Developing and Evaluating the DiabetesXcel Mobile Application for Adult Patients With Type 2 Diabetes Free

Participants were recruited over 1 month from Montefiore Medical Group Diabetes Clinic sites and/or primary care sites. Participants were recruited through flyers at study sites and via provider enrollment. Participants had to be at least 18 years of age, speak English, and have a diagnosis of type 2 diabetes (defined as a formal type 2 diabetes diagnosis by a health care provider, treatment with at least one antihyperglycemic medication, or an A1C >6.5%). Participants were also required to receive diabetes care through Montefiore outpatient clinics and to have access to a smartphone with an iOS or Android operating system. Exclusion criteria included pregnancy, advanced psychiatric or cognitive conditions that would prevent study completion, and chronic illness either 1) requiring chemotherapy or steroids or 2) complicated by organ failure, including advanced liver disease, heart failure, or chronic kidney disease in stage 3 or 4 or requiring dialysis.

Individuals eligible for the study provided written informed consent, and Health Insurance Portability and Accountability Act authorization was obtained. The study was approved by the Einstein Institutional Review Board (2018-9590).

Participants downloaded the app, were taught how the app functions, and answered baseline questionnaires in the project’s RedCAP (Research Electronic Data Capture) database.

The DiabetesXcel app was developed using a visual interface adapted from the ASTHMAXcel app, a successful asthma control app previously developed at Albert Einstein College of Medicine and Montefiore through user-centered design (3,9). DiabetesXcel opens with a Start screen, then follows with 1) description and instructions, 2) a Terms of Use and Privacy Policy legal screen that must be read and acknowledged before proceeding, 3) a Table of Contents that displays, with links, each of the nine app topics, and 4) each of the chapters in sequence from one to nine (Table 1). These chapters correlate with diabetes education topics and include short, animated videos (Figure 1). The chapters were created with input from our team’s diabetes specialist physician and diabetes educators based on national diabetes guidelines. A checkbox was displayed in DiabetesXcel upon chapter completion.

Our mobile platform’s app usage and click-through data, available through the platform’s web-based administration panel, were used to evaluate user-level and aggregated trends in participant engagement. Individuals received tailored push notifications from their DiabetesXcel app in the form of weekly supportive messages and check-in messages to 1) find out how well participants adhered to the guidelines (e.g., missing medication doses, having a foot exam in the past year) and 2) indicate to participants whether they should review any topics they are having difficulty with and redirect them to the relevant videos. All study visit data were securely stored in our project’s REDCap database and the DiabetesXcel app’s secure Cloud-based server.

A cohort of 55 adult participants with type 2 diabetes were recruited from 16 November 2020 to 16 November 2021 to test the DiabetesXcel app. At baseline, sociodemographic and clinical characteristics were collected, including age, sex, race/ethnicity, highest education level, insurance type, diabetes medications received, years since diabetes diagnosis, type of diabetes complications, latest A1C and LDL cholesterol levels, number of missed work/school days in the past month, number of other comorbidities, number of diabetes-related ED visits and inpatient hospitalizations in the past 2 months, and whether participants had had a diabetes health maintenance exam in the past month. Other comorbidities obtained from the electronic health record included asthma, arthritis, heart disease, liver disease, hypertension, chronic obstructive pulmonary disease, thyroid conditions, renal disease, multiple sclerosis, and peripheral arterial disease.

At baseline, 4 months, and 6 months, quality of life with diabetes was gauged with the Diabetes Quality of Life (DQoL) questionnaire and ability to self-manage diabetes was evaluated with the Diabetes Self-Management Questionnaire (DSMQ). The DQoL was measured on a five-point Likert scale (10,11) and separated into three domains: satisfaction, impact, and worry. The DSMQ included 16 items, each scored from 0 to 3 (12). Diabetes knowledge was assessed with the Diabetes Knowledge Questionnaire-24 (DKQ-24) (13), with scores ranging from 0 to 24. Self-efficacy was assessed with the Stanford Diabetes Self-Efficacy Scale (DSES), with eight items scored from 0 to 10 (14). Depression was evaluated with the nine-item Patient Health Questionnaire (PHQ-9), with scores ranging from 0 to 27 (15). Higher scores on the DSMQ, DKQ-24, and DSES determined improved outcomes. Lower scores on the DQoL and PHQ-9 indicated improved quality of life and depression, respectively. Measurement bias was minimized by using validated questionnaires. A1C and LDL cholesterol levels, and diabetes-related ED visits and hospital admissions were also assessed at baseline, 4 months, and 6 months. Lowering of A1C and LDL cholesterol levels meant improved glycemic and lipid control, respectively.

