Throughout this paper, we follow the definition of eHealth by Eysenbach [15]. We use eHealth interventions in T2DM as a term to refer to all mobile Health (mHealth) interventions—those delivered via personal digital assistant (PDA), tablet, computer, Internet, and other forms of ICT—implemented to improve the management and outcomes of T2DM.

To address the objectives of this scoping review, we followed the 5 steps described in the framework by Arksey and O’Malley [52]: (1) identifying the research question, (2) identifying relevant studies, (3) selecting relevant studies, (4) charting data from the selected studies, and (5) summarizing and reporting the results [52]. Unlike systematic reviews, scoping reviews do not quantitatively aggregate the evidence but rather collate and summarize the evidence by mapping the related literature and examining the extent, breadth, nature, and characteristics of the available evidence [52]. Levac et al [53] recommended additional substeps to deal with the challenges encountered while conducting scoping studies. The details of the main challenges in each stage and the recommended substeps can be read elsewhere [53].

The research questions were developed after a rapid scan of the eHealth literature regarding the areas of prevention, self-management, and long-term medical care for persons with T2DM. We hypothesized that eHealth interventions play an important role in supporting patients who are under diabetic care. We also hypothesized that eHealth interventions targeting persons with T2DM include behavioral components. To search the relevant evidence for our hypothesis, we formulated the following research questions: Which outcome measures are used to assess the effectiveness of eHealth interventions in poorly controlled T2DM patients? What are the active ingredients of the eHealth interventions in poorly controlled T2DM?

PubMed, PsycINFO, and Web of Science were searched for relevant studies. During a preliminary search, we did not observe major differences in search results when using Excerpta Medica (EMBASE), Cumulative Index to Nursing and Allied Health Literature (CINAHL), and Cochrane Library. Therefore, we concluded that PubMed and Web of Science covered the relevant articles. Articles containing results pertaining to eHealth interventions targeting patients with poorly controlled T2DM (HbA_1c ≥7.0%) published in peer-review journals from January 1990 to June 2016 were considered as potentially relevant for the review. To be included in the review, articles had to report findings of studies with quasi-experimental or pre-/post-designs or of RCTs and had to have a focus on eHealth interventions and poorly controlled T2DM. Articles were excluded if they were published in languages other than English, if only titles were available, and if they were study protocols for future or ongoing evaluations of eHealth interventions. The screening process and identification of the relevant studies are shown in Figure 1.

The key word search strategy employed to identify relevant literature is described in Multimedia Appendix 1. All search results of PubMed, PsycINFO, and Web of Science were exported to EndNote version X 7.3 reference software (Clarivate Analytics).

Two authors (MK and TM) independently examined the titles and abstracts of eHealth intervention studies targeting persons with T2DM to assess their relevance for the review. We used the American Diabetes Association definition of poorly controlled T2DM as having an HbA_1c level of ≥7.0% [54,55]. In line with the framework by Arskey and O’Malley [52], a quality appraisal or quality assessment was not performed because it is not essential for scoping studies. Hence, methodological rigor of the published articles was not a criterion for inclusion or exclusion. The titles and abstracts previously selected by the 2 independent reviewers were merged and further screened for duplicates and following predefined inclusion criteria.

After charting the relevant data from the studies in spreadsheets, the results were collated and described using summary statistics, charts, figures, and tables. First, the types of eHealth interventions were charted into categories. Second, the outcome measures using studies examining the role of eHealth interventions in poorly controlled T2DM were categorized. Third, by exploring the contents of the intervention and cross-checking them with the definitions and examples of the 93 techniques in the BCTTv1, the active ingredients of the interventions were coded (Multimedia Appendix 3).

Keyword searches in PubMed, Web of Science, and PsycINFO resulted in 624, 775, and 5 articles respectively (Multimedia Appendix 1), with a total of 1404 articles.

Removing the duplicates, subsequent screening, and eligibility assessment of the titles and abstracts led to 227 potentially relevant articles. Screening the full texts of these 227 articles and applying the eligibility criteria resulted in 32 studies [6,20,25,31-33,56-81] being included in the review (Figure 1).

