App Design Features Important for Diabetes Self-management as Determined by the Self-Determination Theory on Motivation: Content Analysis of Survey Responses From Adults Requiring Insulin Therapy

Introduction

Background

Achieving treatment goals for patients with diabetes requires sustained behavioral lifestyle changes such as meal planning, monitoring carbohydrate (carb) intake and blood glucose (BG), and exercising. Diabetes apps can function as electronic care plans by helping patients plan and incorporate healthy behaviors into their daily routines [1]. The apps have been shown to lead to the improvement of glycemic control, with hemoglobin A_1c (a blood test measuring average BG over the past 3 months) reduction typically in the range of 0.4% to 1.9% [2-7]. The most common app functions include the documentation of BG reading, diet, and medication use; BG analysis report; data export; and email capability [8]. Visual displays of BG readings help patients link this data to their behaviors, thus facilitating behavior changes to improve glycemic control [9]. Systematic reviews have found that the effectiveness of the apps increased with greater interactivity [10,11]. Interactive feedback could be an automated message from an app algorithm [5] (eg, “you have met your BG goal setting five times this week”), a text message from a dietician who reviewed data and customized a meal plan, [3] or an alert message whenever a BG reading is out of range compared to the goal [3,4,8,12].

Despite more than 1100 apps available on the market, their adoption and use vary, possibly due to design issues [13,14] and variations in technology development [15]. To date, only a few rigorous evaluation studies of app designs have involved patients [16], and most have evaluated the quality of all available apps in the market without involving end users such as patients and clinicians [17,18]. A recent systematic review showed that patient adoption of diabetes apps weighs heavily on patient perception of benefits, ease of use, and clinician recommendation to use diabetes apps [19]. Thus, the Agency for Healthcare Research and Quality stressed the need to understand the patient perspective on the use of diabetes apps [20]. Our research question focused on adults with type 1 or 2 diabetes on insulin therapy: What diabetes app functions are helpful as explained by a theory on motivation, called the Self-Determination Theory (SDT), to promote self-management behaviors? The purpose of this study, therefore, was to describe how patients with diabetes perceive diabetes apps to address the 3 psychological needs of competence, autonomy, and connectivity as described by the SDT [21]. Our analysis also allowed us to provide evidence that would refine this theory on motivation as it applies to the use of mobile apps in the population with diabetes requiring insulin.

Theoretical Framework

Motivation is an important factor in user experience with technology [22,23]. The SDT [21] on motivation, as expanded by Szalma [24] for motivational design on effective human-technology interaction, guided this study. The SDT posits that people are driven to engage in behaviors because they believe those behaviors will personally benefit them [25]. According to the theory, humans have 3 basic psychological needs that influence behaviors [21]. Competence is the need to master tasks and learn skills [26]. Autonomy is the need to feel in control of one’s behaviors and goals [27]. Relatedness or connectivity is the need to feel attached to other persons [26,28]. The SDT has been used in educational, business, and health care settings [29-31]. It is used to explain the human-technology interaction [24]. Ryan et al [32] reported that the ease of technology use directly and positively affected the satisfaction of psychological needs. This theory thus provides the basis for this study as we organized participant responses according to the 3 psychological needs outlined in the theory.

Methods

Design

This study was part of a crossover randomized laboratory trial [33] to test 2 top-rated, free commercial apps (OnTrack and mySugr), identified as the “the Best Diabetes Apps 2016” by Healthline [34]. The within-subject design helped control for patient characteristics because the same individual tested the 2 apps in random order. Quantitative measures of these diabetes apps’ usability, including user satisfaction, time, success, and accuracy rates, have been reported elsewhere [33]. The data for the analysis presented here include field notes of observations during app use, audio recordings taken during the tests, and participant responses to an electronic survey with open-ended questions that queried what app functions patients perceived as being the most useful and most important in supporting diabetes self-management.

Ethics Approval

This study was approved by the University of Minnesota Institutional Review Board (MOD00001221).