Baseline characteristics were reported as mean ± SD for normally distributed continuous variables, median (interquartile range [IQR]) for nonnormally distributed continuous variables, and raw numbers and percentages for categorical variables. Paired t tests were used to compare outcomes, including scores on the DQoL, DSMQ, DKQ-24, DSES, and PHQ-9 at baseline and at each subsequent visit (at 4 and 6 months). The outcomes of A1C, LDL cholesterol, and diabetes-related ED visits and hospital admissions were also compared between visits with paired t tests.

Linear mixed-effect models tested the association of app usage with questionnaire scores and other outcome trends over time, while allowing for adjustment of potential confounding variables, including age, sex, insurance type, years since diabetes diagnosis, diabetes complications, and number of comorbidities. Participant effect was the random effect, and the other effects were all fixed effects. Confounders were chosen a priori based on known biological associations.

Statistical significance was set at P <0.05. Analysis was performed with SAS, v. 9.4, statistical software (SAS Institute, Cary, NC).

Of the original 55 participants, five did not complete the baseline visit and were excluded. Fifty eligible participants completed the baseline visit and were included in the study. Of these patients, 65% were follow-up visits, and 35% were new-patient visits. All patients were enrolled from the diabetes specialty clinic within the Endocrinology Department. Information was available at 4 and 6 months for 48 individuals (Figure 2).

At baseline, participants were a mean age of 47 ± 9 years. Most identified as female (70%), roughly half identified as Black or African American (48%), one-third were high school graduates (30%), and approximately half had private insurance (46%), whereas 40% had Medicaid. The median duration of time since diabetes diagnosis was 12 years (IQR 6–20 years), and although many (40%) did not have diabetes complications, many (22%) did report neuropathy. More than half of the participants (56%) used insulin. At baseline, the median latest A1C was 7.8% (IQR 7.0–9.9%), and most participants (58%) had uncontrolled diabetes, which was defined as an A1C >7.5%. Most individuals (88%) had had a diabetes eye exam, and 52% had had a foot maintenance exam in the past year. The majority of individuals (92%) did not have a diabetes-related ED visit or a diabetes-related hospital admission (94%) in the past 2 months from baseline, but most had more than one comorbidity at baseline (52%). The most common comorbidities were asthma (26%), hypertension (16%), and heart disease (12%) (Table 2).

There was significantly improved quality of life from baseline to 4 months (score 50.06 vs. 37.08, P <0.0001), from 4 to 6 months (score 37.08 vs. 28.52, P <0.0001), and from baseline to 6 months (score 50.06 vs. 28.52, P <0.0001) (Table 3). When the DQoL was separated by domain, there was a significant improvement from baseline to 4 months in satisfaction (score 23.72 vs. 16.63, P <0.0001), impact (score 15.78 vs. 12.94, P <0.0001), and worry (score 10.56 vs. 7.52, P <0.0001); from 4 to 6 months in satisfaction (score 16.63 vs. 12.35, P <0.0001), impact (score 12.94 vs. 11.27, P <0.0001), and worry (score 7.52 vs. 4.90, P <0.0001); and from baseline to 6 months in satisfaction (score 23.72 vs. 12.35, P <0.0001), impact (score 15.78 vs. 11.27, P <0.0001), and worry (score 10.56 vs. 4.90, P <0.0001). Self-management was significantly improved from baseline to 4 months (score 21.06 vs. 23.35, P = 0.001), from 4 to 6 months (score 23.35 vs. 24.88, P = 0.02), and from baseline to 6 months (score 21.06 vs. 24.88, P <0.0001). Self-management was compared for those with and without diabetes complications. There was no significant difference in self-management for those with and without complications (score 21.27 vs. 22.65, P = 0.54).

