixed interactions with the gamified and social features, with greater use of the collections and the digital coach, respectively, but rarely read stories or completed quizzes. Fully engaged (32/214, 15% of users) comprised high users of most app features and included children who engaged the most with the active ingredients of the app (ie, the behavioral features such as setting aims and completing tasks).

Figure 5B shows 5 parental digital phenotypes (N=214): unengaged, socially engaged, independently engaged, partially engaged, and fully engaged. Unengaged (103/214, 48.1% of users) included parents who did not use most of the features, with the exception of check-ins. Socially engaged (35/214, 16.4% of users) regrouped parents who engaged with the social features of the app by creating posts on the social wall, answering poll questions, and interacting with the live health coach. However, socially engaged parents had low use of the behavioral and gamified features and, in particular, did not complete any tasks within the app. Independently engaged (18/214, 8.4% of users) comprised parents who made little use of the social features (the only social feature they used involved direct messages with the live health coach but did not interact with other parents). Instead, independently engaged parents focused their attention on the behavioral features of the app and mostly set aims, read articles, and completed check-ins; however, they also interacted with all the behavioral features to some degree. Partially engaged (32/214, 15% of users) included parents who had a mixed use of most app features, indicating that their engagement with some behavioral (eg, aims and check-ins) and social (eg, posts) features was evenly split between low and high use. Partially engaged parents tended to be high users of the article feature, low users of the answer poll feature, and nonusers of the digital coach feature. Hence, their overall engagement with the behavioral features tended to be greater than with the gamified and social features. Finally, fully engaged parents (26/214, 12.1% of users) included users who interacted extensively with all app features, except the digital coach feature.

Table 2 and Table 3 show the distribution of demographic factors across child and parent digital phenotypes, respectively, with relative risk ratios and significance levels available in Multimedia Appendix 2 (Tables S1 and S2). The results are presented separately for children and parents. Children in the fully engaged phenotype were 1 to 1.5 years younger than children belonging to the dabblers (P=.04) and unengaged (P=.003) phenotypes. Furthermore, children from the dabblers phenotype were more likely to be in a household with an income >CAD $80,000 (US $63,771) than children belonging to the fully and partially engaged phenotypes (P=.03 and P=.047, respectively). Parents in the socially engaged phenotype were 2 to 3 years older than parents in the fully engaged (P=.02) and unengaged (P=.01) phenotypes. Moreover, fully engaged parents were more likely to be married, common law, or living with a partner than parents belonging to the independently engaged (P=.02) and unengaged (P=.01) phenotypes, who were more likely to be single, divorced, or widowed.

Figure 6 shows the distribution of parental digital phenotypes across children’s phenotypes, highlighting how their phenotypes were strongly associated. At one end of the spectrum, fully engaged children were more likely to have fully and partially engaged parents, and at the other end, unengaged children were more likely to have unengaged parents.

Table 4 summarizes the 3-month changes in zBMI, diet, physical activity, and screen time across children’s and parents’ digital phenotypes, with statistically significant (P<.05) or borderline significant (P<.10) comparisons shown in Figure 7, where panels A to C show child outcomes across child phenotypes, and panels D and E show child outcomes across parent phenotypes.

Multiple group comparisons showed that changes in the total sugar intake of children differed across phenotypes (P=.01; Figure 7A-C). Children belonging to the fully engaged (P=.01; f²=0.04) or partially engaged (P=.004; f²=0.05) phenotypes reduced their total sugar intake over 3 months compared with children in the unengaged phenotype (reference group), who increased their total sugar intake over time. Regarding children’s total daily energy intake and energy intake from sugary beverages, we found borderline differences (P=.07 and P=.09) that became significant in individual pairwise comparisons. Children from the fully engaged (P=.03; f²=0.01), partially engaged (P=.04; f²=0.03), and dabblers (P=.03; f²=0.00) phenotypes decreased their total energy intake over 3 months compared with the unengaged children who increased their daily energy intake over time. Finally, children from the partially engaged phenotype decreased their intake of sugary beverages compared with unengaged children who did not (P=.01; f²=0.02). In this case, fully engaged children did not differ significantly from unengaged children; however, as shown in Figure 7C, children’s intake of sugary beverages in the fully engaged group trended downward, whereas unengaged children’s intake trended upward (P=.12).

Differential changes in outcomes among children were also observed across the parental phenotypes (Figure 7D and 7E). Multiple group comparisons showed borderline significant changes in children’s zBMI and total daily energy intake across parental phenotypes (P=.06 and P=.08, respectively), which became significant in individual pairwise comparisons. Specifically, children whose parents were fully engaged significantly decreased their zBMI (P=.01; f²=0.05) compared with children with unengaged parents (reference group) whose zBMI slightly increased. Similarly, children whose parents belonged to the independently engaged phenotype decreased their daily caloric intake (P=.02; f²=0.03) compared with children with unengaged parents whose daily caloric intake increased over 3 months. Figure 7E also shows trends of decreased caloric intake among children with fully and partially engaged parents compared with children with unengaged parents; however, these trends were not statistically significant (P=.11 and P=.07, respectively).

This is the first digital phenotyping study of an mHealth intervention targeting health behavior changes among children with overweight or obesity and their parents. We evaluated user typologies based on how children and parents interacted with different features of the Aim2Be app. We found 4 child (unengaged, dabblers, partially engaged, and fully engaged) and 5 parent (unengaged, socially, independent, partially engaged, and fully engaged) phenotypes, which illustrate the ways in which participants used the behavioral, gamified, and social features of the Aim2Be app. As expected, based on Aim2Be’s conceptual framework [25], our results demonstrated that specific patterns of use supported behavior change, whereas others did not, meaning that greater engagement with the active ingredients of the app improved children’s dietary and weight outcomes.

