FormalPara Key Points

Type 1 diabetes is treated with exogenous insulin, and recent advances in diabetes technology have culminated in the development of a commercially available, hybrid artificial pancreas (AP).

Adjunctive therapy to an insulin-only AP including glucagon, amylin analogues and glucagon-like peptide 1 receptor agonists has shown promise.

An AP has also been investigated in various cohorts including in pregnancy and type 2 diabetes and in critically ill individuals, but larger studies with evaluations of the psychosocial impact of an AP in these different cohorts are warranted.

1 Introduction

Type 1 diabetes mellitus (T1D) is characterised by insulin deficiency caused by autoimmune destruction of the pancreatic islet beta cells. Worldwide, 425 million people have diabetes and T1D accounts for 5–10% of all cases [1]. The treatment for T1D is exogenous insulin, delivered in the form of either multiple daily injections of insulin or continuous subcutaneous insulin infusion (CSII), also known as an insulin pump. Insulin requirements throughout the day vary depending on time, activity, carbohydrate contents of meals, menstruation, stress and illness. Achieving optimal glycaemic control in this context can be physically and psychologically challenging.

The landmark Diabetes Control and Complications Trial showed that intensive insulin therapy results in a lower glycosylated haemoglobin (HbA1c) compared with conventional treatment. It is also associated with a lower risk of diabetic microvascular complications [2]. The 30-year follow-up on Diabetes Control and Complications Trial participants also showed that intensive insulin therapy has long-term beneficial effects on the incidence of cardiovascular disease in T1D [3]. However, in achieving glycaemic control closer to target, and limiting future complications with intensive insulin therapy, the risk for hypoglycaemia is increased [4] with potential acute complications including seizures, unconsciousness and death [4, 5].

Technology aiding the self-management of T1D has advanced exponentially over the past few decades. Insulin pumps are becoming increasingly accessible worldwide for people with T1D, enabling flexibility that fits around the individual’s lifestyle. Meta-analyses of CSII in T1D have shown that CSII improves glycaemic control [6, 7], whilst at the same time is associated with significantly reduced rates of severe hypoglycaemia [6] and improved quality of life [8] compared with multiple daily injections of insulin.

Similar advances have been observed in glucose-monitoring devices since the development of blood glucose meters in the 1970s to the emergence of continuous glucose monitoring (CGM) in the 1990s. Continuous glucose monitoring provides continuous access to real-time glucose data, information on the direction and rate of change of glucose. Some CGM devices may be used without the need for calibration or confirmatory capillary blood glucose for decision making. The accuracy of CGMs in the recent years has improved compared with previous iterations [9]. Factors contributing to CGM accuracy include its calibration and software algorithm, which improve the MARD (mean absolute relative difference) by filtering and smoothing the signal [10]. A lower % MARD corresponds to better sensor performance, with a MARD of under 10% representing sufficient accuracy for CGM data to make an insulin dosing decision [11]. The MARDs of commercially available, subcutaneous CGM systems at present range between 9.0 and 13.6% [12].

Continuous glucose monitoring devices are also equipped with real-time alerts and an alarm for impending hypo- and hyperglycaemia and can therefore be beneficial in individuals with T1D at high risk of hypoglycaemia, such as those with recurrent severe hypoglycaemia and hypoglycaemia unawareness [13]. With the development of continuous glucose sensors, CSII technology has evolved. Sensor-augmented pump (SAP) therapy, which combines the technology of an insulin pump with a continuous glucose monitoring sensor, was developed, and subsequently the ability to suspend insulin delivery when glucose is low (low-glucose suspend or LGS) or when glucose is predicted to become low (predicted low-glucose suspend or PLGS) was added.

The next evolutionary step from a SAP with PLGS is the artificial pancreas (AP), also known as a closed-loop system or automated insulin delivery, which aims to mimic the endocrine function of a healthy pancreas for glucose homeostasis. The AP system incorporates a sensor for CGM, an insulin pump to deliver insulin and an algorithm connecting the two devices, which directs the pump to deliver insulin based on the real-time glucose readings from the sensor (Fig. 1). Several AP systems are currently in development at different stages [14]. Various aspects of the AP have been studied in clinical trials in the last decade, including its use in the outpatient and home setting, single vs. dual-hormone systems and its use in different cohorts. Recently, the US Food and Drug Administration approved the first hybrid AP system, the Medtronic 670G (Medtronic, Northridge, CA, USA), for use by people with T1D over 14 years of age [15]. It is also CE Mark approved for use in people with T1D over 7 years of age within Europe. The hybrid system is able to deliver and adjust basal insulin automatically without user input when used in the Auto Mode. However, the user must still manually deliver bolus insulin during meals.

