Recent systematic reviews on the global prevalence of psychiatric disorders in children and adolescents have produced varying results ranging from 6.8% to a notably higher pooled rate of 13.4% [1,2]. Emotional disorders as well as significantly distressing subthreshold emotional problems are among the most common psychiatric problems reported in adolescent populations [1-3]. According to the Centers for Disease Control and Prevention, over 6% of US children between 12 and 17 years old have been diagnosed with depression and over 10% have been diagnosed with anxiety disorders [4]. Globally, it is estimated that 10% to 20% of youth experience mental health problems [5-7].

Smartphone ownership is growing fast worldwide. In the United States, smartphone ownership or access among adolescents was estimated at 95% in 2018, representing an increase from 73% in 2015 [8,9]. Similar development is evident elsewhere, with youth smartphone ownership surpassing the 90th percentile in the majority of developed economies [9]. Not only are smartphones widely distributed but people also tend to carry their phones with them, spending an estimated 170 minutes per day using smartphone apps [10]. Some studies indicate that daily smartphone usage among adolescents is often more than 270 minutes [11]. The number of mobile health (mHealth) interventions available has risen steeply in a steady fashion since first appearing roughly a decade ago, with an estimated 325,000 apps available on the market in 2017 [12]. However, only 8% of adolescents use health apps and relatively few studies have documented how they specifically function in adolescent populations [13]. Although mental health problems disproportionally burden minority and lower socioeconomic status groups in terms of receiving evidence-based interventions, smartphones may be used as a tool to diminish such disparities [14,15]. For example, in the United States, adolescent smartphone ownership is not related to gender, race, parental educational levels, or socioeconomic status [9].

Smartphones offer possibilities of a uniquely personalized platform to tailor the many aspects of treatment to individual patients. Patient treatment through support by smartphones or other mobile devices such as tablets, patient monitoring devices, and personal digital assistants have been collectively labeled “mHealth” [16]. mHealth interventions have shown promising cost-effective outcomes related to lowered anxiety and depression symptoms in youth populations despite recurring issues of high attrition rates [15,17-22]. Attrition is here defined as leaving treatment before obtaining a required level of improvement or completion of intervention goals [23,24]. Treatment attrition is common, costly, and important, although varying definitions of the term have challenged research on the matter [17,24].

Studies on mHealth interventions targeting emotional disorders or subthreshold emotional problems in adolescent populations have frequently lacked time-specific data alongside a lack of accurate definitions and analysis of treatment attrition [15,25]. Usability data in mHealth studies targeting adolescent populations are frequently presented with attrition rates from the initiation of intervention, either at the time of recruitment or launch of the intervention’s first section. For instance, average weighted attrition rates prior to commencing online treatment and after treatment have been reported to be 21%, whereas the rate was 8% from treatment completion to follow up [23].

Attrition in mental health care interventions has been shown to be up to twice as common compared to that in other medical fields [24]. For example, attrition from cognitive behavioral therapy (CBT) was reported to range from 20% to up to nearly 44%, although some indications are toward lower attrition in CBT at the group level [26]. Recent research has found that online CBT programs are effective in treating adolescent mental health problems such as anxiety and depression [5,15,25,27]. However, attrition from these programs remains an issue, with a high degree of variation and attrition rates reaching up to 50% [15,28]. To ultimately lower attrition in adolescent mental mHealth interventions, a better understanding of what factors explain mHealth usage in different adolescent subgroups and time-specific attrition is direly needed.

Adolescents with significant emotional problems (ie, anxiety and depression) were reported to have lower general self-efficacy than their peers and were more likely to either not seek treatment or drop out [3,29,30]. Originating from social cognitive theory, self-efficacy is defined as an individual’s belief that ability is sufficient to succeed or accomplish a task and has been used extensively to guide a theoretical framework for interventions targeting health behaviors [3,29-31]. Individuals with higher levels of self-efficacy are more likely to seek treatment and persist longer in their efforts to change behavior [32]. A partial reason for this appears to be effective use of self-regulatory skills such as planning, problem-solving, and self-incentives [32]. Research has shown a positive relationship between higher levels of self-efficacy and successful health behavior change in terms of weight management and exercise behavior [33]. Studies have also revealed a mediating relationship between higher levels of self-efficacy and treatment adherence in diverse chronic illnesses [34]. Attrition rates in adolescent mHealth interventions could perhaps be better accounted for by an increased understanding of the relationship between self-efficacy levels and detailed descriptions of time- and content-based usage. The purpose of this study was to assess time-specific attrition rates in an adolescent mHealth intervention, as well as to describe usage and the intervention’s feasibility in relation to self-efficacy levels and participants’ emotional and physical health.

