On July 24, 2012, we searched within PubMed, EMBASE, Cochrane, PsycInfo, and Cinahl. Because of the size of the search term used, the search terms can be found in Multimedia Appendix 1. Some keywords used in the search were diabetes mellitus; diabetes mellitus, type 2; Internet; eHealth; online; and Web-based. The systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement where applicable [30].

The retrieved articles were screened using the following inclusion criteria: written in English, published after 1994 (the introduction of the Internet), about T2DM, included patients aged 18 years or older, and concerned Web-based (online) self-management programs for which participants had to use the Internet to connect to the intervention. We only included RCTs to establish whether the BCTs used in the programs were associated with significant improvements. We defined self-management programs as systematic approaches to assist patients in their diabetes self-care, and where in some way or other patients were actively engaged and prompted to make decisions for themselves and have responsibility over their own actions [8,10]. Articles were excluded if they were not related to diabetes, reported only on technology testing, were not Web-based programs, did not target a self-management behavior, or only included type 1 diabetes mellitus (T1DM). Book chapters, abstracts, and pilot studies were not included.

Two researchers (MvV, WHJJC) independently reviewed the articles and extracted data on demographics, care setting, type of study, duration, measurements, nature of the intervention and control condition, applied inclusion criteria, used theory or model, BCTs, target behavior(s), outcome parameters, results, limits, and adherence. The risk of bias was assessed for all included studies using a quality assessment tool as proposed by van Tulder et al [31] and can be found in the Multimedia Appendix 2. The BCTs were categorized based on the checklist as proposed by Michie et al [27] which can be found in Multimedia Appendix 3. Disagreements regarding defined BCTs between the researchers were resolved by discussion within the research group. The BCTs used and the statistically significant outcomes were uncovered for each study. For each study with an improved study outcome (health behaviors, clinical outcome measures, and psychological outcomes), we looked if a BCT was present in the intervention for improving that particular outcome. We used Microsoft Excel 2003 to cross-reference this data and generated a list of frequently used BCTs associated with significant improvements in defined behavioral, clinical, and psychological outcomes.

Figure 1 shows a flowchart of the screening process. The search query resulted in 17,885 articles. After removing duplicated articles, titles and abstracts were screened for inclusion and exclusion criteria. After the first draft, 16,998 articles were excluded because they did not meet the inclusion criteria. We categorized the remaining 306 articles as: (1) studies on Web-based self-management programs, (2) reviews, (3) telehealth, telecare, or telemedicine studies [32-34], and (4) nonrelevant studies. This resulted in 13 articles reporting on 8 different Web-based self-management interventions for T2DM patients. These articles were individually read, screened for the BCTs used, and then discussed to reach consensus. Most articles provided only a short basic description of the intervention that was used. For 1 study [35], an additional article was consulted to uncover the content of the intervention [36].

Tables 1 and 2 provide an overview of the included studies and their results. Of the 13 RCT studies, 10 were performed in North America [37-46], 1 in Asia [47], and 2 in Europe [35,48]. The combined total patient sample size was 3813. Demographically, on average 54.8% of the participants were female and the average age was 57.2 years (SD 7.20). Average program completion rate was 81.7% (SD 15.2%). Four studies recruited their participants from the community by using flyers and newspapers [35,39,45,48]. Five studies recruited their participants from primary health care [37,42-44,46], 1 study recruited their participants from secondary health care [47], and 3 studies recruited their participants from primary and secondary health care [38,40,41]. All the studies included patients who had been diagnosed with T2DM for longer than a year. Five studies also included patients with T1DM [35,38,40,41,48] and 1 study also offered the intervention to people with diagnoses of chronic heart disease and chronic lung disease [45]. Average study duration was 6.69 months (SD 4.92). Adherence for all studies was high, which in itself contributes to the overall high quality of the included studies.

