An adaptation of the program-planning model developed by Kreuter et al [53] was conceptualized to adjust the model to a Web-based intervention. Kreuter’s model was initially designed for print-based communication and includes the following nine steps: (1) analyzing the health problem, (2) developing a program framework, (3) developing the tailoring assessments, (4) designing feedback, (5) writing tailored messages, (6) creating tailored algorithms, (7) automating the tailoring process, (8) implementing the program, and (9) evaluating the program.

We felt it necessary to adapt the model to the needs of our Web-based intervention because it appeared more logical for our team to follow the original steps in a different order (see Table 1). First, we created the algorithms in Step 4 because we deemed it more logical to decide on which circumstances participants should receive which messages prior to any message writing activities. Making this step prior to writing the message booklet allowed us to have a concrete overview of how many messages should be written, which message participants should receive, and under which circumstances they should receive it. By creating the algorithms during Step 4, we were also able to assign a coding variable to each message, which later facilitated the automation of the tailoring process (Step 7). Finally, the original Step 4, designing feedbacks, became Step 6 because our team agreed on the difficulty of envisioning the appearance of the feedbacks without knowing what their content would be. Since messages are at the core of the intervention, it once again seemed more intuitive for our team to write the messages before designing their appearance to adapt the design of the Web platform to the messages’ length and style.

In addition to changing the order of the steps, we divided Step 9 (Kreuter’s), evaluation of the program, into two different steps: (1) usability evaluation, which comes prior to implementing the intervention, and (2) process, impact, and outcome evaluation, which is subsequent to the implementation of the intervention. Usability evaluation, a new step added to the model, will be described and discussed later in this paper. The adapted model is presented in Table 1.

The questionnaire was developed using the constructs selected for the tailored content of the intervention. In addition to these constructs, sociodemographic and anthropometric data (self-reported) were collected including height, weight, age, occupation, origins, and marital status. Items related to each tailored construct are presented in Table 5.

This tailoring assessment questionnaire will be incorporated into the website, within the registration process, which participants will have to complete to officially be part of the intervention. This registration process can be completed during the 2-week enrollment period and includes five main steps: (1) creating a website account, (2) completing the selection criteria phase, (3) reading the detailed information about the project, (4) giving informed consent, and (5) completing the tailoring assessment questionnaire.

Moreover, some constructs were reassessed during motivational session 4 and motivational session 8 to provide a dynamic tailoring experience and comparative-progress feedback to participants. The constructs that were reassessed during motivational session 4 are PA behavior, intention, and importance ruler. During motivational session 8, the reassessed constructs were PA behavior, intention, and confidence ruler. There are two major reasons that confidence ruler and importance ruler were used instead of self-efficacy and attitude to provide comparative-progress feedbacks. The first reason is that fewer items are necessary to measure MI-constructs compared to those of the I-Change Model, which eases the task of answering questions for participants and may make the motivational sessions more enjoyable. The other reason is that MI constructs were developed specifically for intervention purposes. In other words, scores obtained on importance ruler and confidence ruler allow the program to provide an appropriate tailored feedback aimed at eliciting change-talk and therefore better foster autonomous motivation toward change in participants.

Tailoring algorithms are the logic statements or decision rules that specify which messages should be given to which participants under which circumstances [53]. These algorithms were created for the following intervention components: (1) enrollment process, (2) all motivational sessions, and (3) the action planning tool. In order to have a clear view of the algorithms required, we used LucidChart to create flowchart representations of the algorithms, giving an overview of all the processes and decisions to be made for each component. All flowcharts can be examined in detail in Multimedia Appendix 2.

A flowchart can be viewed as a type of diagram that uses symbols to graphically represent the processes, decisions, operations, and other concepts of a program [101]. Although there is a standardized use of specific symbols when drawing flowcharts [101], it is important to mention that we did not fully adopt the correct use of these symbols due to the fact that this methodology comes from a computer science and engineering background, which is not our primary expertise. However, all flowcharts created for the intervention are detailed and designed in a way that health professionals who are not familiar with the flowchart method will still be able to understand the algorithms of each component.

With the flowchart representations completed, the next phase in officially creating the algorithms is to transform the representations into real computation formulas. This phase was executed in Step 7, automating the tailoring process, with the TailorBuilder software.

