Because theory-based interventions are associated with higher effect sizes than interventions not based on theory [6,46], defining the theoretical framework for the intervention is necessary. To do so, prominent health behavior theories and scientific literature were reviewed.

Social cognitive theory (SCT) was adopted as a basis for the theoretical framework as it is one of the most prominent behavior change theories used to inform interventions targeting health behavior change [10,47,48], and a recent meta-analysis reported that SCT concepts may explain 31% of variance in PA [10]. SCT addresses both individual and social factors and recognizes the reciprocal relation between individuals and their context or environment. For these reasons, SCT thus guided and informed the intervention’s theoretical framework; insights from other theories and evidences were added for specific topics. Figure 1 shows the structural pathways of Bandura’s SCT [49], and Figure 2 shows the specific theoretical framework [49] that is used for the Active2Gether intervention. In Figure 2, the bold lines and boxes represent the elements that are based on the Social Cognitive Theory, and the dotted lines and oval boxes represent behavioral determinants added to the theoretical framework.

We first identified evidence-based and relevant BCTs and linked these with the behavioral determinants of the theoretical framework by means of a review of the relevant literature, based on an existing taxonomy of BCTs [42,43]; see Multimedia Appendix 1 [9,22,42,43,50-53]. To explore which BCTs were used in already existing PA promotion apps, a systematic content analysis of such apps available in iTunes and Google Play was conducted [30]. Additionally, focus group discussions with the target population were conducted. The methods and results of these focus groups have been published in more detail elsewhere [45]. Finally, a Web-based cross-sectional survey was conducted among 179 young adults to assess their ratings with respect to the importance of specific BCTs applied in apps and their preferences for personalized tailoring [44].

The foundation of the intervention is the definition of the program’s outcomes and objectives [54,55]. Therefore, specifying who and what will change as a result of the intervention is necessary [54,55]. Intervention objectives were based on the Dutch guidelines for physical activity for adults, which included the following: 30 minutes of moderate PA for at least 5 days a week or 20 minutes of vigorous PA for 3 days a week [56].

Based on steps 1 and 2 and the research team’s expertise, a general framework was developed. The aim was to develop a highly tailored intervention that contains a self-monitoring tool, goal setting, social comparison, and motivational and context-specific messages.

Tailoring and personalization of the intervention content is realized in the following 6 ways: determining the personally appropriate type of support (ie, education, coaching, or feedback), selecting the personally relevant and preferred domain of PA for user coaching (ie, sports participation, stair use, or active transport), suggesting a weekly goal, selecting the personally appropriate behavioral determinants for coaching, sending only relevant coaching messages and filtering out nonrelevant messages, and tailoring and personalization of the app content.

First, after considering functionalities, validity, and costs of a range of available activity trackers, the user’s activity was monitored using Fitbit One, which includes monitoring of steps and stairs climbed. Fitbit One was chosen because of its functionalities and small size [57]. The activity monitor communicates with the Fitbit app and website that display the collected data for example by showing a color-coded chart indicating the proximity to the step goal, which is set to a default of 10,000 steps per day.

Fitbit allows developers and researchers to access Fitbit data and thus integrate the Fitbit data into health behavior interventions such as Active2Gether. To access Fitbit data, Fitbit offers an application programming interface (API). Fitbit One was validated using smaller time intervals (ie, minutes, hours, and days) relevant for real-time feedback and instant behavioral insights to its users. Healthy young adults (N=34) wore the ActiGraph GT3x+ and a Fitbit One for one week. Detailed information on the methodology can be found elsewhere [58].

Literature was reviewed for relevant, existing, and validated questionnaires to assess behavioral determinants. Behavioral determinants are assessed by means of a questionnaire with both its long and short versions, which were selected based on validations of such questionnaires. The long version is part of an “intake” questionnaire before the actual intervention and as a point of departure for the tailored intervention, whereas the short version is used repeatedly throughout the intervention period to dynamically tailor the intervention content to the user.

A questionnaire was designed for the purpose of assessing information on significant places. In addition, the Active2Gether app was built in a way that enabled the collection of the user’s location data.

To increase the users’ engagement, we assessed whether users’ friends were also participating in the Active2Gether intervention. Because Facebook is very popular among Dutch young adults—93% of Dutch adults aged 18-24 use Facebook [59]—Facebook was used to find connected friends that were also participating in the study.

