By 2045, 693 million people will be living with diabetes, the majority with type 2 diabetes (T2D) [1]. Diabetes self-management behaviors, including blood glucose monitoring, healthy eating, being physically active and taking prescribed medications, can improve diabetes-related outcomes, reduce complications, and improve quality of life, but these behaviors can be difficult to initiate and sustain [2]. Diabetes self-management education and ongoing support are critical for establishing and maintaining self-care routines [3]. However, the uptake of face-to-face educational programs is low because of several factors, including difficulty in attending because of medical, financial, or transport issues; lack of perceived benefits; and shame and stigma [4-7]. Furthermore, the provision of ongoing support is difficult because of resource constraints and issues of reach and scalability [5]. An increasingly common strategy to address these challenges has been to use smartphone apps as a means to deliver diabetes education and self-management support to complement clinical care.

The evidence for the efficacy and acceptability of diabetes self-management apps is increasingly robust [8-11]. However, research trials typically focus on overall efficacy, not individual differences in user experiences, and cannot shed light on factors that influence engagement [12-14]. This is a gap that needs to be addressed if apps that demonstrate efficacy in controlled trial settings are to be translated into effective real-world interventions [15,16].

The lower engagement, or lack of thereof, with diabetes self-management apps is often attributed to a mismatch between what people with T2D want and the functions provided by apps, loss of motivation, and the difficulty integrating app use into everyday life [17-22]. Research suggests that multiple factors, including treatment, attitudes to self-management, and existing knowledge, influence the needs and preferences of people with T2D [22]. For example, people with newly diagnosed diabetes favor apps that educate them about diabetes, whereas those with more experience of living with and managing diabetes express frustration with basic education materials and are keen to see more cutting edge news and links for further reading [23-25]. Those who have been living with diabetes for longer engage with technology to refine care routines, whereas those less experienced use diabetes self-management tools to establish routines, for example by troubleshooting out-of-range blood glucose readings [20,26]. Finally, those with more experience are less willing to explore new options, including apps, especially if the benefits are uncertain, and the effort is substantial [27]. Unfortunately, participants in these studies were asked either to give feedback on apps they had not used before or to use unfamiliar devices. These limitations precluded an in-depth examination of user experiences over time and in the context of participants’ everyday lives.

Therefore, this study aimed to investigate users’ experiences of a diabetes self-management app (My Diabetes Coach [MDC]) accessed via personal devices and used in the context of everyday life over a prolonged period and to understand the interplay between users’ characteristics, needs, and preferences and engagement with a diabetes self-management app.

This qualitative study was a substudy of a randomized controlled trial testing the efficacy of a T2D self-management app MDC. The trial was conducted from 2014 to 2018 (Australia New Zealand Clinical Trials Registry ID ACTRN12614001229662) [28,29]. The University of Melbourne’s human research ethics committee approved this study (HREC number: 1442433). In-depth, semistructured interviews were conducted to evaluate the MDC app in terms of users’ experiences. We used a qualitative approach to explore subjective perspectives constructed from the experience of people with T2D using a self-management app in the context of their everyday lives [30]. This report is consistent with the consolidated criteria for reporting qualitative research checklist (Multimedia Appendix 1) [31].

The MDC app was designed to provide education, support, and feedback on diabetes self-care using weekly sessions or appointments with an embodied conversational agent Laura (Figure 1). Laura had human-like characteristics and mimicked human conversation using interactive voice recognition (IVR) and a database of prerecorded conversational elements. Laura conversed with users either via spoken voice or text, using sophisticated script logic. The app’s script logic was personalized by incorporating information and targets provided by users’ health care professionals (eg, blood glucose monitoring targets). Users were able to respond to Laura by speaking, inputting text, or touching an option on the screen. The program was designed to enable responses made in a preceding session to dictate the direction of the next session with the user, enabling a high degree of personalization.

The first appointment with Laura was scheduled to suit the user and thereafter occurred at the same time every week, with some flexibility, enabling users to complete their appointment up to 48 hours after the planned time. Users could choose a particular module from those available but were required to complete the module over a series of sessions before moving to a new one. Available modules included blood glucose monitoring, nutrition, physical activity, medication taking, and foot care. The app applied several gamification elements, including goal setting, monitoring of progress, feedback, and quizzes [32].

Throughout the trial, users had access to a program coordinator to assist them with technical difficulties. They were also given an Accu-Chek Advantage blood glucose monitoring device with Bluetooth capabilities (Roche Diabetes Care), enabling the automated upload of glucose data to the MDC app. Finally, the app had inbuilt links to a website with diabetes resources and a user guide for the app.

