The FGM data-based Remote IntervEntion for adults with insuliN-dependent Diabetes (FRIEND) trial was a 12-week, investigator-initiated, open-label, parallel, randomized controlled study conducted at the CHA Bundang Medical Center in Korea. The study protocol ( Supplementary Figure 1 ) was approved by the Institutional Review Board of CHA Bundang Medical Center (no. 2021-03-032) and performed in accordance with the Declaration of Helsinki and the Good Clinical Practice Guidelines of the International Council for Harmonization. All the participants provided written informed consent before any trial-related activity. The study was registered at ClinicalTrial.gov (trial number, NCT04936633).

Eligible subjects were adults with type 1 diabetes aged 19–75 years who had been on intensive treatment with MDI or CSII therapy for more than one year, had a HbA1c level of 7.0% or higher, and a desire to use the FGM system. The exclusion criteria were non-insulin-dependent diabetes, diabetes duration <1 year, a history of using a rtCGM or FGM within the previous 12 weeks, pregnancy, end-stage renal disease and on dialysis, current treatment for severe cognitive impairment or psychiatric problems, a history of substance abuse or alcoholism within the previous 12 weeks, a history of corticosteroid therapy for more than seven consecutive days within the previous four weeks, and participation in other clinical trials within the previous four weeks. The flowchart of study participants is shown in Supplementary Figure 2 .

The participants were randomly assigned in a 1:1 ratio to either the intervention group or the control group. The intervention group used FGM for 12 weeks with remote intervention by medical staff and the control group used FGM without intervention. The participants were stratified at randomization according to their baseline HbA1c level (<9.0% or ≥9.0%) and age (<47 or ≥47 years).

All participants were provided with a FGM system (FreeStyle Libre; Abbott Diabetes Care, Witney, UK) with basic instructions on how to use it. In the intervention group, the remote intervention was conducted over a phone call at 2-week intervals for a total of five times during the study period if one or more of the following criteria were met in the previous two weeks; i.e., active time of sensor <70%, the number of scans per day <4, time with glucose in the range of 70–180 mg/dL <70%, time with glucose below 70 mg/dL ≥4%, time with glucose below 54 mg/dL ≥1%, time with glucose above 180 mg/dL ≥25%, time with glucose above 250 mg/dL ≥5%, coefficient of variation ≥33%, and mean glucose level ≥140 mg/dL.

The remote intervention lasted about 10 minutes and was based on CGM data in LibreView. Its contents were as follows: education on the insulin to carbohydrate ratio and the insulin sensitivity factor; carbohydrate counting training; insulin management training including basal dose adjustment, bolus dose titration based on meal content and current glucose level, and use of a sliding scale; identifying the causes of hypoglycemia, hyperglycemia, or glycemic variability; advice on lifestyle modifications such as diet and exercise; and how to use FGM system including using glucose trend arrows. When the sensor activation time was less than 70% during the previous two weeks, both groups received phone calls or text messages encouraging FGM use.

At baseline and week 12, blood samples were taken from all participants to determine HbA1c levels and questionnaires were completed on following characteristics: treatment satisfaction and perception of hyperglycemia or hypoglycemia (Diabetes Treatment Satisfaction Questionnaire [status version and change version, DTSQs and DTSQc], Korean ver. 8.3.06, licence ref CB1202) (13), depression (Patient Health Questionnaire-9, PHQ-9) (14), and anxiety (General Anxiety Disorder-7, GAD-7) (15). The DTSQ questionnaire consists of eight questions with six querying treatment satisfaction, one querying perceived frequency of hyperglycemia, and one querying perceived frequency of hypoglycemia. Higher scores on six items asking about treatment satisfaction indicate greater satisfaction with treatment. Lower scores on two items asking about the perceived frequency of hyperglycemia and hypoglycemia indicate that blood glucose levels were closer to the ideal, while higher scores indicate problems. The PHQ-9 and GAD-7 questionnaires consist of 9 and 7 questions, respectively, with higher scores indicating severe depression or anxiety.