To realize such tailored coaching, we developed a system that combines detailed behavior monitoring with intelligent data interpretation and model-based predictions. Thus, by combining data from the different sources, the system enables personalization of the coaching strategies to try to achieve the most positive effect on behavior change. Detailed information on the system and the development of the system can be found elsewhere and is not described in the Results section [60].

We decided that the communication channel of the Active2Gether intervention should be a smartphone app. The app shows the website in a format that is viewable on smaller screens. Thus, the intervention content was accessible through the app or the website. The research team developed the design template of the smartphone app. Information on the development of the app can be found elsewhere [60].

As a result of Step 1, a theoretical framework was built based on the relevant scientific literature (please see further details below). The theoretical framework was subsequently used to develop the content of the intervention and predict the PA behavior of the users so that the intervention content could be tailored to each individual user.

Self-efficacy, a key construct within SCT (and in other health behavior theories) [18,49], was adopted as a key construct in Active2Gether. Self-efficacy is defined as someone’s beliefs in his or her own capabilities to perform certain actions needed to achieve a desired outcome. Self-efficacy affects PA both directly and indirectly, as seen in Figures 1 and 2. Self-efficacy may influence outcome expectations—one’s beliefs about the positive and negative consequences of one’s behavior, such as participating in physical activities [18,49]. In other words, people who are more efficacious about being physically active will also be more likely to expect the favorable outcomes of participating in physical activities. [49] Moreover, self-efficacy may also influence how people perceive potential obstacles and impediments [49] and may also influence intentions to engage in PA behaviors [61]. Goal setting was adopted as a second important basis for change, where goals can be either proximal (ie, shorter-term intentions to act) or distal (ie, longer-term goals to achieve something) [49,62]. Distal goals are goals set for the longer term and they set the course for personal change [62]. According to Bandura [49], distal or long-term goals can initiate behavior change but are not sufficient to change behavior directly, as seen in Figure 1. Goal setting is dependent on the levels of self-efficacy and perceived barriers and opportunities. In line with this notion, a meta-analysis inspired by the action-control framework indicated that 48% of the participants who intended to be physically active failed to do so. Therefore, forming intentions is often not sufficient to realize behavior change; self-regulatory and action-control techniques are needed to support behavioral enactment [63]. A further meta-analysis on effective techniques in healthy eating and PA interventions concluded that interventions that offered self-monitoring and addressed self-regulation were more successful in increasing PA than interventions not including those techniques [8]. SCT posits that when individuals adapt and revise their behavior, they may adjust their beliefs and goals regarding this behavior [49]. In our theoretical framework, we therefore included “satisfaction,” which is defined as an evaluation of the PA behavior.

In line with SCT, we also recognized that the social environment influences behavior through social norms and that performing certain behaviors can evoke social reactions, both positive and negative [49]. In the Active2Gether intervention, we address not only intrapersonal (eg, lack of motivation and tiredness) and social barriers (eg, lack of support) but also contextual impediments (eg, lack of time, weather and travel distance), as seen in Figure 2. Lastly, it was decided that users will be categorized based on their awareness of their personal PA levels before they will be coached; people who are overly optimistic about their PA levels (ie, who believe they engage in adequate amounts of PA while their data show insufficient levels) will be much less likely to be motivated to increase their PA levels [64].

Content analysis showed that the apps available to date generally lack sufficient incorporation of evidence-based BCTs [30]. BCTs that were applied most often were providing feedback on performance, prompting self-monitoring of behavior, prompting specific goal setting, and planning social support or social change [30]. Additionally, focus group discussions with the target population indicated that participants preferred self-monitoring, goal setting, and a ranking feature but were not willing to share their accomplishments on social media for social comparison and initiating social support [45]. The focus groups further suggested that the Active2Gether app should be highly personalized, have an easy-to-use design and format, include a coaching feature that provides tailored feedback to self-set goals, enable competition with friends by ranking or earning rewards, and include the option to personally customize the application [45]. Finally, a Web-based cross-sectional survey among 179 young adults to assess their ratings with respect to the importance of specific BCTs applied in apps and their preferences for personalized tailoring confirmed the need for a personal coaching feature and showed that BCTs addressing goal setting, goal reviewing, feedback, and self-monitoring were rated as important to be incorporated in an app, whereas social support and social comparison were considered less important [44]. The combined results of the literature review, focus group discussions, and survey guided the selection of BCTs to be included in Active2Gether (Multimedia Appendix 1).