Fig. 1
figure 1

Model of an artificial pancreas, comprises a continuous glucose sensor and a continuous subcutaneous insulin infusion that are connected by a control algorithm. Insulin delivery is dependent on the glucose level and the algorithm, which in turn affect the final glucose levels

Full size image

This review article aims to explore the rationale behind the AP, discuss the development status of various closed-loop systems, including their benefits and limitations, report on their efficacy and tolerability in different cohorts, and explore the future outlook in the treatment of T1D. The search terms used to identify publications on PubMed included “artificial pancreas”, “closed-loop insulin” and combinations of these with “bi-hormonal”, “glucagon”, “pramlintide”, “GLP-1 agonist”, “pregnancy”, “type 2 diabetes”, “critically ill” and “psychosocial”.

2 Artificial Pancreas: How Does it Mimic a Pancreas?

The main hormones involved in the regulation of glucose homeostasis are insulin and glucagon. Insulin is synthesised and secreted by the pancreatic beta cells in response to rising glucose levels, and increases glucose uptake into skeletal muscle and fat, inhibiting gluconeogenesis, and stimulating glycogen synthesis. In contrast, glucagon is produced by the alpha cells of the islets of Langerhans and when secreted in response to low blood glucose levels, stimulates hepatic glycogenesis and activates gluconeogenesis. As with most hormonal feedback loops, the interaction between these hormones is constantly in motion and tightly regulated.

The ideal treatment for T1D would be an intervention that can mimic the glucose-regulating function of the pancreas. The development of an AP can be traced back to the 1960s when the possibility for external blood glucose regulation was established in studies in people with T1D using intravenous glucose measurement and infusions of insulin and glucose [16]. Multiple studies have since been conducted to develop a system that is portable, safe and efficient to use in the outpatient setting.

3 Control Algorithm

The control algorithm within an AP is arguably the most important piece of the system. Several control algorithms have been developed and studied, including model predictive control (MPC), proportional-integral-derivative (PID) control and fuzzy logic (FL) control. The former two approaches are more commonly used in clinical studies and the development of an AP [17].

The basic principles of the MPC approach involve a model, which is used to predict the outcome of control moves (insulin infusion) on future outputs (glucose) over a defined prediction horizon [18]. Model predictive control is a general control paradigm and is flexible, allowing it to be used in a dual-hormone AP [19].

The PID control algorithm was initially modelled on the pancreatic beta-cell response and is also referred to as physiological insulin delivery [20]. It calculates insulin delivery based on three set-points: (1) proportional: insulin delivery is adjusted in response to current measured glucose; (2) integral: insulin delivery adjusted corresponding to the area under the curve between measured and target glucose levels; (3) derivative: insulin is delivered based on the rate of change of glucose over time.

A randomised crossover study comparing personalised MPC and PID control algorithms in 30 participants was conducted for 27.5 h with an unannounced 65-g meal in a supervised outpatient suite. The results showed good overall performance in both groups. Model predictive control showed a significantly greater improvement in glucose control with a greater mean time in the range of 3.9–10.0 mmol/L (74.4 vs. 63.7%, p = 0.020), lower mean glucose during the entire trial duration (7.7 vs. 8.9 mmol/L, p = 0.012) and 5 h after the unannounced 65-g meal (10.1 vs. 12.2 mmol/L, p = 0.019). Percentage of time in hypoglycaemia (< 3.9 mmol/L) was minimal in both MPC (4.6%) and PID (2.9%) with no significant differences between the groups [19].

Other types of control algorithm used in studies of an AP include FL control and bio-inspired control algorithms. Although FL is not used as frequently as MPC or PID, its use has increased in recent years. The FL algorithm in an AP system modulates insulin delivery based on rules that attempt to replicate diabetes clinical practitioners [21]. A bio-inspired control algorithm is based on a mathematical model of pancreatic beta-cell physiology [22] but its use in AP studies has been limited thus far [23].

