Safe-D development conformed to eHealth, persuasion, and interactivity best practices. A literature review was conducted to align development with current literature, presented thematically in this section. The aim of Safe-D was to persuasively encourage UV exposure. Persuasive technologies are those to prompt behavior change and reinforce learning [9]. A variety of concepts in the literature [2-4,9,10,24-28] were considered in the development of the app, including persuasion through monitoring and tracking, simplicity, rewarding behaviors, gamification, tailored messages, and improving user experience (UX).

Eysenbach [2] found that participants liked to monitor, track, and review records, and he established that “simply viewing records…could prompt healthier behavior.” Along similar lines, it has been found that the interactivity of tracking physical activity led to an increase in steps through incremental changes [27]. Therefore, young women tracking other quantifiable events may do the same. Interactive eHealth apps for complex messaging should enable exposure tracking and performance indicators.

Dennison et al [2] also uncovered competing participant desires for accurate records without frequent data entry. Where entry is complicated and time-consuming, users discontinued tracking [10,26]. The premise behind these findings is expressed by Fogg [3] who argued that barriers that could prevent desired behavior must be removed for the behavior to occur. Therefore, as much as interaction is important, it seems that data entry by the users also has to be quick, easy, and convenient, while accurate.

Rewarding positive behaviors, rather than punishing non-compliance, leads to continued engagement [2,8,26,29]. Consolvo and Landay [8] found negative reinforcement does not encourage behaviors, nor does lack of credit for compliance. Therefore, it was important to consider how to reward behaviors while not punishing non-compliance.

Gamification is the use of game elements in non-game contexts to provoke competition [30]. Calvo and Peters [31] found that provision of information alone only motivates change in 10% of individuals and suggested that interactive gamification may increase motivation and intervention compliance, a strategy also suggested by Segerstahl et al [10]. Given convergent recommendations, interactive eHealth apps for complex messaging should consider gamification to encourage use and improve compliance.

It is well published that tailored messages are most effective at motivation [2,8,26,28]. While messages are vital for an eHealth intervention to aid learning, doing so excessively dissuades users [2]. The contexts of target users must be considered to deliver relevant messages at opportune times for desired behavioral change to occur [3]. Clinicians also should be involved in complex message construction and messages tested for suitability and relevance with target users.

UX encompasses “reactions and feelings, which arise from interactions with a system” [10]. UX elicitation can be accomplished by applying visually appealing designs [8,28]; a competent look and feel [28]; creative, fun, and humorous elements [26]; ease of use; error tolerance; and system responsiveness [9]. The literature indicates that for successful persuasion, UX must be considered [2,9,10]. Therefore, interactive eHealth app development should focus on user-centered design by involving users and leveraging researcher expertise in UX and human computer interaction.

As is commonplace in qualitative research, development began relatively theory-free [34] in “round-table”-style discussions and ideation meetings with the team. Initial meetings were then used to fully appreciate and articulate the complex and confounding factors of achieving adequate vitamin D synthesis, which the app would have to cater for. These discussions were important to establish the range of variables to be considered, the diverse contexts of the target users, and, just as importantly, common understanding between the different groups of experts at the table. This was not only an important part of the app development process but was also the beginning of the participatory action research process.

A preliminary survey was conducted to further understand what young women wanted in an app, to inform design decisions. The survey was conducted online and advertised via a Melbourne Health media release, the “Netball Scoop” forum, VGen Core group, the Youth Affairs Council of Victoria monthly e-newsletter, and an email to past participants in the Young Female Health Initiative (a collaborative comprehensive health study of 16-25 year-old women in Victoria, Australia). A total of 71 respondents commenced the survey. Of these, 57 passed all eligibility criteria to participate, that is, being English-speaking females aged 18-25.

Developers were hired, and development followed stages from the software development lifecycle, selected due to its applicability to both business and health care scenarios [35]. The stages of our software development lifecycle were iterative and tailored to incorporate agile development ideologies [36]. These are described in Table 1. An existing algorithm [37] was selected to calculate recommendations for UV exposure due to applicability in New Zealand, similar in latitude to Australia.

Throughout each stage, available recommendations from the literature (as previously outlined) were used to help create the best experience in the app.

Researcher reflections and observational notes were used to highlight recommendations for future developments. This analysis was triangulated through data collected during retrospectives. Each member (ie, information systems and medical researchers, medical professionals, beta testers and app developers) shared what they thought went well, what did not, and what they would do differently next time based on this experience. Post-development, retrospective data were organized thematically through affinity mapping to check against initial researcher observations. This process condensed the data in a manageable format to identify any themes and relationships between themes, using an inductive approach, as per Neuman [33]. The data analysis enabled theoretical abstractions and pattern detection for identification of overarching principles for development of interactive eHealth apps for complex messaging.