Geographically, most of the studies included in the review were conducted in the United States (46.9%), followed by Canada (15.6%) and Europe (12.5%) (Figure 2).

Regarding the study design, 16 studies were RCTs, 4 were parallel and 3 cluster RCTs, 3 had pretest/posttest designs, 2 were 3-arm randomized trials, 1 was a prospective randomized trial, and 1 was a nonrandomized controlled intervention.

Among the 32 eHealth interventions investigated in the included studies, 24 (75%) were tailored to the health and behavioral characteristics of the individual patient. According to the evidence (seen in Figure 3), an increasing trend for individualization of intervention content was observed.

Only 8 interventions were designed following theories or models of behavioral change. The theories/models used for designing the interventions were cognitive behavioral therapy [31,78], the reach out problem-solving model [31,78], motivational interviewing [31,61,78], the universal model of behavioral change [6], Green and Kreuter’s PRECEDE-PROCEED model [63], the health belief model [20], the community model [32], and Wagner’s chronic care model [71].

Of the 32 interventions, 18 (56%) were PDA-, tablet-, computer-, or mobile phone–delivered or Web-based interventions [6,25,31,33,58,60,61,68-71,74-80], 7 interventions (22%) were telehealth interventions delivered via landline telephones [57,59,63,66,67,73,81], 6 (19%) used text messaging [20,56,62,64,65,72], and 1 employed videoconferencing [32].

Changes in HbA_1c level were used as the primary outcome in the majority (28/32, 88%) of the studies. In addition, outcomes such as changes in lipid profiles (ie, total cholesterol, high-density lipoprotein [HDL], low-density lipoprotein [LDL], and triglyceride levels), changes in dose and quantity of antidiabetic drugs, use of drugs, adherence to treatment, and changes in diabetic knowledge were used as primary outcomes. Examples of secondary outcomes used in the interventions include patient satisfaction, medication adherence, performance of self-care tasks, and quality of life. The detailed list of primary and secondary outcomes employed in the included intervention studies are outlined in Multimedia Appendix 2. The outcome measures were broadly categorized and a framework was then developed (Figure 4) including all outcomes and suggesting pathways between different outcomes.

Outcomes included in this category were service satisfaction, remote home-monitoring device use, program compliance, patient interaction, perception of the program, and provider satisfaction.

This category included behaviors pertaining to disease self-management (ie, self-management score, summary of diabetes self-care activities, and performance of self-care tasks).

The outcome measures were the intermediate outputs that were considered to lead to the long-term effects of the interventions.

Long-term effects of the intervention were quantified by changes in diabetic quality of life, bodily pain, general health, vitality, role functioning, general well-being, diabetes dependent impairment, and the cumulative incidence of diabetic complications, including incidence of microangiopathic complications (ie, diabetic retinopathy, diabetic neuropathy, diabetic nephropathy, diabetic foot ulcer, eye complications, macrovascular complications, microvascular complications).

In most of the studies (25 out of 32), a statistically significant change in HbA_1c percentage was used as a primary measure of eHealth intervention effectiveness in changing glucose levels in persons with poorly controlled T2DM. However, changes in diabetes knowledge [69], knowledge about antihyperglycemic medications, patient-reported medication decisional conflict [61], and cumulative incidence of diabetic complications [82] were also used as a primary outcome measures for assessing eHealth intervention effectiveness. In addition, achieving the target of fasting blood glucose <120 mg/dL, fasting and postprandial blood glucose levels [64], changes in physical functioning and role limitations [32], self-efficacy, medication adherence [20], proportion of patients achieving HbA_1c <7% without hypoglycemia [65], adherence to treatment prescriptions, and use of drugs [72] were used to determine intervention effects.

These outcomes were combined in the framework displayed in Figure 4. This framework was developed after careful examination of the nature of each outcome and hypothesizing its relationship in the pathway.

The types of behavior change techniques (BCTs) identified in the selected interventions are described in Multimedia Appendix 4. All of the 32 interventions included multiple BCTs. Of the 16 overarching thematic categories, 11 (69%) were addressed in interventions: goals and planning, feedback and monitoring, social support, shaping knowledge, natural consequences, comparison of behavior, associations, repetition and substitution, reward and threat, regulation, and antecedents (Figure 5). No BCTs from the following 5 overarching categories were identified: comparison of outcomes, identity, scheduled consequences, self-belief, and covert learning. Of the 16 hierarchical clusters of the BCTTv1, feedback and monitoring was included in 27 of the 32 studies.