Among all invited participants in the intervention group, 70% (male:female ratio 8:6) began the intervention. Participants’ descriptive characteristics at baseline are summarized in Table 1. Retention after 6 weeks of the intervention was 65% among those who began the intervention (male:female ratio 5:4). No significant gender difference was evident in retention rates (χ²_{2, 14}=0.83, P=.36). The mean total score on the SUS was satisfactory (78.09, SD 9.82). There was no gender difference in total score on the SUS (P=.85) or in-app activity (P=.72), although female participants showed a higher frequency of usage on average than male participants in all health behavior categories (Table 2).

There was a significant 76% decrease in the total number of in-app health exercises from week 1 to week 2. However, from week 2 to study end, there was a 22% increase in the total number of exercises. The weekly individual mean number of in-app exercises is shown in Figure 2 and the average usage throughout the intervention is summarized in Table 2. Participants who dropped out of the intervention were significantly more likely to use the app between midnight and midday than those who completed the intervention (R_t=0.43, P<.001). Details of when the participants used the app and the types of health exercises they completed are shown in Table 3.

Among the 23 participants who answered the baseline questionnaire, 13 (57%) participants were either interested or very interested in using the app to increase health behavior. Four participants (17%) had never downloaded a health app prior to the intervention. However, 9 participants (39%) had never used health apps, which indicates that 5 (22%) participants had downloaded a health app that they never used. All participants had downloaded and used social media apps. Daily active screen time did not significantly differ between the intervention and control groups or between genders at baseline and at study end (χ²_1,23=3.73, P=.16). At baseline, the intervention group had more problems with daytime sleepiness than the control group (χ²_5,20=11.29, P=.04) but this difference was no longer detected at study end (χ²_5,23= 2.35, P=.80). There were significantly greater problems with disruptive wake-ups during night sleep for participants in the intervention group than for participants in the control arm (χ²_5,20=11.87, P=.04) at baseline, although no such difference was evident at study end. Mean hours of sleep did not differ significantly between the intervention group (6.90, SD 1.29) and the control group (7.15, SD 0.69; t₂₁=0.4, P=.64) at study end.

There were no significant differences in daily portions of vegetables or fruits and berries between the intervention and control groups. There were also no differences regarding the frequency and amount of sugary or sugar-free soft drinks at pre and post measures, as well as for the consumption of salted chips, French fries, or popcorn and candy or sweets. In addition, there were no differences between the intervention and control groups in terms of the consumption of energy drinks at baseline or at study end. The perceived amount of physical exercise was increased by nearly 20% in the intervention group between baseline and study end but decreased by 26% in the control group. Total anxiety scores did not differ significantly between groups at baseline; however, an 8% decrease was found for the intervention group compared to a 4% increase in the control group (Multimedia Appendix 2). Symptoms of depression were also more evident in the SidekickHealth app intervention group at baseline. However, these differences were not apparent at study end. There was also a significant difference between the intervention group and control group in terms of negative self-esteem at baseline, which was not evident after the intervention (Multimedia Appendix 2). The total score on the GSE scale revealed a roughly 8% increase in self-efficacy scores in the intervention group compared to a 3% decrease in the control group from baseline to study end.

Correlations were found between the count of app exercises during the intervention period and weight difference from baseline to intervention end. However, none of the correlations found was statistically significant due to the small sample size. When underweight participants were excluded from the calculations, the decrease in BMI-SDS from baseline to study end appeared to be significantly more apparent in the intervention group (mean 0.48, SD 0.36) compared to that of the control group (mean 0.08, SD 0.44; t₁₇=–2.14, P=.04).

