Single-Hormone Artificial Pancreas Use in Diabetes: Clinical Efficacy and Remaining Challenges

The first AP clinical trials date back to 2006 and only included SH-AP systems until 2010, when glucagon was added in dual-hormone systems. Some research groups have tested systems with a fully closed-loop design (requiring no patient intervention), whereas others adopted hybrid designs (requiring patients’ active input of information such as the timing of exercise and meals to the control algorithm). Despite the relatively recent history of AP testing, two hybrid SH-AP systems have already hit the market. The first was the MiniMed 670G (Medtronic, Northridge, Calif.) (3), which was approved in 2017 in the United States and at the end of 2018 in Canada. The second was the DBLG1 system (Diabeloop, Grenoble, France), which has recently cleared some early hurdles in the European approval process (4).

SH-AP studies to date have been of relatively small size, and very few have lasted 3 months. These trials have tested either hybrid or fully automated systems in different settings (e.g., research centers or outpatient environments), mostly in patients with type 1 diabetes, with a few involving pregnant women with diabetes and people with type 2 diabetes. The following sections describe the main results of these trials, which were heterogeneous in terms of design, duration, reported parameters, types of algorithms used, and patient populations included.

The clinical efficacy of SH-AP in type 1 diabetes has been proven in most studies to be superior to conventional or sensor-augmented pump (SAP) therapy. Although A1C was not the main outcome in most trials due to their short duration, clinical efficacy was reflected by an increase in time spent in the target glucose range and a decrease in time spent in hypoglycemia.

In 2017 and 2018, two meta-analyses of outpatient randomized controlled trials (RCTs) were published examining AP effects on glucose control in children and adults with type 1 diabetes (5,6). The first, by Weisman et al. (5), had stricter inclusion criteria and included 27 trials with a total of 585 participants; the second, by Bekiari et al. (6), analyzed 40 trials involving 1,027 patients. Both meta-analyses reported a high degree of heterogeneity among included studies. Although these meta-analyses integrated both single-hormone and dual-hormone AP studies, post-hoc analyses on each version were also performed and presented. Most of the SH-AP studies used SAP therapy as a comparator.

These meta-analyses reported the superiority of AP to either sensor-augmented or conventional insulin pumps (5,6). The percentage of time in range (70–180 mg/dL) was increased in both reports by means of 12.6% (95% CI 9.0–16.2%, P <0.0001) and 9.6% (95% CI 7.5–11.7%, P <0.001), respectively, which are equivalent to 2.5 and 3 additional hours in range per 24 hours, respectively. The percentage of time spent in hypoglycemia (<70 mg/dL) was decreased by means of 1.5% (P <0.001) and 2.5% (P = 0.0003), respectively, which was equivalent to a decrease in hypoglycemia by 20 and 35 minutes over 24 hours, respectively. These effects were consistently more pronounced during the nighttime. Post-hoc analysis for SH-AP–specific effects revealed an increase in time in range of 11.1% (95% CI 6.9–15.2%, P <0.0001, 22 trials) and 8.5% (95% CI 6.3–10.7%, P not provided, 26 trials). Time in hypoglycemia was also reduced with SH-AP by 1.9% (95% CI –3.4 to –0.36%, P = 0.02, 16 trials) and 1.3% (95% CI –1.6 to –0.9%, P not provided, 24 trials).

Only three trials to date have been long enough to report A1C outcomes, two of which were included in the two meta-analyses. Thabit et al. (7) compared SH-AP to threshold-suspend insulin pump therapy for 12 weeks in 25 pediatric patients (using SH-AP only at night) and 33 adults (using SH-AP day and night) with type 1 diabetes. Time in range increased by a mean of 11% (P <0.001) in adults and 9% (P <0.001) in adolescents and children, and A1C decreased by 0.3% (P = 0.002) in adults. Similarly, Kropff et al. (8) reported a decrease in A1C of 0.2% (P = 0.003) in 32 adults using SH-AP overnight for 8 weeks.

