Glycemic and Psychosocial Outcomes of Advanced Hybrid Closed-Loop Therapy in Youth With High HbA_1c: A Randomized Clinical Trial

Advanced hybrid closed-loop (AHCL) therapy improves glycemic outcomes and has consistently been found to be superior to conventional therapy in children and adolescents in various clinical trials (1,2). The real-world outcomes are encouraging, with a higher proportion of users of closed-loop systems achieving the recommended glycemic targets (3). These positive outcomes may be less representative of youth struggling with daily diabetes care with suboptimal glycemic outcomes. Studies in this relatively high-risk cohort are needed to determine whether the improvement in glycemic outcomes with new technology can be translated to this vulnerable group. It is well established that high glycated hemoglobin (HbA_1c) is associated with a greater risk of diabetes-related micro- and macrovascular complications (4) with reduced life expectancy compared with the general population (5). Furthermore, the burden of disease care negatively impacts mental health (6) and places a large socioeconomic burden on families and community (7,8). An increase in HbA_1c by one percentage point is expected to lead to a 6.0% increase in diabetes-related medical costs for type 1 diabetes (9). Hence, it is vital to support this group with interventions that can minimize the risk of diabetes complications and reduce the burden of disease on both the individual and community.

With the advent of closed-loop devices designed to improve glycemia and minimize user-related decisions, this study explored the hypothesis that automated insulin delivery will result in improved glycemic outcomes and reduce diabetes burden. Closed-loop technology was found to be beneficial in youth with type 1 diabetes with suboptimal glycemia in single-arm 3-month prospective studies transitioning from insulin injections to closed-loop therapy (10,11). There was also demonstration of sustained improvement in glycemic and psychosocial outcomes with 12-month AHCL use in a small cohort of adolescents with suboptimal glycemia and experiencing considerable diabetes burden (12). However, there was no control group in these studies. Likewise, there are no studies reviewing outcomes of individuals on insulin pump therapy with suboptimal glycemia who have commenced AHCL therapy. Hence, this study is a 6-month randomized clinical trial (RCT) to determine whether AHCL will be of benefit in improving glycemic and psychosocial outcomes in this high-risk subgroup of youth living with type 1 diabetes on insulin pump therapy. This is an important cohort to study, as they have the greatest potential for improvement in glycemic outcomes, improved well-being, and reduced burden of disease from diabetes complications.

This was an open-label, multicenter, parallel, randomized controlled phase 3 in-home trial conducted at four tertiary pediatric diabetes centers in Australia. Participants were randomly assigned to “treatment as usual” on continuous subcutaneous insulin infusion (CSII) with or without continuous glucose monitoring (CGM) as the control group or to the intervention group on AHCL therapy over 24 weeks. Ethics for all sites was approved by the Child and Adolescent Health Service Human Research Ethics Committee under the National Mutual Acceptance (RGS 0000000886, initial approval date 19 April 2018). Safety aspects were overseen by an independent data safety monitoring board at Perth Children's Hospital. The trial was prospectively registered with Australian New Zealand Clinical Trials Registry (ACTRN12619001452189).

The study recruited participants aged between 12 and 25 years with type 1 diabetes for more than a year, fasting C-peptide of <0.1 nmol/L, and mean HbA_1c over 6 months and the most recent HbA_1c >8.5% (65 mmol/mol) on CSII with or without CGM (CSII ± CGM). Participants were excluded if they were on any form of closed-loop system, experienced severe diabetic ketoacidosis (DKA) in the 6 months prior to the screening visit, had used any noninsulin glucose-lowering agent within the preceding 3 months, commenced CGM in the 3 months prior to the screening visit, had uncontrolled celiac or thyroid disease or clinically significant gastroparesis, were pregnant or planned pregnancy, were unable or unwilling to meet protocol requirements, and/or had an unstable medical or psychological condition which, in the opinion of the treating physician and/or investigator, would compromise the ability to meet protocol requirements.

Written informed consent was obtained from participants aged ≥18 years, and written parent consent and participant assent was obtained for those <18 years.

