The benefits of exercise are irrefutable [1] and have been demonstrated for individuals of all ages [2]. In Canada, the government encourages engagement in exercise through the publication of the Canadian Physical Activity and Sedentary Behaviour Guidelines Handbook [3]. While it is often assumed that knowledge of the benefits of exercise will increase exercise levels, informational campaigns have been rendered as ineffective for sustaining behavior change [4,5]. This is exemplified by the steady decline in physical activity and fitness levels among Canadians [2] despite the promotion of these physical activity guidelines, with only 15% of Canadian adults currently meeting daily physical activity recommendations [6]. Similar levels of engagement are seen within the United States [7] and Australia [8], with less than 50% of the respective populations engaging in adequate levels of physical activity.

One strategy that may provide an effective medium to target physical inactivity at the population level is mobile health technology. To date, there are nearly 7 billion mobile phone subscriptions worldwide, with the use of mobile devices reaching 90% in developing countries and 96% globally [9]. This widespread use of mobile devices has led to the creation of mobile health-based (mHealth) products. Simultaneously, advances in technology have shifted traditional means of health promotion materials from read-only (eg, pamphlets and websites with read-only content) to interactive and responsive means (eg, mobile apps). mHealth apps offer many advantages over traditional informational materials. Data from users can now be analyzed in a context-appropriate, timely, and sophisticated manner [7]. More than ever before, there is opportunity to provide real-time support to the masses, outside of costly traditional personal training or counseling appointment times.

As of 2012, 84% of mobile phone owners had downloaded at least one app to their phone; 19% of those individuals had downloaded an app specifically related to tracking or managing a health-related behavior[10]. The continually increasing prevalence of app use further demonstrates the potential reach for mHealth exercise interventions. While app use is increasing in popularity, existing apps are not without limitations.

The majority of health-related apps currently available are developed on the traditional dissemination model. Generic, automated text messages are typically sent to individual users on a standardized time of day or week, or access to content is on a dedicated website in which users are referred to go read. While these apps provide valuable information to the user, such apps have neglected to integrate evidence-based strategies from established health behavior change theories [7,11-13]. In a 2012 review of the Health and Fitness category in the Apple App Store, Cowan and colleagues [11] concluded that there was an overarching lack of theoretical constructs used within 127 surveyed apps. Similar findings were reported in Direito and colleagues’ 2014 study evaluating the presence of 26 behavior change techniques in the 40 most popular physical activity and dietary apps from the Apple App Store [14]. While some incorporation of behavior change techniques is evident, Direito and colleagues' conclusions remained consistent with Cowan’s review, in that an absence of behavior change strategies exists in physical activity and dietary apps [14]. Cowan, Direito, and colleagues are not alone. Several other reports have also strongly recommended that future apps in the health domain be improved by incorporating evidence-based practices that are known to enhance health behavior change [11,15-17]. These findings highlight the need for collaboration between health behavior change experts and app developers.

Although rare to find, research-derived programs such as the Heart Exercise And Remote Technologies (HEART) mobile phone trial [18] demonstrate the benefits of integrating evidence-based behavior change strategies. Having established utilizing principles of behavior change from social cognitive theory [19], HEART uses a personalized, automated package of text messages aimed at increasing levels of exercise behavior in individuals with ischemic heart disease. HEART short message service(SMS) texts were developed to assist users with goal setting, exercise scheduling, and self-efficacy to overcome exercise barriers and engage in regular exercise in a positive and cost-effective manner.

As demonstrated in the HEART trial, social cognitive theory is particularly well suited for mHealth interventions, as the tenets of the theory are grounded in (1) self-monitoring, (2) self-evaluation, and (3) modification of current behavior based on this self-reflection [20,21]—tasks that mHealth apps have the capacity to assist the user with. The majority of apps allow the user to record their exercise sessions as a form of self-monitoring. Relatedly, many apps allow the user to look back at past exercise sessions in a summative format (eg, number of sessions completed last week)—an opportunity for self-evaluation and reflection. A main tenet within social cognitive theory is self-efficacy. Self-efficacy is a set of beliefs one has about his/her ability to organize and complete a task in order to accomplish a certain task that is crucial for eliciting health behavior change [22]. Self-efficacy is an important predictor to exercise adherence, with numerous trials demonstrating the significant association between improvements in self-efficacy and exercise adherence [23]. Together, these tasks provide the opportunity for the user to modify current behavior in order to meet one’s goals. As such, most mHealth apps have the capacity to allow the user to self-regulate behavior based on past experience and future goals, if guided appropriately. Results of the HEART trial support the continued use of SMS texts to increase exercise engagement through a significant main effect for leisure time physical activity in those receiving the SMS theory-based texts, which was mediated by task self-efficacy.

