Automated Insulin Delivery Systems: Today, Tomorrow and User Requirements

Abstract

Automated insulin delivery (AID) is the most recent advance in type 1 diabetes (T1D) management. It has the potential to achieve glycemic targets without disabling hypoglycemia, to improve quality of life and reduce diabetes distress and burden associated with self-management. Several AID systems are currently licensed for use by people with T1D in Europe, United States, and the rest of the world. Despite AID becoming a reality in routine clinical practice over the last few years, the commercially hybrid AID and other systems, are still far from a fully optimized automated diabetes management tool. Implementation of AID systems requires education and support of healthcare professionals taking care of people with T1D, as well as users and their families. There is much to do to increase usability, portability, convenience and to reduce the burden associated with the use of the systems. Co-design, involvement of people with lived experience of T1D and robust qualitative assessment is critical to improving the real-world use of AID systems, especially for those who may have greater need. In addition to this, information regarding the psychosocial impact of the use of AID systems in real life is needed. The first commercially available AID systems are not the end of the development journey but are the first step in learning how to optimally automate insulin delivery in a way that is equitably accessible and effective for people living with T1D.

Introduction

Self-management of type 1 diabetes (T1D) is challenging, with established glucose targets flanked by acute hypoglycemia and its associations with distress, morbidity, and mortality, and hyperglycemia and its longer-term consequences of micro- and macrovascular complications.^1-3 Maintaining glucose close to, or within, the euglycemic target range in the face of the constant perturbations caused by food, exercise, insulin, stress, intercurrent illness, hormonal changes, and many other factors, is the major goal of medical management of T1D, and evidence-based approaches to achieve this goal include support, multidisciplinary education, ⁴ pharmacology, ⁵ and devices.

Automated insulin delivery (AID) is the most recent advance in type 1 diabetes devices and is achieved by combining a continuous glucose sensor, ⁶ a control algorithm and an insulin delivery device. It has the potential to enable and empower people living with T1D to achieve their glycemic targets without disabling hypoglycemia, to improve quality of life and reduce diabetes distress, and to address some of the burden of self-management.

In this review we discuss the state of the art in automated insulin delivery systems and consider how future AID systems may address unmet needs. We will also consider the needs of the user, their requirements for safe and effective use of the system, and practicalities of operation.

Efficacy, Safety, and Usability of the Available Systems in the Market

Several AID systems are currently licensed for use by people with T1D in Europe. The first to market was the Medtronic 670G system (recently upgraded to the 780G system). The algorithm used is a proportional integral derivative controller with insulin feedback for the 670G with some elements of fuzzy logic control in the new 780G device. Both use the Guardian 3 sensor with a sensor duration of 7 days and requiring capillary blood glucose monitoring. The time-in-range (TIR) 70 to 180 mg/dl reported in the pivotal trials was 67% in adolescents and 74% in adults with 670G, with 73% achieved in adolescents and 75% in adults using the 780G system. Less than 1% of time in hypoglycemia <54 mg/dl was observed for both devices. There is limited randomized clinical trials evidence of efficacy and safety, especially in specific groups who may benefit most, however during the last few months some evidence has been published in randomized controlled scenarios using the Medtronic system. A RCT in which the 780G hybrid closed-loop system has been compared to sensor-augmented pump therapy with predictive low-glucose suspend over a 4-week study included people with T1D aged 7 to 80 years. The study demonstrated an improvement in TIR by 12.5 + 8.5% with no increase in hypoglycemia. ⁷ Additionally, the FLAIR study compared 2 hybrid closed-loop systems in adolescents and young adults with T1D. The 670G system was compared to the advanced AID system including fuzzy logic regulation. Hyperglycemia was reduced without an increase in hypoglycemia. ⁸ Finally, a recent study compared the use of the 670G system with an insulin pump therapy in a randomized study involving people with T1D with impaired awareness of hypoglycemia. This 8-week study was not able to demonstrate an improvement in counter-regulatory hormonal response. However, higher hypoglycemia symptom scores during controlled hypoglycemia, better self-reported hypoglycemia awareness and less time spent in hypoglycemia suggest potential benefits for this subset of patients with hypoglycemia unawareness. ⁹