Most control algorithms include safety modules to constrain insulin delivery, limiting the amount of insulin on board or the maximum rate of insulin delivery, and suspending insulin delivery when glucose levels are low or decreasing [24]. Individual parameters that guide insulin delivery (such as basal rates of insulin, insulin: carbohydrate ratios and insulin sensitivity factor) are not fixed, but change over time in people with T1D. Some AP algorithms have been developed to incorporate adaptive features that enable automatic adjustment of basal insulin delivery and insulin: carbohydrate ratios/insulin sensitivity factor in response to changes seen in insulin sensitivity and post-prandial glucose responses. Different approaches to AP adaptation have been explored including the run-to-run approach [25].

4 Types of Closed-Loop Systems

The Artificial Pancreas Project launched by JDRF in 2006 developed a six-stage pathway that defines the different stages of development and types of AP, based on automation of insulin with or without glucagon delivery (Fig. 2).

Fig. 2
figure 2

JDRF’s six-step Artificial Pancreas Project development pathway

Full size image

4.1 Sensor-Augmented Pump with Automated Insulin Suspension

A sensor-augmented insulin pump combines the technology of CSII with a CGM sensor, which transmits the glucose readings to the insulin pump. This comprises steps 1–3 of the JDRF six-stage AP pathway, and includes LGS and PLGS systems.

An LGS system interrupts insulin delivery when the glucose level reaches a predefined threshold value (e.g. 4.0 mmol/L). Insulin delivery is suspended for 2 h if the user does not respond to the low-glucose alarm and resumes automatically after 2 h irrespective of the glucose level, although it can be restarted beforehand by the user. An example of this is the Medtronic MiniMed Paradigm VEO. In a study of 247 people with T1D comparing the use of a SAP with the threshold suspend feature with a standard sensor-augmented insulin-pump therapy over 3 months, the results showed a reduction in nocturnal hypoglycaemic events (by 31.8%) in the threshold-suspend group (1.5 ± 1.0 vs. 2.2 ± 1.3 per patient-week, p < 0.001) without an increase in HbA1c [26].

Subsequently, insulin pumps are available with a PLGS function that reduces or suspends insulin delivery when the glucose reading is predicted to be low. As well as reducing or suspending insulin delivery, PLGS also automatically resumes basal insulin infusion after up to 2 h in the absence of intervention. Insulin delivery can automatically restart after 30 min if the glucose level rises above a predefined threshold value. Randomised studies using PLGS in children, adolescents and adults compared to a SAP have demonstrated a reduction in hypoglycaemia without an increase in hyperglycaemia [27, 28] and no difference in HbA1c at 6 months [29].

Later generation systems in steps 4 and 5 comprise automated insulin-only delivery systems. Step 6 is a fully automated multi-hormone closed loop system, using glucagon in addition to insulin.

4.2 Insulin-Only Artificial Pancreas

An insulin-only system controls the glucose level by increasing or decreasing the amount of insulin infused based on CGM data. This includes hybrid systems that automatically adjust basal insulin with manual bolus insulin delivery at meal times managed by the user, and fully closed-loop systems that automatically adjust basal and prandial insulin.

Most of the AP systems being developed and investigated use a single-hormone (insulin-only) system. Insulin-only AP has been trialled in various settings including research facilities, diabetes camps and home conditions across different age populations from children to adults with type 1 diabetes.

Meta-analyses and systematic reviews on the use of AP in the adult and paediatric populations have shown a better mean glucose level [30], a higher percentage of time in the target range [31, 32] and reduced time in hypoglycaemia and hyperglycaemia [31]. A meta-analysis and systematic review on the use of an AP in 2018 included eight studies and 354 participants of which seven studies (318 participants) were based in the home setting. The AP significantly maintained a better mean glucose level (weighted mean difference [WMD] − 1.03, 95% confidence interval [CI] − 1.32 to − 0.75; p = 0.00001) compared with the control group over 24 h. Time spent in the hypoglycaemia state was also significantly lower (WMD − 1.23, 95% CI − 1.56 to − 0.91; p = 0.00001) [30].