Of the individual 93 BCTs of the BCTTv1, 31 (33%) were employed in interventions to change behavior to manage poorly controlled T2DM. On average, 6.7 BCTs (SD 2.0) were included in interventions. The BCTs and the specific content of interventions with examples are displayed in Table 1. The maximum number of BCTs included in 1 intervention was 11 and the minimum was 3 (Multimedia Appendix 4).

The purpose of this review was to identify the relevant outcome measures reported in studies examining the effects of eHealth interventions in persons with poorly controlled T2DM and characterize the active ingredients of eHealth interventions among persons with poorly controlled T2DM using the BCTTv1.

Most of the studies (25 out of 32) measured the effectiveness of eHealth interventions using a statistically significant change in HbA_1c percentage as a primary outcome measure. This is similar to the review reported by Vorderstrasse and colleagues [29]. A review from the Cochrane Library by Pal and colleagues [27] found that all 16 RCTs included in its review used HbA_1c percentage as a primary outcome measure of effectiveness. Lipska and Krumholz [83] challenged this glucocentric approach, reporting that the effectiveness indicator of interventions in T2DM is moving away from the historic surrogate marker (ie, HbA_1c) to cardiovascular outcomes.

The identification of the active ingredients of the behavioral interventions is a basis for synthesizing evidence, building on evidence, and replicating interventions targeting behavioral change. The development and use of the EPOC taxonomy and BCW models contribute to the homogeneity in characterizing the contents of different interventions and in quantifying intervention effects (eg, by aggregating effect sizes). However, we observed that these 2 frameworks were not sufficiently comprehensive to characterize the content of interventions in detail. BCTTv1, in contrast, appeared suitable for in-depth analysis of the active ingredients of interventions. It offered a means of handling heterogeneity and provided a baseline for meta-analysis or the estimation of effect sizes for quantifying effects of behavior change interventions.

In our scoping review, only 31 (33%) of the 93 BCTs were identified in interventions. Similarly, Presseau et al [42] identified less than a quarter of the 93 BCTs in 23 interventions. BCTs such as credible source, reward (outcome), focus on past success remain underused in interventions. Innovative eHealth interventions employing these BCTs need to be tested with regard to their impact in changing patient behavior and affecting T2DM outcomes. Of the 31 BCTs identified in interventions included in this review, the most frequently used were instruction on how to perform behavior, adding objects to the environment, social support (practical), feedback on outcomes of behavior, self-monitoring on outcomes of behavior, and prompts/cues. Van Vugt and colleagues [49] identified BCTs such as providing feedback on performance of behavior, providing information on consequence of behavior, problem solving, and prompts/cues as the most commonly used BCTs in Web-based self-management programs for patients with T2DM. Pal [27] demonstrated that among the most frequently used BCTs, prompt self-monitoring of behavioral outcome and provide feedback on performance were reported to have significant effects on HbA_1c levels. However, frequency of inclusion of an individual BCT is neither proof for it significantly improving patient outcomes nor proof of a proper design of interventions [84].

