Abstract
Introduction
This year we are celebrating the 100th anniversary of the discovery of insulin—the remarkable event that marked the beginning of an era for using diabetes technology for the treatment of diabetes. A century later, advanced and innovative tools, such as connected insulin pens and pumps as well as self-monitoring blood glucose systems, are in common use by people with diabetes for maintaining their glucose levels. In recent years, continuous glucose monitoring (CGM) has become a routine clinical practice; the first automated closed-loop control systems, known as the “artificial pancreas” (AP) are on the market; and new upcoming closed-loop (CL) systems are being tested in pivotal trials. However, despite recent development of new tools, people with diabetes still need help in navigating their own metabolic control. Decision support systems (DSS) have been developed with the aim to help healthcare practitioners during patients' office visits and to help patients between office visits. DSS mobile toolbox includes a wide range of algorithms and apps to guide diverse aspects of diabetes management and insulin dosing, such as tools to use in times of physical activity and eating; real-time alarms and glucose prediction apps that help to prevent hypoglycemia and hyperglycemia events; various bolus calculators for meal-time dosing; and sophisticated DSS to guide insulin dosing for people using insulin pumps, multiple daily injections, and closed-loop systems. DSS for healthcare providers and patients is becoming a subject for intensive research, development, and adoption in the clinical practice (1,2). A number of advisory systems have been introduced this year, some using sophisticated data science methods, including machine learning and artificial intelligence (3). In view of this increase, the U.S. Food and Drug Administration (FDA) released in October 2019 new guidance for development of clinical decision support software for healthcare providers, patients, and caregivers.
Arguably, the development of subcutaneous automatic glucose control is the most prominent example of the merger of medicine and engineering for the treatment of diabetes, and the past year was critical to its success. Several reports of new closed-loop systems dominated the news at ATTD in February 2020 and at the 80th Scientific Sessions of the American Diabetes Association (ADA) in June 2020. The Control-IQ (Tandem Diabetes Care) system has become the second system, after MiniMed 670G, to be approved by the FDA in December 2019 for clinical use by children and adults 14 years or older, becoming the first system to receive the designation interoperable automated insulin dosing controller. The system uses a Dexcom G6 sensor and Tandem t:slim X2 insulin pump, and a control algorithm regulating basal rate and administering automated correction boluses. A 4-month study of Control-IQ involving 100 children with T1D ages 6–13 was published in the New England Journal of Medicine, showing 11% increase in time in range (TIR) (4). As a result of this study, in June 2020 the FDA approved Control-IQ for ages 6 years or older. Further development in CL systems includes the advanced hybrid closed-loop system (AHCL, Medtronic MiniMed 780G) that was presented at ADA 2020. This system includes technology from DreaMed Diabetes—algorithm enhancements and automated-correction boluses that have Bluetooth connectivity, which will enable users and their care partners to see real-time glucose data and trends on compatible iOS and Android smartphones via apps. The 90-day pivotal trial of the 780G, which was a 16-site, single-arm trial including 157 participants, was also presented during the ADA (5). TIR improved by 5.7%, and HbA1c was reduced by 0.5% (5). The new MiniMed AHCL was compared to its predecessor 670G in the FLAIR trial using a cross-over design that included 126 participants with T1D, ages 14–30 years old. The new system resulted in a TIR increase of 4% over the 670G (6). The Medtronic 780G received CE marking in June 2020. Another CL system was reported at ADA 2020—the new Omnipod 5 (Insulet) automated glucose control system powered by Horizon. The first outpatient study of this system, together with preliminary results from the pivotal trial, was presented (7). In this system, the Omnipod pump communicates with Dexcom G6 sensor; the basal control algorithm is built into the pod and has a user-selectable target between 110 and 150 mg/dL, and a smartphone app or a personal data manager is used to deliver boluses and as a controller interface (7).
Between July 1, 2019, and June 30, 2020, PubMed included over 150 publications on closed-loop systems and more than 140 on DSS were published. In the article studies of pivotal, real-life data, we chose to include new systems and mobile closed-loop systems and to focus on clinical studies evaluating DSS.
Spänig S, Emberger-Klein A, Sowa JP, Canbay A, Menrad K, Heider D
Wilson LM, Tyler N, Jacobs PG, Gabo V, Senf B, Reddy R, Castle JR
El Fathi A, Palisaitis E, von Oettingen JE, Krishnamoorthy P, Kearney RE, Legault L, Haidar A
Palisaitis E, El Fathi A, Von Oettingen JE, Krishnamoorthy P, Kearney R, Jacobs P, Rutkowski J, Legault L, Haidar A
Mosquera-Lopez C, Dodier R, Tyler NS, Wilson LM, El Youssef J, Castle JR, Jacobs PG
Murphy ME, McSharry J, Byrne M, Boland F, Corrigan D, Gillespie P, Fahey T, Smith SM
Brown SA, Kovatchev BP, Raghinaru D, Lum JW, Buckingham BA, Kudva YC, Laffel LM, Levy CJ, Pinsker JE, Wadwa RP, Dassau E, Doyle 3rd FJ, Anderson SM, Church MM, Dadlani V, Ekhlaspour L, Forlenza GP, Isganaitis E, Lam DW, Kollman C, Beck RW, for the iDCL Trial Research Group
Messer LH, Berget C, Vigers T, Pyle L, Geno C, Wadwa RP, Driscoll KA, Forlenza GP
Akturk HK, Giordano D, Champakanath A, Brackett S, Garg S, Snell-Bergeon J
Kovatchev B, Anderson SM, Raghinaru D, Kudva YC, Laffel LM, Levy C, Pinsker JE, Wadwa RP, Buckingham B, Doyle FJ 3rd, Brown SA, Church MM, Dadlani V, Dassau E, Ekhlaspour L, Forlenza GP, Isganaitis E, Lam DW, Lum J, Beck RW for the iDCL Study Group
Deshpande S, Pinsker JE, Church MM, Piper M, Andre C, Massa J, Doyle FJ 3rd, Eisenberg DM, Dassau E
Sherr JL, Buckingham BA, Forlenza GP, Galderisi A, Ekhlaspour L, Wadwa RP, Carria L, Hsu L, Berget C, Peyser TA, Lee JB, O'Connor J, Dumais B, Huyett LM, Layne JE, Ly TT
Lee MH, Vogrin S, Paldus B, Jones HM, Obeyesekere V, Sims C, Wyatt SA, Ward GM, McAuley SA, MacIsaac RJ, Krishnamurthy B, Sundararajan V, Jenkins AJ, O'Neal DN
Ekhlaspour L, Nally LM, El-Khatib FH, Ly TT, Clinton P, Frank E, Tanenbaum ML, Hanes SJ, Selagamsetty RR, Hood K, Damiano ER, Buckingham BA
Dovc K, Piona C, Yeşiltepe Mutlu G, Bratina N, Jenko Bizjan B, Lepej D, Nimri R, Atlas E, Muller I, Kordonouri O, Biester T, Danne T, Phillip M, Battelino T
The virtual doctor: an interactive clinical-decision-support system based on deep learning for non-invasive prediction of diabetes
Spänig S1, Emberger-Klein A2, Sowa JP3, Canbay A3, Menrad K2, Heider D1
1Department of Bioinformatics, Faculty of Mathematics and Computer Science, Philipps-University of Marburg, Marburg, Germany; 2Weihenstephan-Triesdorf University of Applied Sciences/TUM Campus Straubing for Biotechnology and Sustainability, Straubing, Germany; 3Department of Gastroenterology, Hepatology and Infectious Diseases, Otto-von-Guericke University Magdeburg, Magdeburg, Germany
Background
A new era in medicine will arise due to artificial intelligence (AI). However, current technology for AI systems does not yet allow for interaction with a patient and thus is only used by the physicians for predictions in diagnosis or prognosis, for which these systems are already widely in use.