These findings were supported by another meta-analysis on AP treatment for outpatients with T1D [31]. Forty randomised controlled trials of any AP systems (single and bi-hormonal) compared to any manual insulin treatment in non-pregnant outpatients with T1D were included. The use of both single and bi-hormonal APs is associated with a modest but significantly higher proportion of time in the target range (3.9–10.0 mmol/L) over 24 h. In particular, the single-hormone system favoured a higher percentage of time in target both overnight and over 24 h (WMD 12.77, 95% CI 9.82–15.71 and WMD 8.53, 95% CI 6.34–10.72, respectively). Time in hypoglycaemia (glucose < 3.9 mmol/L, WMD − 1.28, 95% CI − 1.65 to − 0.92) and hyperglycaemia (glucose > 10.0 mmol/L, WMD − 7.52, 95% CI − 10.38 to − 4.66) are also reduced.

In the paediatric population, a recent systematic review and meta-analysis examined 25 studies comparing AP and open-loop interventions for children with T1D. Twenty-one out of 25 studies were conducted in the outpatient setting. In a total of 305 paediatric participants with T1D, the percentage time in the target range was increased by approximately 12% in the AP group compared with SAP therapy [32]. The findings also showed that the closed-loop system was associated with significantly reduced percentage times in the hypoglycaemic and hyperglycaemic range (− 0.67% and − 3.01%, respectively).

4.2.1 Approved Hybrid Artificial Pancreas for Clinical Use

In 2017, the US Food and Drug Administration approved the use of the Medtronic MiniMed 670G system (Medtronic), the first commercially available, hybrid AP system in people aged 14 years and above with T1D. This approval was expanded in August 2018 to be used within an older paediatric group aged 7–13 years with T1D. More recently, it received CE Mark approval for use within the same age group in Europe. A hybrid AP is partially automated in that it only delivers basal insulin automatically and requires the user to manually input carbohydrate content into the bolus calculator for the insulin pump to deliver insulin at meal times.

The pivotal study evaluating the safety of a hybrid AP system in T1D included 123 people aged 14–75 years in ten investigational sites [15]. Each subject wore the system for 3.5 months in three study phases. Although there were no statistically powered endpoints in the study, it did demonstrate a reduction in mean HbA1c from 7.4% ± 0.9 to 6.9% ± 0.6 with an increase in the mean percentage of time in range (3.9–10.0 mmol/L) from 66.7% ± 12.2 to 72.2% ± 8.8. There were 24 severe hyperglycaemic events reported (defined in the protocol as a glucose level of > 16.7 mmol/L with blood ketones > 0.6 mmol/L or accompanied by symptoms of nausea, vomiting or abdominal pain). However, there were no reports of diabetic ketoacidosis or severe hypoglycaemic events.

The latest published study evaluating the safety of the Medtronic MiniMed 670G system in 105 children (ages 7–13 years) with T1D over 3 months showed that in-home use of this system was safe and associated with reduced HbA1c levels (from 7.9 ± 0.8 to 7.5 ± 0.6%, p < 0.001) and increased time in the target glucose range (from 56.2 ± 11.4 to 65.0 ± 7.7%, p < 0.001) compared with baseline [33].

Although hybrid AP systems are a significant advance in the development of an AP, they are not a fully closed-loop system. Barriers remaining to full automation include the slow pharmacokinetics of subcutaneous insulin, sensor accuracy and the impact of other factors such as activity. Further developments including the addition of glucagon, better accuracy of CGM and the availability of insulins with more rapid onsets of action have the potential to improve current AP systems [34].

4.3 Dual-Hormone Artificial Pancreas (Glucagon)

Glucagon is secreted from alpha cells of a healthy pancreas, and acts as a counter-regulatory hormone to insulin by elevating glucose levels through promotion of gluconeogenesis and glycogenolysis. Alpha-cell function and glucagon secretion are impaired in long-standing T1D. The use of glucagon in an AP is therefore logical with the aim to reduce the risk of hypoglycaemia.

Dual-hormone closed-loop systems have been extensively investigated in trials and have been shown to have better outcomes than a SAP, but have not clearly been demonstrated to be superior to an insulin-alone system. A randomised crossover trial involving 39 participants aged 18 years and above, assigned to glycaemic regulation with a bi-hormonal bionic pancreas or usual care (conventional or sensor-augmented insulin pump therapy) showed that bi-hormonal AP can be safely used at home. It also demonstrated a significantly lower mean CGM glucose level in the AP period (7.8 mmol/L, standard deviation [SD] 0.6) compared with the usual care period (9.0 mmol/L, SD 1.6) [difference 1.3 mmol/L, 95% CI 0.8–1.8; p < 0.0001] [35]. Additionally, as meal announcement was optional, therefore not requiring carbohydrate counting, this bi-hormonal AP could reduce part of the user burden associated with the management of diabetes.

