A Novel Dual-Hormone Insulin-and-Pramlintide Artificial Pancreas for Type 1 Diabetes: A Randomized Controlled Crossover Trial Free

More than 70% of people with type 1 diabetes do not achieve glycemic targets despite advances in insulin analogs, educational programs, insulin pumps, and glucose sensors (1). The artificial pancreas is a novel technology that automates insulin pump delivery in type 1 diabetes based on glucose sensor readings and a dosing algorithm (2). Randomized trials have shown that the artificial pancreas reduces hyperglycemia and hypoglycemia compared with sensor-augmented pump therapy (3). However, significant durations of hyperglycemia are still reported with the artificial pancreas (5–8 h/day above 10 mmol/L) (4,5), particularly postprandially.

Pramlintide is an analog of amylin, a hormone that is cosecreted with insulin in healthy individuals but is deficient in people with type 1 diabetes. Pramlintide is indicated to be injected at mealtimes, where it delays gastric emptying, suppresses glucagon secretion, and increases satiety (6). Studies assessing adding prandial pramlintide injections to insulin-alone artificial pancreas (7–9) reported a reduction in postprandial glucose excursions (7,8).

Here, we present a novel dual-hormone artificial pancreas that delivers pramlintide, in addition to insulin, in a glucose-responsive, basal-bolus manner, and with a fixed ratio relative to insulin to mimic a coformulation. Efforts to develop coformulations of insulin and pramlintide are underway (10–12), which would eliminate the need of delivering pramlintide by injection, enhance patient compliance (13), and allow the use of conventional single-chamber insulin pumps to deliver both hormones. We undertook a randomized trial comparing the insulin-and-pramlintide artificial pancreas with insulin-alone artificial pancreas in adults with type 1 diabetes. Two configurations of the dual-hormone artificial pancreas were tested, one that used rapid-acting insulin and another that used regular insulin. Regular insulin was tested because it may match better with meal absorption in the setting of pramlintide.

We conducted an open-label, randomized, crossover study in type 1 diabetes to compare 1) rapid insulin-and-pramlintide artificial pancreas with rapid insulin-alone artificial pancreas and 2) regular insulin-and-pramlintide artificial pancreas with rapid insulin-alone artificial pancreas. Each intervention lasted 24 h. The intervention visits were separated by a median of 14 days (interquartile range [IQR] 14–27).

From February 2017 to July 2018, participants were enrolled at the Research Institute of McGill University Health Centre. Participants were required to be >18 years old and using an insulin pump for at least 6 months. People with gastroparesis were excluded. Other exclusion criteria were applied (Supplementary Data). Participants provided written informed consent. The study was approved by our ethics committee.

We used blocked randomization to generate allocation sequences, which were disclosed after the admission visit. Participants and investigators were not blinded to the allocation. Participants were blinded to hormonal infusions and glucose data during intervention visits. For safety reasons, investigators had access to glucose levels.

Before each artificial pancreas intervention, we optimized participants’ insulin therapy parameters (basal rates, carbohydrate-to-insulin ratios, and insulin sensitivity factors) for 10–14 days (see Supplementary Data for details). During the optimization periods before the rapid insulin-alone interventions, participants used their usual rapid-acting insulin. During the optimization periods before the rapid insulin-and-pramlintide interventions, participants wore a second pump (MiniMed Paradigm Veo or MiniMed 630G) to deliver pramlintide (Symlin; AstraZeneca) (14). During the optimization periods before the regular insulin-and-pramlintide interventions, participants used regular insulin (Humulin R; Eli Lilly) instead of rapid-acting insulin and wore a second pump to deliver pramlintide. At the end of the study, participants anonymously completed a treatment satisfaction survey with regard to the optimization periods (Supplementary Data).

