Performance of an automated insulin‐glucagon delivery system versus automated insulin delivery system during challenging inpatient conditions. A single‐blind randomised controlled crossover trial

Abstract

Aims

To assess the efficacy and safety of an automated insulin‐glucagon delivery system (AIGD) compared with an automated insulin delivery system (AID).

Materials and methods

In a 33‐h, randomised, crossover, inpatient study, 13 participants with type 1 diabetes used the DiaCon system in AIGD and AID modes. Each study period included two overnight stays and standardised challenges: receiving 50% of the calculated insulin bolus for breakfast, 100% bolus for lunch, 130% bolus for dinner, and a 45‐min unannounced bicycle exercise at 50% VO_2max. Co‐primary endpoints were (1) number of 15‐g carbohydrate treatments for plasma glucose <3.0 mmol/L, and (2) percentage of time below 3.9 mmol/L.

Results

The number of carbohydrate rescues was lower with AIGD versus AID (15 vs. 20, p = 0.02). Percent time below range (mean ± SD 3.7 ± 2.5% vs. 3.9 ± 3.1%, p = 0.49), in range (TIR) 3.9–10.0 mmol/L (68.8 ± 14.9% vs. 66.9 ± 10.2%, p = 0.41) and above range >10.0 mmol/L (27.5 ± 14.8% vs. 29.2 ± 10.4%, p = 0.46) were similar. Mean glucose and coefficient of variation were comparable between AIGD versus AID (p = 0.30). The post hoc analysis demonstrated that AIGD had significantly higher TIR 0–3 h after exercise and fewer hypoglycaemia events (<3.9 mmol/L) 0–3 h after each meal. No differences were observed in nausea, headache, hunger, and palpitation.

Conclusions

Under challenging inpatient conditions, the AIGD system provided similar glucose control as AID but significantly reduced the need for carbohydrate rescue and enhanced TIR after exercise.

1. INTRODUCTION

The most technologically advanced therapy for people with type 1 diabetes (T1D) is automated insulin delivery (AID) systems.^{1, 2} These systems automatically deliver insulin subcutaneously through an insulin pump based on the glucose concentrations measured by a real‐time continuous glucose monitor (CGM). ³ In recent years newer generations of AID systems have improved their performance in keeping the user within target glucose range—primarily due to improving the insulin dosing algorithm and the CGM accuracy. ³ Although AID systems demonstrate optimal performance during nocturnal periods, the risk of hypoglycaemia persists during daytime conditions—especially during unanticipated challenging activities, for example, aerobic exercise and incorrect meal insulin doses.^{4, 5} Currently, oral carbohydrate rescue is the only option to treat and prevent non‐severe hypoglycaemia in people using AID. However, increasing carbohydrate intake to avoid hypoglycaemia can raise daily caloric intake and promote weight gain, potentially undermining the cardiometabolic benefits of improving glycaemic outcomes.^{6, 7, 8} An automated insulin‐glucagon delivery (AIGD) system that delivers glucagon to mitigate hypoglycaemia risk may improve glycaemic outcomes without promoting weight gain. ⁹ Several research groups have developed AIGD systems that automatically deliver glucagon in addition to insulin. ¹⁰ Even though the dosing algorithm for insulin and glucagon are different between research groups, these AIGD systems have overall been able to reduce the risk of hypoglycaemia and the need for rescue carbohydrate intake compared with AID. ¹¹

Previously, we demonstrated the safety and effectiveness of our own DiaCon's AID system for overnight blood glucose control using a Model Predictive Control (MPC) algorithm. ¹² Since then, we have developed an automated insulin‐glucagon delivery (AIGD) system, which has been tested in virtual settings but never in adults with T1D.^{13, 14, 15, 16} Simulation studies showed that DiaCon's AIGD was able to improve glucose outcomes and reduce hypoglycaemia events compared with the AID. Therefore, the next step was to confirm the promising findings from our simulation studies in the clinic. We hypothesised that DiaCon's AIGD system was both safe and effective, and superior to the AID system in reducing time in hypoglycaemia and the need for hypoglycaemia‐related carbohydrate rescues.

The objective was to compare the efficacy and safety of both systems in adults with T1D undergoing various challenging activities.

