Safety and Efficacy of Sustained Automated Insulin Delivery Compared With Sensor and Pump Therapy in Adults With Type 1 Diabetes at High Risk for Hypoglycemia: A Randomized Controlled Trial

The occurrence of severe hypoglycemia remains a serious concern in many people with type 1 diabetes (T1D) (1,2). The T1D Exchange reported that 6–10% of adults with T1D in the studied U.S. population experienced at least one episode of severe hypoglycemia during a 3-month period, with slightly more events occurring in those not using an insulin pump or continuous glucose monitoring (CGM) (3). The Study of Adults' Glycemia (SAGE) study, performed with people from several continents, reported that close to 12% of adults with T1D experienced at least one severe hypoglycemic event during a 6-month period (4). Impaired hypoglycemia awareness (IHA) is associated with higher risk for developing severe hypoglycemia and can be identified by using questionnaires, such as the Clarke Hypoglycemia Perception Awareness scale, with a score >3 indicating IHA (5,6). Clarke scores have been shown to correlate positively with the percentage of CGM values <70 and <54 mg/dL. People with T1D and IHA continue to experience frequent hypoglycemia, even when using CGM (7).

Insulin pump use has been reported in a meta-analysis to reduce the occurrence of severe hypoglycemia (8) and, in a pilot study, to improve IHA (9). Nevertheless, although all participants received guidance targeted toward absolute avoidance of biochemical hypoglycemia, insulin pump and real-time CGM uses did not show any significantly higher reduction in severe hypoglycemia and improvement of IHA compared with uses of multiple daily insulin injections and real-time CGM or self-monitoring of blood glucose in the Comparison of Optimized MDI [multiple daily injection] Versus Pumps With or Without Sensors in Severe Hypoglycemia (HypoCOMPaSS) trial (10). Technological innovations in the recent years (e.g., Threshold or Predictive Low Glucose Suspend [PLGS]) of pump insulin delivery, have shown a reduction of hypoglycemic events compared with the CGM and pump combination (11,12). The Study of MiniMed 640G Insulin Pump with SmartGuard (SMILE) trial investigated PLGS in hypoglycemia-prone adults with T1D and showed a significant reduction in the number of severe hypoglycemic events (13). People with sustained problematic hypoglycemia have been using implantable pumps or ports for intraperitoneal insulin delivery with some benefits, prior to the availability of CGM devices (14,15), or have received islet or pancreas transplantation resulting in a dramatic reduction of severe hypoglycemia (16,17). However, the access to intraperitoneal insulin delivery or transplants is limited.

Automated insulin delivery (AID) has been shown to improve glucose control in adults and children with T1D in randomized clinical trials including participants ages 2–72 years (18–20). This improved glucose control was also reported in free-life observation of nearly 20,000 people with T1D ages 1–92 years (21). Benefits include an average increase of approximately 10% in the time in target range (TIR; 70–180 mg/dL), accompanied by a variable reduction of time below the target range (TBR; <70 mg/dL) (22,23). However, there are only few studies of the effect of AID in people with T1D who have history of severe hypoglycemia and/or IHA. Hence, limited data are available on the effect of AID in this high-risk population. An investigational AID system has been assessed in a 4-week randomized control study relative to an sensor-augmented pump. The study included people with T1D at high risk for hypoglycemia based upon several indices (24). Both low blood glucose index (LBGI) (25), computed from CGM-based TIRs, and TBR significantly decreased with the use of AID. Diabeloop for Highly Unstable Type 1 Diabetes (DBLHU), an adapted version of the Diabeloop DBLG-1 AID system, has also been investigated relative to the Medtronic 640G insulin pump connected to an Enlite CGM with safety PLGS software in few patients with highly unstable T1D (26). The percentage of TBR was significantly lower with DBLHU.

In the present study, we aimed to fill the knowledge gap by focusing specifically on the safety and efficacy of a commercially available AID system in people with T1D at high risk for hypoglycemia.