There was a run-in period prior to randomization to collect baseline glycemic data. Sensor glucose data were collected from a 2-week period of masked CGM (Guardian Sensor 3; Medtronic, Northridge, CA). An additional 1–2 weeks of masked CGM were offered to participants who did not meet the required CGM data for at least 70% of readings over at least 10 days (13). If the maximum of four sensors was reached and the required valid sensor glucose readings were not obtained during the run-in period, the participant was withdrawn from the study. This was to ensure that participants demonstrated capacity for appropriate CGM use.

After the run-in period, participants were randomly assigned in a 1:1 ratio, using a computer-generated random sequence, to either CSII ± CGM (treatment as usual) or AHCL. We chose CSII ± CGM as the control group, as this therapy constitutes close to 50% of our clinic cohorts, who were struggling with diabetes care despite adopting diabetes technology. To ensure optimal balance between treatment arms (14), participants were allocated using minimization with the following factors: age, HbA_1c at visit 1, and trial site. Randomization was undertaken by the delegated person at Perth Children’s Hospital. As the study involved a device which required training and use, both the participants and the research team were not blinded to the treatment assigned, although treatment allocation was blinded to analysis.

Following randomization, all participants received reeducation from a diabetes nurse educator and dietitian, with duration of the session dependent on the prior knowledge of the participant, and adjustments to pump settings were made as needed. Control group participants continued their usual treatment (CSII ± CGM). All AHCL group participants received training on the MiniMed 670G Version 4.0 AHCL pump (Medtronic) equivalent to the MiniMed 780G system (without Bluetooth connectivity) with Guardian Sensor 3 and transmitter. In view of the recall on insulin pumps with clear retainer rings in October 2021, the study switched to the MiniMed 780G insulin pump with the same algorithm, without the CareLink app, to maintain similar user interface for the study. Participants were commenced with glucose target set at 5.6 mmol/L (100 mg/dL) and active insulin time (AIT) of 3 h. The support provided to participants in the study reflected routine clinical care. A diabetes educator provided ongoing support for both groups. All participants commenced on AHCL had additional weekly follow-ups for 4 weeks via phone call or e-mail. Participants in the control group (treatment as usual) and AHCL were encouraged to attend their routine clinical visits and had scheduled research visits at 12 weeks (midstudy) and 24 weeks (study end) postrandomization. All participants had access to clinical and/or technical support during the study.

Point-of-care HbA_1c (DCA Analyzer) was measured on finger-prick capillary blood at randomization and 12 and 24 weeks. Masked CGM data were collected at baseline, prior to midstudy, and at study-end visits. Real-time CGM data were collected in AHCL participants during the entire study period. Age-appropriate validated psychosocial questionnaires (15) were administered at baseline, midstudy, and study end. Retinal fundus photograph scans were taken at each site at randomization and study end to evaluate for diabetic retinopathy, a potential risk when glycemic control is rapidly optimized (16). At the completion of the primary phase of the 6-month study, all participants were given an option to continue into the extension phase of another 6 months, during which all participants in the intervention group on AHCL therapy continued AHCL, while participants in the control arm were offered AHCL therapy. We report the results from the primary phase of the 6-month study. Semistructured interviews were also conducted after 6 months of AHCL at the primary site, the results of which are reported separately.