In adjusted linear mixed-effect models, quality of life and self-management were significantly better from baseline to 4 months (β = −13.24, P <0.0001, and β = 2.25, P = 0.0007, respectively) and from baseline to 6 months (β = −21.61, P <0.0001, and β = 3.70, P <0.0001, respectively) (Table 3). Similarly, the DQoL score domain categories showed a significant improvement from baseline to 4 months (satisfaction β = −7.19, P <0.0001; impact β = −2.93, P <0.0001; and worry β = −3.03, P <0.0001) and from baseline to 6 months (satisfaction β = −11.52, P <0.0001; impact β = −4.46, P <0.0001; and worry β = −5.54, P <0.0001). No other covariates in these models were significant.

There was a significant improvement in knowledge from baseline to 4 months (score 16.94 vs. 20.06, P <0.0001), from 4 to 6 months (score 20.06 vs. 21.48, P <0.0001), and from baseline to 6 months (score 16.94 vs. 21.48, P <0.0001). Self-efficacy was significantly improved from baseline to 4 months (score 51.28 vs. 62.21, P <0.0001), from 4 to 6 months (score 62.21 vs. 67.83, P <0.0001), and from baseline to 6 months (score 51.28 vs. 67.83, P <0.0001). Participants reported significantly lower depression from baseline to 4 months (score 4.54 vs. 2.50, P <0.0001), from 4 to 6 months (score 2.50 vs. 2.15, P = 0.03), and from baseline to 6 months (score 4.54 vs. 2.15, P <0.0001). There were significantly better A1C levels from baseline to 4 months (8.64 vs. 7.74%, P <0.0001), from 4 to 6 months (7.74 vs. 7.56%, P = 0.01), and from baseline to 6 months (8.64 vs. 7.56%, P <0.0001). There was also a significant improvement in LDL cholesterol levels from baseline to 4 months (124.50 vs. 107.71 mg/dL, P <0.0001), from 4 to 6 months (107.71 vs. 96.15 mg/dL, P <0.0001), and from baseline to 6 months (124.50 vs. 96.15 mg/dL, P <0.0001). There were significantly fewer diabetes-related ED visits from baseline to 6 months (0.08 vs. 0, P = 0.04) but not from baseline to 4 months (0.08 vs. 0.02, P = 0.18) or from 4 to 6 months (0.02 vs. 0, P = 0.32). There were significantly fewer diabetes-related hospital admissions from baseline to 6 months (0.06 vs. 0, P = 0.03). No participants had an admission at 4 or 6 months (Table 4). There was no significant difference for those with and without complications in knowledge (score 18.33 vs. 18.55, P = 0.81).

In adjusted linear mixed-effect models, the models with outcomes of knowledge, self-efficacy, depression, A1C, LDL cholesterol, and diabetes-related hospital admissions all had significant improvement between baseline and 4 months and between baseline and 6 months, and the model of diabetes-related ED visits was showed a significant difference between baseline and 6 months (Table 4). The presence of diabetes complications was a significant predictor in the models for A1C (β = −1.24, P = 0.02), self-efficacy (β = 6.71, P = 0.02), and depression (β = −2.70, P = 0.03), and years since diabetes diagnosis was a significant covariate in the models for DKQ-24 (β = 0.10, P = 0.02), self-efficacy (β = 0.33, P = 0.04), and depression (β = −0.15, P = 0.045).

In this study, participants with type 2 diabetes in Bronx, NY, used the DiabetesXcel app over 6 months (1). After only 4 months of app use, participants had improved scores on questionnaires evaluating diabetes quality of life, self-management, knowledge, and depression and also had better A1C and LDL cholesterol levels. After 6 months, participants had fewer diabetes-related admissions and ED visits.

To our knowledge, this is the only app of its kind incorporating educational videos for type 2 diabetes within the United States. DiabetesXcel is also one of the few tools of its kind to be piloted in a non-White majority, urban population with prevalent health disparities within the United States. This is important in that our population, a high-risk group greatly affected by diabetes and other chronic conditions, may reap great benefit from having a mobile app modality of diabetes management external to a formal health care setting. Although our participants were rather well connected to health care, having a mobile app like DiabetesXcel for home use could potentially be an effective management tool for those with difficulty accessing health care, such as individuals without insurance, because only a smartphone is needed to use the app. Throughout the coronavirus disease 2019 pandemic, having reliable methods of care outside of a physical health care setting was necessary and beneficial (16–18). DiabetesXcel could be helpful for continuing diabetes care in future pandemic waves.

