This study adopted the principles of a systematic review process to identify and select apps, including using an app quality assessment tool, a search strategy, predefined inclusion criteria to screen apps, app rating and selection, and data extraction for qualitative analysis.

Two reviewers performed the data extraction. One reviewer evaluated Apple App Store’s apps, and the other reviewer evaluated Google Play apps. Before reviewing the apps, the reviewers conferred and decided to include all the critical features that the apps provide. We specifically focused on medication reminders and the adherence features that the apps offer. The reviewers extracted information on the app features using a predetermined Excel (Microsoft Inc) sheet. The Excel sheet was developed on the basis of a pilot app assessment exercise with 5 apps with extensive features; the Excel sheet was refined for this study. We excluded apps that could not be assessed owing to country-specific or other restrictions. We included additional features of free apps that were available upon subscription only.

We grouped the apps on the basis of the operating system (ie, Android and iOS), presented the mean MARS rating, and summarized the main features. We also presented the MASS and app-specific rating for medication adherence based on awareness, knowledge, attitude, intention to change, help-seeking, and behavior change.

We identified 249 apps in Google Play and 209 apps in the Apple App Store. The initial inclusion criteria resulted in 63 and 39 Apps for Google Play and Apple App Store, respectively. Finally, 8 apps with a MARS greater than 4 were included. The mean MASS for the included apps was 4.2, and the average app rating by users was 4.7. Figure 2 depicts the app selection methodology followed for both Apple App Store and Google Play for selecting the relevant apps that satisfy the selection criteria.

All 8 apps were available in the Apple Store [49-56], 7 were available in Google Play [49-55], and one app had the option to be accessed through a web application and Amazon Alexa [56]. Given that all the apps were available in the Apple App Store, we extracted information for the apps from the App Store. Table 1 presents an overview of the apps. In contrast, Multimedia Appendix 2 presents detailed information for each app.

The Apple App Store apps were categorized as medical [49-53,55,56] and health and fitness [54] apps. Including English language, the apps were available in around 5 [51,56], 10 [49,54,55], 20 [52], and 30 languages [50,53]. Additionally, all the apps had family sharing options (ie, the app could be shared with and used by 6 family members) [49-56]. Apart from one app [56], the apps defined their privacy policy explicitly, defining the identified [49-52,54,55] and deidentified data [49,50,53-55], although it differed between the apps. For example, data such as health and fitness information, contact information, identifiers, diagnostics, location, user content, and usage data were considered “identified” data by one app [52]. In contrast, the same data were considered “deidentified” data by another app [53]. Further, the apps were sold by software companies [54,56], digital health companies [49,51,52,55], health and fitness companies [53], and individuals [50]. Finally, although all the apps were free [49-56], they had an in-app purchase option [49-52,54]; that is, users had the option to pay a prescribed amount to access additional features, such as premium subscription features [49-52] and access to a coach [54].

The compatible operating system (OS) version for the apps varied drastically between the apps. For the apps to run smoothly, 9.0 [49], 10.3 [56], 11.0 [51], 12.0 [50], 13.0 [53-55], 13.2 [52] or later versions of the OS were required.

Among people with diabetes, optimum glycemic control is essential, and good adherence is associated with a lower risk of all-cause mortality and hospitalization [58]. Many models have been studied to explore the acceptance and use of technologies. One such model is task-technology fit (TTF). TTF states that a technology should be effective in completing the assigned task, which will lead to an increase in user performance and adoption [59]. TTF rates the task characteristics and technology characteristics of apps, which affects the effectiveness and adoption of the apps by the user. Accordingly, in this section, we used the TTF and quantified the effectiveness of apps in diabetes adherence to evaluate whether the apps promote adherence to diabetes medication [60,61]. A recent study used TTF to evaluate the effectiveness of the mHealth app in delivering health care services [62].

Primary prevention and mitigating the severity of diabetes revolve around regular BG checks, food consumed, physical activity, other vital readings, and adherence to medication [63]. Hence, they are the primary factors considered in this study. Although apps have been found to improve awareness of medication adherence and reduce self-reported barriers to medication adherence among medication-nonadherent patients with diabetes [64], a study observed that a large proportion of diabetes self-management apps lacked features for enhancing medication adherence and safety [65]. Consequently, among the app features, we considered the overview (report generated and graphically presented for a period), reminders (customized alerts to perform tasks), notifications (alerts when BG shoots up or drops below the threshold), assistance (additional support provided by various stakeholders), and motivation (assistance from loved ones and challenges through gamification) as the imperative app features needed to support an individual in diabetes medication adherence.

Table 2 presents a TTF matrix for diabetes medication adherence to evaluate the apps against the defined primary factors and app features. In the matrix, we have marked “yes” when factors and features are available. We then totaled the score for primary factors and app features separately. We defined a score of 5 in both primary factors and app features as apps assisting in very high adherence, 4 and above in either primary factors or app features as apps assisting in high adherence, and a score of 3 and above in either primary factors or app features as apps assisting in moderate adherence. We further defined a low adherence app as apps scoring 2 and above in either primary factors or app features and any score below 2 for primary factors or app features as apps resulting in poor adherence.