The primary outcome was HbA_1c collected at 24 weeks. The secondary outcomes included CGM metrics derived from masked CGM and reported according to the recommendation for artificial pancreas clinical trials (17,18). CGM metrics included time in range (TIR) (70–180 mg/dL), time in hypoglycemia (<70 mg/dL, <60 mg/dL, <54 mg/dL, and <50 mg/dL) and hyperglycemia (>180 mg/dL, >250 mg/dL, and >300 mg/dL), hypoglycemic events, mean sensor glucose, SD, and coefficient of variation, along with fasting capillary glucose. The study also evaluated CGM metrics overall and separately for day (0600 to 2400 h) and night (0000 to 0600 h). Other clinical outcomes included change in auxological parameters and change in insulin doses (total daily insulin, proportion of basal insulin, and insulin carbohydrate ratio [ICR]). Validated questionnaires were also administered to evaluate patient-reported outcomes. These included quality of life, diabetes distress, fear of hypoglycemia, anxiety and hypoglycemia awareness assessed using the Diabetes-specific Quality of Life Questionnaire (19), Problem Areas In Diabetes (PAID) Survey (20), Hypoglycemia Fear Survey (21), Generalized Anxiety Disorder (22)/Patient Health Questionnaire (23), and Gold score (24). Safety outcomes were the frequency of severe hypoglycemia, DKA, device-related adverse events, and any other untoward medical occurrence. Retinal fundus photographs were reviewed by the ophthalmologist at the lead site and were assessed and graded for changes associated with diabetic retinopathy according to the International Clinical Diabetic Retinopathy Severity Scale (25).

We determined that a sample size of 44 (22 in each group) would be required to provide 80% power for a two-tailed test to detect a difference between groups of 1.5 percentage points in HbA_1c at 6 months, assuming an SD of 1.7 percentage points and with α set at 0.05. Allowing for 15% predicted dropout, the study planned to recruit 50 participants. Estimates used in the power calculation were based on a previous study using insulin pumps in adolescents with suboptimal glycemic control (26) and unpublished data from our team's previous studies.

Data were analyzed on an intention-to-treat basis for all randomized participants. To test for an effect of treatment group on the primary outcome and other continuous secondary outcomes, an ANCOVA including treatment arm and baseline of the outcome was conducted. Where ANCOVA assumptions were violated and were unable to be addressed through log transformation for the outcome measure, rank sum tests were performed. For the primary outcome analysis, a multiple imputation approach using predictive mean matching was employed to account for missing data (27); auxiliary variables included in multiple imputation were selected based on their correlation with the outcome, and 40 iterations were conducted. Additional analyses, including a complete case analysis (with no imputation) and an ANCOVA model including minimization factors, were also conducted. All secondary measures were analyzed on a complete case basis. Intervention effects on CGM metrics were also examined separately for day (0600 to 2400 h) and night (0000 to 0600 h). The association between time spent in closed loop and end of study HbA_1c was explored using ANCOVA, adjusted for baseline HbA_1c. Tests of significance were two sided, and P values of <0.05 were considered statistically significant. No adjustments for multiple comparisons were made. All analyses were conducted using R (version 4.1.1) (28).

Forty-six participants were enrolled in the study between June 2020 and June 2022. Three participants did not meet the prerandomization criteria of the required sensor readings of >70% for 10 days, and one participant withdrew because of perceived study burden. Forty-two eligible participants were randomized, with 21 each assigned to the control and intervention groups. Four participants (two control, two intervention) withdrew postrandomization (Fig. 1). Table 1 provides the baseline characteristics of 42 participants by group allocation. The overall cohort had a mean (SD) age 16.2 (2.5) years and diabetes duration of 9.7 (4.2) years. The mean PAID score was 50.3 (19.8), with 71% of the participants experiencing significant distress (29). HbA_1c of participants at enrollment ranged between 8.7% (72 mmol/mol) and 13.7% (126 mmol/mol). The overall mean HbA_1c at the first study visit was 10.3 (1.3)% or 89 (14.2) mmol/mol and 9.8 (1.1)% or 84 (12) mmol/mol at the time of randomization. Seventy-four percent of the participants were current CGM users, with mean CGM use of 43% in the 3 months prior to the study. Seven participants were on Tandem t:slim pump, while the remaining participants were on Medtronic pumps (MiniMed 770G, 670G, 640G, and Veo). None of the participants were utilizing the predictive low-glucose suspend features.