Methods
We developed an AI that uses a speech recognition and speech synthesis system and thus can autonomously interact with the patient, which is particularly important for situations such as rural areas, where the availability of medical care is limited by low population densities.
Results
As a proof-of-concept, the system is able to predict type 2 diabetes mellitus (T2DM) using noninvasive sensors and deep neural networks. The system also delivers an easy-to-interpret probability estimation for T2DM.
Conclusion
In addition to developing the AI, we analyzed the acceptance of young people for AI in healthcare to help determine the impact of such a system in the future.
Comment
It is clear that artificial intelligence tools and decision support systems are needed to help cope with the increased number of people with diabetes and the relative low number of healthcare professionals (HCPs) available to them.
In the present study the authors developed an AI that is able to interact with a patient and predict type 2 diabetes mellitus based on noninvasive sensors and deep neural networks as well as provide an easy-to-interpret probability estimation for type 2 diabetes mellitus for a given patient. I am not sure this is a question that I would choose to ask a “virtual doctor,” and I do not see any attempt to compare the predictions provided by the AI to the answers provided by a real physician. How much time or money was saved? AI studies should include a control group.
Patient input for design of a decision support smartphone application for type 1 diabetes
Wilson LM1, Tyler N2, Jacobs PG2, Gabo V1, Senf B1, Reddy R2, Castle JR1
1Department of Medicine, Harold Schnitzer Diabetes Health Center, Oregon Health & Science University, Portland, OR; 2Department of Biomedical Engineering, Oregon Health & Science University, Portland, OR
Background
Smartphone applications that include decision support integrated with continuous glucose monitors may improve glycemic control in T1D. In this study, we conducted a survey to inform the design of decision support technology concerning the needs of potential users.
Methods
October 2017 through May 2018, a 70-question survey was distributed from a specialty clinic and T1D Exchange online health community (
Results
The majority (84.2%) of the 1542 responses (mean age 46.1 years [SD 15.2], mean duration of diabetes 26.5 years [SD 15.8]) had never used an app to manage diabetes; however, a large majority (77.8%) expressed interest in doing so. The majority of the respondents identified that the ability to predict and avoid hypoglycemia would be the most important feature, with 91% of respondents indicating the highest level of interest in these features. Management of glucose during exercise was determined to be the most difficult task for respondents (only 47% of participants were confident in this task); 85% of respondents were interested in the possibility of features that would help them manage glucose during exercise. The responses determined that integration and interoperability with peripheral devices/apps and customization of alerts as desirable. Responses from participants were generally consistent across stratified categories.
Conclusions
These results provide valuable insight into the needs of patients regarding decision support applications for management of T1D.
Comment
Patients' input is essential for those developing solutions for the care of diabetes, whether it is a device, a decision support algorithm, or a new medication. The present study is, therefore, extremely relevant. The main limitation of the present study, however, is that the sampling of participants was not broad enough. It is important to obtain the opinion and desire of various age groups, different socioeconomic groups, varied geographical areas, etc. We would like to encourage researchers and development groups from both academia and industry to listen to and analyze the wishes and needs of people with diabetes and to shape their innovations accordingly.
A pilot non-inferiority randomized controlled trial to assess automatic adjustments of insulin doses in adolescents with type 1 diabetes on multiple daily injections therapy
El Fathi A, Palisaitis E, von Oettingen JE, Krishnamoorthy P, Kearney RE, Legault L, Haidar A
1Department of Electrical and Computer Engineering, McGill University, Montreal, Canada; 2Department of Biomedical Engineering, McGill University, Montreal, Canada; 3Montreal Children's Hospital, Pediatric Endocrinology, Montréal, Canada; 4The Research Institute of McGill University Health Center, Montréal, Canada
Background
Both basal and bolus insulin doses are used with multiple daily injections (MDI) therapy for T1D. Nonoptimal insulin doses contribute to the lack of suitable glycemic control. In this study, we evaluated the feasibility of an algorithm that optimizes daily basal and bolus doses using glucose monitoring systems for MDI therapy users.
Methods
In children and adolescents on MDI therapy who attended a diabetes camp, we performed a pilot, noninferiority, randomized, parallel study comparing basal-bolus insulin dose adjustments made by camp physicians (PA) and a learning algorithm (LA). Participants underwent 11 days of daily dose adjustments in either arm while wearing a glucose sensor. Algorithm adjustments were reviewed and approved by a physician. The last 7 days were assessed for results.
Results
Twenty-one youths (age 13.3 [SD 3.7] years; 13 females; HbA1c 8.6% [SD 1.8]) were randomized to either group (LA [n=10] or PA [n=11]). The algorithm made 293 adjustments with a 92% acceptance rate from the camp physicians. The time in target glucose (3.9–10 mmol/L) in LA (39.5%, SD 20.7) was similar to PA (38.4%, SD 15.6) (P=0.89) in the last 7 days. The number of hypoglycemic events per day in LA (0.3, IQR [0.1-0.6]) was similar to PA (0.2, IQR [0.0-0.4]) (P=0.42). No incidences of severe hypoglycemia or ketoacidosis occurred.
Conclusions
Glycemic outcomes in the LA group were similar to the PA group in this pilot study. This algorithm has the potential to facilitate MDI therapy, and longer and larger studies are warranted.
Comment
This study is a small step in the right direction. Decision support systems (DSS) for both HCPs and people with diabetes are needed. Not every person treated with insulin will be using an automatic insulin device (artificial pancreas), even when these devices are financially attainable and available worldwide. There will always be people with diabetes who will use one of the continuous glucose measurement (CGM) sensors and MDI, such as the group in the present study, or SMBG and MDI. Therefore, DSSs should be developed for adjustment of insulin doses for HCPs during an office visit or to advise the person with diabetes remotely between visits.