Participants installed a glucose sensor (G5; Dexcom) 24–48 h before each artificial pancreas intervention. Participants arrived at the research facility around 0700 h and received interventions until 0800 h the next morning. The glucose sensor was calibrated two to three times per intervention using capillary glucose. Self-selected meals were served at 0800, 1200, 1700, and 2100 h. Meals were standardized between visits of each participant, but were different between participants. Participants were allowed to have walks, which were standardized between visits. Nineteen out of 28 participants had walks (average duration 25 min).

The dosing algorithms of the artificial pancreas systems were initialized using basal rates, carbohydrate-to-insulin ratios, and total daily insulin dose at the end of the respective optimization periods. Every 10 min, the glucose sensor reading was entered manually into a laptop, which ran the dosing algorithms to recommend basal rates every 10 min and boluses at mealtimes. Study personnel delivered basal insulin and pramlintide manually by programming a new temporary basal every 10 min. At mealtimes, study personnel delivered the boluses manually through the pumps. The insulin-and-pramlintide artificial pancreas systems delivered basal-bolus pramlintide and insulin, with a fixed ratio of 6 μg of pramlintide per unit of insulin.

Basal dosing algorithms were based on adaptive model predictive control (15) and were identical in the three interventions. The algorithm switches between dynamical models in real time (16), automatically accommodating different kinetics of insulins as well as the pramlintide effect on meal absorption. The amount of prandial insulin was calculated using carbohydrate content, carbohydrate-to-insulin ratio, and premeal glucose levels. The total amount of prandial insulin was calculated in an identical manner in the three interventions, but the pattern of delivery was different between interventions. The rapid insulin-alone artificial pancreas delivered prandial insulin as a single bolus at the onset of the meals. The rapid insulin-and-pramlintide artificial pancreas delivered three pairs of prandial insulin and pramlintide mini-boluses beginning at mealtime and separated by 10 min. The regular insulin-and-pramlintide artificial pancreas also delivered prandial insulin and pramlintide with three pairs of mini-boluses but beginning 20 min prior to the meals as opposed to mealtime (to account for the slow pharmacokinetics of regular insulin). The immediate bolus was 25–100% of the total bolus, and the two extended boluses combined for 0–75% of the total bolus. The split between immediate and extended boluses depended on the premeal glucose level (lower premeal glucose levels led to lower immediate boluses and higher extended boluses). For small meals, all prandial insulin and pramlintide were delivered as immediate boluses since the amount of prandial pramlintide was small.

Venous blood samples were taken every 10–30 min to measure plasma glucose (2300 STAT Plus Analyzer; YSI). During the three interventions, if plasma glucose fell below 3.3 mmol/L and was accompanied with symptoms, or fell below 3.0 mmol/L irrespective of symptoms, 16 g oral glucose (four tablets × 4 g) was given. Participants were asked 2 h after meals and after the night if they experienced nausea, vomiting, bloating, or heartburn in the preceding hours, and ranked the symptoms as mild, moderate, moderate-to-severe, or severe.

The primary outcome was the time spent in target range (3.9–10.0 mmol/L) (17). Secondary outcomes included times spent below and above the target range, glucose variability, and gastrointestinal side effects. Outcomes were calculated using interpolated plasma glucose measurements, but sensor readings were used when plasma levels were not available (e.g., lack of venous access).

We anticipated that the insulin-and-pramlintide artificial pancreas would increase the percentage of time in target range by 7% (SD 10%) (7,18–21) compared with the rapid insulin-alone artificial pancreas. We intended to do the following pairwise comparisons: 1) rapid insulin-and-pramlintide artificial pancreas with rapid insulin-alone artificial pancreas and 2) regular insulin-and-pramlintide artificial pancreas with rapid insulin-alone artificial pancreas. Therefore, we did a power analysis using the formula for the paired Student t test with 5% significance level (22). We calculated that 19 participants would provide 80% power. Hence, we aimed to include 28 participants to account for any uncertainty in the power calculations.

Our analyses were on a modified intention-to-treat basis. Participants who did not complete the rapid insulin-alone artificial pancreas intervention and at least one insulin-and-pramlintide intervention were not included in the analysis and were replaced in the enrollment process.