The Medtronic algorithm is embedded in the pump software and there is an automated cloud storage for the new 780G system while Carelink manual download is needed for 670G users. The glucose target is conservative with a non-customizable target of 120 mg/dl for the 670G (there is an optional target for exercise mode at 150 mg/dl) meanwhile it can be customized in the 780G to 120, 110 and 100 mg/dl, maintaining exercise mode and introducing as new function the automatic correction boluses starting at 120 mg/dl and considering maximal insulin infusion rate. Both the pump and the sensor are waterproof, and there is only compatibility with rapid acting insulins even though there are some studies published using ultra-fast rapid insulins. The auto mode exits, alarm fatigue and substantial discontinuation rates with the 670G seems to be a challenge for the new 780G device.

The t:slim X2 insulin pump with Control-IQ technology has been available since 2020. The algorithm used is a treat-to-target predictive controller and the sensor is the Dexcom G6 with a duration of 10 days and without the need for calibration which is a major step forward in this technology as patients report. The TIR 70 to 180 mg/dl reported in the trials was 71% for adolescents and adults and 67% in children with less than 1% of time <54 mg/dl. The sensor accuracy, ease of use, attractive interface, pump size, efficient connectivity and improved glucose have been recently reported as specific factors related to high trust in the system. In this case the algorithm is also embedded in the pump software, but the automatic cloud storage only exists for glucose data, while the pump needs to be manually downloaded using a third-party platform. The glucose target considered is from 112 to 160 mg/dl and uses a horizon of prediction of 30 minutes to drive changes in basal insulin delivery and deliver correction boluses when required. Additionally, there are manually activated fixed ranges for sleep of 112 to 120 mg/dl and for exercise of 140 to 160 mg/dl. Both the pump and the sensor are watertight to 3 feet, and there is only compatibility with rapid acting insulins. The system cannot revert to Basal IQ (manual mode with basal suspend mode when hypoglycemia is predicted) if Control IQ is not suitable.Information from RCT is available. Control-IQ technology was compared to SAP in adult patients with T1D. An increase in a 10% of TIR was observed while remained unchanged in the control group. The characteristics of the participants in this trial may reflect an interest in and willingness to use AID among patients who were already using devices as part of their treatment (70% using CGM and 79% using CSII at enrollment). However, the results appear to be identical to others in which the participants were not on pump therapy before the study.^10-12

CamAPS has been developed by CamDiab Ltd (www.cambiad.com, Cambridge, UK) and approved for its use in the UK and the EU. The algorithm runs in an Android operating system smartphone and is a treat-to-target adaptive model predictive control (MPC) controller driven by the Dexcom G6 sensor and the Dana Diabecare RS and DANA-i insulin pumps. The TIR 70 to 180 mg/dl reported in the trials was 76% adults and 65 to 68% in adults and children with glucose above target at baseline with less than 1% of time <50 mg/dl. This is the only system currently approved for use during pregnancy and the data are published using an ultrarapid insulin analog. The full system can push data automatically to the cloud and to a third-party platform. The glucose target considered is 105 mg/dl but can be customized between 80 and 200 mg/dl with an optional target set for exercise. Both the pump and the sensor are waterproof and is approved for use with both rapid and ultra-fast insulin analogs. Data from RCT are available from an open, multicentre, multinational, parallel study in suboptimally controlled people with T1D older than 6 years compared to SAP use.^13-15

Open-source AID systems, developed by the community including people with diabetes are available for people to build but are not regulated and, while their safety and effectiveness have not been systematically assessed, data suggest similar times in ranges are achieved in selected study participants. ¹⁶ The safety and effectiveness of the Loop Do-It-Yourself automated insulin delivery system has been also evaluated in a prospective real-world observational study including adults and children. TIR increased around 7% with a reduction of 0.33 points in HbA_1c after 6 months. There was also a reduction in the incidence rate of severe hypoglycemia. ¹⁷