The largest trial to date in outpatients with type 1 diabetes (children >6 years of age and adults) was recently published (9). Participants were randomized to hybrid SH-AP (n = 46) or SAP therapy (n = 40) for 12 weeks. Better glucose control was achieved in the SH-AP group, with a mean difference in A1C change between the two arms of 0.36% (95% CI 0.19–0.53%, P <0·0001). Glucose time in range was higher with AP use (65 ± 8% vs. 54 ± 9%, mean difference 10.8%, P <0·0001). There was no statistically significant difference in time spent in hypoglycemia.

Side effects with SH-AP were not addressed in the meta-analyses. In the trial by Tauschmann et al. (9), no significant side effects were reported except for one episode of hospitalization with diabetic ketoacidosis due to insulin infusion set failure. The rate of severe hypoglycemia could not be assessed in all of the trials to date because of a low overall event rate.

It should also be noted that the findings of these AP studies can not necessarily be extended to all patients with type 1 diabetes since some patient categories were either under-represented or excluded, including patients with high risks of severe hypoglycemia, those with significant comorbidities, and those who were on multiple daily injection (MDI) insulin regimens and therefore not experienced in using an insulin pump.

Given the crucial importance of optimal glycemic control for both pregnant women and their offspring, AP technology could be very useful. However, to date, few studies (with small sample sizes) have addressed AP use in pregnant women with preexisting type 1 diabetes (10,11). In a study by Stewart et al. (11), time spent in the target glycemic range increased by a mean of 15.2% (P = 0.002) over 4 weeks in comparison to sensor-augmented pump therapy in 16 pregnant women with type 1 diabetes. An observation was carried out through labor and 48 hours after delivery for women who elected to keep using SH-AP, showing 84.4% and 82.0% of time in range through the labor and postpartum periods, respectively (12). In a recent randomized cross-over trial of 16 pregnant women with type 1 diabetes who used SH-AP and SAP therapy each for 28 days separated by a washout period, time in range was comparable (62.3 vs. 60.1%, P = 0.47), but less time was spent in the hypoglycemic range and fewer hypoglycemic episodes were observed with SH-AP therapy (13). Therefore, SH-AP is feasible and promising in pregnant women with type 1 diabetes, but larger trials with adequate comparators are needed to accurately assess the extent of potential benefits.

In patients with type 2 diabetes, insulin replacement may be needed with disease progression or during acute events such as hospitalization or surgery. Published reports in type 2 diabetes include one feasibility study in insulin-naive patients (14) and two RCTs in patients who were hospitalized in noncritical wards (15,16). Mean time in range (100–180 mg/dL) was increased from 38% with conventional insulin pump therapy to 59.8% (P = 0.004) with fully closed-loop SH-AP in one study (20 participants in each arm) and from 41.5 to 65.8% (P <0.001) in another (70 vs. 66 subjects) (15,16).

Patients with advanced type 2 diabetes who have significant insulin deficiency are a difficult-to-treat population because many fail to reach glycemic targets despite initiating and intensifying basal-bolus insulin therapy. This population could benefit from AP technology (17,18). Trials investigating AP benefits in patients with type 2 diabetes who require intensive insulin therapy are ongoing.

Besides AP’s favorable effects on glucose control, its psychosocial effects have been brought to attention lately in a few studies (19). AP users report an overall positive experience that includes better sleep quality and self-confidence, more flexibility in life, “time off” from diabetes management, and reduced anxiety (18). On the other hand, technical difficulties and the burden of carrying the multiple components of AP systems have also been raised as drawbacks by many users (20).

AP is now reaching the market; nevertheless, some related challenges remain to be tackled. Aside from extending the testing to a wider population of patients with diabetes—especially those who would be expected to benefit the most from AP use such as those at risk of severe hypoglycemia or with hypoglycemia unawareness, those on steroid therapy, and those with gastroparesis—more work remains to fine-tune the functioning of the various components of AP systems.

As noted above, the beneficial effects of AP are more pronounced at night, and this is attributed to the difficulty of maintaining daily glucose control around meals, exercise, and stress.