2. MATERIALS AND METHODS

2.1. Study design

This was a single‐blind, randomised cross‐over study including 13 adults with T1D, who attended a screening visit and two 33‐h in‐clinic visits with a wash‐period of more than 48 h (Figure 1). The study was conducted at the clinical research unit, Steno Diabetes Center Copenhagen, Herlev, Denmark; monitored by the Good Clinical Practice Unit at Bispebjerg and Frederiksberg Hospital and approved by the Danish Medicines Agency (EudraCT: 2019‐001631‐31 and CIV‐19‐04‐028465), the Regional Committee on Health Research Ethics (H‐19026331), and the Danish Data Protection Agency (P‐2019‐164). The study was registered at ClinicalTrials.gov (NCT04053712) and conducted in accordance with the Declaration of Helsinki.

FIGURE 1.

Study design. Single‐blind randomised controlled cross‐over study. Randomised to start automated insulin delivery (AID) or automated insulin‐glucagon delivery (AIGD) system.

2.2. Participants

Participants were recruited from the outpatient clinic at Steno Diabetes Center Copenhagen from August 2019 to December 2020. Inclusion criteria were age ≥18 years, T1D duration >2 years, use of insulin pumps for ≥1 year, glycated haemoglobin (HbA1c) level ≤8.5% (69 mmol/mol), and current use of faster insulin aspart (Fiasp®, Novo Nordisk, Bagsværd, Denmark). Key exclusion criteria were allergy or intolerance to glucagon, pregnancy or inadequate use of contraception, use of medications affecting glucose metabolism (other than insulin), and concomitant medical or psychological conditions making the individual unsuitable for study participation. Full inclusion and exclusion criteria are available at ClinicalTrials.gov (NCT04053712).

2.3. Screening

After providing an oral and written informed consent, participants went through a screening visit. We recorded the following data to review their eligibility for participation and for baseline characteristics: sex, age, race, diabetes duration, duration of insulin pump use, insulin pump settings, total daily insulin dose (average of previous 7 days), duration of CGM use, allergies, medical history, medications, height, weight, blood pressure, pulse rate, electrocardiography, and HbA1c.

2.4. Randomisation

Participants were block (size = 6) randomised in a 1:1 ratio to each study arm order. An allocation table generated by sealedenvelope.com was uploaded to the Research Electronic Data Capture (REDCap, hosted at the Capital Region of Denmark) system by a person not otherwise involved in the study.

2.5. Procedures

Participants undertook two 33‐h in‐clinic study visits. In random order, the glucose control was managed by DiaCon's system in either AID or AIGD mode (Figure 1).

Prior to the first study visit, 3 days of insulin pump and CGM data were collected for setting up the individual fit of the control algorithms for AID and AIGD.

Two days before each study visit, two Dexcom G6 CGMs (Dexcom, San Diego, CA) were inserted into the abdominal subcutaneous tissue and paired to each Dexcom receiver. After successful pairing and 2‐h sensor warm up, the Dexcom receivers were turned off—though leaving the CGM sensor and transmitter running. Participants were requested not to take paracetamol (acetaminophen) or perform any CGM calibrations prior to study visits. For 24 h before study start, participants refrained from alcohol consumption and strenuous physical exercise. Furthermore, they were asked to consume a low‐fat evening meal containing no more than 60 g of carbohydrates no later than 6 PM on the evening prior to the study. The same meal had to be repeated before the second visit.

Participants arrived in the evening around 9 PM at the research facility (Figure 2). Female participants delivered a urine sample for pregnancy testing. A sampling cannula was placed in the non‐dominant arm's antecubital vein to draw blood every 5–30 min. A capillary glucose level (Contour next®, Ascensia Diabetes Care, Basel, Switzerland) was measured to calibrate both CGMs. The CGM with the closest glucose level to the capillary glucose level was connected to the smartphone (Samsung Galaxy A5 2017 Android phone) that operated the control algorithm for the AID/AIGD system. The other CGM was used as backup in case of CGM malfunctions. Participants' own insulin pumps were removed and replaced with the two study pumps (Dana Diabecare RS®, Sooil, South Korea). The pumps were filled with Fiasp® and GlucaGen® (Novo Nordisk, Bagsværd, Denmark) during the AIGD visit or filled with Fiasp® and isotonic saline during AID, respectively. After 17 h, the infusion set and cartridge for glucagon delivery were changed with freshly reconstituted glucagon. Similar infusion set changes were performed for the saline pump on AID visit. Participants were masked for the contents in the pumps. Afterwards, the wireless connection between the two study pumps and the smartphone was established. Prior to initiating the AID/AIGD system, participants' daily average insulin‐to‐carb ratio, insulin sensitivity factor, and basal rate were registered in the smartphone application (DiaCon APS; Figure S1, Supporting Information).