This is a prospective, open-label, randomized clinical trial with 2:1 randomization to an intervention group using the study AID system versus a control group using usual insulin pump and study CGM sensor (S&P) therapy for 12 weeks, followed by a 12-week optional extension phase in which the initial S&P group had the opportunity to switch to AID and the initial intervention group continued to use AID. The main objective of this trial was to assess whether the study AID system could reduce safely TBR in the enrolled people compared with an open-loop pump with the study CGM, and to assess the sustainability of the results in the optional extension period. The study protocol was approved by the Comité de Protection des Personnes Ouest II, Angers, France, and the French National Agency for the safety of drug and medical devices. The study is registered with ClinicalTrials.gov (identifier NCT04266379).

Screening of participants was based on people with T1D at high risk for hypoglycemia, defined as a Clarke score >3 and/or a history of severe hypoglycemia during the previous 6 months. After informed consent, the participants entered a 2-week run-in phase to confirm their high risk for hypoglycemia by wearing the blinded study CGM (G6; Dexcom, San Diego, CA). A confirmed TBR of ≥5% was an eligibility prerequisite to randomization. Participants were then randomized in a ratio of 2:1, according to a permuted block design, stratified by clinical center, to AID (t:slim ×2 insulin pump with Control-IQ technology; Tandem Diabetes Care, San Diego, CA) or S&P using their usual insulin pump and the nonblinded study CGM (Dexcom G6) for 12 weeks. The participants allocated to the S&P group using an insulin pump connected to a CGM with PLGS software could keep on using it but had to wear the study CGM as well. The primary and secondary study outcomes were assessed at the end of the 12-week randomized trial. At the 12-week visit, participants were offered the option to continue using the AID system (AID group) or to move to the AID system after a 2-week training in open-loop mode (S&P group), for an optional 12-week extension. This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline.

Criteria for enrollment in the study were as follows: age >18 years, having T1D for >1 year, HbA_1c <10.5% (91 mmol/mol), insulin administered by insulin pump for >6 months, trained in carbohydrate counting, a Clarke score >3, and/or experience of severe hypoglycemia during the previous 6 months. A positive pregnancy test in female participants of child-bearing potential, any associated chronic disease or therapy (except insulin) affecting glucose metabolism, use of sodium–glucose cotransporter 2 inhibitors in the previous 3 months, and lack of any social or familial support able to intervene in case of severe hypoglycemic episode were exclusion criteria. Participants were enrolled in the Montpellier and Caen University hospitals in France.

At randomization, CGM data of the run-in period were downloaded, HbA_1c was measured, and study questionnaires were completed as baseline data. Diabetes management was reviewed and insulin pump settings were optimized by the study physician. The AID group was trained to use the study AID system in open-loop mode for 2 weeks before initiating the closed-loop AID mode. The S&P group was trained to use the study CGM in nonblinded mode for the same period.

Training for Control-IQ activation was provided individually during a 4-h outpatient visit. This “hybrid” system uses a model-based control algorithm aimed at keeping CGM in the 112.5–160 mg/dL range, as previously described (18). Participants were provided an emergency phone number for any concerns with the system; there was no remote monitoring. In case a technical issue could not be solved by phone, patients were asked to come to the hospital for device reset or replacement.

All participants were trained to follow the provided guide for the management of hyper- and hypoglycemia. Phone calls were scheduled between investigators and the participants 48 h and 1 week after initiation to the study devices to download device data and resolve potential issues with and/or questions about device use. If needed, pump parameters could be revised, including insulin to carbohydrate ratios. Hospital visits were scheduled after 4, 8, and 12 weeks for the same purposes, plus physical examination, HbA_1c measurement, and completion of study questionnaires at week 12.

During the optional 12-week extension period, participants initially allocated to the AID system had hospital visits scheduled at weeks 16 and 20 for similar follow-up as at weeks 4 and 8, and at week 24 for the end of the study. Participants in initial S&P group were trained to use the AID system and followed as the participants in the initial AID group after 48 h, and 1, 4, 8, and 12 weeks with the end of the study at week 26. HbA_1c measurement and completion of study questionnaires were performed at the end-of-study visit.

The study primary outcome is the change from baseline to the 12-week postrandomization observation in CGM-measured TBR (i.e., CGM percent time [%time] <70 mg/dL). These changes in TBR were compared between the two study groups. Postrandomization CGM metrics were calculated by pooling all CGM readings in the 12-week trial, starting from the randomization visit and ending with the visit at week 12.