In the present study the authors designed a prospective, randomized trial (pilot) of 21 participants in a diabetes camp using the advice of either the camp physician or their algorithm; this is the right design for such a study. The main limitation (in addition to the small number of participants) is the camp environment with fixed mealtimes and supervision very different than real-life conditions.
The efficacy of basal rate and carbohydrate ratio learning algorithm for closed-loop insulin delivery (artificial pancreas) in youth with type 1 diabetes in a diabetes camp
Palisaitis E1, El Fathi A2, Von Oettingen JE3,4, Krishnamoorthy P3, Kearney R1, Jacobs P5, Rutkowski J1, Legault L3,4, Haidar A1,4
1Department of Biomedical Engineering, Faculty of Medicine, McGill University, Montreal, Canada; 2Department of Electrical and Computer Engineering, Faculty of Engineering, McGill University, Montreal, Canada; 3Department of Pediatric Endocrinology, McGill University Health Centre, Montreal Children's Hospital, Montreal, Canada; 4The Research Institute of McGill University Health Centre, Montreal, Canada; 5Department of Biomedical Engineering, Oregon Health and Science University, Portland, OR
Background
Optimizing programmed basal rates and carbohydrate ratios may improve the performance of the artificial pancreas. At a diabetes camp, we tested the efficacy of a learning algorithm that updates daily basal rates and carbohydrate ratios in the artificial pancreas.
Methods
We conducted a randomized crossover trial in campers and counselors aged 8–21 years with type 1 diabetes on pump therapy. Participants underwent 2 days of artificial pancreas alone and 6 days of artificial pancreas with learning. During the artificial pancreas with learning, programmed basal rates and carbohydrate ratios were updated daily based on the learning algorithm's recommendations. All algorithm recommendations were reviewed for safety by camp physicians. The primary outcome was the time in target range (3.9–10 mmol/L) of the last 2 days of each intervention.
Results
Thirty-four campers (age 13.9±3.9, hemoglobin A1c 8.3%±0.2%) were included in this study, and 96% of algorithm recommendations were approved by the camp physicians. Participants were in closed-loop mode 74% of the time. There was no difference between interventions in time in target (55%–55%; P=0.71) nor in hypoglycemia events (0.8–0.9 events per day; P=0.63). This was despite changes in programmed basal rates ranging from −21% to + 117%, and changes in breakfast, lunch, and dinner carbohydrate ratios from −17% to + 40%, −36% to + 37%, and −35% to + 63%, respectively. Moreover, postprandial hyperglycemia and hypoglycemia did not decrease in participants whose carbohydrate ratios were decreased (more insulin boluses) and increased (less insulin boluses), respectively.
Conclusions
In camp settings, despite adjustments to programmed basal rates and carbohydrate ratios, the learning algorithm did not change glycemia, which may point toward limited effect of these adjustments in environments with large day-to-day variability in insulin needs. Longer randomized studies in real-world settings are required to further assess the efficacy of automatic adjustments of programmed basal rates and carbohydrate ratios.
Comment
The present study makes one wonder if the daily adjustment not having an effect on the metabolic control was due to the algorithm performance? The participants? Their ages? The unique environment of a diabetes camp? The numbers of participants and the duration of the study do not give the reader enough information to make that assessment. Longer duration with more participants in normal living conditions and different algorithms will provide us with a more informative answer.
Predicting and preventing nocturnal hypoglycemia in type 1 diabetes using big data analytics and decision theoretic analysis
Mosquera-Lopez C1,2, Dodier R1,2, Tyler NS1,2, Wilson LM1,2, El Youssef J1,2, Castle JR1,2, Jacobs PG1,2
1Artificial Intelligence for Medical Systems (AIMS) Lab, Department of Biomedical Engineering, Oregon Health & Science University, Portland, OR; 2Harold Schnitzer Diabetes Health Center, Oregon Health and Science University, Portland, OR
Background
Despite new glucose-sensing technologies, nocturnal hypoglycemia is still a problem for people with T1D, as symptoms and sensor alarms may not be detected while sleeping. Accurately predicting nocturnal hypoglycemia before sleep may help minimize nighttime hypoglycemia.
Methods
A support vector regression (SVR) model was trained to predict, before bedtime, the overnight minimum glucose and overnight nocturnal hypoglycemia for people with T1D. The algorithm was trained on continuous glucose measurements and insulin data collected from 124 people (22,804 valid nights of data) with T1D. The minimum glucose threshold for announcing nocturnal hypoglycemia risk was derived by applying a decision theoretic criterion to maximize expected net benefit. Accuracy was evaluated on a validation set from 10 people with T1D during a 4-week trial under free-living, sensor-augmented, insulin-pump therapy. The primary outcome measures were sensitivity and specificity of prediction, the correlation between predicted and actual minimum nocturnal glucose, and root-mean-square error. The impact of using the algorithm to prevent nocturnal hypoglycemia is shown in silico.
Results
The algorithm predicted 94.1% of nocturnal hypoglycemia events (<3.9 mmol/L, 95% confidence interval [CI], 71.3–99.9) with an area under the receiver operating characteristic curve of 0.86 (95% CI, 0.75–0.98). Correlation between actual and predicted minimum glucose was high (R=0.71, P<0.001). In-silico simulations showed that the algorithm could reduce nocturnal hypoglycemia by 77.0% (P=0.006) without impacting the time in target range (3.9–10 mmol/L).
Conclusion
An SVR model trained on a big data set and optimized using decision theoretic criterion can accurately predict at bedtime if overnight nocturnal hypoglycemia will occur and may help reduce nocturnal hypoglycemia.
Comment
Indeed, predicting and preventing nocturnal hypoglycemia is still the desire of most insulin-treated people with type 1 diabetes. Closed-loop systems and semi-closed-loop systems that stop insulin delivery when hypoglycemia is predicted were already developed and are already now at clinical use. The authors of the present study used big data to develop their algorithm and in-silico studies to evaluate it. Is reduction of nocturnal hypoglycemia by 77% sufficient? Is it the best that an algorithm can achieve? Head-to-head clinical studies with currently available commercial tools are needed.
Supporting care for suboptimally controlled type 2 diabetes mellitus in general practice with a clinical decision support system: a mixed methods pilot cluster randomised trial
Murphy ME1, McSharry J2, Byrne M2, Boland F1, Corrigan D1, Gillespie P3, Fahey T1, Smith SM1
1Department of General Practice, HRB Centre for Primary Care Research, Royal College of Surgeons, Dublin, Ireland; 2Health Behaviour Change Research Group, School of Psychology, NUI Galway, Galway, Ireland; 3School of Business and Economics, National University of Ireland, Galway, Ireland
Background
We developed a complex intervention called DECIDE (ComputeriseD dECisIonal support for suboptimally controlled type 2 Diabetes mellitus in Irish general practice) that used a clinical decision support system to address clinical inertia and support general practitioner (GP) intensification of treatment for adults with suboptimally controlled T2DM. This study examined the feasibility and potential impact of the DECIDE intervention.