Tables 3 and 4 provide an overview of the interventions. Four of the 8 identified online interventions were developed by adapting existing (group) self-management programs into online self-management programs [35,39,45,47], and 4 interventions were newly created [37,38,40-44,46,48]. Two self-management interventions were developed as adjuncts to routine diabetes care, in which health care providers were able to have either online synchronous and asynchronous communication or telephone contact with the patient [38,40,41,47]. Six interventions were developed as standalone programs [35,37,39,42-46,48]. Five interventions were structured as sequential lessons [35,37,39,45-47] and 3 interventions allowed the participant to navigate freely through the program [38,40-44,48]. All 8 interventions offered some form of online coaching [35,37-48]. Seven of the 8 programs reported using a psychological theory or model as the basis for the self-management program, where some programs used multiple theories [37,42,46]. The theories and models used were: self-efficacy theory, [39,42,45], social support theory [42], transtheoretical model (TTM) [47], social cognitive theory [37,46,48], social-ecological model [37,46], and cognitive behavioral therapy [35].

Only 3 studies explicitly mentioned the BCTs applied [36,37,39]. For the other studies, information on BCTs was extracted from the program description. The frequency of used BCT’s found in the articles is shown in Table 5. The most commonly applied BCT’s were: provide feedback on performance, provide information on consequences of behavior in general, barrier identification/problem solving, provide information on consequences of behavior to the individual, and prompt self-monitoring of behavior. Some of the unused BCTs were shaping, prompting focus on past success, agree behavioral contract, and fear arousal.

Seven of 13 RCTs reported statistically significant improvements in health behaviors (diet, physical activity/exercise, medication use, smoking) [37,42,44-48]. Nine studies reported statistically significant improvements in clinical outcomes measures, such as glycated hemoglobin (HbA1c), fasting blood glucose, cholesterol, and triglycerides [37,39-43,45-47]. Nine studies reported statistically significant improvements in psychological outcomes, such as depression, diabetes distress, psychosocial well-being, self-efficacy, stress, and communication [35,37-39,42-45,48]. Table 6 provides an overview of the frequency of applied BCTs found to be associated with the statistically significant improvement of study outcomes.

The BCTs provide feedback on performance, provide information on consequences of behavior in general, barrier identification/problem solving, prompt self-monitoring of behavioral outcome, provide information on consequences of behavior to the individual, prompt self-monitoring of behavior, and plan social support/social change were all linked with improvements in health behaviors, clinical outcome measures, and psychological outcomes. Additionally, goal setting (behavior) was linked to improvements in clinical outcomes and facilitate social comparison was associated with improvements in psychological outcomes.

To the best of our knowledge, this is the first review of BCT use in online diabetes self-management support programs. This information should prove helpful in designing effective online self-management programs for people with T2DM. We identified 13 RCT studies reporting on 8 different online self-management interventions of which 4 pre-existed as group-based programs. Despite the introduction of the Internet in 1994, only a relatively small number of Internet-based self-management interventions for T2DM patients have been studied. We did find 143 studies on various forms of diabetes telehealth and telecare interventions. These studies were excluded from our review because they did not qualify for our definition of self-management programs. Rather these programs stimulated patients to self-monitor their blood glucose, followed by professional feedback and advice. To our knowledge, these programs do not explicitly prompt or support patients to make decisions [32-34,49].

The majority of the included studies that reported on self-management interventions only gave a very basic description of the program and its background. Indeed, it has been noted before that very few studies provide a detailed description of the actual behavioral change intervention [50-52]. This could be caused by the limited space authors have to describe the intervention in certain journals, making it difficult to replicate the study or allocate an effect size to specific parts of the intervention.