The success of self-regulation is partly dependent on the fidelity, consistency, and timeliness of self-monitoring [24]. Given the instantaneous nature of real-time feedback that mHealth apps can provide, self-monitoring may be carried out promptly and accurately with minimal inconvenience for the individual. However, the process of self-monitoring is not simply an audit of one’s performance [20], and the act of self-monitoring alone is not likely to help an individual self-regulate. Further investment must be taken by looking at past exercise patterns and recognizing barriers. This provision of feedback can provide the individual with an opportunity to evaluate behavior when necessary to remain in line with one’s goal.

While active engagement in self-monitoring and self-regulation are essential for the maintenance of health-related behavior, it is also imperative to provide individuals with personalized feedback on their behavior [25]. Personalized interventions have been demonstrated to be more effective than nonpersonalized interventions at changing health behavior [25]; however, few interventions utilize this technique. Likewise, a systematic review of SMS-based behavior change interventions confirmed the effectiveness of tailored SMS messages for promoting health behavior change (see literature review by Fjeldsoe et al [25]). In the context of mHealth apps, a personalized intervention would allow a health professional to provide tailored feedback to an individual user in a time-efficient manner.

This pilot study sought to examine the utility of a theory-based exercise self-monitoring app for increasing independent exercise adherence over 8 weeks. It was hypothesized that the use of this app would result in (1) more frequent exercise bouts, (2) more frequent self-monitoring, (3) higher perceived self-management of exercise behavior, and (4) higher self-efficacy to self-monitor exercise behavior in comparison to individuals not using the app.

The study was approved by the University of British Columbia: Okanagan Research Ethics Board. A randomized experimental pilot study design was utilized with participants being randomly selected to one of two conditions—the experimental, app-use condition, or the control, no-app condition. Those randomized to the experimental condition used the app for the 8-week investigation, whereas during the same time period, those randomized to the control condition did not have access to the app. The research assistant met with all participants at both baseline and post-testing time points.

Participants were recruited from a local YMCA fitness facility by means of announcements in fitness classes, posters located throughout the facility, and an information booth in the lobby. In addition, front desk YMCA staff members were instructed to inform individuals about the study opportunity. Eligible participants were current facility members aged 19-70 years, with access to a mobile device. A total of 94 individuals expressed interest in participating. Following initial screening via email, 56 members were deemed eligible (see Figure 1 for detailed information regarding eligibility); they were then randomized through a computer random numbers-generated table to either the experimental condition, which received the app for 8 weeks (18/28, 64% female), or to the control condition, which did not receive the app (20/28, 71% female). While no specific exercise criteria was set, in meeting with the individual participants it became clear that 2 individuals out of 56 (4%) were excessively active and were engaging in competitive athletic events. These 2 individuals were designated not eligible for participation.

Data were analyzed using SPSS Statistics version 21 (IBM Corp). A series of independent t chi-square tests were conducted to examine equivalency between conditions on all demographic, dependent, and independent variables at baseline. Repeated-measures analyses of variance (ANOVA) were used to test the hypotheses that exercise bouts, self-monitoring exercise, self-efficacy to self-monitor, and self-management of exercise will be greater in the experimental condition as compared to the control condition following the 8-week study period. All effects are reported as significant at P<.05. Effect size estimates were calculated using partial eta squared (partial η²) for ANOVA and Cohen’s d for t. Effect size derived from partial η²was interpreted as small (0.01), medium (0.06), and large (0.14) in accordance with conventional practices within the social sciences [32,33]. Likewise, effect sizes derived from Cohen’s d were interpreted as small (0.20), medium (0.50), and large (0.80) [33,34]. Differences between conditions and time × condition interactions yielding medium-to-large effect sizes were further explored with t.

A total of 56 adults—mean age 40 years (SD 13); mean BMI 26.8 kg/m² (SD 5.3)—participated in this study, with 36% of participants having achieved a university level degree or higher, and 50% working either part time or full time. A total of 28 individuals out of 56 (50%) were randomized to the experimental condition—mean age 38 years (SD 14); mean BMI 28.2 kg/m² (SD 6.5)—and 28 out of 56 (50%) were randomized to the control condition—mean age 42 years (SD 11); mean BMI 25.9 kg/m² (SD 3.6). In total, 41 out of 56 participants (73%) provided follow-up data 8-weeks postintervention (see Table 4 for demographic information and Figure 1 for participant flowchart).

There were no statistical differences in demographic, dependent, or independent variables between conditions at baseline with the exception of current exercise self-monitoring behavior. At baseline, participants in the control condition (mean 1.89, SD 2.28) reported a higher frequency of self-monitoring in the past 7 days than participants in the experimental condition (mean 0.52, SD 1.61; t ₃₇=-2.34, P=.02, d=0.69). There was no statistical difference in dropout rate between conditions (χ²₁= 1.2, P=.28).

