In October 2020, we conducted a systematic literature search in MEDLINE via PubMed, Cochrane Library, EMBASE, CINAHL, and Web of Science Core Collection in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) strategy [41]. These databases are representative of the entire health-related literature on DM, as they are the five largest databases in this field. The search strategy included the following keywords as Medical Subject Headings or EMBASE Subject Headings terms, as well as title and abstract terms: (“diabetes mellitus”) AND (“smartphone” OR “mobile phone” OR “cell phone” OR “iOS” OR “android”) AND (“mobile applications” OR “app”). The search strategy in PubMed, for example, was as follows: (“diabetes mellitus”[Mesh]) AND (“Smartphone”[Mesh]) OR (“Cell Phone”[Mesh]) OR (“mobile phone”[Title/Abstract]) OR (ios[Title/Abstract]) OR (android[Title/Abstract]) AND (app[Title/Abstract]) OR (“Mobile Applications”[Mesh]).

In addition, we manually searched reference lists and Google Scholar to identify further papers. The search results were filtered in the databases by year (January 2008 to October 2020) and language (German and English). The studies were screened and selected by two independent reviewers.

Since this is a scoping review, we have included several study designs and outcomes to summarize the evidence available on the topic. We included primary research studies (randomized controlled trials, exploratory studies, observational studies, and pre- and posttest design studies) and peer-reviewed studies published between January 2008 and October 2020. Because English is the worldwide scientific language and the authors are native German, we have taken German and English literature into account.

Studies reporting on the clinical effectiveness of DM-specific mHealth apps in DM management in patients with T1DM, T2DM, and GDM that specified the features of the apps and their health effects were included.

We looked for reported significant changes (P<.05) in health-related oucomes such as glycemic control (eg, glycated hemoglobin [HbA_1c], and hypo- and hyperglycemia), blood pressure, cholesterol, body weight, self-care, and self-efficacy. Self-care was defined and measured as DM self-management that included items assessing general diet, specific diet, exercise, blood glucose testing, foot care, and smoking using a questionnaire. Self-efficacy is a predisposing factor that be impaired in chronic diseases like DM. Increased self-confidence levels, measured by questionnaires, can set the stage for improved glycemic control [42].

Furthermore, we excluded posters, comments, study protocols, duplicates, and studies focused on DM diagnosis or prevention.

We extracted the following information about each study: author, year, study design, intervention and control groups, baseline and follow-up HbA_1c values, type of DM, sample size, and main findings related to the outcomes of interest.

We synthesized the studies according to outcomes because the clinical perspective focuses on the improvement of individual outcomes through the intervention. In addition, we conducted a meta-analysis to assess the impact of the interventions on the management of HbA_1c.

HbA_1c is the most important and most studied clinical outcome related to technological therapy for DM, including mHealth apps. To determine the change in HbA_1c, we pooled appropriate studies with intervention groups (using mHealth apps only) and control groups (usual care) and calculated the difference in means, with a 95% confidence interval. We included studies that reported changes in HbA_1c as a percentage from baseline to the end of the study for intervention and control groups.

The database search in October 2020 in the five relevant databases yielded a total of 796 hits. After removing the duplicates, there were 654 citations. Based on the titles and abstracts, we excluded 619 unsuitable papers. The reasons for exclusion can be found in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flowchart (Figure 1). Furthermore, we excluded 8 unsuitable studies based on their full texts. After the additional manual research, which identified 2 papers, there was a total of 27 suitable studies to include in this scoping review. In total, we included 27 papers analyzing 1646 patients in the intervention groups and 1241 in the control groups.

Of the 27 papers, 7 were focused on T1DM (308 patients in the intervention groups and 129 patients in the control groups) [43-49], 12 were focused on T2DM (743 patients in the intervention groups and 645 patients in the control groups) [50-61], and 4 were focused on GDM (339 patients in the intervention groups and 343 patients in the control groups) [62-65]. The remaining 4 papers did not specify the type of DM they looked at or included a mix of DM types (256 patients in the intervention groups and 124 patients in the control groups) [66-69]. Multimedia Appendix 1 gives an overview of the included studies. With regard to the study design, we included 19 randomized controlled trials, 1 randomized crossover trial, 1 exploratory study, 1 observational study, and 5 studies that used a pre-post design (1 of which was controlled). Different diabetes mHealth apps were evaluated in each study. As predicted, the apps had a great variability in their features. Some apps included only one feature, such as digital diaries [47,60], feedback on glucose measurements [51], physical activity promotion [53], data transfer to electronic medical records [61], or educational features [52], while other apps combined multiple features. In the following sections, we present the results of the studies sorted by included outcomes.