A randomised crossover study in 19 children aged 6–11 years investigated the use of a bi-hormonal closed loop system vs. conventional insulin pump therapy in a diabetes camp setting. The study showed a better mean CGM-measured glucose level (7.6 mmol/L [SD 0.6] vs. 9.3 mmol/L [SD 1.7], p = 0.00037) and a lower proportion of time with CGM-measured hypoglycaemia (1.2% [SD 1.1] vs. 2.8% [SD 1.2], p < 0.0001) with the bi-hormonal AP system relative to insulin pump therapy [36]. In a recent randomised crossover study, better glucose control (mean percentage of time spent in the plasma glucose target range over 24 h: 63% [SD 18] vs. 62% [SD 18] in single-hormone AP vs. 51% [SD 19] in CSII) and a significant reduction of time spent in hypoglycaemia (episodes of hypoglycaemic events: 9 vs. 13 vs. 52, respectively) were observed with a dual-hormone closed loop system compared with a single-hormone closed-loop system and CSII [37]. The clinical significance of the time in the range difference is small and is achieved with a greater insulin infusion.

The use of single- and dual-hormone AP in exercise has also been investigated. Using wearable sensors to detect exercise, a dual-hormone AP was shown to have a lower mean time in hypoglycaemia during the exercise period (3.4% [SD 4.5]) compared with a single-hormone AP (8.3% [SD 12.6]; p = 0.009) and PLGS (7.6% [SD 8.0]; p < 0.001) [38].

Including glucagon within the AP system is aimed to theoretically reduce the incidence of hypoglycaemia. This is especially important in high-risk groups such as individuals with impaired awareness of hypoglycaemia, those who exercise frequently and very young children However, it comes with added complexity and with an additional cannula site. Currently available formulations of glucagon are also unstable and further development of a stable glucagon analogue is required. Table 1 summarises 24-h (day-and-night) AP studies performed in the home setting from 2016 onwards.

Table 1 Summary of 24-h (day-and-night) artificial pancreas studies in the home setting (from 2016 onwards)
Full size table

4.4 Intraperitoneal Delivery of Insulin in Artificial Pancreas Systems

To overcome the challenges associated with the pharmacokinetics of insulin delivered in the interstitial space, as well as the constraints of wearing an external device, an implantable AP system (based on intraperitoneal insulin delivery, a PID controller and a venous glucose sensor) was developed by Renard et al. [39]. In 2010, the same study group conducted a 2-day semi-automatic (pre-meal boluses of insulin given) closed-loop trial (n = 8) using a simpler system with a subcutaneous sensor, intraperitoneal insulin delivery and a PID algorithm, which showed that a higher percentage of time was spent in the study glucose target (4.4–6.6 mmol/L) during closed-loop vs. open-loop systems (39.1% vs. 27.7%, p = 0.05) [40]. More recently, a non-randomized 24-h sequential AP study (n = 10) comparing a subcutaneous AP system (using a fast-acting insulin analogue) with an intraperitoneal AP system (using regular insulin via the Diaport system) using an MPC algorithm was conducted. Percentage of time spent within the primary endpoint glucose target range (4.4–7.8 mmol/L) was significantly higher for intraperitoneal delivery than for subcutaneous delivery: 39.8 ± 7.6 vs. 25.6 ± 13.1 (p = 0.03) [41]. The evaluation of the first AP system integrating the Eversense implantable subcutaneous glucose sensor is planned as part of the International Diabetes Closed Loop trial, but outcome data have not yet been published.

5 Adjunctive Therapy in an Artificial Pancreas

Postprandial hyperglycaemia following an unannounced meal remains an issue with a single-hormone AP. This was demonstrated in a study involving ten adults and adolescents investigating the safety and performance of an AP system using a probabilistic estimation of meals to allow for automated meal detection [42]. Participants underwent daily exercise and meal challenges (announced and unannounced meals). The results showed that post-prandial hyperglycaemia was significantly more pronounced for unannounced meals compared with announced meals (4-h post-meal CGM: 11.0 mmol/L ± 2.5 mmol/L vs. 7.8 mmol/L ± 1.9 mmol/L, p < 0.001). This difference in post-prandial glucose with unannounced meals arises from the delay in subcutaneous insulin administration with a meal that is not announced to the controller, and the pharmacokinetics of subcutaneous insulin, which may not be well matched to the absorption of macronutrients.