FIGURE 2.

Study visit procedures for each study visit. Participants arrived in the evening starting either automated insulin delivery (AID) or automated insulin‐glucagon delivery (AIGD) system. They underwent two overnight stays (sleep); received 50% of the calculated insulin bolus for breakfast (underbolus), 100% bolus for lunch (normal bolus), 130% bolus for dinner (overbolus); and performed a 45‐min unannounced bicycle exercise at 50% VO_2max.

At 10 PM the initial blood samples were drawn, and the AID/AIGD system was initiated to automatically adjust the infusions of the two pumps based on the 5‐min CGM readings. Except the intervention (AID versus AIGD), the study visits were identical (Figure 1), including two overnight stays, eating breakfast (energy distribution: 66% carbohydrate, 17% protein, and 17% fat), lunch (energy distribution: 58% carbohydrate, 23% protein, and 19% fat), dinner (energy distribution: 55% carbohydrate, 34% protein, and 10% fat), and bicycling for 45 min on a stationary bike with an estimated 50% VO_2max by Karvonen et al. ¹⁷ The meal sizes were prepared according to participants' body weight, that is, <70 kg: 60 g carbohydrates per meal; 70–80 kg: 75 g carbohydrates per meal; and >80 kg: 90 g carbohydrates per meal. Prior to the meal intake, the AID/AIGD system was informed about the participant's assessed carbohydrate content of the meal. For breakfast, 50% of the estimated carbohydrate was announced to the system (underbolus); for lunch, 100% was announced; and for dinner, 130% was announced (overbolus). During the study visit, participants could move around freely. At night‐time, they were encouraged to sleep in bed.

Plasma glucose (PG) and plasma lactate were measured every 30 min on a YSI 2900 STAT (Yellow Springs, OH) during day and night, but every 10 min during exercise. During hypoglycaemia (PG <3.9 mmol/L (70 mg/dL)), PG was measured every 5 min. During hyperglycaemia (PG >16 mmol/L (288 mg/dL) or PG >12 mmol/L (216 mg/dL)), PG and blood ketones (Precision Xceed ketone meter, Abbott, Abbott Park, Illinois) were measured every 15 min. During exercise, PG was measured every 10 min. Serum insulin and plasma glucagon were measured at predefined timepoints using a Cobas 600 analyzer (Roche Diagnostics GmbH, Mannheim, Germany) and using a sensitive C‐terminal glucagon specific radioimmunoassay, ¹⁸ respectively.

Participants experiencing symptomatic hypoglycaemia or a PG <3.0 mmol/L (54 mg/dL) received 15 g oral glucose (dextrose tablets) rescue treatment and received another rescue treatment if not resolved after 15 min. In contrast, if PG remained >16.0 mmol/L (288 mg/dL) for 2 h despite solving any closed‐system malfunctions, an insulin bolus was administered based on the participant's insulin sensitivity factor (ISF) to aim for a PG of 7.0 mmol/L (126 mg/dL).

Potential side effects to glucagon (nausea, headache, stomachache, and palpitations) were scored using a 0–100 visual analog scale (VAS) every 4 h from 7 AM to 11 PM.

Regardless of the mode of AID/AIGD control, the study visit lasted for 33 h. The study pumps and the CGMs were then disconnected, and the participants were allowed to reconnect their own pump and CGM.

2.6. DiaCon AID and AIGD system description

The DiaCon automated delivery system is based on model predictive controller (MPC) and consists of two main components: an insulin controller and a glucagon controller (Figure S1). The system can operate with either the insulin controller alone (AID) or with both controllers activated simultaneously (AIGD). However, even with both controllers activated, the system does not deliver insulin and glucagon simultaneously. In this specific study, the control algorithms were running on a Samsung Galaxy A5 2017 Android phone which communicated with Bluetooth Low Energy to the two Dana‐RS insulin pumps to adjust delivery of insulin and glucagon based on 5 min readings from a Dexcom G6 CGM (Figure S1). To uphold the blinding, similar procedures were done for saline, even though the saline pump functioned as a “dummy” pump by not infusing anything nor being controlled by the system.

2.7. Study outcomes

The primary outcomes were the percentage of CGM‐derived time below range (TBR1: <3.9 mmol/L or 70 mg/dL) and the number of rescue carbohydrate treatments (15 g carbohydrate counted as one rescue) during the entire study duration of 33 h.