Future Commercial Hybrid AID Systems

The efficacy and safety of Diabeloop single-hormone, hybrid AID automated insulin delivery system which runs an original algorithm, has been assessed in several studies. The system comprises a smartphone running the algorithm (DBLG1; Diabeloop, Grenoble, France), a Kaleido pump (ViCentra, Utrecht, Netherlands) and a Dexcom G6 sensor (Dexcom, San Diego, CA, USA). The system has been compared with sensor augmented pump therapy in a randomized, controlled, cross-over trial. ¹⁸ During the study, 68 patients were remotely monitored in an ambulatory setting for 12 weeks. At the end of the study period, TIR was 9% greater in the participants using DBLG1 (69% vs 59%) with a significant reduction of time in hypoglycemia; 2.0% vs 4.3% (<70 mg/dl). More recently, DBLG1 efficacy and safety has been evaluated in real world conditions over 6 months in 25 patients. At the end of the study, there was a significant improvement in glucose control (TIR >70% in 58% of patients and time <70 mg/dl <4% in all participants) with a final HbA_1c of 7.1% (54 mmol/mol) without any serious adverse event. The AID system was in functional mode an average of 85% of the time with only one participant leaving the study. ¹⁹ In November 2018, Diabeloop’s first medical device, DBLG1, was CE marked. The company is working on commercializing, in Europe (France and Germany at first) and later in the United States. Recently, Diabeloop announced a partnership with Roche in a step towards an interoperable automated insulin delivery system.

The Omnipod 5 (Insulet, Billerica, MA, USA) system uses an MPC algorithm. Omnipod’s 3-day wear patch pump communicates with Dexcom’s G6 sensor while allowing users to control the system directly from their smartphone. The efficacy and performance have been evaluated in adults, adolescents, and children with T1D under free-living conditions. It was safe and performed well over 5 days and 4 nights with some improvement in TIR and percentage of time <70 mg/dl in some age-ranges. ²⁰ The system safety and effectiveness is now being evaluated in a pivotal trial before its launch. This is also the case for the Beta Bionics insulin-only iLet hybrid AID system, comprising the iLet BionicPancreas System and Dexcom G6 sensor (Beta Bionics, Boston, MA, USA).

The addition of glucagon to an AID system may confer additional protection from the most dangerous and frequent adverse event of any type of insulin treatment, hypoglycemia, reducing the need to take carbohydrates. Its use in T1D may allow more aggressive insulin administration to increase the time in glucose targets. However, bi-hormonal AID systems are mechanically more complex, requiring 2 separate infusion modules and a stable room-temperature glucagon preparation for long term administration is not yet approved. It is worth to mention that at the time of writing this manuscript 2 stable glucagon formulations have been approved to treat severe hypoglycemia (including dasiglucagon).

There are some examples of bi-hormonal AID systems, although none are currently commercially available. A dual hormone AID system has been evaluated for 11 days in 43 adult patients in an unrestricted home use setting in comparison with usual treatment (conventional or sensor augmented pump therapy) in a random order cross-over study. ²¹ The AID was initialized only with body weight and meal announcements were optional with no carbohydrate counting requirement. The use of the system was associated with an increase in TIR, a reduction of time in hypoglycemia and reduced the need for oral carbohydrates to treat hypoglycemic episodes. There were no serious adverse events during the study, although with the use of visual analogue scale, nausea was higher during the dual hormone AID period. An outpatient study compared dual hormone with insulin only AID in 23 adults with type 1 diabetes during 60 h (day and night) and showed no difference in TIR or in time <70 mg/dl. ²² A feasibility study to evaluate the function of the bi-hormonal configuration of the iLet bionic pancreas delivering dasiglucagon when compared to the insulin-only configuration of the iLet bionic pancreas in a home-use study in adults with type 1 diabetes has been already performed (NCT03840278 clinicaltrials.gov). The superiority of the dual hormone approach over an insulin-only AID approach is still to be demonstrated.