Table 1 shows all of the results according to HbA_1c values. Effects based on the comparison of HbA_1c levels between the intervention and control groups at the study end points were investigated. Findings are presented in Multimedia Appendix 2. The meta-analysis revealed an effect size, compared with usual care, of a mean difference of –0.54% (95% CI –0.8 to –0.28) for T2DM (8 suitable studies) and –0.63% (95% CI –0.93 to –0.32) for T1DM (2 suitable studies) (Multimedia Appendix 3).

In general, specific mHealth apps clearly improved glycemic control by effectively reducing HbA_1c values in patients with T1DM (mean difference: –0.63%, 95% CI –0.93% to –0.32%) and T2DM (mean difference: –0.54%, 95% CI –0.8% to –0.28%). While no significant improvements in blood pressure or cholesterol were found in patients with T2DM, a few studies showed positive tendencies toward improved self-care and self-efficacy with regard to patients with DM in general.

The studies were diverse with respect to the type of DM, study design, number of participants, and app features. Often, different app features were combined or the app was used in conjunction with web portals, feedback from HCPs, or Bluetooth-enabled devices. Because of that, it was not possible to distinguish a relationship between specific app features and health outcomes.

However, some effects were clearly demonstrated from the use of DM-specific mHealth apps in general. We categorized the outcomes included in the studies into HbA_1c, hypo- and hyperglycemia, further glycemic control outcomes, blood pressure, cholesterol, body weight, self-care, self-efficacy, and further outcomes.

Nearly all of the studies (22 of 27 studies) included HbA_1c level as an outcome, with a total of 2352 patients analyzed. For patients with T1DM, 3 studies reported significant improvements within the intervention groups, with a mean difference of –0.57%, yielding HbA_1c levels of 7.73% on average, and 2 studies reported significant differences between groups with a mean difference of –0.73, favoring the intervention groups. Those results are consistent with other reviews. Sun et al [71] reported on 3 studies that showed a significant improvement in HbA_1c levels, ranging from –0.50% to –0.58%, in people diagnosed with T1DM. Hou et al [72] reported a significant improvement of –0.49% in HbA_1c level but rated the grade of evidence to be low. Moreover, Kitsiou et al [73] investigated the effect of mHealth interventions in general and reported an improvement of –0.3% in HbA_1c levels in people with T1DM.

For T2DM, one of the included studies found a significant improvement in HbA_1c levels, approximately –1.1%, yielding a mean HbA_1c of 7.64% in the intervention group [56], and 7 studies determined a significant difference between intervention and control groups, with a mean difference of –0.78%, favoring the intervention group. Furthermore, Kim et al [61] showed a significant improvement for users who were highly satisfied with the mHealth app. This could be problematic in light of the results of Fu et al [74], who found that patients rated the usability of T2DM-specific apps to be “moderate to catastrophic”. However, Fu et al [74] also reported similar significant improvements in HbA_1c values, based on the results of 4 studies, ranging from –1.9% to –0.4% [74]. In addition, they highlighted that people with poor glycemic control (HbA_1c >9%) achieved greater reductions and that apps with interactive features (eg, receiving feedback) were especially likely to show highly significant improvements [74]. The importance of receiving feedback, for example from HCPs, was also reported by Hou et al [72]. In their review, they reported that the higher the frequency of HCP feedback was, the greater was the reduction in HbA_1c [72]. All in all, they reported a mean difference of –0.57% in HbA_1c for patients with T2DM using mHealth apps [72]. In other reviews, such as one by Cui et al [75], a significant mean difference of –0.4% of HbA_1c was found between DM-specific mHealth app intervention groups and usual care groups in favor of the intervention groups.