The study main secondary outcomes included CGM-measured TIR, time above range (TAR), mean sensor glucose (SG) reading, and percentage of CGM time <54 mg/dL. To preserve statistical significance, a hierarchical analysis of these variables in the following order was used: if a nonsignificant comparison was encountered, all subsequent comparisons in the hierarchical list were considered exploratory. Other CGM metrics were included in exploratory analysis as well, including %TIR 70–140 mg/dL, glucose variability measured by the coefficient of variation (CV), SD, %time <60 mg/dL, LBGI, hypoglycemia events (defined as ≥15 consecutive minutes at <70 mg/dL), %time >250 mg/dL and >300 mg/dL, and high blood glucose index (HBGI) (25), all derived from downloaded CGM data. They also included change of HbA_1c levels measured by high-pressure liquid chromatography in blood samples and scores of study questionnaires between baseline and week 12. The same efficacy outcomes were assessed at the end of the optional extension, and changes were compared with those at week 12 between study groups.

Information on events occurring from the run-in period until the end of the study visit, including the randomized trial and the optional study extension, was collected and analyzed according to treatment mode. Key safety outcomes included the frequency of severe hypoglycemia and diabetic ketoacidosis.

Study questionnaires were completed by the participants at randomization, at the end of the randomized trial, and at the end of the extension study. They included the Clarke Hypoglycemia Awareness Scale (6); total and emotional burden scores on the Diabetes Distress Scale (27); total score and behavior and worry subscales scores from the Fear of Hypoglycemia Survey-II (28); Hyperglycemia Avoidance Survey total score (29); hypoglycemia confidence scale total score (30); and the Insulin Delivery Systems: Perceptions, Ideas, Reflections, and Expectations (INSPIRE) survey score (31).

The intervention was considered effective if the AID use was superior to the S&P use, using a statistical significance of α = 0.05 and the model specified below (i.e., P < 0.05).

The null hypothesis was that there is no difference in the change in mean CGM-measured TBR from baseline to the 12-week postrandomization period (i.e., the treatment effect), between AID and S&P. Sample size has been computed for the primary outcome. Data from the Anderson et al. (24) pilot trial conducted with the same algorithm in the same high-risk population were used. The estimated effect size for the primary outcome derived from these data using baseline treatment changes in the AID versus S&P groups was d = 1.06. For the power calculations, this effect size was conservatively reduced by 22% to d = 0.83. We also assumed 2:1 randomization to the AID and S&P groups, respectively, two-sided α = 0.05 and power = 90%. Primary outcome change in CGM-measured TBR yielded a total sample size 72. In our experience with previous AID protocols, few participants dropped out of studies; nevertheless, the recruitment quota was increased to 120 to allow for proper participant selection and risk-status confirmation prior to randomization.

Changes in the treatment effect were compared using a repeated-measures ANOVA (within or between factor). A point estimate, 95% CI, and two-sided P value were reported for the treatment effect based on the linear regression model, and a 5% level was used to declare statistical significance. Residual values were examined for an approximate normal distribution. If values were highly skewed, a transformation or robust statistical methods were used instead. However, previous experience suggested that the residual values for TBR follow an approximately normal distribution. Imbalances between groups in important covariates were not expected to be of sufficient magnitude to confound the analysis.

Other efficacy end points were similarly analyzed, but the false discovery rate was controlled using the adaptive Benjamini–Hochberg procedure (32). The false discovery rate–adjusted P values were calculated separately for the following categories: 1) all other CGM metrics, 2) HbA_1c analyses, and 3) questionnaire scores. The study main secondary outcomes were submitted to a hierarchical analysis, as described above, and the other secondary outcomes were considered for exploratory analysis.

All randomized participants were included in the safety analyses and all their postrandomization safety-event information was collected. The circumstances of all reportable cases of the following were tabulated by treatment group: severe hypoglycemia, diabetic ketoacidosis, and ketone events defined as day with ketone level >1.0 mmol/L.