A linear mixed model was fitted to the data while adjusting for the period effect. To examine for carryover effect, a model was fitted with the treatment by period interaction term. Residual values were examined for normality, and if skewed, the data were transformed using the square root function. P values <0.05 were regarded as significant (22). Results are reported as median (IQR) or mean (SD).

Supplementary Figure 1 shows the flow of participants. Twenty-eight participants completed the rapid insulin-alone artificial pancreas intervention and at least one insulin-and-pramlintide artificial pancreas intervention, and were included in the analysis (43% female, mean age 25 years [13], HbA_1c 7.8% [0.9] [62 (10) mmol/mol], duration of diabetes 23 years [14], total daily insulin 0.65 units/kg [0.16]) (Supplementary Table 1). Out of those 28 participants, 1 did not complete the rapid insulin-and-pramlintide intervention and 2 did not complete the regular insulin-and-pramlintide intervention.

Mean basal rate at the end of the rapid insulin-alone optimization period was 0.99 units/h, at the end of the rapid insulin-and-pramlintide optimization period was 0.94 units/h, and at the end of the regular insulin-and-pramlintide was 1.01 units/h. The mean carbohydrate-to-insulin ratio at the end of the rapid insulin-alone optimization period was 9.8 g/unit, at the end of the rapid insulin-and-pramlintide optimization period was 10.4 g/unit, and at the end of the regular insulin-and-pramlintide was 9.1 g/unit. Figure 1 compares the glucose profiles during the artificial pancreas visits. Sample tests are in the Supplementary Data.

The rapid insulin-and-pramlintide artificial pancreas increased the mean percentage of time spent in the target range compared with the rapid insulin-alone artificial pancreas from 74% to 84% (P = 0.0014), reduced mean glucose from 7.9 to 7.4 mmol/L (P = 0.0053), reduced time spent >10.0 mmol/L from 22% to 12% (P = 0.00012), reduced glucose coefficient of variance from 30.3% to 26.8% (P = 0.035), and reduced SD from 2.4 to 2.0 mmol/L (P = 0.0053) (Table 1). There were no benefits associated with the regular insulin-and-pramlintide artificial pancreas compared with the rapid insulin-alone artificial pancreas in time spent in target range (P = 0.22), mean glucose (P = 0.95), time >10.0 mmol/L (P = 0.49), glucose coefficient of variance (P = 0.09), or SD (P = 0.17) (Table 1). No treatment by period interaction was found, and no difference was observed due to the order of interventions (data not shown).

The benefits of the rapid insulin-and-pramlintide artificial pancreas were due to improved glucose control during the day. During the day (0800–2300 h), the rapid insulin-and-pramlintide artificial pancreas increased the time in target range compared with the rapid insulin-alone artificial pancreas from 63% to 78% (P = 0.0004), reduced mean glucose from 8.7 to 7.9 mmol/L (P = 0.0011), reduced glucose coefficient of variance from 29.3% to 25.6% (P = 0.017), and reduced glucose SD from 2.5 to 2.0 mmol/L (P = 0.0019) (Table 1). During the night (2300–0800 h), the rapid insulin-and-pramlintide artificial pancreas and the rapid insulin-alone artificial pancreas achieved a similar time in target range (94–95%) (Table 1).