Prespecified secondary outcomes for the entire study duration were the percentage of time with CGM and PG readings in range (TIR: time in range of 3.9–10.0 mmol/L or 70–180 mg/dL), time below range level 2 (TBR2 <3.0 mmol/L or 54 mg/dL), above range level 1 (TAR1: >10 mmol/L or 180 mg/dL), and above range level 2 (>13.9 mmol/L or 250 mg/dL); mean CGM and mean PG; CGM glucose variabilities measured as coefficient of variation (CV); mean amount of insulin delivered per day, and the mean amount of glucagon delivered per day; and mean VAS‐measured nausea, headache, palpitations, and hunger.

Post hoc analyses included abovementioned glycaemic metrics for 0–3 h after exercise, 0–3 h after each meal, and between 10 PM and 07 AM to cover the overnight sleep.

3. STATISTICAL ANALYSIS

In this study, 13 participants were needed to detect a difference of 2.3%‐points (~30 min) in time with CGM values <3.9 mmol/L (70 mg/dL) (primary outcome) with 80% power, a 5%‐significance level, and a presumed 2.7%‐points standard deviation. ¹⁹ Participants dropping out of the study were replaced to ensure a data set of 13 participants completing both study visits.

Data were analysed in an intention‐to‐treat approach including periods with any malfunction of the infusion set (e.g., occlusions, kinking of the tube), missing CGM values, and loss of connectivity between study devices to run the AID/AIGD systems.

Repeated‐measures ANOVA with a compound symmetry covariance structure was used to test the differences between the two study visits, adjusting for the study order. If data had a skewed distribution, logarithmic transformations were used. If the transformations could not normalise the distribution, nonparametric Wilcoxon signed rank tests were used. Spearman correlations were used to determine relationships. The difference of the time course of PG, insulin, and glucagon was tested with a linear mixed model with random effects. A p‐value of <0.05 was considered statistically significant. Non‐predefined outcomes were Bonferroni adjusted for multiple comparisons within each parameter. Data are presented as median (interquartile range) or as mean ± SD if not otherwise stated.

4. RESULTS

4.1. Baseline characteristics

Thirteen participants (7 females) with T1D were included and had a median (range) age of 50 (26–64) years, diabetes duration of 26 (17–45) years, BMI of 27.8 (22.2–36.6) kg/m², total daily insulin dose of 44.0 (25.7–75.8) IU with basal insulin covering 49 (49–58) %; and HbA1c level of 7.1 (6.2–8.5) % or 54 [44–69] mmol/mol. The 14‐day CGM readings showed a TIR of 64.5 (39–76) %, TBR1 1.2 (0.3–4.3) %, TBR2 0.2 (0.0–1.7) %, TAR1 27.0 (15.9–56.7) %, and TAR2 5.9 (2.3–32.6) %. All participants were using Fiasp®. None of the abovementioned baseline characteristics differed significantly across sex or visit order of AID and AIGD.

4.2. Glucose

Regarding 33 h AID/AIGD glucose control, no significant differences were observed between AIGD and AID on CGM‐ and PG‐calculated mean glucose, CV, TIR, TBR1, TBR2, TAR1, and TAR2 (Table 1). Similarly, no differences were seen from 0 to 3 h after each meal nor during the two overnight periods from 10 PM to 07 AM. However, during and after exercise from 0 to 3 h, AIGD had significantly higher TIR and significantly lower TAR1. Further, from 0 to 3 h after each meal, CV and the number of CGM‐registered hypoglycaemia (<3.9 mmol/L) were significantly higher for AID than for AIGD (Figure 3).

TABLE 1.

Mean ± SD values of glucose outcomes and number of carbohydrate rescues during each in‐clinic visit comparing 33 h of automated insulin delivery (AID) and automated insulin‐glucagon delivery (AIGD) control; stratified into exercise, postprandial, and overnight periods.