A message booklet was developed for the tailored and non-tailored components of the intervention. The parts of the booklet related to the tailored components were developed based on the flowcharts created in Step 4. For these tailored components, a variable name was assigned to each possible message block to facilitate the later automation of the tailoring process (Step 7). For the non-tailored components, messages were written for the website home page, main menu (when participants are logged in to the website), all emails that participants will receive, the self-monitoring tool, and Safety Tips, FAQs, and Contact tabs. Because of many updates made during the development of the intervention, no official message booklet can be offered as a reference in this paper. However, a sample of the message booklet is provided in Multimedia Appendix 3.

Considerable efforts were also made to write every intervention message in a needs-supportive style of communication to correctly apply the SDT and MI concepts outlined in Table 2. By considerable efforts, we mean that messages were written with careful attention by MM and FB to support basic needs, based on knowledge gained from our literature review on MI and SDT, MI workshops attendance, and guidance from our MI expert collaborator. However, once written, the message booklet was reviewed only by the research team members. Neither our MI expert collaborator nor an SDT expert reviewed the message booklet. To verify if the messages adequately support the basic needs of competence, autonomy, and relatedness, participants will be asked to answer an adapted version of the 6-item modified Health Care Climate Questionnaire [102] 1 month after the intervention (first follow-up).

The development of the intervention components design was executed in two parts. First, non-tailored components were designed by a member of the research group (MM) with LucidChart. Web pages corresponding to each of these components were drawn for a first time by MM and then discussed and revised with the Web solution company and FB to refine them. All mock-up templates created with the software can be seen in Multimedia Appendix 4. Second, tailored components were designed with TailorBuilder via an integrated hypertext markup language (HTML) editor, combined with external cascading style sheets (CSS) used to design certain parts of the messages. The intervention logo was designed by the communications service at the Université du Québec à Trois-Rivières.

It is also important to mention that in addition to the discussions held with the Web solution company, efforts were made to respect some of the general design principles described in the program-planning of Kreuter et al [53]: (1) preferably have one objective per Web page, (2) choose a single feel for the tailored feedback provided (eg, authoritative, empathetic), (3) do not overfill pages with content and leave empty space, (4) be consistent with illustrations or photographic style, (5) limit the design to a few font styles, and (6) use bold effects and color sparingly. Examples of the application of these principles are presented in Table 6.

The Web solution company created most of the non-tailored intervention components using the previously designed mock-up templates of Step 6. These non-tailored components were thus transformed into functional Web pages and include the following: (1) website home page, (2) account creation page, (3) a more info page, (4) a forgotten password page, (5) safety tips tab, (6) FAQs tab, (7) contact tab, and (8) a self-monitoring tool.

All other components were generated into functional Web pages using TailorBuilder and include the following: (1) tailored motivational sessions, (2) action-planning tool, (3) main menu when participants are logged in, (4) follow-up questionnaires, (5) pre-programmed emails, and (6) an enrollment session including selection criteria, detailed information about the project, consent of participants, and the tailoring assessment questionnaire. These components were generated in two major steps. First, messages and questions related to each component were linked with their appropriate coding variable and integrated into their HTML/CSS format. Second, by using the variable codes assigned to the entire set of messages and questions, all the algorithms were translated into their computer format, which is mainly composed of decision rule statements (eg, IF…ELSE…THAN…) implemented with a simplified programming language integrated in TailorBuilder. An overview of all the programming code written with TailorBuilder is presented in Multimedia Appendix 5.

Once the two parties involved in this step finished creating their Web pages, the Web solution company merged all the Web pages into one website, thus creating the complete DEF Web platform ready for Step 8, evaluating usability.

Our usability evaluation not only assessed the users’ ability to perform tasks successfully on the DEF website but also examined the users’ broader interaction with the website comprising their thoughts, feelings, and perceptions [103]. The method used to execute this step was a first experience for the research team and was based on regular practice in the field of usability evaluation [103,104]. Detailed activities of this step are described below.