A repeated-measures ANOVA examining self-reported exercise frequency revealed no main effect for time (F _1,38=2721, P=.11, partial η²=.067) or condition (F _1,38= 2.45, P=.13, partial η²=.061). The time (pre, post) × condition (experimental, control) interaction, although not meeting statistical significance, yielded a medium-to-large effect size (F _1,38= 3.87, P=.06, partial η²=.092). Exploratory post hoc analysis revealed a significant difference between conditions, such that those in the experimental condition (mean 7.24, SD 3.40) were engaging in significantly more bouts of exercise per week than those in the control condition (mean 4.74, SD 3.70; t ₃₈= 2.23, P=.03, d=0.70) (see Table 5).

A repeated-measures ANOVA examining self-monitoring frequency showed a main effect for time (F _1,33= 59.55, P<.001, partial η²=.643) and condition (F _1,33= 15.38, P<.001, partial η²=.318). These main effects were superseded with a significant time (pre, post) × condition (experimental, control) interaction (F _1,33= 49.39, P<.001, partial η²=.599). Post hoc analysis revealed a significant difference between conditions at postintervention, such that those in the experimental condition (mean 6.00, SD 0.93) were engaging in a significantly higher frequency of self-monitoring compared to the control condition (mean 1.95, SD 2.58; t ₄₀= 6.88, P<.001, d=2.10) (see Table 5).

A repeated-measures ANOVA examining self-management of exercise behavior revealed no main effect for time (F _1,38= 1.91, P=.18, partial η²=.048) or condition (F _1,38=.408, P=.53, partial η²=.011). The time (pre, post) × condition (experimental, control) interaction was not significant (F _1,38=.039, P=.85, partial η²=.001) (see Table 6).

A repeated-measures ANOVA examining SESM was conducted, revealing no main effect for time (F _1,39=.092, P=.76, partial η²=.002) or condition (F _1,39=.665, P=.42, partial η²=.017). Further, the time (pre, post) × condition (experimental, control) interaction was not significant (F _1,39=.021, P=.89, partial η²=.001) (see Table 7).

This preliminary pilot study investigated the utility of a theory-based self-monitoring app for improving exercise adherence. To our knowledge, this study is the first to integrate behavior change theory in an app, using personalized goals and interaction with a virtual exercise counselor for the promotion of exercise behavior. Findings provide preliminary evidence that, after 8 weeks, individuals with access to such an app engage in a higher frequency of exercise behavior in comparison to individuals who did not have the app. Specifically, app users reported engaging in 7.2 bouts of exercise per week after 2 months, whereas individuals without use of the app reported engaging in 4.7 bouts of exercise per week at this time period. Although this did not reach statistical significance (P=.06), this difference between conditions on exercise behavior represents a medium-to-large effect size (partial η²=.092), findings that are similar to those reported in other mHealth trials [35,36]. Possible reasons for these positive findings is that mHealth apps allow feedback messages to be sent in a time-sensitive manner, designed around the individual user to further facilitate communication [36]. Use of an app is found to be a simple self-monitoring tool, serving as a means of encouragement and motivation. When combined with feedback, both visual and verbal, the use of an app encourages users to work toward their activity goals [35].

Findings from this pilot study also provide partial support for our secondary hypothesis that use of a theory-based self-monitoring app will result in a higher frequency of self-monitoring in comparison to individuals without access to an app. From baseline to 8 weeks later, app users’ self-monitoring frequency increased from less than one event per week, to an average of six self-monitoring events per week. Self-reported self-monitoring of exercise behavior was unchanged from baseline to post-testing in the control condition (see Table 5). Such an increase in self-monitoring can be partially accredited to the electronic nature of our mHealth app. Electronic diaries facilitate (1) the instantaneous transfer of data between user and counselor or health care provider [37] and (2) have been shown to be associated with higher rates of adherence when compared to traditional self-monitoring via paper and pen diaries [38].

Despite increases in both exercise and self-monitoring behavior, our hypotheses that use of the app would result in an improved self-management of exercise behavior, or self-efficacy to self-monitor exercise behavior was not supported. Use of the app did not result in a significant effect on self-management of exercise from pre- to post-testing time points (see Table 6) or self-efficacy to self-monitor exercise behavior (see Table 7). In regard to the control condition, participants showed no significant change in self-management of exercise or self-efficacy to self-monitor across time points. In light of our nonsignificant findings for self-management of exercise behavior, further evaluation is warranted to understand the manner in which our intervention targeted self-regulatory principles. The purpose of this intervention was to increase the practice of self-monitoring as a key component of self-regulation, and not overcoming exercise barriers. As such, one plausible explanation for the failure to change perceived self-management of exercise behavior in the experimental condition is that this construct was not adequately addressed in the intervention content.