Of the 117 enrolled patients from 13 May 2020 to 15 October 2020, 45 were not eligible for randomization because of having <5% TBR during the run-in phase while using blinded CGM in 44 cases, and a case of severe hypoglycemia with vertebral compression fractures during the run-in phase; the latter participant declined to further participate in the study. Hence, 72 enrolled patients were randomized from 26 May 2020 to 9 November 2020, 49 of whom were allocated to the AID group and 23 to the S&P group. Table 1 presents the demographic and descriptive baseline characteristics of the participants. The single significant difference at baseline was the sex distribution, with a higher number of male participants in the S&P group. Of note, close to half of the randomized participants were using an S&P with PLGS software at baseline, with a similar distribution between the randomization groups. All those allocated to the S&P group agreed to stop using the PLGS software during the study. Slightly more participants allocated to the S&P group had experienced severe hypoglycemic events during the prior 6 months. Supplementary Fig. 1 shows participant flow through the 12-week randomized trial. One person allocated to the S&P group declined to further participate immediately after the randomization. All other participants completed the randomized trial between 9 September 2020 and 15 February 2021.

The primary and main secondary study outcomes of the randomized trial derived from study CGM data are presented in Table 2. The change in TBR (primary outcome) was significantly different between the study groups, with a decrease of −3.7 percentage points (95% CI −4.8, −2.6; P < 0.001) with AID versus S&P. The prespecified hierarchical analysis of the main secondary outcomes showed that TIR and TAR were also significantly different between the two study groups, with an increase in TIR of +8.6% (95% CI 5.2, 12.1; P < 0.001) and a decrease in TAR of −5.3% (95% CI −8.7, −1.8; P = 0.004) related to AID effect; the mean SG reading was similar. The P value for change in percentage of SG readings <54 mg/dL (−0.8; 95% CI −1.2, −0.5) was not computed, in accordance with the hierarchical analysis.

Figure 1 illustrates the progression from baseline through the 12-week study intervention for the S&P and AID groups. Figure 1A presents weekly box-and-whisker plots of TBR; Fig. 1B presents box-and-whisker plots for TIR. A rapid improvement in both TBR and TIR occurs when patients switch to AID, and the glycemic benefits of AID are sustained thereafter.

The exploratory analyses of the other secondary outcomes are presented in Table 3, showing that the %time in the tight 70–140 mg/dL range was significantly higher with AID, whereas the SG reading CV and SD, as well as %time with SG level <60 mg/dL, LBGI, and the number of events with SG level <70 mg/dL for ≥15 min were all significantly lower in the AID group. The %time with SG level >250 or 300 mg/dL, HBGI, and HbA_1c levels were not statistically different between the groups, but a trend for lower %time in the AID group was observed for SG levels >250 mg/dL and the HBGI.

The study outcomes based on submitted self-reported questionnaires showed no significant difference in scores between the two study groups at week 12, as presented in Supplementary Table 1.

All participants in the original AID group agreed to continue with the 12-week extension phase; 16 of the 22 participants in the original S&P group continued with the optional study extension. The results of the extension phase are presented in Supplementary Table 2. Although the participants who kept using the AID system sustained their results of the randomized phase, those who moved from S&P to AID reached similar results as the AID group. These data show the sustainability of the effects of AID use for 24 weeks and the reproducibility of these effects by those newly exposed to AID. Interestingly, the Clarke score gradually decreased in the 49 participants who used AID for 24 weeks, moving from 4.41 (SD ±1.76) at baseline to 3.82 (SD ±1.69) after 12 weeks, and 3.35 (SD ±1.85) after 24 weeks, leading to an AID versus S&P condition difference of −0.71 (95% CI −1.30, −0.21; P = 0.019), but this decrease may not be clinically significant, because the average score remained >3. Of note, the percentage of participants in the AID group with a Clarke score >3 at baseline decreased from 88% to 57% after 12 weeks and to 53% after 24 weeks. Scores on the hypoglycemia confidence and INSPIRE surveys were similarly improved: P = 0.037 and P = 0.045, respectively.