Our cohort was representative of the typical population in Bronx, NY. We had fewer people identifying as Hispanic than in the general Bronx population (8), but most of our sample included non-White individuals, similar to the Bronx population overall (8). This diversity is important to recognize because there are racial disparities in type 2 diabetes prevalence (19) and care (20). We also had fewer people in our cohort receiving Medicaid compared with the general population, with most of our cohort using private insurance (21). Compared with the 9% of individuals <65 years of age in the Bronx population who are without health insurance, only 2% of our cohort did not have health insurance, signaling that our population may not be as representative of insurance disparities and the impact of insurance on health care access (8).

Type 2 diabetes mobile apps offering multiple features from education to glucose tracking have been tried in various studies. Our app is part of a growing group of mobile apps that go beyond just the app itself and can incorporate a variety of other integrative features. A few trialed educational apps did not show a significant change in quality of life (22), self-efficacy (22), A1C levels (22–24), LDL cholesterol levels (25,26), depression (27), or health care utilization (22). Some studies reduced A1C over a 12-week period, but the difference attenuated by 26 weeks (28). However, multiple studies have shown that type 2 diabetes apps with educational modules were effective in improving quality of life (24), self-efficacy (23,29–31), and A1C (25–27,30–36). We reported a similar improvement in these metrics with the DiabetesXcel app.

The change in A1C levels of this study was a decrease of slightly more than 1%. Prior studies have shown that a 0.9-point reduction in A1C is related to a 25% lower risk of microvascular complications of diabetes (37). Because oral antidiabetic medications also reduce A1C anywhere from 0.5 to 1.25%, there is a question of whether a similar reduction in A1C with DiabetesXcel could indicate the app’s use as a “digital therapeutic” (38). A mobile app yielding an improvement in A1C similar to that of medication, without the side effects and cost of medications, could be an incredibly powerful diabetes management tool.

Although apps for type 2 diabetes can be beneficial, they are potentially underutilized. One study showed that only 8% of Australian participants with type 2 diabetes used diabetes management apps because of the perception that apps would not be helpful in type 2 diabetes management (39). Our study demonstrates that apps can be useful in supporting this condition.

This study has multiple strengths, including completion of the baseline, 4-month, and 6-month visits by most participants (87.27%). We also were able to assess multiple outcomes over 6 months. This study was located in a region with poor type 2 diabetes outcomes and shows the advantages of the DiabetesXcel app in a high-risk and diverse population.

However, there are a few limitations to our study. Information from questionnaires was self-reported and could have been affected by recall and social desirability bias. This study did not include a control group, so we were not able to determine how outcomes associated with DiabetesXcel compared with those in people who did not use the app. However, this could be addressed in a future randomized controlled trial. Still, we saw an improvement in all outcomes with whatever frequency and duration participants used the app. We also required individuals to be present for multiple visits and were not able to determine the effectiveness of the app for those who access health care services less frequently. As with any mobile app, the use of DiabetesXcel is limited by access to and ability to use a smartphone or device with similar capabilities. Currently, this app is available in English only and therefore is inaccessible for those who are not proficient in the English language. We also did not include qualitative analysis of narrative comments regarding the app.

Overall, our study found that using the DiabetesXcel app for only 4 months significantly improved diabetes outcomes in a region with a diverse population, high diabetes rates, and poor diabetes health outcomes. Our findings support the growing level of evidence that type 2 diabetes mobile apps are effective interventions for individuals with type 2 diabetes, improving well-being and health outcomes, particularly in regions with limited medical resources and for individuals with difficulty accessing health care.

This study was supported by National Institutes of Health/National Center for Advancing Translational Science Einstein-Montefiore Clinical and Translational Science Award grant number UL1 TR002556. This work was also funded by a National Institute of Diabetes and Digestive and Kidney Diseases P30 grant (P30DK111022).

L.P.M. is a medical advisor for SindyXr, a virtual reality–based fitness app. No other potential conflicts of interest relevant to this article were reported.

A.A.B. and P.S.C. wrote the manuscript and assisted in data analysis. S.K. and L.P.M. participated in data collection and analysis and manuscript preparation. W.M. and L.G. performed the data analysis and manuscript preparation. J.S.G. and J.W.-R. participated in project conceptualization and data analysis. S.P.J. participated in project conceptualization, data collection and analysis, and manuscript preparation. S.P.J. is the guarantor of this work and, as such, had full access to all the data presented and takes responsibility for the integrity of the data and the accuracy of the review.