	Entire study period					Exercise + 3 h					Postprandial 0–3 h					Overnight (10 PM–7 AM)
	AID		AIGD		p‐value	AID		AIGD		p‐value	AID		AIGD		p‐value	AID		AIGD		p‐value
Sensor glucose (Dexcom G6)
TAR2, %	11.7	6.7	8.16	6.6	0.26	9.5	27.9	0	0	‐	20.8	22.2	18.4	16.2	1.0W	9.66	9.72	4.08	6.61	0.083W
TAR1, %	29.2	10.4	27.5	14.8	0.46	16.2	21.9	2.6	5.8	0.040W	25.2	9.6	31.6	10.9	0.052W	10.56	8.55	15.90	13.90	0.150W
TIR, %	66.9	10.2	68.8	14.9	0.41	63.1	26.0	87.6	11.2	0.008Y	50.9	18.0	48.9	23.5	0.405W	77.42	14.24	76.95	15.20	0.86
TBR1, %	3.9	3.1	3.7	2.5	0.49	9.3	11.9	9.4	10.7	0.53P	2.9	3.9	1.02	1.7	0.096W	2.22	3.02	2.17	3.40	0.60W
TBR2, %	0.6	0.5	0.6	0.7	0.46	1.9	2.8	0.5	1.2	0.10W	0.2	0.6	0.16	0.58	‐	0.15	0.53	0.91	1.91	0.083W
Start glucose, mmol/L	10.5	3.6	10.2	2.5	0.73	8.4	5.1	6.7	2.6	0.23W	7.8	3.7	7.8	2.9	0.72	‐	‐	‐	‐	‐
Mean, mmol/L	8.6	1.0	8.3	1.2	0.30	7.8	2.0	5.9	1.4	0.046LX	10.2	2.8	10.4	1.9	0.79	8.2	1.3	7.6	1.3	0.016W
CV, %	39.0	9.0	39.0	5.0	0.30	28.3	13.1	24.3	6.6	0.24	37.3	7.8	30.1	8.7	0.009L	35.6	13.0	32.9	8.2	0.63W
No. events, <3.9 mmol/L	33		29		0.13MC	9		10		0.07 MC	12		5		0.03MC	12		14		0.007MC
No. events, <3.0 mmol/L	6		7		0.13MC	4		2		0.13 MC	1		1		0.13MC	1		4		0.125MC
Plasma glucose (YSI)
TAR2, %	8.4	11.8	7.2	7.4	0.38	2.8	10.1	0	0	0.31W	15.5	18.2	16.8	17.2	0.74 W	6.03	8.87	4.06	7.31	0.74 W
TAR1, %	25.7	15.6	24.2	18.7	0.41	8.4	10.8	3.5	8.7	0.26W	27.4	11.3	25.9	15.2	1.0 W	13.68	11.25	13.60	16.78	0.56 W
TIR, %	64.9	12.1	66.9	15.1	0.36	65.0	24.0	70.2	21.1	0.78W	50.3	13.8	52.2	26.8	0.78 W	72.66	14.62	74.53	17.03	0.78 W
TBR1, %	9.4	6.7	8.9	7.2	0.41	11.2	9.2	19.9	15.1	0.13W	5.5	7.3	4.2	5.4	1.0 W	6.31	6.76	7.37	6.74	0.40 W
TBR2, %	1.7	2.4	1.7	2.6	0.48	4.9	12.9	6.4	10.8	0.65W	1.3	2.6	0.9	2.5	0.56 W	1.33	2.22	0.45	1.09	0.56 W
Start glucose, mmol/L	9.8	3.0	9.3	2.5	0.63	5.2	1.5	5.4	2.5	0.78	6.9	3.6	6.7	3.0	0.80	9.3	4.4	8.9	3.8	0.75
Mean, mmol/L	8.2	1.9	7.8	1.6	0.36	7.1	3.2	5.4	1.1	0.037LY	9.6	2.5	9.7	2.3	0.95	7.5	1.5	7.2	1.8	0.49
CV, %	42.0	7.0	41.0	10.0	0.70	32.4	12.5	30.9	8.7	0.65	39.2	9.1	33.3	10.0	0.023	37.5	12.6	33.5	9.6	0.29
No. carbohydrate rescues	20		15		0.02MC	7		7		>0.99 MC	8		4		0.82MC	5		4		0.30 MC

Note: Measurements include: time above range (TAR: TAR2 >13.9 mmol/L, TAR1 >10.0 mmol/L), time in range (TIR: 3.9–10.0 mmol/L), time below range (TBR: TBR1 <3.9 mmol/L, TBR2 <3.0 mmol/L), and coefficient of variation (CV). Plasma glucose were measured by Yellow Springs Instrument (YSI). Statistical comparison between study visits were performed and following approaches are denoted in the table: Log transformation (L); non‐significant after adjustment for starting glucose (X); significant after adjustment for starting glucose (Y); Wilcoxon test for non‐normally distributed data (W); McNemar's test (MC).

FIGURE 3.

Time profiles of sensor glucose (A), plasma glucose (B), serum insulin (C), and plasma glucagon (D) during 33 h of automated insulin‐glucagon delivery (AIGD; red line) and automated insulin delivery (AID; grey line) control.