Participants

Using a flier posted on a bulletin board or on the web, 92 participants were recruited from the following venues: Facebook (n=46); participant referrals (n=8); Federally Qualified Health Center clinic (n=7); university campus (n=6); public housing (n=6); Craigslist (n=5); veteran’s clinic (n=4); diabetes support groups (n=3); and miscellaneous sites from a state fair, church, and library (n=7). Inclusion criteria were (1) aged ≥18 years; (2) having type 1 or type 2 diabetes; (3) having used an Android phone for 6 months or longer; (4) having used insulin therapy for 6 months or longer; (5) adequate English proficiency; and (6) smartphone proficiency (ie, they used the device for more than phone calls, emails, texting, or taking pictures). Exclusion criteria were (1) inability to read or speak English and (2) prior use of the OnTrack or mySugr app or use of any diabetes app in the past 6 months. Individuals were screened for eligibility on the phone, and written informed consent was obtained prior to the start of each study session.

Procedures

From July to November 2017, we conducted 92 sessions of in-person tests of the apps that lasted an average of 1 hour. The testing took place in a private meeting room inside a public library or building. Participants viewed a YouTube training video posted by each app developer. They then practiced using the apps by the following protocol: (1) enter a carb intake; (2) enter an exercise activity; (3) enter an insulin dose; (4) enter a BG reading; (5) locate a BG report for specific days of the week; (6) locate a BG report for each meal; and (7) email a BG report. Then, each participant tested the 2 apps in a randomized order to carry out the same tasks listed in the practice protocol. Each participant received a US $50 gift card upon study completion.

Data Collection

The first author (HF) kept field notes detailing her observations of participant reactions during the test of the apps and audio recorded the tests. The field notes and audio recordings were transcribed verbatim in a Microsoft Word file by a research assistant. The survey was administered on an iPad (Apple Inc.) and included questions on demographic characteristics, technology use, and diabetes history. In addition, based on the SDT [21], the survey also included questions about motivation for self-management and psychological needs for competence, autonomy, and connectivity. Details of these measures are reported in prior publication [33]. To explore participant responses to the app, the survey queried participants about their perceptions of app usability and satisfaction, preferences for a “dream” app and indications of what function(s) would be the most useful, and identification of the most important functions in a diabetes app.

Data Analysis

Field notes, audio recordings, and survey responses were analyzed based on key constructs from the SDT [21]. The analytic team, consisting of 4 members (HF, JFW, CJP-M, and TJA), analyzed the transcripts with the aid of Dedoose [35], a web-based, qualitative data analysis software. Directed content analysis, as described by Hsieh and Shannon [36], was used. With this approach, an existing theoretical framework (SDT) was used to organize data according to predetermined categories that are aligned with key constructs in the theory: competence, connectivity, and autonomy. Data that failed to contribute to the categories were coded and used to suggest modifications or extensions of the theory. A codebook was developed based on the initial reading and updated with independent coding from an analysis team. The team reached consensus on the code definition that were clear and mutually exclusive (see Table 1 for conceptual and operation definitions for codes used).

Competence was conceptually defined as app features to help patients gain skills to keep BG in the target range [24]. Competence was operationalized as app functions to help patients understand the meaning of their data. This refers to how the app records data, analyzes data, and provides reports on which numbers are not in the target range and why. Autonomy was conceptually defined as app features that help patients set safe goals on diet, insulin dose, or activity level based on personal trends of BG and carb intake [24]. Autonomy was operationalized as app data visualization to help patients identify abnormal highs or lows, which are important for setting up reasonable targets to change behaviors associated with those abnormal readings. Connectivity was conceptually defined as app features to facilitate interactions between persons and the technology involved, which means enabling the sharing of home-monitored data and communicating with clinicians [24]. Connectivity was operationalized as app print report options, exports of data and analysis reports, and reports sent to clinicians or others through email.

Analysis occurred in several steps consistent with content analysis procedures as described by Miles et al [37]. First, based on the SDT [21], the team reviewed the conceptual definitions of the 3 main categories (eg, competence, autonomy, and connectivity) and, through discussion and consensus, developed operational definitions of each that were clear and mutually exclusive. See Table 1 for the conceptual and operational definitions of each of the categories. Second, a codebook was developed that outlined rules for coding data to each of the categories. The codebook was refined through several iterations of coding. Third, a table was developed that included each of the 3 categories as column headings and a column heading labeled “other” for codes that did not align with any of the categories. Data from each participant were placed on a row that was identified with the participant’s ID number. Fourth, all data were read by all team members and divided into text units (eg, coherent phrases or sentences relevant to the study purpose). The text units were coded with a label that captured the essence and, based on the coding rules, placed in the appropriate cells on the table. Fifth, the analytic team met to gather similar codes from each column into subcategories through a process of discussion and consensus. The subcategories in the 3 main columns (ie, competence, autonomy, connectivity) were described.