5.1 Amylin Analogues

Amylin is a peptide produced in pancreatic beta cells and co-secreted with insulin. It affects glucose control by slowing gastric emptying, regulating postprandial glucagon, and reducing food intake [43] and is deficient in T1D [44]. Pramlintide is an amylin analogue that can be administered subcutaneously at meal times.

In a 52-week randomised study evaluating the use of pramlintide vs. placebo as an adjunct to insulin therapy in T1D, treatment with pramlintide led to a mean significant reduction in HbA1c from baseline to week 13 (0.67% vs. 0.16%, p < 0.001) without inducing weight gain or increasing the overall risk of severe hypoglycaemia [45].

Because of its effect in lowering post-prandial glucose excursions, and potentially eliminating the need for meal announcement, the use of pramlintide has been evaluated in AP clinical trials. One of the earlier studies showed that the use of pramlintide (30-µg pre-meal injections) in addition to AP was associated with an overall delayed time to peak blood glucose and a reduced magnitude of glycaemic excursion compared with a closed-loop system only [46].

A more recent clinical trial investigated the effects of adjunctive pramlintide with AP in ten participants over a 24-h period [47]. Pramlintide was shown to delay the time to peak plasma glucose excursion (AP 1.6 ± 0.5 hour vs. AP + pramlintide 2.6 ± 0.9 hour, p < 0.001). Pramlintide with AP was also associated with blunting of peak post-prandial increments in plasma glucose (p < 0.001) and reductions in post-meal incremental plasma glucose area under the curve (p = 0.0002).

Even though pramlintide is associated with side effects such as nausea, vomiting and abdominal bloating, the major barrier to its use with AP is the requirement for manual subcutaneous administration at meal times, which may increase the treatment burden of the AP. The effect of co-administration of insulin and pramlintide within an AP in T1D is currently being investigated.

5.2 Glucagon-Like Peptide-1 Receptor Agonists

Glucagon-like peptide-1 is an endogenous hormone that regulates secretion of insulin and glucagon in response to meals. It also slows gastric emptying, inhibits inappropriate post-meal glucagon release and reduces food intake. Glucagon-like peptide-1 receptor agonists are established treatment options in the type 2 diabetes management pathway [48]. Evidence for its use in T1D is limited. In general, HbA1c lowering with glucagon-like peptide-1 receptor agonists in T1D has been modest, with a relative decrease in HbA1c of 0.1–0.2% when tested against a control group [49]. However, it is associated with weight loss in all trials and may be considered in people with T1D who are overweight or obese.

The use of glucagon-like peptide-1 receptor agonists in an AP has been investigated. A randomised crossover trial comparing insulin monotherapy with adjuvant subcutaneous liraglutide 1.2 mg and insulin in an AP system was conducted in 15 participants. The liraglutide arm was associated with overall significantly decreased mean blood glucose levels and better 2-h post-breakfast and post-lunch glucose profiles [50]. There was no difference in hypoglycaemic episodes between the groups.

Similar results were seen in another study evaluating the use of adjunctive liraglutide with an AP [47]. Liraglutide with an AP was associated with marginal reductions in peak glucose excursions (p = 0.05) and incremental peak glucose area under the curve (p = 0.004). There was also a 26% reduction in total daily insulin dose (p = 0.05) and weight loss of 3.2 ± 1.8 kg (p = 0.003) in the liraglutide arm.

6 Artificial Pancreas Use in Specific Groups

6.1 Pregnancy

Type 1 diabetes in pregnancy is associated with an increased risk of foetal and maternal adverse outcomes including congenital malformations, miscarriage, preterm delivery, preeclampsia, macrosomia and perinatal mortality [51]. Maintaining tight glycaemic control during pregnancy minimises risk but can be challenging as insulin requirements increase during the later trimesters [52]. Hypoglycaemia can also occur more frequently during pregnancy [53].