4.3. Oral carbohydrate rescues

Overall, the number of carbohydrate rescues were significantly higher during AID versus AIGD (p = 0.02) with no difference in number of rescues during 0–3 h after exercise (p > 0.99), 0–3 h after each meal (p = 0.82), or overnight (p = 0.30) (Table 1).

4.4. Insulin

The total insulin delivery for the 33 h of AID were 62.5 ± 16.8 IU compared to 56.3 ± 19.8 IU for AIGD (p = 0.07) with a basal/bolus‐percentage of 68.5 ± 9.2% compared with 71.0 ± 9.7% (p = 0.32), respectively. However, total insulin delivery was higher during AID than AIGD in both the 0–3 h post‐exercise period and overnight, primarily driven by differences in basal insulin rates. The starting, mean, and peak insulin concentrations did not differ between AID and AIGD 0–3 h after exercise; however, during the overnight period, both the starting and peak insulin concentrations were significantly higher in AID compared to AIGD. Otherwise, the total AUC as well as the starting, mean, and peak insulin concentrations were similar between the interventions (Table 2, Figure 3).

TABLE 2.

Mean ± SD values of insulin, glucagon, and lactate outcomes during each in‐clinic visit comparing 33 h of automated insulin delivery (AID) and automated insulin‐glucagon delivery (AIGD) control; stratified into exercise, postprandial, and overnight periods.

	Entire study period					Exercise + 3 h					Postprandial 0–3 h					Overnight (10 PM–7 AM)
	AID		AIGD		p‐value	AID		AIGD		p‐value	AID		AIGD		p‐value	AID		AIGD		p‐value
Insulin
Total delivery, IU	62.5	16.8	56.3	19.8	0.07	6.1	4.7	2.8	2.2	0.03	33.8	14.3	33.7	17.4	0.97	22.6	6.1	19.8	7.0	0.03
Basal, IU	42.7	11.9	39.6	13.6	0.18	4.2	2.1	2.3	1.6	0.01	16.5	6.1	18.2	10.1	0.39	22.0	6.5	19.0	7.2	0.02
Bolus, IU	19.9	8.4	16.7	8.9	0.17	1.9	3.7	0.5	1.3	0.16	17.3	11.0	15.5	10.0	0.43	0.7	1.4	0.7	1.8	0.93
Basal/Bolus, %	68.5	9.2	71	9.7	0.32	85.4	27.8	92.3	19.2	0.38	53.0	18.5	56.2	15.9	0.40	96.8	7.0	96.2	9.3	0.86
Total AUC, U/L*min	44.8	14.1	47.3	19.7	0.91	−0.26	0.9	0.2	0.6	0.89	0.4	8.9	6.1	19.5	0.11	39.8	21.3	49.5	2.7	0.16
Start insulin, mU/L	23.2	8.8	27.2	16.4	0.24	32.5	14.1	33.8	16.0	0.24	32.5	14.1	33.8	16.0	0.65	20.9	11.9	28.9	15.5	0.003
Mean insulin, mU/L	27.8	7.9	29.1	8.8	0.72	27.8	7.9	29.1	8.8	0.72	29.4	7.2	28.6	7.3	0.77	24.5	11.0	20.6	9.7	0.08
Peak insulin, mU/L	69.8	23.4	55.9	17.1	0.12	55.9	17.1	69.8	23.9	0.12	63.7	26.9	51.8	16.5	0.27	33.4	14.9	25.9	12.3	0.03
Nadir insulin, mU/L	12.1	6.1	10.8	6	0.48	12.1	6.1	10.8	6.0	0.49	12.7	5.9	11.2	6.4	0.51	15.3	6.7	16.3	9.6	0.56
Glucagon
Total delivery, μg	‐	‐	521	360	‐	‐	‐	179	149	‐	‐	‐	94	141	‐	‐	‐	249	257	‐
Total AUC, pmol/L*min	6342	2271	28 823	15 922	<0.001	1060	1576	5589	4479	0.002	7703	4893	12 378	9706	0.06	5388	2255	16 853	13 630	0.004
Mean glucagon, pmol/L	3.5	1.3	25.2	19	<0.001	2.9	1.5	72.8	68.1	<0.001	4	1.8	8	7.3	0.005	3.1	1.2	10.1	8.3	0.002
Peak glucagon, pmol/L	12	5.8	182	194.3	<0.001	4.8	4.4	165.7	204	<0.001	10.9	6.03	36.6	51.1	0.001	6.6	4.9	43.2	50.5	0.003
Lactate
Start, mmol/L	0.87	0.18	0.87	0.17	0.97	1	0.3	1	0.3	0.64	0.7	0.2	0.7	0.2	0.68	0.8	0.2	0.8	0.2	0.67
Mean, mmol/L	0.84	0.13	0.85	0.13	0.86	1.4	0.5	1.4	0.4	0.65	0.9	0.1	0.9	0.1	0.74	0.6	0.1	0.6	0.1	0.78
Peak, mmol/L	2.9	1.5	2.9	1.2	0.42	1.5	1.3	2.9	1.3	0.64	1.3	0.2	1.5	0.3	0.15	1	0.2	1	0.2	0.92
Total AUC, mmol/L*min	1504	207	1478	209	0.66	422	105	424	88	0.86	1418	225	1362	255	0.58	1214	244	1203	249	0.92