Methods
This was a pilot cluster, randomized controlled trial conducted at 14 Irish general practices. The DECIDE intervention was aimed for general practitioners (GPs), who applied DECIDE to patients with suboptimally controlled T2DM, which was defined as a glycated hemoglobin (HbA1c) ≥70 mmol/mol and/or blood pressure ≥150/95 mmHg.
The intervention included training and a web-based clinical decision support system that advocated: (1) medication intensification actions and (2) nonpharmacological actions to support care. Control practices delivered the usual care.
Feasibility and acceptability were established using thematic analysis of semistructured interviews with GPs combined with data from the DECIDE website. HbA1c, medication intensification, blood pressure, and lipids were included in clinical outcomes.
Results
We recruited 14 practices and 134 patients. After 4 months, all practices and 114 patients were followed up. GPs indicated that decision support was helpful in navigating increasingly complex medication algorithms; however, the majority of GPs thought that the target patient group had poor engagement with GP and hospital services for various reasons. At follow-up, there was no difference in glycemic control (−3.6 mmol/mol [95% CI −11.2 to 4.0]) between intervention and control groups or in secondary outcomes such as blood pressure, total cholesterol, medication intensification, or utilization of services. Continuation criteria advocated proceeding to a definitive randomized trial with some modifications.
Conclusion
The DECIDE study was feasible and acceptable to GPs, but larger impacts on glycemic and blood pressure control should be considered for this patient population going forward.
Comment
Failure to optimize glycemic control in T2DM is commonly attributed to therapeutic inertia: a combination of patient nonadherence to treatment and clinician delays in adjusting therapy. Many factors contribute to therapeutic inertia, but one of the physician's main barriers is the complexity of decision making related to evidence-based algorithms for intensification of treatment. A wide array of factors needs to be considered in order to choose the appropriate personalized treatment out of the growing number of glucose-lowering medications. Therefore, there is certainly a need for the presented DECIDE decision support system, especially as most people with T2DM are treated by primary healthcare providers. A short intervention period of 1 month showed no advantage in glycemic control, lipid profile, or blood pressure for the use of this tool by general practitioners, compared to the control arm, for suboptimal-controlled people with T2DM (HbA1c ≥8.6%, 70 mmol/mol). The study design is lacking in several aspects: the short intervention and follow-up period could not capture the influence of treatment adjustments on glycemic control, patient engagement intervention was not added, half of the intervention study population was already treated at hospital and the GPs avoid changing treatment, and only modest intensification of treatment was executed. Therefore, the outcomes of the study may not reflect the value of such a DSS, and a well-designed randomized study is needed.
ARTIFICIAL PANCREAS: PIVOTAL TRIALS
Six-month randomized, multicenter trial of closed-loop control in type 1 diabetes
Brown SA1, Kovatchev BP1, Raghinaru D2, Lum JW2, Buckingham BA3, Kudva YC4, Laffel LM5, Levy CJ1,6, Pinsker JE7, Wadwa RP8, Dassau E9, Doyle 3rd FJ9, Anderson SM1, Church MM7, Dadlani V4, Ekhlaspour L3, Forlenza GP1,8, Isganaitis E5, Lam DW6, Kollman C2, Beck RW2, for the iDCL Trial Research Group
1University of Virginia Center for Diabetes Technology, Charlottesville, VA; 2Jaeb Center for Health Research, Tampa, FL; 3Department of Pediatrics, Division of Pediatric Endocrinology and Diabetes, Stanford University School of Medicine, Stanford, CA; 4Division of Endocrinology, Diabetes, Metabolism and Nutrition, Department of Internal Medicine, Mayo Clinic, Rochester, MN; 5Research Division, Joslin Diabetes Center and Department of Pediatrics, Harvard Medical School, Boston, MA; 6Division of Endocrinology, Icahn School of Medicine at Mount Sinai, New York, NY; 7Sansum Diabetes Research Institute, Santa Barbara, CA; 8Barbara Davis Center for Diabetes, University of Colorado, Anschutz Medical Campus, Aurora, CO; 9Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA
This manuscript is also discussed in the article on Diabetes Technology and Therapy in the Pediatric Age Group, page S-113.
Background
In patients with type 1 diabetes, closed-loop systems that automate insulin delivery may improve glycemic outcomes.
Methods
Patients with T1D were assigned in a 2:1 ratio to be treated using a closed-loop system (closed-loop group) or a sensor-augmented pump (control group) in this 6-month, randomized, multicenter trial. The main outcome, measured by continuous glucose monitoring, was the percentage of time that the blood glucose level was within the target range of 70 to 180 mg per deciliter (3.9 to 10.0 mmol per liter).
Results
A total of 168 participants were randomized into the closed-loop group (n=112) and the control group (n=56). The patients were 14 to 71 years old, and the glycated hemoglobin level ranged from 5.4% to 10.6%. All 168 patients completed the trial. The mean (±SD) percentage of time that the glucose level was within target range increased in the closed-loop group from 61±17% at baseline to 71±12% during the 6 months and remained unchanged at 59±14% in the control group (mean adjusted difference, 11 percentage points; 95% confidence interval [CI], 9 to 14; P<0.001). The main secondary outcomes included percentage of time that the glucose level was >180 mg per deciliter, mean glucose level, glycated hemoglobin level, and percentage of time that the glucose level was <70 mg per deciliter or <54 mg per deciliter (3.0 mmol per liter); these outcomes all met the prespecified hierarchical criterion for significance, favoring the closed-loop system. The mean difference (closed-loop minus control) in the percentage of time that the blood glucose level was lower than 70 mg per deciliter was −0.88 percentage points (95% CI, −1.19 to −0.57; P<0.001). The mean adjusted difference in glycated hemoglobin level after 6 months was −0.33 percentage points (95% CI, −0.53 to −0.13; P=0.001). After 6 months, the closed-loop group's median percentage of time that the system was in closed-loop mode was 90% over 6 months. No serious hypoglycemic events occurred in either group; one episode of diabetic ketoacidosis occurred in the closed-loop group.
Conclusions
A closed-loop system was correlated with a greater percentage of time spent in a target glycemic range than a sensor-augmented insulin pump.
Comment
This report in the NEJM presents the largest-to-date trial of automated insulin delivery—a 6-month, multicenter, randomized trial testing a new closed-loop system (Control-IQ, Tandem Diabetes Care, San Diego, CA). This study was Protocol 3 of the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)–funded International Diabetes Closed-Loop (iDCL) Trial, which involved seven prominent research centers in the United States. The study was followed by a 12-month extension (8).