Figure 2 shows postprandial glucose profiles stratified by premeal glucose levels. During rapid insulin-alone artificial pancreas visits, postprandial glucose increased after the meals and peaked after 60 min, irrespective of premeal glucose levels. However, during insulin-and-pramlintide artificial pancreas visits, postprandial glucose profiles depended on premeal glucose levels. When premeal glucose levels were >10 mmol/L, postprandial glucose levels immediately decreased toward euglycemia. When premeal glucose levels were between 5 and 10 mmol/L, postprandial glucose profiles were relatively flat. When premeal glucose levels were <5 mmol/L, postprandial glucose levels immediately increased toward euglycemia. These tailored postprandial glucose profiles resulted from the use of pramlintide and the quasi-dual-wave prandial dosing algorithm that splits immediate and extended boluses (pramlintide and insulin) depending on the premeal glucose level. During rapid insulin-and-pramlintide visits, premeal glucose levels >10 mmol/L led to mean 81% of the prandial doses as immediate (and 19% as extended), whereas premeal glucose levels between 5 and 10 mmol/L led to 66% as immediate (and 34% as extended), and premeal glucose levels <5 mmol/L led to 34% as immediate (and 66% as extended). During regular insulin-and-pramlintide visits, the immediate components were 99%, 75%, and 45%, respectively.

Insulin delivery was similar on rapid insulin-and-pramlintide and rapid insulin-alone artificial pancreas visits. However, insulin delivery was higher on regular insulin-and-pramlintide artificial pancreas visits compared with rapid insulin-alone visits (Table 1).

On rapid insulin-and-pramlintide artificial pancreas visits, total pramlintide delivery per day was 279 µg (92) (3.5 µg/kg [0.9]) and per night was 53 µg (19) (0.66 µg/kg [0.18]). The mean prandial pramlintide dose (immediate plus extended boluses) for a meal was 34 µg (17). Twelve percent of the prandial pramlintide doses were <15 µg, 35% were between 15 and 30 µg, 27% were between 30 and 45 µg, and 26% were >45 µg.

On regular rapid insulin-and-pramlintide artificial pancreas visits, total pramlintide delivery per day was 320 µg (107) (4.1 µg/kg [1.1]) and per night was 52 µg (16) (0.66 µg/kg [0.18]). The mean prandial pramlintide dose for a meal was 39 µg (21). Thirteen percent of the prandial pramlintide doses were <15 µg, 24% were between 15 and 30 µg, 29% were between 30 and 45 µg, and 34% were >45 µg.

There were 11 hypoglycemic events requiring oral treatment on rapid insulin-alone artificial pancreas visits, 12 on rapid insulin-and-pramlintide artificial pancreas visits, and 18 on regular insulin-and-pramlintide artificial pancreas visits (Table 2). On rapid insulin-alone artificial pancreas visits, there was no nausea, vomiting, bloating, or heartburn. On rapid insulin-and-pramlintide artificial pancreas visits, there were 108 meals, 6 of which were followed by gastrointestinal symptoms (3 mild and 3 moderate, all were transient) (Table 2). On regular insulin-and pramlintide artificial pancreas visits, there were 104 meals, 11 of which were followed by gastrointestinal symptoms (2 mild, 6 moderate, and 3 moderate to severe). No gastrointestinal symptoms were reported during the nights in any intervention. The three participants that reported moderate gastrointestinal symptoms on the rapid insulin-and-pramlintide artificial pancreas visits also reported symptoms on the regular insulin-and-pramlintide visits.

There was no elevated ketones (>1.0 mmol/L) in any artificial pancreas visits. During insulin-and-pramlintide interventions, there was no failure in pramlintide delivery via the pump. One participant experienced irritations at the pramlintide infusion sites during the optimization periods.

Participants reported higher treatment satisfaction with the rapid insulin-and-pramlintide therapy than the rapid insulin-alone therapy (Supplementary Data). Eighty-two percent (23 of 28) of participants strongly agreed (8 of 28), agreed (13 of 28), or slightly agreed (2 of 28) that using rapid insulin and pramlintide made their blood glucose control more even or predictable. Seventy-nine percent (23 of 29) of participants strongly agreed (9 of 29), agreed (9 of 29), or slightly agreed (5 of 29) that using rapid insulin and pramlintide provided benefits that rapid insulin-alone has not provided them. Ninety percent (26 of 29) of participants strongly agreed (12 of 29), agreed (8 of 29), or slightly agreed (6 of 29) that if a coformulation product of rapid insulin and pramlintide was available on the market, they would use it. Treatment satisfaction was not improved with the regular insulin-and-pramlintide therapy compared with the rapid insulin-alone therapy (Supplementary Data).