The team used several procedures to enhance the trustworthiness of the study findings based on criteria outlined by Lincoln and Guba [38]. First, participants were carefully chosen based on comprehensive inclusion criteria that ensured they had sufficient backgrounds to fully engage with the app testing. Second, expert consensus was achieved with a 4-member research team experienced in diabetes self-management, the SDT [21], and app use, working together to reach consensus in the interpretation and grounding of the theory of the SDT. Third, transferability was enhanced with detailed descriptions of the study population and context. Fourth, auditability was ensured with a detailed audit trial maintained in the Dedoose software chronicling all analytic decisions of the study. Finally, research bias was addressed through frequent team discussions that encouraged researcher reflexivity.

Results

Sample Characteristics

In all, 92 persons participated in the study. Their mean age was 54 (range 19-79) years. The majority were female (54/92, 59%), White (57/92, 62%), and college educated (61/92, 66%; Table 2).

Most (64/92, 70%) participants had type 2 diabetes and had used insulin for an average of 12 (SD 12) years. The participants reported a wide variety of diabetes complications including short-term memory loss; retinopathy; mobility impairment with the use of a cane, walker, or wheelchair; hemiparesis related to stroke; hand tremor; and peripheral neuropathy affecting hand dexterity. The majority (57/92, 62%) were comfortable or very comfortable using a smartphone. Additionally, 60 participants reported whether they were working (n=35) or not working (n=25)—student (n=3), retired (n=13), homeless (n=2), and disabled (n=7). Participants reported the most important app functions related to promoting competence as described by the SDT; on the other hand, what they reported as dream app functions were general app designs unrelated to the SDT (Figure 1). Of the 436 text units that were highlighted, 292 (67%) were coded to 1 of the 3 categories of needs based on the SDT [21]: competence (n=212, 48.6%), autonomy (n=47, 10.8%), and connectivity (n=33, 7.6%). The remaining 144 (33%) text units were not aligned with any of the 3 categories. The categories are discussed below.

‎

Competence

Participants found that the apps could improve their sense of competence by helping them monitor data (ranked 1st), create analysis reports (ranked 2nd), gain knowledge about reasons for out-of-range BG (ranked 4th), and plan behavior changes in self-management activities (ranked 5th), including counting carbs linked to a food library (ranked 7th; see Table 3).

Some appreciated receiving information that guided them in adjusting their insulin doses. One participant stated, “It helps me know my high and low blood sugar reading so I can adjust insulin dose. If it is real high in the morning, then at night I take more insulin. Now I do trial and error. My way is not the best.” Participants liked the automatic carb counting function. One said, “[You] take a picture [and let it] analyze for you and tell you how many carbs and everything it is.”

Autonomy

Participants found that the apps improved their sense of autonomy. They felt more self-sufficient because the apps showed if their BG was trending high or low in relation to the time (ranked 9th) and in relation to the day of the week (ranked 10th). Being provided with a data visualization of these personal patterns increased their sense of empowerment and assisted them in identifying short- and long-term goals for changing behaviors. One participant explained, “a function that easily helps me find when I most commonly have hypoglycemia.” Information provided by the apps aided their decision-making regarding how and when to change behaviors to keep BG in the target range. This could be done with data visualization; one participant stated the benefit to see “how my trends are changing.”

Connectivity

Participants found the apps enhanced a sense of connectivity because the clinicians could receive emails or print reports on home-monitored data to better understand patients’ self-management behaviors (ranked 6th). One participant said, “An app that can send my numbers directly to [the doctor] if there is a concern [about frequent] lows or highs.” Participants also felt connected because of the bidirectional messaging functions of the apps. These functions supported monitoring of BG, and readings could be compared to hemoglobin A_1c laboratory readings in the clinic. Connectivity was also enhanced by informal coaching support from others (ranked 14th). One patient stated, “help people share what other people not understanding. (1) report, (2) sharing - support for other patients with diabetes.”

Top General App Design

Most participants reported the necessity for a diabetes app to save time regardless of functions. They described that the app needs to be efficient and “easy,” requiring minimal user effort. They desired the app to use patient-friendly terminology and display easy-to-understand reports (ranked 3rd; see Table 4).