Towards a Fully Automated System

Despite AID becoming a reality in routine clinical practice over the last few years, the commercially hybrid AID systems are still far from a fully optimized automated diabetes management tool. ²³ In addition to this, there is much to do to increase usability and the burden associated with the use of the systems, including work to simplify (or remove the need for) meal boluses, carbohydrate counting, exercise management, auto-mode stability, wireless connectivity, software issues, data management platforms, interoperability, size, sensor accuracy, calibration, longer wear-time, and many other features. ²⁴ Despite impressive clinical effectiveness, the acceptance of an AID system by users relies more upon the usability, portability, and convenience aspects than on the reported clinical effectiveness of a particular control algorithm. ²⁵

Work to improve algorithms is underway to enable automatic management of meals and exercise without announcement to the system using additional physiological and non-physiological signals such as heart rate, accelerometry, gyroscope, and galvanic skin response. While supporting reduced self-management burden increased automation would be particularly important for people with barriers to manual management which may include disability, visual impairment, dexterity limitations, and psychological barriers to adherence. The introduction of new ultra-fast acting insulins analogues with the required modification of algorithms may also contribute to the improvement in minimizing hyperglycemic excursions. ²⁴ Finally, artificial intelligence-based AID glucose controllers may also contribute to optimizing glucose in people with acute and critical illness, and in surgery. ²⁵

User Requirements

Living with T1D poses some unique challenges for medical devices. In contrast to almost any other long-term condition there is an absolute requirement for continuous treatment and that uninterruptable supply of insulin means that AID systems must be able to function constantly while body-worn and subject to the stresses of movement, temperature, impact, water, and other environmental factors. These challenging conditions limit design form and function but systems additionally need to be inclusive and accessible. ²⁶ Inclusive for adults and children with variable levels of numeracy and language, and accessible for people with permanent disability which may include visual impairment from retinopathy, limited dexterity from cheiroarthropathy, or disability unrelated to diabetes; and for people with transient disability related to hypo- or hyperglycemia, or intercurrent illness.

Successfully addressing these unique requirements to enable people with T1D to extract the maximum value from AID systems is critical to their real-world adoption and effectiveness but is challenging. Data assessing clinician views of AID systems have been published^27,28 and similar work has been undertaken to explore the views of existing and future users.^29-31 Co-design, involvement of people with lived experience of T1D and robust qualitative assessment is critical to improving the real-world use of AID systems, especially for those who may have greater need.

Despite the well-documented benefits for glycemic control, the information regarding the impact of technology, including AID, on psychosocial aspects is scarce and little understood. Given that technology is still imperfect and may increase perceived burden, its psychosocial impact on the users and their families is an important consideration in routine clinical practice. ³² A recent review including studies on children and their families, indicates that there are positive and negative psychosocial aspects related to diabetes technology use among youth and their families. The strongest evidence for benefits of technology seems to be for quality of life and for an increased sense of control and freedom. ³³ Studies evaluating young people who have used AID systems point to positive psychosocial benefits, including reduced fear of hypoglycemia, reduced worry about hyperglycemia with less burden from the demands of daily diabetes care.^34-36 The impact of AID devices on caregivers has been also evaluated. Parents reported improvements in their own anxiety, sleep conditions and confidence in the use of technology. ³⁷ In the future, further research is needed with primary outcomes focused on the psychosocial impact of the use of AID systems.

Implementation of AID systems requires education and support of healthcare professionals, as well as users and their families. Data suggest that initial education burdens may be higher with reduced subsequent need for input ³⁸ but importantly people with T1D and their carers must continue to be able to safely self-manage without an AID in the case of sensor, algorithm, or pump failure. These potentially complex modes of failure may be challenging to manage, and it remains to be seen how best to deliver continued support and education to maintain self-management skills and to keep pace with technology developments.

In the light of the profound changes to care delivery in 2020 during the coronavirus pandemic, clinical services continued to support self-management remotely with the use of downloaded and remotely viewed glucose and insulin data. Understanding the best way to deliver remote multidisciplinary care for people using AID systems is an important component to future safe and effective implementation and involves some complex issues for data sharing and privacy, especially for emerging adults.^39,40

Conclusions

In this narrative we summarize the available AID technologies, the future system developments, their potential psychosocial impact and consider the challenges for education and for the user. In clinical studies AID systems improve metrics of glucose compared to usual care, and increase the time spent in the target glucose range. However, these studies often recruit people at lower risk of problematic hypoglycemia with HbA_1c values close to target at baseline. As data are gathered in people with higher HbA_1c values and in those at greater risk of hypoglycemia, evidence-based improvements to usability, implementation, education, and support will be needed, and optimized metrics to report these improvements may be needed. The first commercially available AID systems are not the end of the development journey but are the first step in learning how to optimally automate insulin delivery in a way that is equitably accessible and effective.