The primary outcome of the study was changes in HbA1c levels from baseline to week 12. Secondary glycemic outcomes included changes in CGM metrics, such as time with glucose in a range of 70–180 mg/dL (TIR), time with hypoglycemia (<54 and <70 mg/dL), time with hyperglycemia (>180 and >250 mg/dL), mean glucose level, and coefficient of variation (CV). CV (%) was calculated using the following formula: dividing the standard deviation (SD) of glucose levels by the mean glucose level and multiplying by 100. The CGM metrics data of the first two weeks and two weeks before week 12 were compared. In addition, changes in the psychosocial, behavioral, and physical variables were assessed as outcomes; i.e., scores of the DTSQs, DTSQc, PHQ-9, and GAD-7 questionnaires, total daily doses of insulin, the number of scans per day, lifestyle factors such as diet and exercise, and anthropometric variables.

Among the baseline characteristics of study participants, comparisons of categorical variables were performed with either Pearson’s chi-squared test or Fisher’s exact test, as appropriate. For continuous variables, either the independent t-test or the Wilcoxon-Mann-Whitney test was used, as appropriate. The Shapiro-Wilk test and skewness/kurtosis test were used to test for the normality of data. The outcomes comparing baseline and 12-week follow-up data for each group were analyzed with the paired t-test or the Wilcoxon signed-rank test. The analysis of covariance (ANCOVA) was used to compare changes in continuous variables between groups after adjusting for the baseline values, and the rank transform ANCOVA was used when data violated the ANCOVA assumptions. To test the relationship between two variables, either Pearson’s or Spearman’s correlation analysis was used. For the CGM metrics, post hoc analysis was performed separately for daytime (6:00 AM–11:59 PM) and nighttime (12:00 AM–5:59 AM) as well as for 2-week durations at baseline and at weeks 4, 8, and 12. The trend of changes in the CGM metrics with an increasing number of study weeks was evaluated using the one-way analysis of variance and test for linearity. Data are presented as number (%), mean ± SD, or median (interquartile range [IQR]). Statistical significance was defined as 2-sided P values <0.05. Statistical analyses were performed using R Statistical Software (v4.1.1; R Core Team 2021).

Participants were recruited between June 2021 and December 2021. A total of 36 participants were randomly assigned to the intervention group (n = 18) or the control group (n = 18; Supplementary Figure 2 ). A total of 34 participants (94%) with 17 in each group completed the study and were analyzed for the per-protocol population. All participants in the intervention group had five times of remote intervention because they met ≥1 intervention criteria every two weeks.

The baseline characteristics of participants are shown in Table 1 . There were no significant differences in baseline demographics and clinical characteristics between the two groups. The mean age of participants was 44.3 (SD, 13.3) years and 52.8% were female. The mean duration of diabetes was 17.1 (SD, 10.4) years and the mean baseline HbA1c level was 8.9% (SD, 1.6).

The changes in HbA1c levels from baseline to week 12 were significant in both groups (from 9.2% ± 2.0% to 7.8% ± 1.0%, P = 0.003 in the intervention group; from 8.6% ± 1.1% to 7.8% ± 0.9%, P = 0.004 in the control group; Table 2 and Figure 1A ). The mean reduction in HbA1c levels was greater in the intervention group compared to the control group (−1.4% and −0.8%, respectively), although the difference between groups were not statistically significant (P adjusted for baseline values = 0.506; Figure 1B ).

Although changes in HbA1c levels were not significantly different between groups and both groups showed significant changes in HbA1c levels, only the intervention group showed significant correlations between changes in HbA1c levels and changes in CGM metrics, such as TIR (R = 0.640, P = 0.006), time with glucose >180 mg/dL (R = −0.710, P = 0.001), and mean glucose level (R = −0.670, P = 0.005), whereas the control group did not show a correlation ( Supplementary Figure 3 ). This result suggests that HbA1c levels were improved along with the CGM metrics in the intervention group.