Control-IQ uses a Dexcom G6 CGM without fingerstick calibration and a t-slim X2 insulin pump. The control algorithm was originally developed at the University of Virginia and then transferred to industry. Its distinguishing features include: (1) automated insulin correction boluses administered using CGM-based patient state estimation, in addition to basal-rate modulation; (2) dedicated hypoglycemia safety system that attenuates smoothly or discontinues insulin delivery using CGM and insulin-on-board information; and (3) gradually intensified control overnight, sliding the algorithm target down to achieve blood glucose levels of approximately 110–120 mg/dL by the morning.
Compared to the sensor-augmented pump, Control-IQ resulted in 11% increase in the time in target range of 70–180 mg/dL (9% during the day and 16% overnight) and in simultaneous reduction in the time <70 mg/dL by 0.9%. All 168 randomized participants completed the trial; there were no episodes of severe hypoglycemia.
After reviewing the data, on December 13, 2019, the FDA authorized the clinical use of Control-IQ, which became the first system to receive the designation “interoperable automated insulin dosing controller.” In other words, this controller can connect to an alternate controller-enabled insulin pump (ACE pump) and integrated continuous glucose monitor (iCGM), thereby providing more choices to patients looking to customize their diabetes management.
REAL-LIFE USE
Real-world hybrid closed-loop discontinuation: predictors and perceptions of youth discontinuing the 670G system in the first 6 months
Messer LH1, Berget C1, Vigers T1,2, Pyle L1,2, Geno C1, Wadwa RP1, Driscoll KA1,3, Forlenza GP1
1Barbara Davis Center for Childhood Diabetes, School of Medicine, University of Colorado Denver, Denver, CO; 2Department of Biostatistics and Informatics, Colorado School of Public Health, Aurora, CO; 3Department of Clinical and Health Psychology, University of Florida, Gainesville, FL
This manuscript is also discussed in the article on Diabetes Technology and Therapy in the Pediatric Age Group, page S-113.
Background
The aim of this study is to describe predictors of hybrid closed-loop (HCL) discontinuation and perceived barriers in youth with T1D.
Method
In a 6-month observational study, youth with T1D (eligible age 2–25 years; recruited age 8–25 years) who initiated the MiniMed 670G HCL system were followed prospectively.
Demographic, glycemic (time in range, HbA1c), and psychosocial variables (Hypoglycemia Fear Survey [HFS]; Problem Areas in Diabetes [PAID]) were gathered for all participants. Participants who discontinued HCL (<10% HCL use at clinical visit) completed a questionnaire on perceived barriers to HCL use.
Results
In total, 92 youth (15.7±3.6 y, HbA1c 8.8±1.3%, 50% female) initiated HCL, and 28 (30%) discontinued it, with the majority (64%) discontinuing between 3 and 6 months after starting. Baseline HbA1c predicted discontinuation (P=0.026), with the odds of discontinuing 2.7 times higher (95% CI: 1.123, 6.283) for each 1% increase in baseline HbA1c. Youth who discontinued HCL determined that difficulty with calibrations, number of alarms, and too time-consuming to make the system work were their biggest problems. Qualitatively derived themes included technological difficulties (error alerts, not working correctly), too much work (calibrations, fingersticks), alarms, disappointment in glycemic control, and expense (cited by parents).
Conclusions
Youth with higher HbA1c are at greater risk for discontinuing HCL than youth with lower HbA1c and should be the target of new interventions to support device use. The main reasons for discontinuing HCL are associated with the workload required to use HCL.
Comment
The Medtronic MiniMed 670G system was the first HCL approved by the FDA for clinical use 3 years ago. Since then this system accumulated significant user experience (UX), which is reflected by the present report in Pediatric Diabetes.
To understand the significant 30% discontinuation rate 3 to 6 months after initiating the HCL, we need to assess key UX features of this first system. In particular, three design decisions stand out: First, the MiniMed 670G automates basal rate but does not automate any boluses, which means that the user is responsible for any correction boluses that may be needed. Second, the Guardian 3 continuous glucose monitor requires fingerstick calibration at least twice a day. And third, the user must respond promptly to system alerts. In addition, auto mode is discontinued in various situations, for example, if there is prolonged hyperglycemia according to the sensor, if the pump has delivered minimum or maximum insulin rates for certain periods of time, if the sensor is not calibrated twice a day, or if glucose data are missing or considered inaccurate.
As confirmed in this paper, these UX features contributed to the discontinuation of HCL by nearly one-third of the users within months of initiation. Indeed, as reported: “Youth who discontinued HCL rated difficulty with calibrations, number of alarms, and too much time needed to make the system work as the most problematic aspects of HCL. Qualitatively derived themes included technological difficulties (error alerts, not working correctly), too much work (calibrations, fingersticks), alarms, disappointment in glycemic control, etc.” (Messer et al., 2020). In terms of user characteristics, higher baseline HbA1c was identified as a predictor of HCL discontinuation, which perhaps hints that a lower degree of user motivation to maintain tight control might be a contributing factor.
Long-term real-life glycaemic outcomes with a hybrid closed-loop system compared with sensor-augmented pump therapy in patients with type 1 diabetes
Akturk HK, Giordano D, Champakanath A, Brackett S, Garg S, Snell-Bergeon J
Barbara Davis Center for Diabetes, University of Colorado, Aurora, CO
Background
We aimed to compare glycemic metrics at 3 and 6 months in T1D patients on a 670G HCL system after using a sensor-augmented pump (SAP) for at least 3 months.
Methods
A retrospective study was conducted from a center with the largest number of 670G users in the United States. Data were evaluated from 202 SAP users. A total of 61 patients were excluded—2 for steroid use, 4 for pregnancy, 27 for previous Enlite use, and 28 for noncontinuous use of 670G. Out of 141 patients who met the inclusion criteria, 127 (aged 21–68 years) completed the data.
Results
HbA1c levels decreased by 0.4% at 3 months and were maintained at 6 months (7.6±0.07 vs 7.2±0.08, P<0.001) with no weight gain. Time in range (70–180 mg/dL) increased from 59.5%±1.1% to 70.2%±1.2% and 70.1%±1.1% at 3 and 6 months (P<0.001), respectively. At 6 months, time spent in hypoglycemia (<70 mg/dL) and time spent in hyperglycemia (>180 mg/dL) were reduced by 30% (2.2%±0.2% vs 3.2%±0.2%; P<0.05) and 26% (28.3%±1.2% vs 38.1%±1.2%; P<0.001), respectively. More time in auto mode was correlated with improved continuous glucose monitoring metrics, lower HbA1c, and decreased glycemic variability. Time in auto mode decreased in men after 3 months, while women maintained similar auto mode use throughout the study.