The reported improvements in HbA_1c in people with T1DM and T2DM are consistent with the results of the studies that did not specify the type of DM. Of the studies that did not specify the DM type, one study found a significant improvement in HbA_1c within the intervention group [66] and the other study found an increase in the proportion of participants with HbA_1c <7% [68]. No clear effect on HbA_1c could be seen in the studies that focused on patients with GDM because of limited data.

The problem of limited data also applies to the study outcomes of hypo- and hyperglycemia and further glycemic control parameters because the studies included different kinds of outcomes. Thus, no clear conclusions can be drawn from them. Other reviews reported an improvement of glycemic control through mHealth app interventions [71,74,75] but mainly based their conclusions on HbA_1c improvements.

The outcomes of blood pressure, cholesterol, and body weight were only included in studies focusing on T2DM. No effect could be determined for blood pressure, total cholesterol, or HDL or LDL cholesterol because the studies predominantly reported nonsignificant differences. With regard to body weight, no effects could be determined either because of inconclusive study results. This is consistent with the review by Cui et al [75], which did not report on the effects of T2DM-specific mHealth apps on blood pressure, cholesterol, or body weight.

Although the data on the outcomes of self-care and self-efficacy were also limited for all types of DM, the studies showed a trend toward improvements in both. Other studies reported improved DM self-management skills as well [71,76]. However, Hoppe et al [77] criticized the lack of inclusion of behavior change techniques in DM-specific mHealth apps.

Other than the effects on health-related outcomes, different aspects of DM-specific mHealth apps should be taken into account for further research and development. For example, Höchsmann et al [53] highlighted that not just the content of an app is important but also the way it conveys the content. They created their app as a game and found significant effects on HbA_1c level and steps per day as a result of the intervention [53]. In addition, Boels et al [52] reminded us that the different needs of people with DM—for example, if someone requires insulin or not—need to be considered. Also, the age of the patients appears to matter. Hou et al [78] showed in their subgroup analyses that young people with T2DM are more likely to benefit from apps. Moreover, elderly people diagnosed with DM may have special needs, such as a larger font size because of reduced eyesight, and not all apps are able to meet these needs [79]. This goes hand in hand with the conclusion of Meister et al [2] that living in the digital world demands a kind of digital literacy. But despite the widespread use of smartphones, digital literacy barriers are common in vulnerable populations, which could reduce the effectiveness of diabetes technologies [80]. Moreover, a lack of standards and regulations lead to potential health risks, for example via misinformation through an mHealth app [39]. Certified medical apps are more trustworthy and should therefore be preferred. However, in the field of DM, they are still rare, and additional online libraries of high-quality DM-specific mHealth apps should be taken into account for recommendations [28]. In addition, data safety in mHealth apps is a serious concern, as they deal with sensitive data [28,39,81]. These issues need to be addressed in future studies.

Although the results of this paper show some possible improvements achieved by using mHealth apps in the treatment of DM, some limitations need to be addressed. A major limitation is the small sample size, especially regarding GDM. Only 4 studies that focused on GDM were included, and they in turn reported predominantly on different outcomes. Thus, no effects of mHealth app use could be determined for patients with GDM. To resolve this issue, we must increase our knowledge of which outcomes are affected by DM-specific mHealth apps and include these outcomes in further studies. In addition, it appears that for patients with GDM, a separate assessment of mHealth app effectiveness is reasonable because outcomes that are important to patients with GDM do not apply in general to patients with T1DM or T2DM, such as different aspects of pregnancy and childbirth. Another limitation of this paper is that the quality of the included studies was not assessed. Therefore, we cannot judge whether an effect was based on poor study quality.

Overall, this review clearly shows how the use of DM-specific mHealth apps results in improvements in glycemic control by effectively reducing HbA_1c levels in patients with T1DM and T2DM. However, a few studies found no significant effects of app use on blood pressure or cholesterol in patients with T2DM. With regard to the other outcomes, only a few suitable studies could be identified. In addition, a handful of studies showed positive tendencies toward improved self-care and self-efficacy as a result of mHealth app use in patients with any type of DM. This suggests a need for further research on the clinical effectiveness of DM-specific mHealth apps.