The TIR significantly increased from baseline to week 12 in the intervention group (from 49.8% ± 15.7% to 60.9% ± 7.9%; P = 0.001), but not in the control group (from 50.0% ± 15.7% to 54.0% ± 13.9%, P = 0.151; Table 2 and Figure 2A ). Participants in the intervention group showed significant decreases in time with hyperglycemia (P = 0.002 and P = 0.026 for >180 and >250 mg/dL, respectively) and mean glucose level (P = 0.017), whereas those in the control group did not show significant changes. The changes in time with hypoglycemia (<54 and <70 mg/dL) and glycemic variability measured by CV from baseline were not significant in either group. When we compared the CGM metrics between two groups, changes in the TIR (adjusted mean difference, 7.0%, P = 0.019), time with glucose >250 mg/dL (adjusted mean difference, −7.2 mg/dL, P = 0.019), and CV (adjusted mean difference, −6.4%, P = 0.018) were significantly improved in the intervention group compared to the control group ( Table 2 ). Moreover, the ambulatory glucose profile (AGP) at week 12 compared to baseline showed that the IQR and the interdecile range (5th to 95th percentile) of glucose levels were narrower and the median was stabilized in the intervention group, indicating a reduction in glycemic variability ( Figures 2B–E ).

The CGM metrics at daytime and nighttime and at 4-week intervals were obtained. Overall, the changes in glycemic metrics during both daytime and nighttime were similar to those of all day ( Supplementary Table 1 ). During the daytime, significant decreases in time with glucose >180 mg/dL and mean glucose level were found in the intervention group. During the nighttime, a significant increase in TIR and a significant decrease in time with glucose >250 mg/dL were observed in the intervention group. Although statistical significance was lacking due to a low percentage of time with hypoglycemia, time with nocturnal hypoglycemia (<70 mg/dL at nighttime) was decreased in the intervention group, but it was increased in the control group.

The changes in glycemic variability measured by CV from baseline were decreased in the intervention group, contrary to the increase in the control group, but the changes were not significant in either group. However, the between-group difference of the change in CV was significantly reduced in the intervention group compared to the control group both during the daytime and nighttime (adjusted mean difference, −5.6% and −7.0%, P = 0.039 and P = 0.022, respectively; Supplementary Table 1 ).

Furthermore, in the analysis of glycemic metrics at 4-week intervals, unlike the control group, there was an increasing trend of TIR (P for trend = 0.013) and a decreasing trend of time with hyperglycemia >180 mg/dL (P for trend = 0.034) as the length of time participants in the intervention group used FGM with remote intervention increasing ( Figures 2F, G ; Supplementary Table 2 ). Mean glucose levels and CV decreased gradually in the intervention group, although the linear trend was not statistically significant ( Figures 2H, I ).

Regarding DTSQs questionnaire items, the satisfaction-related scores increased from baseline at week 12 in both groups, though not significantly ( Table 3 ). The perceived hyperglycemia-related scores decreased significantly in the intervention group (P = 0.015) and the perceived hypoglycemia-related scores increased significantly in the control group (P = 0.041), although the between-group differences were not significant. However, the DTSQc results at week 12 showed that the satisfaction-related scores were significantly higher in the intervention group than the control group (P = 0.016). Although the decrease in PHQ-9 scores was significant in both groups and the decrease in GAD-7 scores was not, the intervention group showed a greater decrease than the control group, indicating that anxiety and depression were further reduced in the intervention group.

The increase in total daily insulin doses was greater in the intervention group, even though there was no statistical significance ( Table 4 ). The number of scans per day was reduced significantly in the control group (P = 0.034), but not in the intervention group. The frequency of meals, snacks, and exercise were not changed in either group, and the hours of exercise decreased significantly more in the control group (P = 0.042). Body mass index and waist circumference increased in both groups, and only the increase in the body mass index in the control group was statistically significant (P = 0.015).