Conclusions
The HCL system improved HbA1c levels and time in range and decreased time spent in hypoglycemia and hyperglycemia at 6 months. Auto mode use was significantly correlated with continuous glucose monitoring metrics and glycemic outcomes.
Comment
During the last few years, a number of clinical trials have established the feasibility of long-term HCL control. Most of these studies pointed out the superiority of HCL over continuous subcutaneous insulin delivery (CSII) in terms of: (1) increased time within target range (typically 70–180 mg/dl); (2) reduced incidence of hypoglycemia, and (3) better overnight control. The real-life clinical practice of HCL began with the introduction of Medtronic's MiniMed 670G system, approved by the FDA for clinical use 3 years ago.
This paper reports glycemic outcomes of people with type 1 diabetes, 3 and 6 months after switching from SAP therapy to HCL. All expectations set by clinical trials of various HCL systems were met by the MiniMed 670G during routine clinical practice and were sustained at 3 and 6 months, for those who continued to use the system in auto mode (n=127 patients had complete data, after excluding 28 patients prior to study initiation and another 14 during the study for noncontinuous use of the 670G).
For example, time in range (TIR) increased from 59.5% on SAP to 70.2% after 3 months of HCL use, and remained unchanged (70.1%) after 6 months of HCL use. Similarly, the time below range was reduced from 3.2% at baseline to 2.6% at month 3 and 2.2% at month 6. TIR overnight increased from 60.6% to 76.2% at 3 months and remained steady thereafter (75.9% at 6 months). It is therefore evident that, if an HCL system is used continuously, it improves key CGM-based metrics as well as HbA1c.
MOBILE AP SYSTEMS
Randomized controlled trial of mobile closed-loop control
Kovatchev B1, Anderson SM2, Raghinaru D3, Kudva YC4, Laffel LM5, Levy C6, Pinsker JE7, Wadwa RP8, Buckingham B9, Doyle FJ 3rd10, Brown SA2, Church MM7, Dadlani V4, Dassau E10, Ekhlaspour L9, Forlenza GP8, Isganaitis E5, Lam DW6, Lum J3, Beck RW3 for the iDCL Study Group
1Center for Diabetes Technology, University of Virginia, Charlottesville, VA; 2Division of Endocrinology and Metabolism, Department of Medicine, University of Virginia, Charlottesville, VA; 3Jaeb Center for Health Research, Tampa, FL; 4Division of Endocrinology, Diabetes, Metabolism and Nutrition, Department of Internal Medicine, Mayo Clinic, Rochester, MN; 5Joslin Diabetes Center, Harvard Medical School, Boston, MA; 6Division of Endocrinology, Icahn School of Medicine at Mount Sinai, New York, NY; 7Sansum Diabetes Research Institute, Santa Barbara, CA; 8Barbara Davis Center for Diabetes, University of Colorado, Anschutz Medical Campus, Aurora, CO; 9Division of Pediatric Endocrinology and Diabetes, Department of Pediatrics, Stanford University School of Medicine, Stanford, CA; 10Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA
Objective
The aim of this study was to assess the efficacy of inControl AP, a mobile closed-loop control (CLC) system.
Methods
This protocol, NCT02985866, is a 3-month, parallel-group, multicenter, randomized unblinded trial intended to compare mobile CLC with SAP therapy. Eligibility criteria were T1D for at least 1 year, use of insulin pumps for at least 6 months, age ≥14 years, and baseline HbA1c <10.5% (91 mmol/mol). The study was designed to measure two coprimary outcomes: superiority of CLC over SAP in CGM-measured time below 3.9 mmol/L and noninferiority in CGM-measured time above 10 mmol/L.
Results
Between November 2017 and May 2018, 127 participants were randomly assigned 1:1 to CLC (n=65) versus SAP (n=62); 125 participants completed the study. CGM time below 3.9 mmol/L was 5.0% at baseline and 2.4% during follow-up in the CLC group vs 4.7% and 4.0%, respectively, in the SAP group (mean difference −1.7% [95% CI −2.4, −1.0]; P<0.0001 for superiority). CGM time above 10 mmol/L was 40% at baseline and 34% during follow-up in the CLC group vs 43% and 39%, respectively, in the SAP group (mean difference −3.0% [95% CI −6.1, 0.1]; P<0.0001 for noninferiority). One severe hypoglycemic event that was unrelated to the study device occurred in the CLC group.
Conclusions
The study met its coprimary endpoints—superiority of CLC over SAP in CGM-measured time below 3.9 mmol/L and noninferiority in CGM-measured time above 10 mmol/L—and has established that mobile CLC is feasible and could offer certain usability advantages over embedded systems, provided the connectivity among system components is stable.
Comment
Prior to 2012, all artificial pancreas (AP) trials, including the first outpatient studies, used systems running the controller on laptop computers wired to CGMs and insulin pumps. Evidently, this was cumbersome and unsuitable for outpatient use; thus, a major challenge was to make the system portable and wireless. The first mobile AP system was DiAs, developed at the University of Virginia in 2011, which used a smartphone as the computational hub receiving CGM data, running the AP control algorithm, and commanding the pump's insulin delivery.
The present report in Diabetes Care introduces the largest-to-date study of mobile AP—a 3-month, multicenter, randomized trial testing the commercial descendant of DiAs, inControl, which is a mobile AP system developed by TypeZero Technologies (Charlottesville, VA). This system used Roche Accu-Chek Spirit Combo insulin pump, a Dexcom G4 CGM, and the DiAs control algorithm built into a smartphone. A total of 127 participants were randomly assigned 1:1 to mobile AP (n=65) versus SAP; 125 of them competed the entire protocol. This study was Protocol 1 of the NIDDK-funded International Diabetes Closed-Loop (iDCL) Trial.
The study achieved its primary objectives, showing superiority of mobile AP over SAP in terms of both prevention of hypoglycemia and reduction of hyperglycemia. Specifically, compared to SAP, inControl resulted in 1.7% decrease in the time below 70 mg/dL, accompanied by 3.0% decrease in the time above 180 mg/dL, thereby confirming that mobile AP is feasible, provided the connectivity between system components is stable. iDCL Protocol 1 and inControl, presented in this paper, became predecessors of iDCL Protocol 3 and Control-IQ discussed previously.
Randomized crossover comparison of automated insulin delivery vs conventional therapy using an unlocked smartphone with scheduled pasta and rice meal challenges in the outpatient setting
Deshpande S1,2, Pinsker JE2, Church MM2, Piper M2, Andre C2, Massa J3, Doyle FJ 3rd1,2, Eisenberg DM3, Dassau E1,2,4
1Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA; 2Sansum Diabetes Research Institute, Santa Barbara, CA; 3Department of Nutrition, Harvard T.H. Chan School of Public Health, Boston, MA; 4Joslin Diabetes Center, Boston, MA
Background
Automated insulin delivery (AID) hybrid closed-loop systems have not been well studied in the context of prescribed meals. We evaluated performance of our interoperable artificial pancreas system (iAPS) in the at-home setting, running on an unlocked smartphone, with scheduled meal challenges in a randomized crossover trial.