The analyses of glucose levels during the optimization periods are reported in Supplementary Table 2. The time in target range was higher during the rapid insulin-and-pramlintide therapy compared with the rapid insulin-alone therapy (difference 5% [12], P = 0.039), and mean glucose was lower with the rapid insulin-and-pramlintide therapy compared with the rapid insulin-alone therapy (difference −0.7 mmol/L [1], P = 0.037). However, time spent <3.9 mmol/L was higher during the rapid insulin-and-pramlintide therapy compared with the rapid insulin-alone therapy (difference 1.2% [−0.1 to 2.8], P = 0.0092). None of the outcomes differed between the regular insulin-and-pramlintide therapy and the rapid insulin-alone therapy.

Out of 103 days when participants were asked about gastrointestinal side effects during the rapid insulin-and-pramlintide run-in periods, participants reported 5 days with gastrointestinal symptoms (3 mild and 2 moderate). Out of 104 days when participants were asked during the regular insulin-and-pramlintide run-in periods, participants reported 14 days with symptoms (3 mild, 5 moderate, and 1 severe).

Insulin-alone artificial pancreas systems improve glucose control compared with conventional pump therapy (3), but hyperglycemia (>10 mmol/L) remains common and occurs on average 5–8 h per day (4,5). In this study, we compared two novel insulin-and-pramlintide artificial pancreas systems with rapid insulin-alone artificial pancreas. The rapid insulin-and-pramlintide artificial pancreas increased the time in target range, reduced mean glucose level, and reduced glucose variability. The regular insulin-and-pramlintide artificial pancreas did not improve glucose control compared with the rapid insulin-alone artificial pancreas.

Prior to this study, we conducted preliminary experiments (data not shown) where we noted a risk of early postprandial hypoglycemia after pramlintide boluses (23,24). A 20–30% reduction of the insulin-to-carbohydrate ratios to mitigate this risk led to late postprandial hyperglycemia (6,25,26). We thus adopted a strategy that spreads prandial insulin and pramlintide over 20 min, mimicking a dual-wave bolus, to avoid early postprandial hypoglycemia without the need to excessively reduce the insulin-to-carbohydrate ratios (27).

As a physiological mechanism against hyperglycemia, high glucose levels delay gastric emptying in healthy individuals and in type 1 diabetes (28,29), although this mechanism is weakened in type 1 diabetes due to the absence of endogenous amylin (28,29). Conversely, as a physiological mechanism against hypoglycemia, low glucose levels accelerate gastric emptying in healthy individuals (30) and in type 1 diabetes (30,31). Our pramlintide prandial dosing logic aligns with these two physiological defense mechanisms (Fig. 2). At high premeal glucose levels, more immediate pramlintide is delivered to delay gastric emptying and prevent aggravation of postprandial hyperglycemia (28,32). At low premeal glucose levels, less immediate pramlintide is delivered to avoid interference with the physiological hypoglycemia-induced acceleration of gastric emptying.

Somewhat surprisingly (23,33), we observed only a modest increase in gastrointestinal side effects with the rapid insulin-and-pramlintide artificial pancreas. This may have been due to various reasons. First, participants were exposed to pramlintide for 10–14 days prior to the insulin-and-pramlintide artificial pancreas visits. Second, our prandial delivery strategy spread the delivery of pramlintide over 20 min, which must have led to lower peak plasma levels compared with if pramlintide was delivered as a single bolus. Third, our prandial pramlintide doses decreased with smaller meal sizes and were lower for participants with low insulin doses. Fourth, continuous basal pramlintide delivery may have increased the participants’ tolerance to pramlintide (34).