Safe-D aims to safely improve vitamin D status, through individualized UV exposure recommendations, messages, and learning. Safe-D will be presented by the major actions that users take within the app, which are initial setup, tracking UV exposure, monitoring progress, recording missed exposure, and viewing messages.

Initial app setup is required when the user accesses the app for the first time. Safe-D prompts for (1) push notification and GPS permissions needed to provide interactive personalized recommendations based on real-time locational UV, (2) answers to the Fitzpatrick skin-type questionnaire [38] to determine skin type (avoiding self-selection error), which is input once, stored, and used in calculations, and (3) avatar skin, hair, and eye color and hairstyle configuration. These selections are used to present an avatar that will remind the user of themselves, without tying skin tone to skin type, thus increasing attention and persuasion [3,9,10,25,26,39]. The avatar configurations enable users to comfortably represent themselves for greater persuasion through imitation. There were extensive UX considerations regarding avatar and clothing options. The avatar does not propagate unrealistic body image, and cultural awareness was considered through inclusion of the hijab and niqab options.

Tunneling, the practice of guiding users through 1 question per screen, was implemented to reduce cognitive load. All data entry is simple to encourage usage, and logon details can be remembered for automatic sign-in to remove barriers to future use. As Safe-D is intended for initial evaluation in an RCT, access must be restricted to only participants through password protection.

To record UV exposure, the avatar must first be dressed. This is a seamless and fun interaction used to record exposed skin and sunscreen protection. Sunscreen and clothing are required for exposure calculations. Therefore, they are explicitly prompted. Exposed skin entry is simple: the avatar enables easy, visual, touch entry of clothing to calculate exposed skin. For usability, the avatar defaults to clothing from the last exposure, removing a step if exposed skin remains stable exposure-to-exposure.

These inputs, with locational UV levels and skin type, are used to calculate and display safe exposure recommendations. As exposure has a safe limit, the timer cannot simply count down. Exposure counts down to zero, then up in red to indicate, alert to, and dissuade overexposure. Gamification promotes usage, inspired by Fish n’ Steps metaphor [30]: a fish whose health improved or deteriorated with exercise. In the study, steps did not increase in the prestudy pedometer-only phase, yet when using Fish n’ Steps there was a positive impact on 74% (14/19) of participants [30]. In Safe-D, a sunflower was selected to reiterate connotations of sun exposure. As exposure reaches target, the sunflower health display grows healthier and happier. If healthy exposure is exceeded, health declines. When the user stops timing, they are provided with interactive feedback regarding whether more exposure is required or exposure has been met, or exceeded.

There are two ways to monitor progress on the app: viewing today’s exposure or overall exposure. Today’s exposure is visible on the homepage by assessing the sunflower’s health state. The sunflower health state can also be assessed while the timer is running. Overall progress is monitored via statistics, displaying the current consecutive days-of-use in the form of a streak, and longest consecutive use. Streak here is used in the sporting sense to refer to a running total of consecutive days when the app was accessed. To encourage reaching exposure, without exceeding the defined safe limit, Safe-D enables easy entry and viewing of exposure, making it apparent when the safe limit is exceeded. The progress display of how much exposure the user has had each day allows them to adjust behavior accordingly. This visual display takes into account that exposure can be too high, too low, or just right for the most number of users. Inspiration was garnered from discussions with other app designers [40] at the Melbourne Quantified Self group meetup, which also points to the concept of a “goldilocks” zone (the zone for just the right level of access to information for users, or an ideal level of desired behavior). In this case, the goldilocks zone refers to just the right level of exposure to sunlight for each user, which is integrated with a traffic light display system. Therefore, the “just right” level of sunlight exposure is denoted as green, “too high” is red to signal danger, and “too low” is yellow to indicate it is suboptimal without reprimanding the user. This “just right” goldilocks zone is dependent on the acceptable statistics based on the algorithm behind the app.

Missed exposure can occur for two reasons: a user may be unable to get sun or may forget to record it. When unable to get sun, the “I Can’t Get Sun Today” feature records the reason. If users forget to start or stop the timer, they can retrospectively enter or delete using “My Sun Exposure.” Users may be unable to get sun for a number of reasons, and this function handles this sensitively without punishing a user by forcing them to lose their in-app streak.

Safe-D aims to safely improve long-term vitamin D status through UV exposure. Thus, to prevent indefinite app reliance, information is required for learning. The complex messaging occurs on a per scenario basis (see Table 2), and different messages are delivered in different ways. These messages are complex because, while the same scenarios trigger messages for all users, the content of these messages varies based on the user’s skin type, how much skin they have exposed and whether they are using sunscreen or not, the current UV based on their current location, and the exposure they have already had for the day. Messages can be delivered in one or multiple modalities: in-app messages to deliver point-of-interaction information, push notifications requiring action and garnering attention to trigger this action as per the Fogg Behavioral Model [3], and mail messages available for later retrieval. Table 2 shows the scenarios that trigger messages to send and how these messages are delivered.