Methods
Ten adults with type 1 diabetes completed 2 weeks of AID-based control and 2 weeks of conventional therapy in random order. They consumed regular pasta or extra-long-grain white rice as part of a complete dinner meal on six different occasions in both arms (each meal three times in random order). Surveys assessed satisfaction with AID use.
Results
Postprandial differences in conventional therapy were 10,919.0 mg/dL x min (95% CI 3,190.5 to 18,648.0, P=0.009) for glucose area under the curve (AUC) and 40.9 mg/dL (95% CI 4.6 to 77.3, P=0.03) for peak CGM glucose, with rice showing greater increases than pasta. White rice resulted in a lower estimate over pasta by a factor of 0.22 (95% CI 0.08 to 0.63, P=0.004) for AUC under 70 mg/dL. These glycemic differences in both meal types were reduced under AID-based control and were not statistically significant, where 0–2 hour insulin delivery decreased by 0.45 units for pasta (P=0.001) and by 0.27 units for white rice (P=0.01). Subjects reported high overall satisfaction with the iAPS.
Conclusions
The AID system running on an unlocked smartphone improved postprandial glucose control over conventional therapy in the setting of challenging meals in the outpatient setting.
Comment
In the previous section, we discussed DiAs and inControl—mobile AP systems running their control functions on a smartphone. The smartphones these systems used were locked down to prevent interference with the work of the controller and to meet regulatory requirements, meaning that the phone could not place or receive calls or run applications other than the AP.
The next logical step in mobile AP development was to unlock the phone and effectively turn the AP into a standard application that can communicate with a variety of CGM sensors and insulin pumps—that is, to develop an interoperable artificial pancreas system (iAPS). This report in Diabetes Technology and Therapeutics presents the first outpatient trial of mobile iAPS.
The iAPS consisted of Tandem t:AP insulin pump and Dexcom G6 CGM wirelessly connected to an unlocked Google Pixel 2 smartphone when in the closed-loop. The study required specific meal choices with relatively high carbohydrate content. The primary outcome was percent time in range (TIR, 70–180 mg/dL) as measured by CGM. A number of secondary outcomes focusing on hypoglycemia and the post-meal performance of the system were considered as well. User satisfaction with the Mobile iAPS was assessed by a questionnaire.
The study was by nature a safety trial and with 10 participants was not powered to show superiority of iAPS vs SAP in terms of its primary outcome. Several other significant differences were found, including less postprandial insulin on the iAPS, accompanied by marginally lower postprandial glucose excursions. Overall, the study concluded that an unlocked smartphone could be a feasible host for a mobile AP system.
FIRST TRIALS OF NEW SYSTEMS
Safety and performance of the Omnipod hybrid closed-loop system in adults, adolescents, and children with type 1 diabetes over 5 days under free-living conditions
Sherr JL1, Buckingham BA2, Forlenza GP3, Galderisi A1, Ekhlaspour L2, Wadwa RP3, Carria L1, Hsu L2, Berget C3, Peyser TA4, Lee JB5, O'Connor J5, Dumais B5, Huyett LM5, Layne JE5, Ly TT5
1Division of Pediatric Endocrinology & Diabetes, Department of Pediatrics, Yale University, New Haven, CT; 2Division of Pediatric Endocrinology, Department of Pediatrics, Stanford University, Stanford, CA; 3Barbara Davis Center, University of Colorado Anschutz Medical Campus, Aurora, CO; 4ModeAGC LLC, Palo Alto, CA; 5Insulet Corporation, Acton, MA
This manuscript is also discussed in the article on Diabetes Technology and Therapy in the Pediatric Age Group, page S-113
Background
The objective of this study was to assess the safety and performance of the Omnipod® personalized model predictive control (MPC) algorithm in adults, adolescents, and children aged ≥6 years with T1D under free-living conditions using an investigational device.
Methods
A 96-hour hybrid closed-loop (HCL) study was conducted in a supervised hotel/rental home setting following a 7-day outpatient standard therapy (ST) phase. Eligible participants were aged 6–65 years with A1C <10.0% using insulin pump therapy or multiple daily injections. Meals during HCL were unrestricted, with boluses administered per usual routine. There was daily physical activity. The primary endpoints were percentage of time with sensor glucose <70 and ≥250 mg/dL.
Results
Participants were 11 adults, 10 adolescents, and 15 children aged (mean±standard deviation) 28.8±7.9, 14.3±1.3, and 9.9±1.0 years, respectively. Percentage time ≥250 mg/dL during HCL was 4.5%±4.2%, 3.5%±5.0%, and 8.6%±8.8% per respective age group—a 1.6-, 3.4-, and 2.0-fold reduction compared to ST (P=0.1, P=0.02, and P=0.03). Percentage time <70 mg/dL during HCL was 1.9%±1.3%, 2.5%±2.0%, and 2.2%±1.9%, a statistically significant decrease in adults when compared to ST (P=0.005, P=0.3, and P=0.3). Percentage time 70–180 mg/dL increased during HCL compared to ST, reaching significance for adolescents and children: HCL 73.7%±7.5% vs ST 68.0%±15.6% for adults (P=0.08), HCL 79.0%±12.6% vs ST 60.6%±13.4% for adolescents (P=0.01), and HCL 69.2%±13.5% vs ST 54.9%±12.9% for children (P=0.003).
Conclusions
The Omnipod personalized MPC algorithm was safe and performed well over 5 days and 4 nights of use by a cohort of participants ranging from youth aged ≥6 years to adults with T1D under supervised free-living conditions with challenges, including daily physical activity and unrestricted meals.