Note: Outcomes were expressed as total, starting value, mean, peak, nadir, and area under the curve (AUC).

4.5. Glucagon

During the 33‐h AIGD period, 521 ± 360 μg glucagon was delivered—of which 179 ± 148 μg was used for the 0–3 h post‐exercise period and 248 ± 257 μg for the two overnight periods. The mean, peak, and AUC for glucagon concentration in plasma were – due to the external glucagon supply—higher for AIGD compared to AID (Table 2, Figure 3).

4.6. Lactate

Lactate levels—starting, mean, peak and AUC—were similar between study visits (Table 2).

4.7. Device malfunctions

In total, the AID system malfunctioned 10 times across 8 participants: 5 insulin infusion set occlusions, 2 app crashes, and 3 disconnections—between sensor and smartphone, and between pump and smartphone app. For the AIGD system, 18 malfunctions occurred among 8 participants: 4 insulin infusion set occlusions, 6 glucagon infusion set occlusions, 2 app crashes, and 6 disconnections. All insulin infusion set occlusions led to set change and administration of insulin from an insulin pen.

4.8. Side effects

VAS assessment for nausea, headache, palpitation and hunger were performed seven times during each study visit and showed no significant difference between the study days (Table S1).

5. DISCUSSION

In this randomised, single‐blind crossover 33‐h in‐clinic study of 13 adults with type 1 diabetes, there were no significant differences in the percentages of time in ranges between DiaCon dual‐hormone AID/AIGD system and single‐hormone AID/AIGD system. However, the need for oral carbohydrate rescue was significantly lower for AIGD compared with AID.

While the overall percentage of time in hypoglycaemia and frequency of hypoglycaemia was not different between AIGD and AID, the significantly lower carbohydrate rescue frequency supports the ability of AIGD to mitigate hypoglycaemia risk. Moreover, mean glucose levels were slightly but significantly lower with AIGD without a corresponding increase in glucose variability, indicating that AIGD not only protects against hypoglycaemia but also modestly improved average glycaemia.

The clearest benefit of AIGD in the study was observed during and post‐exercise. From the start of exercise to 3 h post‐exercise, AIGD significantly increased time in range (87.6 ± 11.2% vs. 63.1 ± 26.0%; p = 0.008), while time in hypoglycaemia and number of carbohydrate rescues remained comparable (Table 1). This suggests that AIGD provides better glycaemic stability during exercise without increasing the risk of hypoglycaemia. During the exercise session of 45 min, the TIR and the number of CGM registered hypoglycaemia tended to be better for AIGD compared to AID, but did not reach statistical significance (p = 0.08; Table S2). These results are also supported by a recently published systematic review and meta‐analysis on the use of low‐dose glucagon during exercise, showing that the risk of hypoglycaemia was reduced to a half, and that the percentages of time spent in hypoglycaemia were reduced by 4%‐points when using low‐dose glucagon compared to a non‐glucagon comparator arm. ¹⁰