Basal pramlintide did not improve glucose control at night, due to the sufficiency of the rapid insulin-alone system, but the increased basal pramlintide in response to increasing glucose levels prior to the meals (10–12, 15–17, and 19–21 h) (Fig. 1) may have contributed to improved glucose control during the day. This glucose-responsive pramlintide delivery aligns with natural physiology (in healthy individuals, increasing glucose levels from 5.0 to 11.0 mmol/L led to higher plasma amylin levels from 2.1 to 18.9 pmol/L [28]). Nevertheless, our study does not compare glucose-responsive basal pramlintide delivery against constant basal pramlintide delivery, nor does it compare basal-bolus pramlintide delivery against bolus pramlintide delivery. It may be possible for our findings to be reproduced by delivering pramlintide at mealtimes only, using a dual-chamber pump, but this will necessitate an additional infusion set and an additional drug manipulation.

A required long optimization would be a barrier to adoption. However, our system should not normally require an optimization period, and our previous studies (19–21) with the rapid insulin-alone system without prior optimization had comparable results to this study. The optimization periods were needed in this study to mitigate against two potential biases. First, the systems are initialized using daily insulin dose, carbohydrate-to-insulin ratios, and basal rates, but these parameters prior to the study are tailored for rapid insulin-alone therapy and thus are not appropriate to initialize the two dual-hormone systems. To initialize the dual-hormone systems properly, we need parameters that result from a prolonged use of insulin and pramlintide concomitantly, and this requires a run-in period. Second, without a run-in period, the gastrointestinal outcomes during the dual-hormone artificial pancreas interventions would reflect the acute effect of pramlintide. However, most gastrointestinal symptoms are transient, and we are interested in estimating the nontransient effects as they better reflect prolonged use in real-world settings.

At the end of the optimization phase of the rapid insulin-and-pramlintide intervention, we reduced the insulin-to-carbohydrate ratios compared with the rapid insulin-alone by only 6%. This finding appears at first glance to contradict those of other studies reporting a reduction of 15–25% (6,26) when adding pramlintide to prandial insulin, but our study uniquely delivered prandial insulin and pramlintide as dual-wave boluses. In another study that added pramlintide to square-wave prandial insulin boluses, a minimal decrease of the insulin dose (7.8%) was also needed (27), despite using large pramlintide boluses (60 µg).

On the other hand, during the artificial pancreas interventions, nocturnal basal insulin rates were not lower with the rapid insulin-and-pramlintide system compared with the rapid insulin-alone system, despite delivering basal pramlintide and achieving a similar mean glucose level (4–8 h) (Fig. 1). Other studies with basal pramlintide also reported little (27) or no reduction (14,34) in basal insulin needs. Moreover, one study showed that nocturnal basal pramlintide delivery had no effect on glucagon levels (33). Another study reported that endogenous glucose production (a surrogate of basal insulin needs) is not reduced with pramlintide boluses (28), despite a reduction in glucagon levels. These findings and the results of our study indicate that basal pramlintide delivery has little or no effect on basal insulin needs.

Our data did not support the use of regular insulin and pramlintide in the artificial pancreas. Other studies reported benefits of delivering regular insulin and pramlintide, but these studies used regular insulin and placebo as a comparator (25,26,33,35), were limited to prandial delivery (25,35), reported late postprandial hyperglycemia (25,26), or did not deliver insulin and pramlintide at the same time (i.e., not mimicking a coformulation) (35). In our study, postprandial control was improved with regular insulin and pramlintide (Fig. 2), but this benefit was offset by the slow glucose-lowering effect of regular insulin between meals and during the night. Compared with rapid-acting insulins, regular insulin is associated with ∼0.5-fold lower plasma insulin levels 60 min after delivery, and 1.5- to 1.8-fold longer time-to-maximum glucose excursion (36).