In order to receive guidance on the best way to evaluate our website, we first recruited a field expert who assisted the team in the development of an evaluation framework, but not for the realization of the evaluation itself. Concretely, by examining the project’s needs and constraints for the project, this expert provided an interview guide, concrete guidelines to follow when running this kind of evaluation, and advice on tools we can use to facilitate the evaluation. Since we also recruited an external interviewer for the evaluation, some guidelines were discussed extensively with this professional to standardize the interview procedures.

Participants for the usability evaluation were recruited from a diabetes education center located in Trois-Rivières. A small sample of the study population (n=11) was recruited. Some characteristics of the participants were (1) 6 females, (2) average age of 53.5 years, (3) all but one not meeting PA recommendations, with a group average of 50 minutes per week of moderate aerobic PA, (4) average years with type 2 diabetes diagnosis was 8.11, (5) all but one using the Internet at least daily, for an average of 1 hour per session, (6) 5 participants with elementary school as their highest education level, 2 participants with high school, and 4 with a university degree or higher, and (7) all participants followed by at least one health professional. Although small, this sample seems representative of the population targeted and the larger population of Canadian adults with type 2 diabetes [105,106].

Participants were invited to the facilities of the Université du Québec à Trois-Rivières, where one-on-one interviews were conducted in a research lab by the external interviewer (a clinical nurse) using a think-aloud protocol and a software allowing for sound, video, and computer screen recording. During the interview, participants were asked to execute some predefined tasks on the website, while all of their facial expressions, comments and actions on the website were recorded. While participants executed the tasks, the interviewer used the interview guide to make sure that the participants voiced their thoughts and feelings to be able to measure their comprehension and enjoyment of specific parts of the website. The data collected during the recorded interviews was used for analysis and usability evaluation on various metrics including time on task, number, and type of errors made, bugs, and relevant issues [103]. All metrics were useful in making website improvements.

It is also important to mention that the evaluation was conducted iteratively. That is, a first evaluation period took place about 5 weeks before the intervention was implemented. During this first period, six interviews were conducted over a 2-day period after which a first set of improvements was made to the intervention. Four weeks prior to the intervention implementation, a second evaluation period took place during which five interviews were again conducted over a 2-day period. After this final evaluation period, a month was granted for final improvements before implementing the intervention. Due to the project’s time restrictions and because we felt quite confident about launching the intervention with the data collected during the two evaluation periods, no other usability testing was conducted. See Figure 2 for iterations of the usability evaluation of the DEF intervention.

Following this step, numerous improvements (ie, more than 40) were made, ranging from minor (eg, changing a word that was hard to understand on a page to an easier one) to major (eg, redesigning a whole page). Here are some of the major improvements made to the website: (1) reduction of the number of pages that must be viewed to complete the enrollment process (ie, 5 pages fewer, from 20 to 15 pages), (2) replacement of the neutral background of videos to backgrounds inspiring PA behavior, (3) redesign and reorganization of the information on the self-monitoring tool page, (4) reduction of the length of the information messages and summary messages by about 50% for almost every tailored motivational session, (5) significant reduction in the number of choices for MI-based multiple choice questions in tailored motivational sessions 3 and 5, (6) redesign of the emails sent to recruit participants, and (7) reduction in the number of times participants have to scroll to almost zero.

In the Web-based context of the DEF intervention, data from participants are handled automatically through interactions between the website and TailorBuilder to provide appropriate feedback. This step therefore involves only establishing (1) how participants will be recruited and (2) how the data will be collected and exported to data-analysis software.

For the recruitment phase, the DEF tailored intervention was implemented in partnership with Diabète Québec, a recognized organization for Canadians with type 2 diabetes (particularly in the province of Quebec). Consequently, following a joint agreement between both parties, Diabète Québec sent two invitation emails over 2 weeks to nearly 2300 Canadian francophone adults with type 2 diabetes, who were members of its newsletter.

Sociodemographics and data related to each construct of our theoretical model were collected automatically by TailorBuilder. All data collection questionnaires related to the intervention were automated, so all answers given to each question of the questionnaires were automatically saved under their appropriate variable code and assigned to the appropriate participant. Data exportation was executed easily with TailorBuilder, which allows for the creation of Statistical Package for the Social Sciences (SPSS) files that merge all the data collected in a ready-for-analysis format.