There has been a steep and steady increase in mHealth interventions since their appearance roughly a decade ago, with an estimated 325,000 mHealth apps available on the market in 2017 [12]. The purpose of this study was to report attrition rates in a pilot study of an adolescent mHealth intervention and to begin to depict different attrition periods. Time-specific data alongside accurate definitions and analysis of attrition in adolescent mHealth interventions (ie, online CBT programs) targeting emotional disorders or subthreshold emotional problems in adolescent populations are lacking in the literature [15]. Studies on attrition in mHealth interventions targeting adolescent populations frequently present attrition rates in a two-point fashion, from initiation of the intervention to the end of treatment. This study offers some insight into mHealth usage among adolescents in this regard with an observed 35% attrition rate from program initiation to termination, which is similar to reported rates in previous online youth CBT programs [5]. However, as this study focused on usage and attrition rates on a weekly basis throughout the intervention, some interesting findings emerged regarding usage of the app. There was a 22% increase in usage between the second week of the intervention and termination, suggesting a sensitive attrition period shortly after instigation of mHealth interventions treating adolescents that warrants further examination. A better understanding of the timeline from treatment instigation to attrition in adolescent mHealth interventions and how emotional disorders or subthreshold emotional problems are connected to that timeline would be a reasonable next step.

Daily time of usage seemed to play a contributing role to gaining an increased understanding of attrition rates. Participants who dropped out of the intervention were significantly more likely to use the app between midnight and midday than those who completed the intervention. Those who completed the intervention were both more likely to use the app from midday to midnight as well as to complete more in-app exercises on average than those who dropped out. The latter finding may seem somewhat rudimentary but is of importance, since this indicates that participants who dropped out are not less motivated to use the app at initiation of treatment. These findings do perhaps highlight the importance of examining time of usage through survival analysis in subsequent studies assessing the factors contributing to attrition from interventions.

An integral measuring factor in the development and implementation of mHealth interventions for adolescent populations is assessing the intervention’s feasibility and usability for the desired research population. Participants reported on the app’s adequate usability on the SUS and seemed willing to engage in health exercises, completing over 21 in-app exercises on a weekly basis throughout the intervention. A significant decrease in average exercises performed between the first week of the intervention and subsequent intervention weeks was evident, although average usage leveled off at roughly 15 weekly in-app exercises throughout the intervention period. Interestingly, roughly 39% of the participants had never used an mHealth solution prior to this study, although all participants were accustomed to apps since all had downloaded and used social media apps on their smartphones. These findings reveal higher usage rates among participants as previous studies have shown that merely 8% of adolescents use health apps [13].

This study also highlights noteworthy health behavior changes based on app usage. Indications were toward a positive impact on reported sleep problems, in disruptive night sleep wake-ups, and problematic daytime sleepiness, although these findings need to be studied and documented in a more thorough manner. Reported daily physical exercise was increased by nearly 20% in the intervention group, whereas these numbers dropped by roughly 26% in the control arm. These findings could either point to the fact that in-app usage may heighten perceived levels of physical exercise or simply increase how much the adolescents are physically exercising through the intervention. This could be assessed in a more detailed manner in subsequent studies by comparing the self-reporting and frequency of actual exercise simultaneously. Both factors could aid in increasing health behavior since a perceived increase in behavior can increase self-efficacy levels among adolescents in general mHealth interventions or online CBT interventions targeting emotional problems [30]. These findings are related to the fact that self-efficacy levels increased by 8% among adolescents in the intervention arm while decreasing by 3% among controls.

In conclusion, this pilot study was designed to assess time-specific attrition rates in an adolescent mHealth intervention, as well as the usage and feasibility in relation to self-efficacy levels and participants’ emotional and physical health. The study was limited to a single research site and the results are based on a small convenience sample, which limits the ability to generalize the findings and determine the intervention’s overall efficacy. However, the results revealed interesting findings regarding sensitive attrition periods that warrant further examination. The obtained attrition rates point to a sensitive period during the first week, and indicate that adolescents who use the app in the afternoons and evenings are less likely to drop out. More research in this area seems called for as this could be studied in relation to app features such as the timing and frequency of notifications and instructions.