According to our evaluation, none of the apps assists in very high adherence, and there are no poor-adherence apps. In addition, we found that 3 apps each assisted in high adherence (25%) [49,55] and moderate adherence to diabetes medication (25%) [54,56]. In contrast, 4 other apps were in the low adherence category in assisting with diabetes medication adherence (50%) [50-53].

In this systematic review, we identified 8 high-quality apps publicly available in the Apple App Store and Google Play free of charge. The mean MASS of the apps was 4.2, while the average user ratings was 4.7. Only 2 apps showed high adherence and 2 apps showed moderate adherence, while half of the apps showed low adherence. However, since the global market is poised for rapid growth and could have widespread implications in delivering personalized health care [66], interoperable mHealth solutions available in the Apple App Store and Google Play are generally designed and developed for widespread adaptability. Our systematic review suggests evaluating apps using a standardized framework before prescribing them to patients with diabetes, and using behavioral theories for improving medication adherence.

All the apps were available in English and more than 5 other languages [49-56]. Moreover, bilingual English-speaking patients expressed the need for language translation to understand and communicate using the apps [67], and the selected apps could address this concern through the multilanguage option provided. However, globally, 79% of adults with diabetes live in transitional countries [3]. An increase in mobile phone subscriptions over the years, including in transitional countries [68], and the constant technological advancements in mobile phones [69] promote mHealth as effective for use in transitional countries having barriers, such as lack of infrastructure and equipment and technology gaps [70]. Hence, if the selected apps have languages that predominate in transitional countries with a higher prevalence of diabetes, they might assist in diabetes management, reducing the burden on the health care system and curbing mortality rates.

Deidentification is essential for protecting patient privacy and removing identifiers that directly or indirectly point to a person [71]. However, there is variability in the definition of deidentification [72]. For instance, among the considered apps, health and fitness, contact information, identifiers, diagnostics, location, user content, and usage data were considered identified data by one app [52]. In contrast, the same data were deemed deidentified data by another app [53]. This discrepancy could be due to differences in definitions and legal practice followed in countries where the apps have been developed [72]. Hence, health care apps developed in accordance with the guidelines of legislation, such as the US Health Insurance Portability and Accountability Act and the European General Data Protection Regulation, could have uniformity in defining identified and deidentified data [72]. Accordingly, following global regulatory guidelines could bring uniformity in the use of identified and deidentified data.

In-app purchases permit the user to purchase added services from within an app, and 5 apps considered in this study had this feature [49-52,54]. Remarkably, in-app purchase options work well in promoting health apps wherein the essential functions are offered freely [73]. However, the intention to upgrade to a paid subscription is driven by the subscription features, benefits, and price value [74]. Hence, developing free health apps with in-app purchase options will be necessary to promote the app; however, the content should provide sufficient value to retain subscriptions [74].

A few apps described in this study could be used by all patients with diabetes [50,53,56]. In contrast, other apps were specifically targeted to patients with T2DM [51,54] or to those with T1DM, T2DM, or gestational DM [49,52,55]. Nevertheless, apps with specific descriptions were sought more by users compared to those with general descriptions [75]. Some apps had an age rating of ≥4 [50-54,56] and ≥17 [49,55]. T1DM affects children [3], and although some apps were indicated as suitable for patients with T1DM, the age rating was ≥17 [49,55]. Another app had an age rating of ≥4 and could be used by patients with T1DM [52]. Hence, developing apps with specific descriptions and appropriate age ratings could assist the appropriate target population.

We have considered physiological factors influencing diabetes and app features to develop and assess apps using the TTF matrix.

This review had certain limitations. First, in this study, we included apps that had greater than 500 ratings. Therefore, we could have eliminated several new apps with fewer than 500 reviews, but that might have had a higher MASS. Second, we included only apps in the English language, possibly missing effective apps in languages other than English. Third, our search date was limited to May 2020. However, the constraints associated with the severity of the COVID-19 pandemic could have enabled the development and usage of apps that satisfied our selection criteria. Finally, we limited our search to the apps available in the major app stores: Apple App Store and Google Play. However, these stores account for 80% (4.41 million) of the global apps as of May 2020 [76].

Our findings provide guidelines for app developers to develop an app that assists in diabetes medication adherence. Further, several apps are very likely to be developed and prescribed during the pandemic period [77,78]. Therefore, evaluating the apps using our framework could help the apps to provide high medication adherence for better health and well-being of patients with diabetes. With the advancement in smartphone technology, various health vitals could be captured and transferred in real time for analysis [79-81]. The application of machine learning and big data could provide a wealth of predictive information to the patient and to health care professionals [82-87]. Furthermore, high-quality apps coupled with evidence-based ICT programs using user-centric designs, wearable device, and machine learning approaches could be used to provide personalized interventions for people with diabetes [84,87-93]. Hence, our findings could have practical and research implications for diabetes medication adherence.

Our framework to evaluate smartphone apps in promoting diabetes medication adherence has considered physiological factors influencing diabetes and app features. Therefore, our findings could have positive implications for the design and development of apps for patients with diabetes. Additionally, the available apps could be evaluated in accordance with our framework, and those apps promoting higher medication adherence could be prescribed for better health outcomes.