Our study results suggest that, on average, 6.7 BCTs were included per each intervention. The evidence on whether including many BCTs in an intervention improves patient outcomes is not strong. Systematic reviews by Avery and colleagues [50] and Cradock [85] revealed that only 50% and 60%, respectively, of the most frequently used BCTs were associated with reductions in HbA_1c. An evaluation of diabetes-related apps by Hoppe and colleagues [86] indicated that diabetes mobile phone apps having more BCTs also had significantly higher functionalities and higher user ratings. However, which combination of BCT ingredients had a stronger effect and which BCTs were key moderators of effectiveness in poorly controlled T2DM needs to be further investigated. Customizing eHealth interventions to individual behavioral characteristics and disease progress increases the effectiveness of the intervention [87]. Tailoring or individualizing the communication between patients and providers has gained substantial attention in the past decade. In this review, we observed that more than 75% of the interventions were customized to the individual patient characteristics or needs. In addition, a generally increasing trend of tailored eHealth interventions was noted in the reviewed studies. Strategies used for tailoring vary across studies. Kim [64] and Wayne [80] used pragmatic approaches of tailoring and contextualized the intervention with respect to the individual patient. McFarland and colleagues [67] tailored the intervention to individualize the communication between patients and providers. First, patients self-monitored blood glucose levels by using monitors and transmitted their data using a messaging device. A registered nurse then downloaded the message and contacted the patient via telephone to evaluate whether there were any specific health concerns. Based on specific concerns (eg, with regard to adherence to certain medications or a dietary regimen or hypoglycemic events), patients were given recommendations regarding insulin dosage or lifestyle changes. Ralston and colleagues [71] tailored the Web-based intervention according to the clinical condition of each patient. Accordingly, the care manager responded to specific messages from each patient and reviewed the submitted blood glucose levels of each patient to adjust hypoglycemic medications as needed. Several studies suggest that tailoring may be an effective means of behavioral change and improving self-management skills [88-91]. Tailoring also helps initiate, enhance, and safeguard the partnership between the provider and the patient, increasing shared decision-making and person-centered care which ultimately facilitates the uptake of the desired behavior, such as healthy eating and improved physical activity [92]. However, a recent systematic review reported that there is lack of evidence to suggest tailored eHealth interventions are more effective than nontailored interventions [93]. Therefore, this issue obviously requires more research.

Despite a broad consensus that the use of theories or models to guide the development of interventions leads to greater impact of interventions, the current review showed that only 8 (25%) of the 32 eHealth interventions were theory-based. The finding of our review is consistent with the claim that undertheorization of eHealth interventions and underutilization or an inadequate application of behavioral science and health education theories is still a major issue in the eHealth intervention literature [38,47,84,85,94]. The evidence on effectiveness of designing and implementing interventions through the use of theories is mixed. Some evidence suggests that theory-grounded eHealth interventions are more likely to be associated with positive outcomes of patients with T2DM. Theories can enhance the uptake of the desired behavior by supporting providers and patients to collaboratively set targets, enhance the motivation of intervention participants, and provide a roadmap for behavior and treatment modification [47-51]. The impact and processes by which eHealth interventions influence outcomes are not directly comparable to the impact of pharmacological drugs that are administered into the body and bring a change within a certain period of half-life of the ingredient. eHealth interventions, in part, impact cognitive processes (eg, by improving knowledge) and may help intervention participants internalize the advantages of performing the target behavior, such as improving self-management, dietary, or physical activity behavior, leading to long-term maintenance of these behaviors. Behavior maintenance can then be subsequently linked with changes in biological markers and long-term changes in quality of life and a lower incidence in complications.

Our scoping review had several limitations. The definition of poorly controlled diabetes was based on that of the American Diabetes Association. However, other guidelines, such as the one from the National Institute for Health and Care Excellence, consider an HbA_1c level up to 7.5% as a good indicator of glycemic control. This should be taken into consideration while interpreting our results. In some cases, it was challenging to crossmatch intervention contents described in the articles with the BCTs. For example, there were interventions that included motivational messages or calls to induce the uptake of a target behavior. However, motivation was not explicitly described in BCTTv1. In addition, there were interventions with poor descriptions. For instance, interventions provided education but there was no information available regarding the type of education. Coding the poorly described interventions was therefore challenging. Hence, we were forced to develop assumptions to deal with poorly described interventions. Another limitation of our scoping review was that the correlation between the 2 reviewers coding the BCTs was not systematically assessed. Rather, 2 coders independently analyzed contents of interventions, and where they disagreed, a third person was consulted to reach consensus.

For most interventions, changes in HbA_1c levels were reported as a primary measure of effectiveness. Overall, the BCTTv1 appeared practical and helped identify the active ingredients of interventions. Our results suggest that one-third of the 93 BCTs were employed in eHealth interventions targeting persons with poorly controlled T2DM.

Developing theory-based interventions and considering BCTs during the intervention design phase is desirable for obtaining effective interventions and transparently reporting the results of these interventions in the future and possibly in other chronic disease contexts.