This single-arm, retrospective, real-world study evaluated the change in glycemic control and engagement in people with T2D that participated in the combined BlueStar digital health and Dexcom RT-CGM program for 3 months. As this was a feasibility study under real-world conditions, the sample size was not predetermined. Participants were recruited from 3 health care clinics in the United States and were current or former enrollees in an employer-sponsored health program. Study enrollment occurred between February 2021 and January 2022.

Eligibility criteria included being aged 18 years or older, being willing to use the BlueStar app and the Dexcom G6 RT-CGM system, and having a diagnosis of T2D treated with basal insulin, oral medications, noninsulin injectables, or lifestyle management. Key exclusion criteria included prandial insulin use, pregnancy, cancer, dialysis, or the presence of a major psychiatric disorder.

Initial enrollment included downloading the BlueStar and Dexcom G6 mobile apps (compatible with Apple and Android phones), provision of a CGM transmitter and three 10-day sensors (1 month’s supply), and optional training on CGM insertion and use. Participants were required to return to the clinic each month to obtain additional sensors. The BlueStar app is an integrated digital health platform that can sync with numerous devices, provide personalized feedback and digital coaching to its users, and help users track their medication, sleep, exercise, and other health behaviors. The Dexcom G6 mobile app allows users to view their glucose levels in real time with updates every 5 minutes. Trend arrows and access to retrospective data using Clarity reporting software (Dexcom, Inc) can help users identify short- and long-term patterns in their glucose levels.

Ethical review and full waiver of Health Insurance Portability and Accountability Act (HIPAA) authorization were obtained from Advarra Institutional Review Board (Pro00069142).

The program originally enrolled 122 participants. To be included in the analysis cohort, participants were required to have 10 contiguous calendar days of CGM readings with 70% data sufficiency within 30 days from activation on the BlueStar platform and a follow-up measurement of 10 contiguous calendar days of CGM readings with 70% data sufficiency between 80 and 110 days after activation on the BlueStar platform. After excluding participants treated with bolus insulin and applying the baseline and follow-up CGM data sufficiency criteria, the final analysis cohort was 52 participants. Average engagement per user per week was calculated for users who logged medication-taking, exercise, food, weight, sleep, and blood pressure events. To be included in the engagement analysis, a user had to have ≥1 engagement for a given event between the baseline and follow-up glycemic outcome measurement periods as noted above.

CGM metrics and engagement measures were then calculated for two population segments: (1) participants with mean baseline glucose >180 mg/dL (suboptimal control) and (2) participants with mean baseline glucose ≤180 mg/dL. The cutoff of 180 mg/dL was chosen because it is the upper bound of the International Consensus TIR guideline metric [10]. Given the small sample size, we tested the CGM data for normality using the Shapiro-Wilk test. The data were not normally distributed so a Wilcoxon signed rank test was used to evaluate 2 distributions of baseline and follow-up groups. The between-groups difference in the rates of engagement with BlueStar app features was evaluated using the Mann-Whitney U test. To test for significant differences in demographic information, we performed a 2-proportion z test for gender and medication regimen and a 2-tailed Welch t test for age, baseline GMI, and baseline mean glucose. Significance for all statistical tests was defined as P<.05.

A total of 122 participants enrolled in the study; 94 initiated RT-CGM and 52 met the data sufficiency requirements at 3 months. There were no significant differences in age or gender between those who met the data sufficiency requirements (n=52) and those who did not (n=42). The combined digital health solution/RT-CGM program is depicted in Figure 1. Baseline demographic and clinical characteristics of participants are presented in Table 1. Participants (n=52) were 37% (19/52) female with an average age of 53 (SD 9) years. In their initial sensor session, 65% (34/52) of participants had a mean glucose ≤180 mg/dL and 35% (18/52) had a mean glucose >180 mg/dL.

The mean proportion of days with sensor wear was 90% (SD 8%). Glycemic outcomes for the 3-month study are presented in Table 2. Among individuals with baseline mean glucose >180 mg/dL, significant improvements in glycemic metrics were observed. Specifically, GMI decreased by 0.8 (IQR –1.0 to 0.1) percentage points, TIR increased by 15 (IQR 1-47) percentage points, and TAR 181-250 mg/dL and >250 mg/dL decreased by 4.4 (IQR –20.2 to 2.1) and 14 (IQR –18.3 to 1.4) percentage points, respectively (all P<.05). This TIR improvement corresponds to an increase of over 3.5 hours per day in target range and is clinically meaningful [10]. Glycemic control among participants with a baseline mean glucose ≤180 mg/dL was maintained [10]. The proportion meeting the international consensus guidelines of >70% TIR [10] increased from 0% to 28% among individuals with baseline glucose >180 mg/dL.

The proportion of individuals with at least one engagement activity per week over the 3-month study period was 29% (15/52). Logging of specific engagement activities overall and stratified by baseline mean glucose is shown in Table 3. The weekly engagement rate per user who logged events is shown in Table 4. Among individuals who logged events, there was a significantly higher total rate of event logging in the cohort with baseline mean glucose ≤180 mg/dL (P=.006). Medication-taking was logged most often in both cohorts, closely followed by exercise and food logging. Between the 2 cohorts, there were significantly higher rates of medication (P<.001), exercise (P<.001), food (P=.007), and sleep (P<.001) logging among participants with baseline mean glucose ≤180 mg/dL.