The mean (SD) HbA_1c decreased from 9.5 (0.9)% or 80 (9.8) mmol/mol at randomization to 8.8 (1.1)% or 73 (12) mmol/mol at 6 months in the AHCL group and from 10.0 (1.3)% or 84 (14.2) mmol/mol to 9.9 (1.2)% or 85 (13.1) mmol/mol in the control group, with a mean adjusted difference between the two groups at study end of −0.77% (95% CI −1.45 to −0.09) or −8.4 mmol/mol (95% CI −15.8 to −1.0); P = 0.027. Similar results were seen in the complete case analysis (adjusted difference of −0.78%; 95% CI −1.46 to −0.09; P = 0.028) and in the model adjusting for stratification factors (adjusted difference of −0.71; 95% CI −1.40 to −0.0071; P = 0.048). Unadjusted mean HbA_1c at randomization and 3 and 6 months is shown in Fig. 2. Change in HbA_1c per participant in both groups is shown in Supplementary Fig. 1.

A statistically significant treatment effect was also seen for secondary CGM metrics (Table 2 and Supplementary Fig. 2). Participants in the AHCL group spent greater TIR (70–180 mg/dL) at study end compared with baseline (AHCL 36.7% to 51.9% vs. control 34.7% to 31.4%), with an adjusted mean difference between the two groups of 19.1% (95% CI 11.1 to 27.1; P ≤ 0.001). Furthermore, participants spent reduced time in hyperglycemia (% time >180 mg/dL [difference −17.7%; 95% CI −26.6 to −8.8; P ≤ 0.001] and % time >250 mg/dL [difference −19.6%; 95% CI −28.3 to −10.8; P ≤ 0.001]). There were no significant differences in rate of hypoglycemia events or time spent in hypoglycemia at all predefined biochemical levels between the two groups. The differential effects of intervention on TIR and time in hyperglycemia were more pronounced during the day than night but were comparable for other metrics (Supplementary Fig. 3).

There was no evidence for an intervention effect on psychosocial outcomes (diabetes-specific quality of life, diabetes distress, anxiety, depression, and fear of hypoglycemia); however, the direction of effects favored the intervention group, as shown in Table 2.

ICRs were adjusted in the study in both participant groups, with no difference in the weighted ICR, total daily, and basal insulin between the two groups at study end. In the AHCL group, the median (interquartile range [IQR]) AIT at the end of the study was 2.45 h (2.3, 3).

Participants in the study had a median (IQR) sensor use of 83 (76, 92)% at 1 month and spent 75 (65,79)% of the time in closed loop. However, at 6 months, the sensor use was 46 (23, 65)%, and time in closed loop was 49 (20,70)%. Supplementary Fig. 4 presents the monthly median (IQR) of sensor wear and the proportion of time in closed loop during the study.

Supplementary Fig. 5 shows the change in HbA_1c from baseline as the function of time spent in closed loop in the last 3 months of the study. Although there appears to be a trend toward improvement in HbA_1c, the association between time spent in closed loop and baseline adjusted HbA_1c did not reach statistical significance; F(1,16) = 1.665, P = 0.215.

There was no evidence of diabetic retinopathy in the entire cohort at randomization and study end. There was one DKA in the control group and one in the AHCL group, with no episodes of severe hypoglycemia in either of the two groups. There were no serious adverse events related to the trial device.

This 6-month multicenter RCT highlights the improvement in glycemia with AHCL in at-risk youth with suboptimal diabetes management, although an improvement in psychosocial outcomes was not observed. Participants in this trial had a mean HbA_1c of 10.3% (89 mmol/mol) at enrollment, which contrasts with previous RCTs that recruited participants with mean HbA_1c between 7.6 and 9.0% (60 and 75 mmol/mol) (1,30–32). In our study, all participants received reeducation and review of diabetes management prior to randomization. However, participants in the control group on standard pump therapy did not show an improvement in HbA_1c. In contrast, participants assigned to closed-loop therapy improved glucose levels, with an adjusted HbA_1c difference of 0.77% or 8.4 mmol/mol between the two treatment groups. This was also reflected in the CGM metrics, with an overall improvement in TIR, and reduction in hyperglycemia, with no increase in hypoglycemia. Overall, the complete lack of improvement in the control group despite reeducation and increased attention from study participation underlines the futility of a standard educational approach and the need to consider user-friendly automated insulin delivery systems in this high-risk group if glycemia is to be improved.