Given the scarcity of research on the digital phenotypes of mHealth users in the context of childhood obesity, it is difficult to compare our findings with those of previous studies. However, our results are similar to a recently published study profiling children’s (but not parents’) engagement with an older version of Aim2Be [33], where the 4 child profiles that emerged were similar, although our study examined 6 additional app features. Interestingly, the results previously observed in the prevention context [33] were replicated in our study using a clinical sample of children. Importantly, users with distinct patterns of engagement obtained different health benefits depending on whether they interacted with the active ingredients of the app. When lifestyle behavior modification interventions required in-person attendance, dose-response studies identified a minimum of 26 hours of contact for the intervention to improve children’s outcomes [4]. However, our digital phenotype analyses illustrate that new approaches are needed to conduct dose-response analyses in the context of mHealth interventions, especially when users have the freedom to select which app features they engage with. As users interact with the Aim2Be app quite differently, this variability must be accounted for when assessing whether the intervention can influence the mediators and outcomes targeted by the app.

In this study, we found that fully engaged children with Aim2Be (eg, set goals, completed tasks, and read articles) experienced more desirable behavior changes than unengaged users. Specifically, children who engaged more fully with the app decreased their intake of total daily calories, total sugars, and sugary beverages. Our findings align with existing research [22,33] suggesting that mHealth interventions have the potential to improve children’s dietary behaviors. Furthermore, in exploratory analyses examining the aims that were most often set and completed among Aim2Be users (data not shown), we found that “Drop sugary drinks” was the most common aim chosen by children, which validates our findings related to lower total sugar and energy from sugary drinks among fully and partially engaged children. These results highlight the importance of increasing engagement with the app’s active ingredients, namely, setting specific goals and completing tasks related to those goals to promote health behavior change among children.

In this study, fully engaged children were more likely to be younger and have fully or partially engaged parents. These associations could indicate that the app was more appealing to younger children, as shown by other research [34], or that parents dedicated more attention to their children when they were younger than when they were older. In addition, younger children might be more easily influenced by their parents, which may explain their use of the Aim2Be app. These findings are aligned with previous studies reporting that parental self-monitoring (a behavioral strategy) and adherence to eHealth interventions were significant predictors of adolescents’ self-monitoring and adherence [33,64].

We also found that children whose parents were fully or partially engaged with the app’s behavioral features decreased their zBMI and total daily energy intake more than children whose parents only engaged with the social features or who did not engage with the app at all. Our findings are consistent with a qualitative study [65] showing that participation as a family is one of the main factors identified by both children and parents to facilitate behavior change. Indeed, current guidelines [4,15,66] for the treatment of childhood obesity include a family-based approach. Taken together, our findings reinforce the critical role that parents play in lifestyle interventions to support their children’s adherence and improvement of health outcomes, even in the mHealth context.

This study also revealed that family structure was associated with parental phenotypes. Fewer single-parent households belonged to the fully and partially engaged phenotypes than parents who lived with a partner or were married, which may reflect that more independent, time-scarce (and therefore task-oriented) parents [67]. Interestingly, single-parent households were also likely to belong to the independently engaged phenotype (ie, parents who only engaged with the behavioral app features such as aims and tasks), and children whose parents belonged to this phenotype reduced more of their total daily energy than other phenotypes. In fact, previous research found that parents of young children decreased their use of mHealth apps when they had limited time or only used the app to find specific information of interest [67]. This could explain why independently engaged parents did not use the gamified or social domains but used the domain exclusively focused on behavioral change and why their children decreased their energy intake over time.

This study had several limitations and strengths. First, our sample was relatively small and not powered to detect significant changes across multiple digital phenotypes in these secondary analyses. This could have limited our ability to detect clinically meaningful changes in health outcomes, although some changes were observed. In addition, overall adherence to the app was low, which limited our ability to detect more phenotypes and perhaps to observe some between-group changes. Moreover, our study included a clinical sample (children with overweight or obesity); thus, our findings are limited to this population. Nevertheless, we used a detailed dietary assessment (24-hour dietary recalls), both self-reported and objective measures of physical activity, and objective measures of app usability through app analytics. Finally, we used a novel approach to examine intervention efficacy, which showed positive effects that are not observed [38] using more traditional analysis.

Overall, 3 key messages from our findings point to future directions in mHealth research. First, even in the mHealth context, parental engagement matters as it can increase children’s adherence to a lifestyle intervention and provide the household environment that supports behavior change. Thus, whether a lifestyle intervention is delivered in person or on the web, parents should be involved as they are active agents of change. Second, dose-response analyses should assess how (and not only how much) the app is being used by the participants, as users’ full engagement with the active ingredients of the app seems to be a critical factor for the success of mHealth interventions. Finally, as participants’ engagement with specific features of an app is key to promoting behavior change, future research should examine how we design program components that ensure users interact with the active ingredients of the mHealth intervention.

This study showed that distinct patterns of use exist among both parents and children who used a family-based lifestyle mHealth app, namely, Aim2Be. Identifying who uses mHealth apps and how can help us understand and develop more tailored interventions to support various users in a health behavior change process. Our findings point to the importance of optimizing users’ full engagement with the active ingredients of the app as a critical factor for the success of mHealth interventions and highlight the need for further research to understand program design elements that can influence participant engagement.