Invitations to participate in the MDC trial were sent by mail to adults with T2D (in New South Wales, Queensland, Victoria, and Western Australia) registered with the National Diabetes Services Scheme (NDSS). Participants were eligible if they were adults aged 18 years or older, diagnosed with T2D, registered with the NDSS for <10 years, had access to a smartphone (with an operating system of at least iOS 8.0 for Apple devices or OS 4.2 for Android), and fluent in the English language. The exclusion criteria were as follows: women who were pregnant or planning to become pregnant; individuals reporting severe comorbid conditions that would prevent participation in the trial; and individuals on nonstable doses of diabetes-related medications.

Interview participants for the qualitative study were recruited from the intervention arm of the MDC trial, all of whom had accessed the MDC app for up to 12 months. Purposive sampling was used to achieve variation in user characteristics, including age, gender, education, occupation location, duration of T2D, and use of the app (operationalized as the number of completed chats).

Participants were sent a plain language statement describing the study and were required to provide written consent. Participant characteristics were collected at baseline via a self-report questionnaire, including demographic and clinical details and current health app use.

An interview guide was developed to include questions about the user’s self-reported diabetes expertise, how they managed their diabetes, when and how they engaged with the app, and their experiences using it. In-depth semistructured interviews were conducted through telephone (by SB) and recorded using SmartInteraction Suite, a cloud architecture voice recording solution (CTI Group). SB has several years of experience in diabetes-related research, including conducting telephone interviews. She worked as a research assistant on the MDC project and was involved with program development, participant recruitment, and data collection. Many of the participants had previously interacted with her. At the beginning of each interview, SB summarized the research and reasons for her interest in it.

The first 2 interviews were analyzed, and changes were made to the interview guide to capture additional information on the context of use and feedback on the timing and delivery of sessions. Data included researcher observations and postinterview notes. Data collection continued until saturation was achieved (19 interviews), as indicated by the recurrence of themes and no new themes emerging. Recordings were stored in a secure cloud-based location and transcribed verbatim by an accredited transcription service with privacy certification. During each interview, SB kept notes of points of interest and used these as prompts. Immediately after each interview, SB prepared a written summary of the interview and relevant observations. These were used to communicate interim findings to the wider research team. When appropriate, additional questions were added to the interview guide, allowing for further exploration of issues raised by participants that were relevant to the research aims. These notes were also used to guide meaningful interpretation of data during data analysis.

Descriptive statistics were computed for demographic and clinical characteristics and current app use using SPSS version 25 (IBM Corp). Data are presented as mean (SD) or number (percentage). Raw interview data were imported into NVivo 11 (QSR International) for coding and analysis. We followed 6 steps for the thematic analysis with the development of themes guided by a priori objectives identified in the aims: (1) data familiarization, (2) identifying initial codes and developing a coding framework, (3) identifying potential themes, (4) matching themes to the supporting data, (5) defining and naming themes, and (6) extracting relevant themes and producing a description of findings [30,33]. SB and GW coded the data. A constructionist approach, focusing on social conditions (user profiles and context of use) and structural conditions (app features and delivery of content), was used to interpret the data.

A total of 19 adults with T2D were interviewed (mean age 60 years, SD 14 years; 42% women). Additional participant characteristics are detailed in Table 1. Interview participants were older, more educated, had a lower baseline hemoglobin A_1c, and used the app twice as much as those in the intervention arm of the MDC trial. The mean duration of the interviews was 51 min (range 29-79 min).

A total of 2 high-level themes were constructed from the data: (1) the moderating effect of diabetes self-management styles on needs, preferences, and expectations and (2) factors influencing users’ engagement with the app: one size does not fit all. These comprised several subthemes, as described in the following sections (summarized in Textbox 1).

This theme describes variations in self-management styles and how these influenced app preferences.

This theme describes how participants engaged with the app, specifically the context, mode, frequency, and duration of interactions and the factors influencing these choices.

This qualitative study investigated users’ experiences of a T2D self-management app accessed via their own smartphones over a 9-month period in the context of their everyday lives. We identified 2 main themes: (1) the moderating effect of diabetes self-management styles on needs, preferences, and expectations and (2) factors influencing users’ engagement with the app—one size does not fit all. We found that the needs, preferences, and expectations of diabetes self-management apps differed based on participants’ self-management styles. The broad implication is that, in addition to previously identified characteristics, such as age, gender, and socioeconomic status, self-management styles also influence engagement and need to be investigated further [16,34,35].

Participants expressing self-directed rather than externally directed self-management styles were more likely to be proactive in seeking diabetes-related information and using other health apps [36,37]. A possible explanation for this finding may be found in the literature on health consciousness, defined as the extent to which an individual takes ownership of their own health condition [38]. Our data are consistent with previous evidence suggesting that individuals who are more health conscious may also be more self-directed in their information-seeking behaviors and more proactive in managing their health [39].