According to Kreuter’s program-planning model, this step serves to determine how effective the program was in meeting its objectives and how the program could be modified for greater performance in the future. Kreuter et al propose their own questions to properly evaluate a program. Our team decided to keep only some of these questions and to add others to better suit the context of our intervention. Questions that the research team will seek to answer through this step are (1) through which processes did behavior change occur in participants? (2) how effective was the program in meeting its behavioral objectives? (3) how effective was the program compared to another type of Web-based intervention using a more general approach (ie, non-tailored) combined with peer-support strategies? (4) was the intervention content well received by the participants? and (5) how can the intervention be improved in the future regarding every dimensions of the Reach Effectiveness Adoption Implementation Maintenance (RE-AIM) framework [107]?

In order to address these questions, it is important to note that this intervention is part of a larger project led by one of the authors (FB). This research project will specifically seek to compare the effectiveness of two types of Web-based interventions having the same behavior change purpose: (1) the intervention described in this paper and (2) a Web-based intervention offering similar but non-tailored content combined with peer-support strategies (ie, a Facebook discussion group).

Readers can access all videos of the intervention in Multimedia Appendix 6. A demo version of the intervention website is also available in Multimedia Appendix 7.

The DEF intervention development cost was about CDN $59,700 and took approximately 54 full-time weeks to develop. Details on the costs and duration of each step are provided in Table 7.

The enrollment period started on September 15, 2014, while the intervention officially started on September 29, 2014. Out of 2300 potential participants targeted for the DEF tailored intervention, approximately 530 people visited the website, 170 people completed the registration process, and 83 corresponded to the selection criteria and were enrolled in the intervention. Four weeks after the intervention started, no dropouts were reported. Similar enrollment results have been obtained for the two other groups of the study (ie, peer-support group and control group). Prior to recruiting participants, a statistical power analysis to determine the required sample size was conducted based on using repeated-measure analysis of variance to detect change in PA. Thus, according to a mixed plan 3 (ie, Web-based tailored intervention group, Web-based peer-support intervention group and control group) x 3 (ie, baseline, 1- and 6-month follow-up), assuming a statistical power of 0.80 and alpha=.05, a total sample size of 204 participants is necessary to detect a conservative effect size of f=.10 [20,22]. This sample size was calculated with the software G*Power 3.1.6.

Through the experience of developing the DEF Web-based tailored intervention, the research team learned several lessons that will help guide future health professionals interested in developing such an intervention.

First, we confirm that the development of a tailored intervention can be quite a complex and time-consuming process, as has been highlighted by previous tailored intervention developers [35,108]. However, in our case, once the development and implementation parts were executed, we found that minimal effort was then required by the research team during the intervention itself. Considering the reach and effectiveness that this kind of intervention can have on health behaviors, our beliefs that this promising approach (ie, Web-based tailoring) can have a significant positive impact on a population level at a decent cost are reinforced. These beliefs are also increasingly supported by evidence [109-111].

Further adding to the complexity of the development, the use of multiple behavior change theories can also require additional efforts, as researchers must be sure to understand all theories and models underlying the intervention and ensure that the integration of these provides a coherent framework for later development steps. We chose to integrate the I-Change Model, SDT, and MI to further explore potential factors that will add to the current effectiveness of tailored interventions targeting PA behavior, as it has been suggested as an area for future research [18]. Similarly, it is also important to note that some evidence suggests that using multiple theories generally results in less effective interventions compared to the use of a single theory [112], but this trend does not seem to be well supported and other evidence suggests that the use of constructs derived from different theories can result in the long-term effectiveness of interventions [60].

Regarding the development method used for the intervention, the program-planning model developed by Kreuter et al was very useful for building an evidence-based tailored intervention. As we mentioned, very small changes were made to the initial framework in order to make it fit with our logic, context, and needs. However, one interesting area for future research could focus on developing new strategies to accelerate the process through which evidence-based tailored interventions are developed, given that this is generally a time-consuming task. For example, in our case, all steps were executed in a sequential manner (eg, once Step 1 was completed, Step 2 was executed), but future developers could examine the feasibility of executing certain steps simultaneously to reduce development time. Or, researchers could also try to determine which systems are faster for automating the tailoring process of interventions, whether it be TailorBuilder [91], other similar tailoring systems [113,114], the use of direct computer programming human resources, or other means.