From our preliminary survey, 64% (7/11) of respondents provided an answer for the acceptable number of notifications per day, reporting 1-2 notifications per day as acceptable. However, beta testing revealed that 1 per day when the user was not actively using Safe-D was annoying and could potentially inhibit use. Push notifications send to a user’s phone only under the following circumstances: the timer is running and/or behavior needs to occur, Safe-D was not used the previous day, the user recorded “Too Much Sun” yesterday, and an educational message sent once a week.

Messages can be viewed either as push notifications or through in-app mail. Messages can be sent to groups to improve effectiveness rather than generic messages sent to all. Other messages are interactive, based on user inputs. Mail was designed in a way that enables tailored messages to be sent to defined groups. Hardcoded events are recorded in the Safe-D dashboard that researchers can access and identify individuals who may need special messages. An example of how this could be implemented is to look at users who continually flag that they cannot get sun today because they are “too busy.” A sensitively tailored message could then be sent to such a participant to encourage them to make time. In more extreme cases if a participant continuously exceeds app-defined safe exposure limits, it would be possible for the non-blinded researcher to identify and contact them directly for their safety.

Educational notifications are delivered weekly. These are designed to be fun, non-clinical, and short. Messages cannot build on previously received messages as participants can start Safe-D at different times. Clinical experts in bone health and dermatology first reviewed all written copy, and expert bodies, such as the Cancer Council, are referenced where appropriate. Messages are framed positively to avoid reprimanding users [8], and they correlate to the state of the sunflower to reinforce the metaphor and learning. Messages were created to be interesting and non-repetitive, cycling through different wording to avoid monotony due to repetition.

As previously indicated, the literature [2-4,9,10,26-29,31] suggested the following design concepts: persuasion through simple tracking, rewarding positive behaviors, not punishing non-compliance, considering message timing, including gamification, and designing UX. We extended these design concepts in the form of the following recommendations based on key learnings from the Safe-D project.

The principles we have presented in this paper were created through drawing inferences from Safe-D development. Safe-D takes multiple inputs and provides individualized and varying messaging to aid learning for users to optimize ongoing cutaneous vitamin D synthesis; continued exposure is required indefinitely to maintain vitamin D status. These principles can be leveraged for development of eHealth apps with similar individualized and complex messaging requirements where a one-size-fits-all message would be harmful. Examples of some potential eHealth apps that would require complex messages are described in Table 3.

Potentially most of these conditions will require individualized and complex messaging with the move towards personalized medicine [42,43]. While these principles will aid in development of other eHealth and interactive apps, they will not all be directly transferable to certain situations. For example, some condition characteristics or demographics may make it impractical to have patients as part of the study team. It is suggested here that these principles do not represent a set procedure, but rather recommended guidelines.

This paper has not presented a detailed description of the Safe-D app to maintain the integrity of the RCT. This information will be included in future reports after the RCT is completed. While Safe-D has been extensively beta tested, resulting in improvements, formal evaluation was not performed as part of this research. The RCT will evaluate whether Safe-D succeeds in its purpose to safely improve vitamin D status in young women and, therefore, the applicability of these principles to complex messaging. The proposed principles may not be applicable to all complex messaging and are not currently ranked in priority. Further, as the findings are based on a reflective analysis from participatory action research, it is possible that the guidelines may have inherent bias. This was mitigated somewhat through the triangulation of data, including using beta testers who were not involved directly in the app development. These limitations can be addressed through future research directions.

Future research, as part of the Safe-D study, should include an evaluation of the usability and utility of the app. The greater study will confirm whether or not the app is effective in safely improving vitamin D status. For example, it is possible to reach exposure limits for the day, to stop sun damage, without synthesizing enough vitamin D when limited skin is exposed. The algorithm may be updated during the study, as well as other changes based on participant interviews during site-visits regarding experiences, to improve effectiveness. A future improvement to Safe-D may include indicating to the user whether they will likely get adequate vitamin D based on what they are wearing, enabling them to change clothing or alter sunscreen use to improve synthesis.

As this research provides recommended principles for developing interactive eHealth apps for complex messaging, the results are largely theoretical at this stage and require validation before their suitability can be entirely understood. The lack of evaluation is a limitation to these principles. While behavior change was experienced anecdotally among the core researchers and beta testers, the Safe-D study will validate the app and therefore the principles applied in its development. Furthermore, future work in the eHealth app area should leverage these principles to validate them.

This research has provided insights into how to develop interactive eHealth apps for complex messaging. This outcome was achieved by drawing on examples from the literature. A set of guidelines to develop interactive eHealth apps generally, and specifically those for complex messaging, was previously missing from the literature. This research has contributed these principles. Safe-D delivers complex messaging to simplify education, explicitly considering user safety. Therefore, it showed a set of recommendations that might help develop a complex eHealth app that is simple for target users to engage with.