Automatic features (ranked 5th, same as to plan behavior change) is the integration between devices so that their data are interoperable. One participant explained, “Have this app be able to read my pump and. An app I reason I don’t use app, having an orange and apple that they don’t talk to each other. An app that easy and talk to my pump.” Voice recognition (ranked 8th) is the elimination of typing text, which was best described by one participant: “speaking function to record all data.” App alerts (ranked 9th) are helpful to remind users to do activities such as retest BG and repeat insulin for elevated BG after eating a meal. App interactivity (ranked 11th) is giving behavior confirmation as one participant explained: “You did it, completed 1 entry.” Other app designs (ranked from 12th to 15th)—color; cost; icon, emoji, or button options; tutorial or self-help; and fun, technical support, and link to pharmacy—were of interests to participants.

Discussion

Principal Findings

The aim of the research question and purpose of the study was to investigate how adults with diabetes requiring insulin therapy perceive diabetes apps based on the 3 key psychological needs described by the SDT [21]: competence, autonomy, and connectivity. Our findings provide evidence on the usefulness of the SDT in mobile health technology and describe specific app functions that address psychological needs. The results are consistent with Szalma’s [24] description of a theoretical model of motivational design based on the extension of the SDT. Newly identified categories about general app design did not fit with the SDT’s psychological needs, but they addressed the technology needs for patients to use an app with minimal effort.

Competence

App functions help patients to record and understand data and plan behaviors as skill to keep BG in the target range. First, the convenience to track electronically whether BG is in the target range (80-130 mg/dL before eating and <180 mg/dL after eating) [39] is highly valued [40]. This is consistent with patient surveys that found diabetes apps are important for BG monitoring [41]. Understandable “Glucose Diary View” is the most practical [42]. Abnormal BG readings should be color-coded [39] and summarized into a 1-page standardized report [43]. An electronic report can increase patient knowledge to plan behavior changes such as eating right (making it easier to count carbs and plan meals) and calculating short-acting insulin dose to lower elevated BG readings due to excessive carb intake. These features are all valuable to patients because they help them to gain insight and understanding about abnormal BG readings so they can achieve competence in diabetes care, which is consistent with a study on the requirements of diabetes apps for underserved patients [44].

Carb counting is a commonly desired app function, where a smartphone takes a picture of the food; analyzes the portion size, carb content, and corresponding insulin dose; and suggests a time for insulin administration. This finding broadly supports app use to improve adherence of medical nutrition therapy [2-4,45]. Currently, many diabetes apps have low-carb diet recipes, multidevice integration, and automatic features, but the cost can be expensive. For example, Glucose Buddy Premium has a subscription cost ranging from US $19.99 to US $59.99 per month to access the full food database [46,47]. Future research should be undertaken to investigate ways to offset the cost of app technology such as subsidizing the expense while the health system could bill insurance for remote patient monitoring, given that the Centers for Medicare and Medicaid Services can reimburse the transmission of home-monitored data and summary report by clinic staff [48]. Offering analysis tool to count carbs and calculate insulin dose is a form of “virtual dietician.” Research is in progress to develop and test apps that leverage machine learning to perform image recognition and automate recommendations of behavior change [49].

Autonomy

App functions of trend analysis help set safe and practical short- and long-term goals by time, day of the week, and month, which aids personalizing options to change. Participants reported the need to visualize the trends or patterns of frequent high or low BG (ie, what) by day of the week and time (ie, when). This finding is consistent with prior research showing that diabetes apps helped patients identify and incorporate healthy behaviors into their daily routine [1]. Seeing demarcations of BG changes between months, weeks, days, and time of the day is very important to show patients when dangerous BG levels occur and to set reasonable goals to change behaviors [50]. Goal or target setting helps patients plan behaviors and provides a warning when they are outside the target [51,52]. Personalizing options should include tracking mental health factors such as mood, stress, and illness, because these factors are associated with hyperglycemia and poor glycemic control. Effective self-management is important economically, since many adults diagnosed with diabetes are not able to maintain work. They exit the work force earlier (30% higher) compared to those without diabetes [53].