The magnitude of glycemic improvement, however, is lower than previously reported in other studies. This could be attributed to a few factors. Our study cohort was not technology naïve and were CSII users and had high diabetes distress. Although most were prior CGM users, sensor wear was low and could have been impacted by the need for calibrations with the G3 sensor. Likewise, settings were less aggressive with AIT, which was not at the current recommendation of 2 h.

As compared with our study wherein participants were on CSII with most of them being CGM users, previous studies were single-arm prospective studies that recruited youth on multiple daily injection (MDI), 50% of the cohort were sensor naïve, and the rest were predominantly on intermittently scanned sensors (10,11). These studies also used the same AHCL system with G3 sensor, but had a different onboarding protocol as participants transitioned from MDI to SmartGuard. In the 3 months after transition to AHCL, HbA_1c decreased from 10.5% (91 mmol/mol) at baseline to 7.6% (60 mmol/mol) in 20 youth (mean age 18 years) (11). After 12 months of AHCL, glycemic improvement was sustained, with favorable outcomes in quality of life and treatment satisfaction (12). Glycemic improvement was also demonstrated with AHCL in a younger cohort of 34 children (mean age 12.5 years) on MDI with suboptimal control, with improvement in HbA_1c from 8.6% (70 mmol/mol) to 6.5% (48 mmol/mol) in 12 weeks (10). Another observation in our study cohort is the baseline high diabetes distress score, categorized as in “clinically significant” range with evidence of mild anxiety/depression. There was no evidence of a strong effect of AHCL on participant-reported outcomes between the two groups, although a trend toward improvement in the psychosocial outcomes was demonstrated.

The improvement in glycemic control was seen despite suboptimal device use. This was seen with a decline in sensor wear over the study period and was also acknowledged by the participants during the semistructured interview (33). The system, guided by real-time glucose levels, has the ability to deliver autocorrections every 5 min as needed, coupled with more stability in closed loop (with less exits) (34), thereby increasing insulin delivery in a cohort of chronically underinsulinized individuals. The benefit derived is proportional to the time spent in closed loop, and hence a reflection of the percentage of sensor wear. Although three-quarters of participants reported using sensors, the use was suboptimal even at baseline. Most participants were on real-time CGM (Dexcom G6) and were not used to calibrating their devices. In our study, the need for finger pricks to be done as part of AHCL therapy (for calibration and for maintenance in closed loop) and wearing an additional masked CGM at the designated time points were added tasks for a cohort not accustomed to routine self-monitoring of blood glucose. This will be reassuringly not an ongoing issue, with the availability of the Guardian Sensor 4, which does not require calibrations. Improved user experiences with reduced burden related to finger sticks and fewer closed-loop exits have been reported from real-world evaluations with transition from G3 to G4 (35). The persistent diabetes burden and the ongoing motivation issues resulted in difficulty in maintaining the behavior change that was needed to optimize diabetes management (33). The study highlights the need for clinicians to assess and address diabetes distress as part of comprehensive psychosocial care in pediatric diabetes clinics (36,37) and not just be reliant on glucose-related outcomes.

The youth in the study group had a mean diabetes duration of 9.4 years and would have commenced sensors as part of diabetes care, after federal government subsidy in 2017. On the other hand, if sensors are used early following diagnosis, the chances of improved uptake and optimal use may be higher (38), which can potentially translate into better outcomes (39). Furthermore, the pump settings with starting AIT of 3 h and a median AIT of 2.45 h at the end of the study are higher than the current recommendation of 2 h in this cohort. Our study commenced in June 2020 and preceded most of the studies to date and hence had used a weaker AIT at the start. Investigators did not have the opportunity to finetune the AIT, given the minimal user interaction with AHCL and the high time in manual mode. As AIT directly influences the autocorrections, participants in this cohort did not receive the strongest autocorrections, and hence this would have influenced the glycemic outcomes.