We observed that 7 of 8 interventions were grounded in a theoretical model, of which one of the used models (TTM), although popular, had limited evidence to support its assumptions [53,54]. Self-regulation theory with monitoring, action planning, and evaluating as its key features [55], and social learning theory, characterized by learning in social context [56], were most commonly used to inform development of the online interventions. However, only 3 studies specifically substantiated their choice for the use of specific BCTs to support their intervention [35,37,39]. After distilling the BCTs from the articles, the BCTs feedback on performance, providing information on consequences of behavior, barrier identification/problem solving, and self-monitoring of behavior and outcomes seemed to contribute the most to the effectiveness of the online self-management programs. These techniques also seemed to be key components for healthy eating and increasing physical activity [57], and were also commonly found in offline T2DM self-management programs [14,58]. However, being used frequently is in itself not a guarantee that these BCTs will actually contribute to the improvement of patients’ self-management in a particular domain. Nor does it mean that these BCTs fit the theory that was chosen to guide the intervention [59]. To further the development of effective online self-management interventions for T2DM and other chronic conditions, it is important to understand the underlying learning process [59]. Appropriate use of theories and supporting BCTs can prevent future interventions to be wrongly interpreted or executed by participants thereby improving treatment fidelity. This is particularly important for online programs, where confusion and misinterpretation on the part of the participant is more difficult to detect and address than in a group setting, for example, because of the more distant and static nature of the Web-based intervention. Therefore, it is crucial that the theoretical framework and BCTs are carefully chosen before a Web-based self-management intervention is created [60].

A number of potentially effective BCTs appear to be used rarely or never in online self-management programs for T2DM despite a good theoretical basis. For instance, only a selection of BCTs derived from social theories, which have a great influence on the self-management of T2DM [61], were represented in the reviewed studies that claimed to use these social theories. Although planning social support and having some form of coaching to provide feedback are frequently used, other BCTs that seem to affiliate with social cognitive theories, such as identification of a role model, model/demonstrate the behavior, and provide information about others’ approval, were not frequently used. The same is true for BCTs such as coping planning and use of imagery that have been shown to be effective in stimulating self-management of T2DM in an offline program [62]. Similarly BCTs that seem to be based in the classical and operant conditioning theories (characterized by associations and rewards), such as prompt rewards contingent on effort or progress toward behavior and teach to use prompts/cues, were also barely used, but have shown to be associated with improving physical activity [63]. Just because these theories and BCTs were not used in the reviewed interventions does not mean that they are of no value to an online self-management program.

The question then arises why researchers only use a limited number of BCTs and why the chosen BCTs do not always match the theories underlying their intervention. One explanation could be that current online interventions are being copied from published successful online or offline interventions based on a selection of theories and BCTs. By copying existing self-management programs, other relevant theories and BCTs are slowly phased out, narrowing the spectrum of BCTs used. Another obvious reason why certain BCTs are not being used could be that they are too complex or too technologically demanding and, therefore, too costly to integrate into an online environment. For instance, integrating elements of social support into the intervention, such as a forum, email messaging, and chatting functionality, demands large databases and continuous moderator involvement. The maintenance costs of these parts could influence the choice of using these elements. Finally, another reason for underuse of effective BCTs may be that the development of Web-based interventions for T2DM patients are driven primarily by technological advancements rather than being based on a BCT [64].

The number of articles uncovered in this review was limited to 13 RCTs, covering 8 different diabetes self-management interventions. By only including English studies there is a possibility we limited the amount of available interventions for this review. This, in combination with multiple BCTs used and a variety of outcome measures, made it difficult to allocate an effect size to a specific BCT. Furthermore, because of the fact that self-management interventions contain multiple modules with interactive components, it is difficult to attribute an improvement in a particular study outcome to one specific BCT.

The development of online self-management interventions for T2DM patients brings with it a responsibility of correctly constructing and choosing the working components to specifically target diabetes self-management goals and outcomes. To avoid a further narrowing of applied BCTs, we recommend developers of online self-management programs to not only copy existing successful programs, but also critically review and consider less frequently used BCTs in the context of their theoretical background and the chosen target behaviors.

Ideally, the creation process should follow the order of choosing a theory first, then matching BCTs, and lastly the technology to support the intervention. BCTs can be selected from the taxonomy of 40 BCTs as proposed by Michie et al [27]. By using this strategy, online theory-based self-management programs for T2DM patients can be developed without making unnecessary compromises or biased choices caused by existing technology. Furthermore, reporting detailed information on used theories and BCTs in research protocols and articles will benefit researchers in the creation and understanding of new effective Web-based self-management interventions for T2DM and other chronic disorders.