In a randomised crossover study comparing an overnight AP with a SAP, 16 pregnant women with T1D were recruited. This was followed by a continuation phase in which an AP was used day and night. An overnight AP was shown to improve glucose control compared with a SAP (74.7% vs. 59.5% in the target range; p = 0.002). There were no significant differences in time spent in hypoglycaemia, insulin doses or in adverse-event rates [54]. Fourteen out of the 16 women also chose to continue using the AP up to an additional 14.6 weeks, including time during labour and delivery. During this period, glucose levels were in the target range 68.7% of the time with a mean glucose level of 7.0 mmol/L [54].

A more recent study by the same group looked at the longer term feasibility of day-and-night AP use. In this randomised crossover trial, 16 pregnant women completed 28 days of AP and SAP insulin delivery and were given the option to continue using the AP up to 6 weeks post-partum. An AP was associated with comparable glucose control and fewer hypoglycaemic episodes than SAP therapy (median 8.0 [range 1–17] vs. 12.5 [range 1–53] over 28 days, p = 0.04) [55].

6.2 Critical Care

Hyperglycaemia and insulin resistance are common in critically ill people, with or without diabetes. Intensive insulin therapy (glucose maintenance of between 4.4 and 6.1 mmol/L) in critical illness, even without previous diabetes, may reduce mortality during intensive care compared with conventional treatment [56]. However, subsequent studies have shown conflicting outcomes with intensive insulin therapy in the critical care setting, mostly reflecting an increased risk of hypoglycaemia [57]. The use of an AP to optimise glucose without hypoglycaemia has therefore been explored.

The effect of an AP device (STG-22; Nikkiso Co. Ltd, Tokyo, Japan) on the maintenance of blood glucose levels was investigated in 280 intensive care participants. The STG-22 system, which monitors blood glucose levels using a dual-lumen intravenous catheter and delivers insulin intravenously, was associated with maintenance of blood glucose between 3.9 and 10 mmol/L for 87.9% of the study period (33.9 ± 42.4 h), with no hypoglycaemic events [58].

The use of an AP, the majority of which use subcutaneous insulin delivery and interstitial glucose sampling, is currently limited in the critical care setting. This is owing to various limitations that may cause inaccuracies in glucose readings and affect insulin delivery in this cohort (e.g. oedema, vasoconstriction) [59].

The feasibility of an automated closed-loop therapy based on a subcutaneous CGM system compared to intravenous sliding-scale insulin in critically ill adults was evaluated [60]. The authors concluded that the AP system is safe and efficacious, and may improve glucose levels without increasing the risk of hypoglycaemia in this cohort.

6.3 Type 2 Diabetes

The efficacy and safety of an automated AP without meal-time boluses compared with conventional subcutaneous insulin therapy were assessed in 40 participants with type 2 diabetes in a non-critical care inpatient setting for a maximum of 72 h. The use of an AP in this setting was associated with a larger proportion of time spent in the target glucose range compared with the control (59.8% vs. 38.1%, p = 0.0004), with no episodes of severe hypoglycaemia or hyperglycaemia in either group [61].

Another study evaluated the feasibility of AP in insulin-naïve people with type 2 diabetes compared with conventional therapy with oral hypoglycaemic agents [62]. Their results showed a greater period of time spent in the target glucose range of 3.9–8.0 mmol/L (median 78 vs. 35%; p = 0.041) and a smaller period of time in hyperglycaemia (22 vs. 65%; p = 0.041) overnight.

This outcome was replicated in a recent two-centre randomised study that investigated the use of an AP vs. conventional subcutaneous insulin therapy in 136 adults with type 2 diabetes in general inpatient wards. The AP was shown to result in significantly better glycaemic control than conventional subcutaneous insulin therapy, without a higher risk of hypoglycaemia [63], clearly demonstrating the feasibility and potential effectiveness of AP in in-patients with diabetes. However, education and resource allocation need to be overcome to implement AP in this environment. The impact on clinical outcomes in this cohort has also not been demonstrated.

7 Psychosocial Aspects of Artificial Pancreas

It is recognised that psychosocial factors, encompassing environmental, social, behavioural and emotional factors, may affect people with diabetes and their outcomes. The American Diabetes Association recommends that psychosocial care should be integrated in patient-centred care and provided to all individuals with diabetes to optimise health outcomes and health-related quality of life [64].