Step 2 resulted in the decision to make the following the primary objective of the Active2Gether intervention: increase total time spent in MVPA for those who do not meet the Dutch guideline, maintain PA levels of those who meet the guideline, or further increase PA levels if they so indicated. The secondary aims were defined as follows: to increase the underlying specific categories of MVPA (ie, minutes of weekly sports participation, weekly numbers of stairs climbed, and weekly minutes of active transport) and to enhance the underlying determinants of the PA behaviors.

The framework contained information on the levels of tailoring and an outline of the steps taken to deliver tailored messages. Detailed information on the framework components can be found in Multimedia Appendix 2.

Our test of the validity of the Fitbit One indicated that Fitbit can be considered a valid device to assess step activity for real-time minute-by-minute self-monitoring, although an overestimation of 677 steps per day by Fitbit was seen compared with the ActiGraph [58]. However, the validation study indicated that Fitbit is less suitable for providing instant real-time feedback and daily feedback on PA intensity levels (ie, minutes of moderate, vigorous, or MVPA) because it substantially and systematically overestimates the time spent per intensity level per hour [58]. For that reason, Fitbit is only used to assess step activity.

Participants need to give permission once for the application to access their activity data. These then can be collected regularly, and a summarized version of the data is stored in the Active2Gether database. These data are utilized in the following several ways: for presenting the activity level (ie, number of steps and number of stairs climbed) to the user, for determining the type of coaching, and for tailoring coaching messages.

We decided to assess behavioral determinants by means of a questionnaire with both its long and short versions, which were selected based on the validations of such questionnaires. The long version is based on existing questionnaires that have previously been validated (ie, Neighborhood Quality of Life Survey and Self-efficacy scales) or questions used in previous studies and were translated and adapted where necessary [20,21,67]. In the short questionnaire, we decided to use single item questions to assess each of the behavioral determinants that are part of the framework and the system. In the short version of the questionnaire, all determinants are specified for each coaching domain (ie, sports participation, stairs use, and active transport). These items were not pretested as such but were based on the long questionnaire. Multimedia Appendix 3 [12,68,69] provides an overview of the questions asked in the long and short versions of the questionnaire, including the answer options.

We also included questions about the participants’ significant places (eg, home address, parental home, sports location, university, work location) in the intake questionnaire. These questions focus on travel options from their home to significant locations, thus information about the active and nonactive transportation options. Additionally, information about the number of stairs available at each location and the maximum number of stairs the participant is willing to climb in one go is assessed as well.

The user’s location (GPS coordinates) is collected using Google’s location services that can be linked with the Active2Gether app. The location data are used to determine whether the user visited his or her significant locations (eg, home, study or work place, and sports club) and to derive information about transport and travels that have been made. In addition, information about the characteristics of locations is used for personalized coaching messages to the user. For instance, if a person is being coached on using the stairs more often at their place of work or study, it is only useful to suggest this when the option to climb the stairs is indeed present at the worksite or university.

Information regarding the participants’ friends is collected using the Facebook API. Users are asked to provide access to their Facebook ID and their connections by logging into Facebook once and giving permission for this. It is important to note that Facebook does not provide personal information about someone’s Facebook connections but only a list of Facebook IDs of their connections. This information can be used to see whether any Active2Gether users are connected on Facebook. If two participants of the current intervention are connected on Facebook, they see a ranking within the app that shows achievements of both users. In this way, the users only share their achievements with a closed group and not with “everybody,” according to the preferences stated in the focus group discussions.

The Active2Gether app shows a nonpersonalized, generic avatar with a welcome message that mentions the user’s current weekly goal. The app displays the current number of daily steps and stairs climbed. In addition, the app shows the following 4 graphs: a bar chart with the step progress toward 70,000 steps per week, a ranking with 6 other Active2Gether users—where possible Facebook friends—based on the step activity over the last seven days, the activity data for each weekday for the current coaching domain (ie, minutes of sport activity, numbers of stair climbed, or minutes of active transport), and the step activity for each weekday. The third and fourth graphs display the user’s own data and the average data assessed within Active2Gether. Moreover, these graphs can be adjusted according to the user’s preferences, that is, they can show data for the last week, last month, or from the first use.

Tailored messages and short questions are sent via push messages through the app. After the user reads the messages, they are displayed at the bottom of the app. Only the 5 messages sent most recently are displayed in the app. Figure 3 shows a screenshot of the app.