It is also possible the items used to measure self-regulation did not adequately measure the construct within the context of exercise. Although Hallam and Petosa’s [29] measure is highly regarded with respect to self-management of a general health condition, it is plausible that the measure was not context appropriate to measure change in exercise behavior. Further, as can be seen by examination of the means for self-efficacy to self-monitor, a possible ceiling effect may have occurred during this intervention, with baseline self-efficacy scores of over 80% being reported by both conditions. Interestingly, Hallam and Petosa [29] also suggested a problematic ceiling effect in their 2004 study integrating social cognitive theory in a work-site intervention. The purpose of this study was to directly impact self-monitoring through tangible use of an app. However, given the widespread use of apps, one possible explanation for the observed ceiling effect is that all participants were familiar with the act of self-monitoring through other generic apps (eg, tracking work or time spent on social media) prior to the investigation, and therefore their belief (ie, self-efficacy) to self-monitor was not significantly impacted through the intervention material.

The integration of theory into the development of a self-monitoring app was the primary strength of this pilot study. To date, principles from theories of health behavior have been used sparingly within mHealth apps [12], despite evidence to suggest the integration of theory (eg, social cognitive theory) lends support to behavior change [18]. In this study, app users received a social cognitive theory-grounded message once per week over 8 weeks. Such an automated strategy could feasibly be incorporated into many existing mHealth apps. Further, each app user set a personalized 8-week goal, allowing for tailored feedback from a virtual exercise counselor. Lack of tailored feedback has been a limitation in previous trials [35]. The current trial was able to integrate the use of tailored, real-time feedback in a nonburdensome manner, facilitated by one exercise counselor. Daily review of users’ self-monitoring was made manageable due to the electronic nature of the app, taking approximately one minute per day per participant to review and respond to users’ questions and comments. As the system is developed with a pre-existing bank of messages for weekly, theory-based content (see Table 1) and an established timeline of when to intervene (see Table 2), contact between the exercise counselor and user is as simple as choosing a situation-specific message and further specifying details based on the individual.

This study is not without limitations. Given the nature of this investigation acting as a pilot trial, and in working with an entrepreneurial developer, a power calculation was not performed out of logistics in working with our industry partner and recruitment time constraints. Recruitment for this study was limited to one fitness facility due to restrictions placed by the app industry partner, resulting in limited power to detect group differences as well as an inability to conduct more complex analyses to better understand why potential differences existed (eg, mediation and multiple mediation). These findings may not be generalizable to individuals who are not able to afford fitness facility memberships; however, it should be noted that the facility utilized in this study offers subsidized memberships based on gross income. Given the general recruitment criteria (ie, 19-70 years of age, access to a mobile phone device), the heterogeneity of our sample may have weakened our ability to draw concise conclusions and apply them to the general population as not all participants were new to exercise and may have had prior experience with mHealth app technology. Self-monitoring behavior was the secondary focus of this intervention. While app users’ self-monitoring frequency was calculated using data from the app, the self-monitoring frequency of participants in the control condition (ie, no app) was based on self-report data, as the control aspect of this study design prohibited measurement via an app of these participants. The use of self-report data is inherent to recall bias [36], potentially resulting in unreliable results in the comparison of conditions. Lastly, this study looked at changes in exercise behavior over the duration of an 8-week intervention. As 40-65% of new exercisers drop out within the first 6 months [4] of a new program, an extended trial of the app is warranted to draw conclusions on long-term efficacy.

Wearable devices (eg, fitness trackers, pedometers, and accelerometers) have become sophisticated, with continued development of technology bringing credibility to such devices. Continued research on the development of mHealth devices could help to establish users' trust in the integration of technology (eg, mobile phone apps and wearable devices) to monitor health behaviors. Overall, an enhanced trust in the use of technology could have a meaningful effect on the ability of a device to impact the health of the public in general, as well as specialized populations [39]. Future studies should look to the integration of a true control group with access to a general health-related self-monitoring app, and objectively measure self-monitoring in the control condition, to provide further understanding of the mechanism under which users are affected by mHealth technology.

A total of 8 weeks of mHealth app use resulted in increased exercise and self-monitoring behavior, providing some support for the use of a self-monitoring app to increase adherence to exercise and self-monitoring of exercise behavior. This study protocol also demonstrates the feasibility of incorporating theory-based messages into existing mHealth apps, although the inclusion of such content did not lead to anticipated changes in self-efficacy to self-monitor or self-management of exercise behavior. Multiple inoculations of theory-based messages may be needed for sizable changes to be made in these constructs. Future research is warranted to understand the long-term efficacy of an mHealth app and its effect on exercise and self-monitoring behavior.