In this real-world study, participation in a digital health program combined with RT-CGM use was associated with significant improvements in glycemic control in adults with T2D and suboptimal glycemic control at baseline. This included a significant decrease in mean glucose and a clinically meaningful increase in TIR of approximately 15 percentage points, corresponding to an increase of approximately 3.5 hours per day time in the target glucose range [10]. Glycemic control was maintained in the group of participants meeting glycemic targets (mean glucose ≤180 mg/dL) at baseline. There was high RT-CGM use and regular engagement with the BlueStar app. Regardless of baseline glucose control, a similar proportion of participants engaged with the BlueStar app and logged self-management events.

There are a variety of digital health programs that incorporate CGM. Some use only live lifestyle coaching; others use live coaching and telemedicine visits; some exclusively use AI-based lifestyle programs; and others use hybrid models of AI-based and live coaching, with or without telemedicine visits. Studies of these programs have reported glycemic and other outcomes; however, it is not possible to directly compare the results to those in this study due to differences in the program models for delivering diabetes educational support and coaching, live clinician management, and CGM wear patterns (continuous vs intermittent wear) [8,9,11-17]. For example, a study by Majithia et al [12] of a digital health program incorporating telemedicine visits found a 10.2 percentage point increase in TIR at 4-month follow-up among participants with suboptimally controlled T2D versus a 15 percentage point increase in this study. However, CGM was worn nearly continuously in this study versus intermittent use in the report by Majithia et al [12]. A retrospective study of people with T2D using CGM continuously as part of an intensive diabetes management program also reported a significant decrease in CGM-derived mean glucose (147.4, SD 59.1 mg/dL to 122.6, SD 33.3 mg/dL; P<.001), and significant improvement in HbA_1C, insulin resistance, and fasting blood glucose at 90 days [15]. As in our study, other studies of combined telehealth and CGM programs in people with T2D have reported greater glycemic improvements among those with a higher baseline value and maintenance of glycemic control in those meeting targets at baseline [13,14].

Beyond glycemic outcomes, participation in digital health programs incorporating CGM has been associated with other beneficial outcomes. Participants in telehealth programs that incorporated RT-CGM reported an increased understanding of their diabetes [13], lower diabetes distress [16,18], and improved quality of life [16]. In a previous study of users in a combined BlueStar and connected BGM application, emergency department visits decreased by 30% and costs decreased by 55% [11].

Engagement metrics are not commonly reported in studies on digital health or telehealth solutions. In one study, two-thirds of participants in a telehealth program incorporating RT-CGM had at least one coaching interaction, with most having greater than 4 during the 26-week follow-up period [14]. In another report, among users of a telehealth mobile app, 84% used coaching and 13% used telemedicine at least once [17]. Notably, those who used CGM had some of the largest improvements at follow-up [17].

It is noteworthy that while the proportion of participants that logged events was similar between groups, among those who logged events, the weekly rate was higher in the group with better glycemic control at baseline. This is reported in other studies where individuals with good glycemic control maintain it through the intervention [13,15]. While our study did not query participants on their engagement behaviors, we hypothesize that individuals with higher average baseline glucose primarily focused on their glucose data and related insights, whereas individuals with better glycemic control had increased availability to log other behaviors. This will be a topic of future research.

The BlueStar digital health solution is designed for high scalability in its use of AI-powered insights rather than live, lifestyle coaching with diabetes educators or telemedicine visits with clinicians. These insights from RT-CGM data can aid users in successful problem-solving, a critical ADCES7 self-care behavior [2], and potentially lead to behavioral change in other ADCES7 self-care behaviors. For example, glucose changes following exercise, a healthy meal, or taking medication as prescribed could clarify the importance of these diabetes self-management behaviors. The combination of a digital health solution and RT-CGM allows for personalized problem-solving and is likely to lead to improvements in diabetes management.

Strengths of this study include the significant improvements in participant outcomes, the use of a real-world, outpatient setting, and the sustainability of the intervention related to the scalability of the digital health solution. Limitations include the all-US population covered under one employee health plan, limited demographic information (such as education level), and the modest sample size, which could have limited statistical significance. In addition, the effect of the digital health solution or RT-CGM system individually could not be quantified in this single-arm study design. Longer studies are needed to evaluate the durability of outcomes. We also acknowledge the potential attrition bias with the use of a 70% data sufficiency requirement; however, our aim was to assess the use of the combined digital health and RT-CGM program and determine its efficacy.

This study suggests that engagement with a combination of RT-CGM and an AI-driven digital health solution helps users improve their glycemic metrics. Those with the highest baseline glycemic metrics were more likely to see significant, meaningful improvements. Behavioral change stemming from AI-derived insights and interaction with RT-CGM data likely contributes to these improvements without the need for live coaching or telehealth visits.