Connectivity

App functions can facilitate supportive interaction by sharing data or app reports with clinicians and “loved ones” to gain support for behavior change. This is consistent with several studies that showed data sharing or showing data from the mobile devices with their clinicians during a medical visit is highly valuable for patients [50,54,55]. Greater app interactivity with a clinician appears to improve glycemic control [11,56]. A simple explanation for this finding may be that successful diabetes self-management takes teamwork [54,55]. Informal coaching support by other people or even a virtual coach in an app is valuable. Artificial intelligence could provide confirmation of positive behavior change, such as reaching a BG value in the target range, to provide immediate feedback to patients. A trial of an artificial intelligence virtual coach with 187 adults with type 2 diabetes, unfortunately, did not demonstrate a difference in changing hemoglobin A_1c but did improve health-related quality of life [57]. Very few long-term studies of diabetes apps have been conducted [58]. However, due to the COVID-19 pandemic, telehealth visits had an unprecedented increase in use from 0.3% in 2019 to 29.1% in 2020 among a 2019 cohort (n=1,357,029) versus a 2020 cohort (n=1,364,522) [59]. Leading companies in web-based diabetes care—Livongo, One Drop, mySugr, Cecelia Health, Steady Health, and Virta Health—noted a rise in subscribers during the pandemic [60]. Future studies using the mobile health platform for telehealth, including a diabetes app, should be undertaken.

Top General App Functions or Features

Themes unsupportive of the SDT emerged that focused on the acceptability of general app design features. These themes did not support the SDT, but they described patients’ technology needs. The theme of being patient-friendly is highly relevant for user-centered app design. A patient-friendly app implies a match between the app and the patient’s real world [61,62], and icons and wording need to speak the users’ languages and concepts. For example, “blood sugar” is preferred to “blood glucose.” Eliminating medical jargon would decrease barriers and make it easy for patients to understand knowledge gained from using apps [50]. Automatic features to integrate devices that test BG and upload results into apps ranked in the top 5, which is consistent with a survey study among patients with type 1 diabetes, 91.6% of whom agreed that it is the most important function (n=167) [51]. Voice recognition decreases the user’s need to type data. Alert notifications can remind patients who are on multiple insulin injections and need frequent BG testing (>4 times a day). Patients desired app alerts to remind them of behavior (eg, repeat BG testing) [63]. An interactive app is about giving the patient a response to promote user interaction, not just data in and data out. A change in emoji, an app message of “good job,” or a sound are ways of interaction between the user and the technology. Color can help customize user experience. An app tutorial or technology support is an important resource to increase user confidence to interact with the app. Overall, these themes around acceptable design features are important for patient engagement to promote a positive user experience and boost patient confidence to use the technology.

Limitations

Three major limitations in this study were (1) the laboratory setting, (2) only 2 top-rated, commercially free apps being tested, and (3) the urban population. The first weakness is that participants only used the apps once in a research visit rather than in their home setting with real data. It is possible that using the apps in the home setting would have changed participants’ opinions about the desired app features. Future work is required to establish the viability of actual app use at home and in other settings (eg, use an app for 2 weeks and attend focus groups to discuss the facilitators and barriers of app use). A second weakness is testing only 2 top-rated free apps, which may not be representative of the diabetes apps on the market. However, mySugr has remained in Healthline’s 2022 list of best diabetes apps [45], and OnTrack has been recommended by educators from the American Diabetes Association [46] and the University of Michigan [47]. Apps requiring payment were not included in this study. Payment for increased functionality may increase patient engagement and potentially create bias to use the app to get a return on the investment [64]. A third weakness is that the results may not be applicable to a rural population who may have no or inadequate internet service. App responsiveness may depend on the type of internet connection. Notwithstanding these limitations, this study offers valuable insight to addressing behavior needs for self-management by adults with diabetes requiring insulin therapy. Several strengths of this study include the diverse sample of racial or ethnic minority participants and a variety of diabetes complications, which increase study generalizability. Additionally, this study had a sample of 92 participants, which is much larger than most usability study sample of 30 participants.

Conclusions

The SDT helped to explain patient perspectives on the roles of diabetes apps as an electronic tool to address their psychological needs of competence, autonomy, and connectivity in diabetes care. Our findings also validated that the 3 concepts of the SDT guided the initial coding, further analysis, and development of operational definitions. Using an app can promote competence in keeping BG in the target range through electronic monitoring of BG, creating analysis reports, and gaining knowledge about reasons for out-of-range BG to plan behavior. The app can promote autonomy to set safe and practical BG goals by showing trends of high and low readings in relation to time, day of the week, and months. An app can promote connectivity by printing reports for clinic visits or emailing reports to a clinician, thereby helping patients receive feedback from clinicians. Patient technology needs, such as being patient-friendly and requiring minimal user effort, are also important. Continued efforts are needed to understand long-term adoption of diabetes apps to support self-management by patients, as well as the integration of diabetes apps in the telehealth setting for clinicians.