Severe adverse events were recorded from enrollment until the end of the extension phase and are listed in Table 3. Two severe hypoglycemic events occurred during the run-in period and resulted in a decision to drop out of the study in one case. One severe hypoglycemia event and two events of hyperglycemia with ketosis occurred in the AID group during the training phase while AID was still not activated. Two severe hypoglycemia events, two ketoacidosis events (including one in the extension phase, which led to the patient leaving the study), and one episode of hyperglycemia >300 mg/dL with no ketosis occurred while AID was in use. All hyperglycemic events, with or without ketosis or ketoacidosis, were related to infusion-catheter occlusions. Two severe adverse events occurred under AID but were not related to diabetes (one case of severe esophagitis and one case of liver carcinoma).

This study, exclusively focusing on people with T1D at high risk for hypoglycemia—a still poorly investigated population with AID systems—shows the significant benefits of AID technology in reducing TBR while increasing TIR and reducing TAR relative to sensor-augmented insulin pump therapy in a randomized controlled trial.

The study participants were almost all affected by various levels of IHA, as shown by a Clarke score >3 at baseline. IHA increases the risk of severe hypoglycemia three-to sixfold (5,6,33), and 50% of our patients had experienced severe hypoglycemia during the previous 6 months despite all being treated with insulin pumps, with concurrent CGM use by 69 of the 72 randomized participants. Of note, nine participants allocated to the S&P group (39%) and 16 allocated to the AID group (33%) reported severe hypoglycemia during the prior 6 months while using an S&P with PLGS software. All patients had been previously assigned to structured education focused on avoidance of hypoglycemia, which has shown efficacy in reducing the risk of severe hypoglycemia, although effectiveness in reducing IHA was variable (34).

The present trial showed a significant reduction of TBR with a commercial AID system, which occurred shortly after AID initiation (Fig. 1) and was sustained over 24 weeks. Of note, the participants allocated to the S&P group also had reduced TBR without using PLGS during the study. This observation could be interpreted as a study effect because all randomized participants received a refreshment of their education to prevent hypoglycemia, used the study CGM with hypoglycemia alarms, and were enrolled in a denser follow-up than in usual practice. Besides TBR, the %time with SG level <60 mg/dL, LBGI, and the number of hypoglycemic events with >15-min SG levels <70 mg/dL were also significantly reduced with AID compared with S&P. Glucose variability assessed by CV and SD decreased significantly, as well, in the intervention group. The reduction of all these metrics during the randomized trial, which was sustained in the study extension, is supportive of an expected decrease in the incidence of severe hypoglycemia in the long term if the effect of AID is further maintained beyond 24 weeks. In addition, the Clarke score tended to improve, although not significantly over 12 weeks with AID compared with S&P, and further decreased during the extension phase in the AID group. This observation suggests that studies aiming at reversal of IHA should last at least several months to result in significant changes.

Nevertheless, two severe hypoglycemic events still occurred during the randomized phase in the participants allocated to AID. The occurrence of these events must be interpreted in the context of participants with a very high risk for severe hypoglycemia, because three similar events occurred during the run-in or the training phase while AID was not active. In one case (Supplementary Fig. 2), the lack of symptoms with SG reading <50 mg/dL led to omitting glucose intake, resulting in hypoglycemic coma. Of note, the SG reading increase after oral glucose for correction of hypoglycemia occurring 2 h before induced an automated correction bolus that could have contributed to the severe hypoglycemia. In the other case (Supplementary Fig. 3), the patient did not use exercise mode with AID at physical exercise. A nonsevere case if hypoglycemia was overtreated with oral glucose, leading to automated correction bolus, which could have contributed to the severe hypoglycemia. As illustrated in these two cases, correction of hypoglycemia must be cautious with this AID system that could be revised to prevent any automated bolus after correction of hypoglycemia. All hyperglycemic events, with or without ketosis or ketoacidosis, were related to insulin-infusion-set failures and should be preventable by reinforcement education (36).

Although TBR was reduced during the trial, AID also significantly increased TIR and %time in the tighter 70–140 mg/dL range, and decreased TAR. Although the mean SG level did not significantly decrease, perhaps because the average HbA_1c level was close to target at baseline, the reduction of TAR contributed to the significantly lower variability of glucose levels that reached, on average, the CV target of <36%. The significant reduction of glucose variability can be seen as a favorable outcome in view of its relationship to severe hypoglycemia and restoration of hypoglycemia awareness, as pointed out by a recent study affirming this association in a large Japanese population (37).