For the automation of the tailoring process and the creation of the website, TailorBuilder has been a very important tool for Web pages including tailored components. This software allowed us to make these tailored Web pages a reality, even if the research team had minimal computer programming background. However, it is important to note that one member possessed some basic skills with the client-side programming languages CSS, HTML, and JavaScript, which we highly recommend acquiring before developing a Web-based tailored intervention, unless one plans on using specialized human resources for the entire website construction.

As noted earlier, the step added to the program-planning model of Kreuter et al, usability evaluation, was found to be very useful during the development of the intervention, as it allowed for several important website improvements. However, more iterations during the usability evaluation phase could have led to more insights on how to improve the website. Also, we might have gained more insights within the same amount of time by using a slightly different method such as the Rapid Iterative Testing and Evaluation method (RITE) [115] in which, under certain conditions, changes to the website can be made as soon as a problem is identified even after observing only 1 participant. In our case, the difference is that we waited until the end of each testing period, including 5 or 6 participants per period, to make any improvement. Plus, the method used did not allow us to test the final improvements on a sample of our target population before starting the intervention. In other words, only the research team was able to pre-evaluate these final improvements.

Additionally, our literature review covered a wide range of fields and provided a strong corpus of evidence on which to base the intervention. However, it is important to note that other fields of research that are less explored in the tailoring field could have further contributed to our evidence basis, such as fields related to message design (eg, the “look and feel” of messages) and message structure (eg, emotional appeals, narrative vs statistical, gain vs loss framing) [23]. In order to further optimize their interventions, future developers may want to consider these fields or others in their literature review and thus broaden their corpus of evidence.

Finally, to our knowledge, this paper presents one of the most exhaustive development descriptions of a Web-based, tailored intervention published to date. However, although we believe this work can help health professionals adopt a deeper view of what the task of developing such an intervention truly represents, we are also of the opinion that greater efforts could be made in future papers to be even more educational, contributory, and reproducible on many levels (eg, clarity of the programming code presented, completeness of message booklet and demo website, use of open source software to facilitate reproducibility). We encourage future developers to embrace this challenge.

Two main limitations related to the development of the intervention are worth mentioning at this stage of the project. First, the assessment of development costs was not performed according to a standardized method [116,117] and should be interpreted with caution. This assessment should be interpreted as including all costs associated with the development of the intervention (Steps 1 to 9) and paid from the grant awarded to FB by the Quebec Health Research Fund. All other potential development costs that could be associated with the project and not paid from the grant were not considered in the assessment, such as the salaries of 2 researchers, MPG and FB. Costs during the intervention and costs associated with Step 10 of our development model (eg, salaries for assistant researchers during these periods and financial compensation for follow-up completions) were also not included because we did not view them as strictly associated with the development phase of the intervention. Although not following a standardized method, and along with the abovementioned details, the cost assessment presented in this paper could guide future investigators on the tangible development costs they will have to pay directly from their awarded research grant, if following a similar development methodology.

Another limitation is that behavior maintenance may currently be minimally addressed by the intervention (ie, as a reference, interventions promoting better maintenance normally last more than 24 weeks [118]), while we clearly stated that it was one of our main goals along with PA behavior adoption. To explain our reasoning, we wanted to determine whether behavior maintenance could be promoted primarily by the use of intervention strategies intended to develop autonomous motivation in participants, thus leading them to keep practicing PA after the intervention for personally relevant reasons and enjoyment of the behavior. Since this is the major reason why we issued behavior maintenance as an intervention objective, we acknowledge that the intervention does not support all the scientific evidence related to PA maintenance available to date.

This paper aimed to fill a gap in the field of Web-based intervention research, which is in need of more in-depth descriptions of the development process of such interventions. It provides a detailed example of a fully automated, Web-based, tailored intervention development for researchers interested in developing similar interventions. Also, as it represents a potential avenue to enhance the effectiveness of tailored interventions, this paper discussed the integration of the I-Change Model, Self-Determination Theory, and Motivational Interviewing as the theoretical framework of the intervention.

Finally, we reemphasize that usability evaluation, a step added to the program-planning model of Kreuter et al, is strongly encouraged prior to the implementation phase in order to better optimize an intervention before disseminating it on a larger scale.