Insulin requirements tended to be lower with AIGD, consistent with glucagon's role in preventing hypoglycaemia and thus reducing compensatory carbohydrate intake and subsequent insulin dosing (Table 2). During the AIGD period, participants received an average of ~500 μg glucagon, with most delivered overnight (~50%) and during exercise (~35%), corresponding to periods of highest hypoglycaemia risk and when carbohydrate rescue intake is more burdensome, respectively. These periods also had high delivery of insulin for AID compared with AIGD without increased need for carbohydrates. Although the insulin dosing algorithm is identical for AID and AIGD, the controller receives information when glucagon is administered to mitigate hypoglycaemia, whereas carbohydrate rescues are unannounced. The observed differences in insulin delivery during these periods may therefore be explained by the controller's ability to anticipate a glucose rise following glucagon delivery, in contrast to the unannounced glucose increase after carbohydrate intake. Importantly, VAS scores for nausea, headache, hunger, and palpitations showed no significant differences between systems, confirming the tolerability of glucagon microdosing. This stands, however, in contrast to the findings of previous studies showing a higher frequency of gastrointestinal symptoms—such as nausea and vomiting—when using glucagon compared to non‐glucagon study arms. ¹⁰

Device‐related issues occurred more frequently with AIGD (18 vs. 10 malfunctions), largely driven by glucagon infusion set occlusions. Both systems also experienced a notable number of disconnections—between sensor and smartphone, and between pump and smartphone app—as well as insulin infusion set failures. It is possible that using commercially available connectivity solutions; insulin aspart (NovoRapid®, Novo Nordisk) instead of Fiasp®; or soluble glucagon instead of powder glucagon could have mitigated these issues. This study used a non‐commercial and in‐house developed automated delivery system; a prototype that was used the first time in humans for the present proof of concept study. Although all malfunctions were manageable, these findings highlight current limitations of pump‐based glucagon delivery and underscore the need for soluble formulations intended for infusion pump use before broader outpatient use. Similar issues were reported in the 12‐month study by Van Bon et al., where frequent insulin and glucagon infusion set occlusions occurred, though these problems diminished after 3 months of use. ²⁰

This study has limitations. First, a single‐blind design has the potential to introduce bias among study staff and provides less robust conclusions than a double‐blind design. However, in this study, specific criteria for intervention from the investigator was predefined, including when to administer rescue carbohydrates, minimising the potential bias. Second, the small sample size and controlled inpatient environment limit the generalisability of the findings. Finally, the non‐commercial AID/AIGD system used in this study had a substantial number of technical malfunctions, indicating that significant improvements are needed before the system could be considered for broader clinical use. Although this study was completed in 2021, only a few studies have since been published on AIGD systems.^{20, 21, 22, 23} In contrast, numerous investigations of more advanced AID systems have emerged, demonstrating substantial improvements in glycaemic outcomes under real‐world conditions. ²⁴ Nonetheless, hypoglycaemia remains a challenge for these AID systems under certain conditions. Achieving a fully AID/AIGD system that functions without meal or exercise announcements may benefit from the addition of glucagon—a concept that still requires further investigation, including assessment of the added technological complexity and the potential economic burden of incorporating a second hormone. ²⁵

6. CONCLUSION

Under challenging inpatient conditions, AIGD achieved comparable glycaemic outcomes to AID while significantly reducing the need for carbohydrate rescue and improving post‐exercise time in range. These findings suggest that incorporating glucagon may help mitigate hypoglycaemia in AID/AIGD systems; however, this requires further investigation, including algorithm optimisation and resolution of the device‐related technical issues observed with the DiaCon AID/AIGD system.

FUNDING INFORMATION

This study was funded by Danish Diabetes and Endocrine Academy, Grant/Award Number: NNF17SA0031406.

CONFLICT OF INTEREST STATEMENT

AGR has nothing to disclose. KN serves as an adviser to Medtronic, Abbott, Tandem, Convatec, and Novo Nordisk; owns shares in Novo Nordisk; has received research grants to the institution from Novo Nordisk, Zealand Pharma, Dexcom, and Medtronic; and has received fees for speaking from Medtronic and Novo Nordisk. CL was employed by the Steno Diabetes Center Copenhagen during the conduct of the study, but as of 1 November 2022, is employed by, and owns shares in, Novo Nordisk. SS has received a speaker fee from Novo Nordisk and Nordic Infucare; serves as an adviser to Novo Nordisk, Abbott, and Danish Diabetes Association; received a consultancy fee from Hedia; and was employed at Novo Nordisk from May 2022 to April 2023. JJH serves on scientific advisory panels and/or as a speaker/consultant for several companies including Novo Nordisk, Eli Lilly, and MSD. He is co‐founder and owns stock in Antag Therapeutics. DB was employed by the Danish Technical University during the conduct of the study, but since January 2021 has been employed by Novo Nordisk. AS was employed by the Danish Technical University during the conduct of the study, but as of April 2023 is employed by Novo Nordisk. JB has nothing to disclose related to the study.

Supporting information

Data S1. Supporting Information.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.