The limitations of the present study include a relatively short exposure to AID for 12 weeks during the randomized phase, although the obtained results were sustained during the 12-week extension period, and a relatively limited number of participants, especially in the S&P group, due to the 2:1 randomization. Although these outcomes were observed in free life, we cannot exclude a study effect, because the participants had frequent phone calls and visits that would not occur in usual practice. The present trial was performed in university hospitals with trained staff dedicated to diabetes education and care that may be unavailable other outpatient settings. Of note, among the study participants, only six had an HbA_1c level ≥8% at baseline. This indicates that our study conclusions may not apply to people who experience both frequent severe hypoglycemia and high HbA_1c levels.

Nevertheless, this randomized controlled trial in which the effects of AID in patients prone to hypoglycemia related to IHA were assessed demonstrates that AID should be considered a potentially effective treatment line in the strategy of care for people with T1D at high risk for hypoglycemia, prior to considering islet or pancreas transplantation (34).

Acknowledgments. The authors thank the participants in this study for dedicating time and efforts to the progress of AID research. The authors acknowledge the help in patient training and follow-up from the diabetes health care teams of Association d’Aide aux Malades Traités par Infusion Médicamenteuse (Montpellier, France) and the Diabetes Care Unit at Caen University Hospital.

Funding. This study is Protocol 2 of the of iDCL Trial, funded by the U.S. National Institute for Diabetes and Digestive and Kidney Diseases (grant UC4 DK108483). Material support was provided by Tandem Diabetes Care (San Diego, CA).

Duality of Interest. E.R. declares consultant and/or speaker fees from A. Menarini Diagnostics, Abbott, Air Liquide SI, AstraZeneca, Becton Dickinson, Boehringer-Ingelheim, Cellnovo, Dexcom Inc., Eli Lilly, Hillo, Insulet Inc., Johnson & Johnson (Animas, LifeScan), Medtronic, Medirio, Novo Nordisk, Roche, and Sanofi; and research support from Abbott, Dexcom Inc., Insulet Inc., Roche, and Tandem Diabetes Care. M.J. declares consultant and/or speaker fees and/or research support from Abbott, Air Liquide Santé International, Amgen, Asdia, AstraZeneca, Bayer, BMS, Boehringer-Ingelheim, Dexcom, Dinno Santé, Glooko, Insulet, Lifescan, Lilly, LVL médical, Medtronic, MSD, Nestle HomeCare, Novonordisk, Organon, Orkyn, Roche Diabetes, Sanofi, Tandem, Vitalaire, and Voluntis. B.D. declares consultant and/or speaker fees and/or research support from Abbott and Sanofi. Y.R. declares consultant or lecture fees from Medtronic, Insulet, Novo Nordisk, Eli Lilly, Sanofi, and Air Liquide. M.D.B. declares research support handled by the University of Virginia from Dexcom, Novo Nordisk, and Tandem Diabetes Care; patent royalties handled by the University of Virginia from Dexcom, Lifescan, Novo Nordisk, and Sanofi; honoraria from Tandem and Sanofi; and consulting fees from Roche, Portal Insulin LLC, and Dexcom. B.K. declares research support handled by the University of Virginia from Dexcom, Novo Nordisk, and Tandem Diabetes Care; patent royalties handled by the University of Virginia from Dexcom, Lifescan, Novo Nordisk, and Sanofi; and honoraria from Tandem. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. E.R. designed the study, collected and interpreted data, and wrote the manuscript. M.J., O.V., B.D., Y.R., A.F., and J.P. collected and interpreted data, and critically revised the manuscript. M.D.B. carried out the statistical analysis and critically revised the manuscript. B.P.K. was the corresponding principal investigator of the iDCL Trial, contributed to the revision of the article, and approved the final submitted version. E.R. 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.

Prior Presentation. Some of the study data were reported at the 14th Virtual International Conference on Advanced Technologies & Treatments for Diabetes, 2–5 June 2021.