Our findings corroborate previous research on the benefits of personalization and tailoring while providing preliminary evidence on how app preferences can be personalized based on a specific user characteristic—diabetes self-management style [21,26,36,40]. For example, participants expressing more self-directed styles value tools that assist them in making independent, informed decisions about their own care. This suggests that to engage these users, messaging within an app needs to be presented as volitional choices rather than explicit directives and needs to acknowledge the user as an expert in their own care [41]. Additional features that are likely to improve engagement by this group include in-depth, current, accurate information on a range of topics related to diabetes care; the ability to track, link, and interpret multiple sources of diabetes-related data; and a high level of flexibility in navigating the app.

Our study also corroborates previous research that shows that people who present a more externally directed self-management style may need additional encouragement to sustain engagement. Previous research suggests that changing attitudes to and motivations for diabetes self-management may be especially important for this group [3,42]. As diabetes is a self-managed condition, successful models of care, especially for those who are not intrinsically motivated, must focus on strategies that promote and maintain autonomy [43]. Strategies to improve engagement in this group could include gamification elements such as quizzes and features that promote accountability, such as goal setting and mechanisms that re-engage users, such as regular feedback [24,32,44,45]. It may also be useful to consider giving these users more customization choices, for example, the device they prefer, because many users were more comfortable with a desktop computer or tablet than with a smartphone.

Almost invariably, the participants did not use IVR because it did not provide any additional benefit. Our findings add to existing research that suggests that features, such as IVR, although novel and interesting initially, can deter or distract from the main objective of using an app over time, especially if they do not improve usability and require additional effort [46]. The implication is that novel features should be used with caution because they can be expensive to implement and may not have the expected benefit. At the very least, users need to choose to turn off features based on personal preferences. Optimizing functionality is key because ease of use and efficiency trump novelty when apps are used in the context of ongoing, real-world self-management of a chronic condition [27,44,47].

Many participants described receiving little to no diabetes education and support following diagnosis, and in some cases, this delayed engagement in self-management [48]. Making time for and having access to adequate face-to-face education and support are often challenging for people with newly diagnosed T2D [4,6,7]. Our data support previous research demonstrating that providing diabetes education and self-management support via an app could be a feasible and acceptable complement to clinical care [8-10,14]. Equally important is the suggestion that this support may be more successful in engaging people when accessed immediately following diagnosis [37,49,50].

Our findings suggest that the proposed contact frequency and duration (ie, weekly sessions of 15-30 min) was acceptable (even for those with busy schedules) and enhanced engagement, potentially through increasing accountability and automaticity [51]. Enabling users to choose a regular time fitting into their schedule and some flexibility in altering that time to fit with competing demands encouraged engagement. However, it was clear that some limits on how the app was used were considered beneficial, even necessary, as many described how complete freedom could result in disengagement. Appointment reminders were useful, but only to those with a busy schedule because those who described themselves as less busy (eg, retired) preferred set appointment times and considered them to be part of an established routine, for which they did not need a reminder.

Finally, our data suggest that although diabetes self-management apps may be helpful in initiating and maintaining self-management behaviors, people with T2D are more likely to engage with an app when it is endorsed by their health care professional. There is some evidence to suggest that although health care professionals think apps may be useful, sourcing evidence-based, high-quality apps from the thousands available on the app stores remains a challenge [50,52,53]. Thus, initiatives are needed to provide health care professionals with reliable resources that enable them to choose quickly from a curated selection of evidence-based diabetes self-management apps while matching them with the individual’s needs.

A key strength of this study is that it was conducted in the context of a randomized controlled trial of the MDC app. In contrast with many previous trials of self-management apps, participants used the app in the wild, that is, in the context of their everyday lives via their own familiar devices, addressing some of the limitations of previous trials. Participants also had access to the app for up to 9 months, making it possible to explore their real-world use and changes over time. This was a significant strength relative to most previous research where participants only used an app once or for a short period (usually less than four weeks). The purposive interview sampling strategy was successful in recruiting participants with a range of experience, facilitating examination of the interplay between user characteristics, app preferences, and engagement. One exception is that expert app users and those expressing a more autonomous self-management style were overrepresented, perhaps because these characteristics made them more likely to want to participate in the interview study. In addition, interview participants used the app twice as much as those in the intervention arm of the MDC trial, suggesting that we were less successful at recruiting less engaged users. We recommend that future research focuses on identifying the experiences and needs of users who are less autonomous and less experienced with technology because they are likely to have different diabetes education and support needs. Finally, our sample did not include younger adults with T2D, a burgeoning cohort with clear unmet needs [54]. Further research is needed to explore the experiences of such a sample.

Our study is one of the first to investigate the use of a diabetes self-management app in the wild. Our findings suggest several ways in which user experiences can be engineered to improve engagement with T2D self-management education and support via an app, such as personalizing app features to user characteristics, recommending a potential optimal time to intervene, developing resources to assist health professionals make evidence-based recommendations for diabetes apps, and recommend potential frequency and scheduling of the intervention. Further research investigating interactions between user characteristics, including self-management autonomy and engagement, is warranted to determine specific strategies to improve engagement with T2D self-management apps if diabetes self-management apps are to become a useful complement to clinical care.