Postprandial glucose excursions are still significant with AP. This is due to several factors, including lag times in interstitial glucose sensing of rapidly rising blood glucose levels, insulin pharmacokinetics with delays in subcutaneous absorption, and other less characterized factors related to patient inter- and intra-variability in insulin sensitivity and in responses to different meal components (i.e., glycemic index and fat and protein content) (21,22). Thus, rather than fully closing the loop, the current prandial approach is to have patients announce meals with insulin bolusing. However, to alleviate the burden of exact carbohydrate counting, Gingras et al. (21,22) have demonstrated the feasibility of classifying ingested carbohydrate and other macronutrients as “snack,” “regular,” “large,” or “very large” meals to guide the algorithm in insulin bolus dosing. New ultra-fast-acting insulins such as faster-acting insulin aspart could be promising and would need to be tested in the AP context (23).

Physical activity has been identified as an important challenge for AP algorithms. This is due, on one hand, to the complex factors that influence glucose homeostasis, including exercise type, duration, intensity and timing, physical fitness, and exercise-related increased insulin absorption and sensitivity (24). On the other hand, delays in interstitial glucose sensing during exercise and the pharmacokinetics of current insulin formulations are additional factors that affect the performance of SH-AP during exercise (25). So far, announcing exercise to the algorithm is the best available option. However, active research includes building in the ability to detect exercise (e.g., via heart rate detectors and accelerometers), which, combined with more advanced algorithms, faster insulins, and more accurate CGM devices, might allow for spontaneous bouts of exercise with AP.

The addition of glucagon offers some advantages in decreasing hypoglycemia risk and duration, but at the expense of increasing the complexity of AP systems (26,27). Future algorithms are expected to be endowed with self-learning adaptation, which would predict glucose responses in different conditions and account for inter- and intra-patient variabilities (28,29). Adjuvant therapies to facilitate treatment of patients with type 1 diabetes will also probably affect closed-loop therapy. Pramlintide (amylin analog) and glucagon-like peptide 1 receptor agonists, known to delay gastric emptying and suppress glucagon release, and sodium–glucose cotransporter 2 inhibitors are potential avenues to alleviate the burden of postprandial glycemic excursions in closed-loop AP technology.

AP systems are revolutionizing glucose management in diabetes. Larger studies extending to wider populations of patients with diabetes are ongoing. These efforts should identify patients and conditions that would benefit most from AP and allow a clear assessment of potential side effects. Fine-tuning the different AP components, including pumps, CGM devices, and control algorithms, will require regular re-assessment and comparison of various algorithms. Other factors that are equally essential to address alongside the development and marketing of AP systems are patients’ perceptions, expectations, and needs, in addition to clear cybersecurity measures (30,31).

This work was supported by an operating grant (NIH 1DP3DK106930-01) and the J-A DeSève research chair. N.T. is a recipient of Canadian Institutes Health Research and Fonds de Recherche Santé Québec scholarships. R.R.-L. holds the J-A DeSève research chair in diabetes.

R.R.-L. has received research grants from AstraZeneca, Eli Lilly, Merck, the National Institutes of Health, Novo Nordisk, and Sanofi Aventis; has contributed to consulting/advisory panels for Abbott, Amgen, AstraZeneca, Boehringer Ingelheim, Carlina Technology, Eli Lilly, Janssen, Medtronic, Merck, Neomed, Novo Nordisk, Roche, and Sanofi Aventis; has received honoraria for conferences from Abbott, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Janssen, Medtronic, Merck, Novo Nordisk, and Sanofi Aventis; has obtained consumable gifts (in kind) from Abbott, Animas, Medtronic, and Roche; has received unrestricted grants for clinical and educational activities from Eli Lilly, LifeScan, Medtronic, Merck, Novo Nordisk, and Sanofi; R.R.-L. also holds patents for identifying type 2 diabetes risk biomarkers and for extending insulin catheter life and has received purchase fees for artificial pancreas from Eli Lilly.

N.T. drafted the manuscript. N.T., S.T., and R.R.-L. conceived the review content and design, revised the manuscript, and agreed on the final submitted manuscript. R.R.-L. is the guarantor of this review and takes full responsibility for the accuracy of data collection, presentation, and analysis.