With the rapid insulin-alone artificial pancreas, more patients will approach HbA_1c levels of 7% (4,5). For HbA_1c levels near 7%, studies in type 2 diabetes have shown that the relative contribution of postprandial glucose to HbA_1c levels is 70% (37,38), compared with 30% for HbA_1c levels near 10%. In other words, as the HbA_1c level decreases, postprandial glucose contributes more than fasting glucose to HbA_1c. Since adding pramlintide to the artificial pancreas targets postprandial hyperglycemia, the rapid insulin-and-pramlintide artificial pancreas may be the next logical treatment in patients approaching, yet not achieving, HbA_1c targets despite using the insulin-alone artificial pancreas.

Participants reported higher treatment satisfaction with the rapid insulin-and-pramlintide therapy during the optimization period than the rapid insulin-alone therapy (Supplementary Data), with the absolute majority stating that they would use a coformulation if it were commercially available. This was paralleled with an increased time in target range and reduced mean glucose, but with a slight increase in hypoglycemia. Unlike during the artificial pancreas interventions, this increased hypoglycemia may be explained by the dual-wave boluses during the optimization period not reducing the immediate insulin and pramlintide boluses at low premeal glucose levels (immediate boluses were always 50%). Gastrointestinal symptoms during the optimization period were low and comparable to those recorded during the artificial pancreas interventions (Supplementary Data).

Our study has several limitations. First, our study was conducted in inpatient settings for only 1 day, and it did not include vigorous exercise. It is unknown how pramlintide and vigorous exercise together affect hypoglycemia and nausea. Second, our study lacked allocation blinding. Third, we delivered insulin and pramlintide using two pumps, and their mixing in a coformulation may alter their action. However, two studies showed that injecting pramlintide and insulin mixed in a syringe did not affect their absorptions or actions (39). Fourth, we used manual control to operate the artificial pancreas systems, but this was unlikely to have affected the clinical outcomes since hormonal deliveries would have been the same if we used an automated system. Fifth, eight enrolled participants were not included in the analysis due to dropout/exclusion (four of whom dropped out before the first intervention) (Supplementary Data).

Our study is the first to propose two novel insulin-and-pramlintide artificial pancreas systems. The rapid insulin-and-pramlintide artificial pancreas improved glucose control compared with the rapid insulin-alone system without increasing hypoglycemia and with a modest increase in transient mild-to-moderate gastrointestinal side effects in some patients. The regular insulin-and-pramlintide artificial pancreas did not improve glucose control compared with the rapid insulin-alone artificial pancreas. Studies with the rapid insulin-and-pramlintide artificial pancreas in free-living outpatient settings are now warranted.

Funding. This study was supported by funding from JDRF (2-SRA-2016-246-M-R).

Duality of Interest. A.H. received research support/consulting fees from Eli Lilly, Medtronic, AgaMatrix, and Dexcom and has pending patents in the artificial pancreas area. M.A.T. received research support from AgaMatrix, consulting fees from Sanofi, and speaker honoraria from Eli Lilly, Novo Nordisk, Boehringer Ingelheim, Janssen, and AstraZeneca. J.-F.Y. received research support from Sanofi, Bayer, and Novo Nordisk and consulting fees and speaker honoraria from Sanofi, Eli Lilly, Novo Nordisk, Boehringer Ingelheim, Janssen, Takeda, Abbott, Merck, and AstraZeneca. A.B. is a nurse clinician and insulin pump trainer for Medtronic Canada and Omnipod Canada. E.P. received consulting fees from Animas and speaker honoraria from Medtronic and Animas. L.L. has pending patents in the field of artificial pancreas, received consulting fees from Dexcom, and has received support for clinical trials from Merck, AstraZeneca, and Sanofi. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.H. and L.L. supervised the study. A.H., M.A.T., J.-F.Y., N.S., and L.L. designed the study. A.H. designed the dosing algorithm. A.H., M.A.T., S.B.-T., J.-F.Y., J.R., A.B., E.P., N.S., and L.L. conducted the study. A.H., J.R., and A.E.F. performed the data analysis, including the statistical analyses. All authors read and approved the final version of the manuscript. A.H. 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.