Previous skepticism around using insulin pump therapy in high-risk individuals with suboptimal glycemia has been largely reexamined, with studies showing glycemic improvement and no increase in serious adverse events (11,26). Youth with high HbA_1c are at increased risk of DKA (40). Two participants had DKA in the 12 months preceding the study. Hence it is reassuring to note that there was only episode of DKA during the study in a participant assigned to AHCL. However, at the time of the episode, the participant was not wearing a sensor and was in manual mode, with DKA secondary to infusion set occlusion compounded by inadequate diabetes management. Likewise, there were no episodes of severe hypoglycemia prior to and during the study. Although the study design excluded participants with a history of DKA in the 6 months prior to enrollment, with the experience of automated insulin delivery systems in research trials and clinical care, previous DKA may not preclude commencement of AHCL therapy.

The strength of the study is the 6-month RCT design that explores the glycemic and psychosocial outcomes in a high-risk cohort of youth who were adopters of technology on CSII with most on CGM, and yet continued to experience significant dysglycemia and diabetes distress. Previous studies in youth with suboptimal glucose control were limited to single-arm short-term studies. The presence of a control group provided the opportunity to not only effectively analyze the impact of intervention on glycemia and psychosocial outcomes but also provide an opportunity for more valid health economic assessments. Blinded CGM data and validated questionnaires were collected to analyze CGM metrics and psychosocial outcomes in both study groups. All participants received reeducation and had reviews at 4 weeks and 3 months to reflect routine clinical care.

Limitations include inability to continue recruitment for the study and reach the required sample size of 50 participants. The study commenced in January 2020 and was impacted by the coronavirus disease 2019 pandemic and recall of the MiniMed 600 series pump in October 2021, which affected the investigational 670G V4.0 used in the study. The study recruitment recommenced following availability of the 780G devices; however, the Bluetooth feature was not used, and hence participants/caregivers did not have the added benefit of the CareLink app, and real-time visibility of glucose levels on their phone was not available. Recruitment was also impacted by the availability of closed-loop systems as part of clinical care in Australia. The Control-IQ hybrid closed-loop system was available from early 2022, and the Medtronic 780G pump with its SmartGuard function was available from mid-2022. Other limitations include the additional reviews that AHCL participants received compared with the control participants because of the use of an investigational AHCL device and the nonavailability of ethnicity and socioeconomic data. Importantly, this was a challenging population to be studied in a clinical trial, to ensure the ability to follow protocol requirements and attendance to research visits and to encourage optimal device use. Hence, the results are applicable to a high-risk cohort of youth with suboptimal diabetes management and high diabetes distress.

To conclude, AHCL should be encouraged in youth with suboptimal glycemia, as AHCL improves glycemia. However, psychological support remains vital, as technology alone may not be able to reduce the burden of diabetes care in this subgroup.

Acknowledgments. The authors thank the participants and families who participated in the study.

Funding. The study was funded by Juvenile Diabetes Research Foundation (JDRF) Australia Type 1 Diabetes Clinical Research Network (4-SRA-2016-350-M-B), a Special Research Initiative of the Australian Research Council, National Health and Medical Research Council (ID APP1078190). In-kind support was provided by Medtronic through the provision of insulin pumps, sensors, and transmitters for the study. Roche Diabetes Care provided the glucometers for the study. JDRF Australia provided input into the study design. M.B.A. was supported by the Department of Health/Raine Clinical Research Fellowship from Western Australia.

This study was investigator designed and led. The devices were provided by Medtronic (as above), but Medtronic was not involved in study design, data analysis, or data interpretation. A copy was provided to Medtronic and JDRF for review prior to submission. The funders of the trial had no role in collection, analysis, and interpretation of the data and in manuscript preparation.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.B.A. contributed to the study design, received approvals, had study oversight across all centers, provided data interpretation, and wrote the manuscript. J.D. coordinated the studies and reviewed the manuscript. G.J.S. conducted the analyses and provided statistical input. K.B. provided support for the psychological outcomes. A.C. performed the retinal photograph grading and reporting. J.M.F., G.R.A., F.J.C., E.A.D., and T.W.J. contributed to the study design, supervised the study at each site, and reviewed the manuscript. T.W.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Kristen J. Nadeau.