As technology implementation in diabetes care continues to grow, so should assessments on the impact of these technologies on the psychosocial aspect of these individuals. A review reported that despite its association with body image and self-consciousness, insulin pump therapy is also associated with a high level of satisfaction, improved or similar levels of depression, reduced anxiety, and improved self-efficacy, family functioning and quality of life [65]. The use of CGM shows generally high levels of satisfaction and reduced fear of hypoglycaemia amongst its users [66]. However, poorer sleep and increased anxiety have also been reported in parents of children with T1D using CGM [67].

Users’ perception of an AP is generally positive, with perceived advantages of stable glucose regulation, less need for self-monitoring, relief of daily concerns and time saving [68]. In a study of overnight AP use within the home setting, adolescents with T1D reported a positive impact on their sleep, improved blood glucose control, and reduced parental fear of hypoglycaemia and anxiety. There are also negatives associated with an AP; these include practical difficulties with carrying and using several devices and feeling that the device controls one’s life [69]. Alarm fatigue has been shown to be an important negative factor that decreases adherence to AP systems [70]. Levels of reported trust in the AP also vary across studies [68]. This may in turn affect the level of anxiety of the user. Future longer term research exploring the use of an AP and its psychosocial impacts in different cohorts should be considered.

8 Challenges, Limitations and the Future of an Artificial Pancreas

The AP is regarded as cutting-edge technology in the management of T1D. Although the development of the AP system is progressing, there are challenges and limitations to current systems that need to be overcome before a fully automated AP can be achieved.

A long-standing concern in the development of an AP has been sensor performance. Continuous glucose monitoring works by measuring glucose levels in the interstitial fluid within the subcutaneous tissue. There is a physiological lag of glucose transport from the intravascular to interstitial fluid compartments, and therefore in CGM measurements. The lag time is at least 6–7 min but may be up to 10 min in people with type 1 diabetes [71, 72].

The pharmacokinetics of currently available, rapid-acting insulin analogues is relatively slow with onset within 10–15 min, and a prolonged duration of action, with time to maximal glucose excursion of 40–60 min and duration of action of 4–6 h [73]. This may limit control of rising glucose and avoidance of hypoglycaemia at times of rapidly changing glucose.

Faster acting insulin aspart (Fiasp; Novo Nordisk) is a newer formulation of insulin aspart with two additional formulation excipients, l-arginine and niacinamide [74]. Pharmacokinetic and pharmacodynamic studies comparing Fiasp and insulin aspart have shown a 5-min earlier onset of first appearance of insulin (4 vs. 9 min), approximately two times higher early insulin exposure and a 74% increased early glucose-lowering effect [75]. Data assessing Fiasp compared to insulin aspart in the closed-loop system (e.g. NCT03554486, NCT03579615, NCT03212950) are pending.

Newer oral anti-glycaemic agents such as sodium-glucose co-transporter inhibitors, which act by increasing renal glucose excretion, may have a potential adjunctive role in future AP trials aiming for improved post-prandial glucose control. The practicality of wearing and carrying several devices of a closed-loop system may be a hindrance to future use of the system. Dual-chamber pumps for the use of the dual-hormone closed-loop system are currently in development and may reduce the burden of wear for the users.

A recent critical review investigated potential ethically problematic situations arising through artificial pancreas use. These include confidentiality and data safety, cost coverage and insurability of care, patient selection, patient coaching and support, and personal identity and agency [76]. More consideration is needed to validate the ethical issues raised to improve our understanding of the implementation of the technology.

Despite these challenges, the future of the AP seems promising. Iterations of the hybrid closed-loop system are commercially available and are being improved. A randomised study using an enhanced performance version of the Medtronic hybrid algorithm including insulin bolus correction and improved automode parameters [77] reported improved time in the target glucose sensor range (3.9–10 mmol/L) with intervention compared with baseline values (74.32% ± 8.41% during study vs. 52.15% ± 9.55% at baseline, relative change 42%) and other systems are imminently available (Tandem Control-IQ).

9 Conclusions

In the future, further evaluation of faster acting insulin in the AP, increased accuracy and reduced lag time of CGM as well as self-learning adapting algorithms will improve the level of automation and effectiveness. Longer term home-setting studies using an AP, single or dual hormone, need to be conducted and extended into more targeted groups of people with T1D for us to understand its overall benefits and, importantly, cost effectiveness in the general population.