Comment
Glycemic control should be managed on an individual basis. Therefore, one CL system would not fit the needs of all people with diabetes. Different CL algorithms with the opportunity to choose the different components of CL devices are needed. This study reports on a feasibility study testing the safety of a personalized closed-loop system based on predictive control algorithm for 4 days in a supervised hotel setting using Omnipod pump and Dexcom G4 sensor. This study is an advanced step in a series of studies relating to safety and efficacy testing of the hybrid CL system currently known as Omnipod 5 Automated Glucose Control System Powered by Horizon. The system was previously tested for 36 hours in in-patient setting (9) and for meal (10) and exercise challenges (11) in a supervised hotel study. This current study compares 4 days of hybrid closed-loop with unrestricted food and physical activity to 7 days of pump and sensor use with pump settings optimization done after the first days. TIR increased for a broad and heterogenous population of children, adolescents, and adults using either pump or multiple daily injection (MDI) therapy before commencing the closed-loop, although statistical significance was reached only for children and adolescents. Transfer from MDI to closed-loop use was demonstrated to be relatively easy. Pump setting adjustments using adaptivity scheme were handled during the closed-loop operation, reflecting the need for any closed-loop system to include adaptation of insulin delivery parameters such as carbohydrate-to-insulin ratio, as insulin sensitivity is constantly changing. Longer studies in a real-life home environment with a true control arm are needed to further evaluate the performance of this system. During the last ADA, the first outpatient studies with the system were presented. Standard therapy (ST) of 14 days was compared to a 5-day pilot use and 4–9 week pivotal trial use: (1) in 18 children and adults 14–70 years old, TIR increased from ST to 5-day and 4–9 week CL by 6.9% and 8.2%, respectively; (2) in 18 children 6–13 years old, TIR increased by 13.9% and 19.1%, respectively. There were no serious adverse events (7).
Glucose control in adults with type 1 diabetes using a Medtronic prototype enhanced-hybrid closed-loop system: a feasibility study
Lee MH1,2, Vogrin S1, Paldus B1, Jones HM1,2, Obeyesekere V2, Sims C1, Wyatt SA1, Ward GM2,3, McAuley SA1,2, MacIsaac RJ1,2, Krishnamurthy B1,2, Sundararajan V1,4, Jenkins AJ1,2,5, O'Neal DN1,2
1Department of Medicine, University of Melbourne, Melbourne, Australia; 2Department of Endocrinology and Diabetes, St Vincent's Hospital Melbourne, Melbourne, Australia; 3Department of Pathology, University of Melbourne, Melbourne, Australia; 4Department of Public Health, La Trobe University, Melbourne, Australia; 5NHMRC Clinical Trials Centre, University of Sydney, Sydney, Australia
Background
Experience from first-generation, closed-loop (CL) systems informs refinements to enhance glucose control and user acceptance. A next-generation prototype enhanced-hybrid CL (E-HCL) system incorporates iterative changes to the Medtronic MiniMed 670G CL system, including automated correction boluses, lower target glucose level, and user enhancements. The aim was to explore safety, system performance, and glucose control using E-HCL in adults with type 1 diabetes.
Methods
Twelve adults underwent this first-in-human feasibility study. After a 1-week run-in using open-loop (OL), E-HCL was activated at the start of a supervised 1-week hotel phase, followed by 3 weeks free living at home. Supervised challenges included two meal interventions (unannounced and late meal bolus) and a sensor calibration intervention. Primary outcome was sensor glucose time in range (TIR); OL run-in and E-HCL at home were compared by Wilcoxon signed-rank test.
Results
Twelve adults (7 men; median [interquartile range] age 48 [39, 57] years; HbA1c 6.8 [6.2, 7.2] %, 51 [44, 55] mmol/mol; diabetes duration 31 [13, 41] years) completed the protocol. E-HCL resulted in greater TIR (85.3 [79.4, 88.4] % vs 75.0 [66.6, 83.7] %, P=0.003) and lower mean sensor glucose (123.0 [119.3, 129.6] mg/dL vs 143.5 [135.8, 154.5] mg/dL, P=0.002) than OL. Time spent <70 mg/dL increased using E-HCL (4.4 [3.3, 6.1] % vs 3.0 [1.8, 3.8] %, P=0.02) with no difference in time <54 mg/dL (P=0.64). Time in CL was 99.98 [99.0, 100.0] %. All participants were satisfied using E-HCL.
Conclusions
In adults with well-controlled HbA1c levels, a prototype E-HCL resulted in high TIR, few CL exits, and positive user experiences at the expense of increased hypoglycemia (<70 mg/dL). E-HCL represents a positive step in the journey toward optimizing glucose control in people living with type 1 diabetes.
Comment
The Medtronic MiniMed 670G was the first system granted regulatory approval, using a conservative approach to glucose control by adjusting basal insulin delivery in response to sensor glucose levels and incorporating several safety layers. The change in basal rate allows for a gradual improvement in glucose levels, while incorporating a bolus correction may achieve a similar result but faster due to the pharmacodynamics and pharmacokinetics of the current available insulins. The new generation MiniMed enhanced hybrid closed-loop has been developed to further reduce hyperglycemia by including automated correction boluses, lowering of the glucose target, and reducing alarms and the workload needed for system operation. The system was tested for the first time in this small feasibility study among well-controlled adults with T1D. While sensor time in range increased by around 10% to reach a median of 85%, time in hypoglycemia increased by 1.4%, but no change was demonstrated in time of significant hypoglycemia below 54 mg/dl between the week of open-loop compared to 3-week closed-loop. Importantly, less exits from closed-loop were observed, with nearly 100% closed-loop time. The higher time in closed-loop enables better control and reduces workload needed to maintain closed-loop control and might lower system discontinuation rate. Longer-term real-life studies with direct comparison to the commercial 670G are needed. The next algorithm version of the enhanced hybrid closed-loop—Minimed Advanced Hybrid Closed-loop—was tested and compared to the commercial 670G in a crossover design NIDDK-funded international FLAIR study.
Feasibility studies of an insulin-only bionic pancreas in a home-use setting
Ekhlaspour L1, Nally LM2, El-Khatib FH3, Ly TT2, Clinton P2, Frank E2, Tanenbaum ML2, Hanes SJ2, Selagamsetty RR3, Hood K2, Damiano ER3, Buckingham BA2
1Diabetes Unit, Massachusetts General Hospital, Boston, MA; 2Division of Pediatric Endocrinology and Diabetes, Stanford University School of Medicine, Palo Alto, CA; 3Department of Biomedical Engineering, Boston University, Boston, MA
Background
The aim of this study was to test the safety and performance of the “insulin-only” configuration of the bionic pancreas (BP) closed-loop, blood-glucose control system in a home-use setting, and in doing so, to determine glycemic results using different static and dynamic glucose set-points.
Method
This open-label, non-randomized study included three consecutive intervention periods in which participants had consecutive weeks of usual care followed by the insulin-only BP with (1) an individualized static set-point of 115 or 130 mg/dL and (2) a dynamic set-point that automatically varied within 110 to 130 mg/dL, depending on hypoglycemic risk. Human factor (HF) testing was handled using validated surveys. The last 5 days of each study arm were dedicated to data analysis.
Results
A total of 13 participants enrolled, with a mean age of 28 years, mean A1c of 7.2%, and mean daily insulin dose of 0.6 U/kg (0.4–1.0 U/kg). The usual care arm had an average glucose of 145±20 mg/dL, which increased in the static set-point arm (159±8 mg/dL, P=0.004) but not in the dynamic set-point arm (154±10 mg/dL, P=ns). There was no relevant difference in time spent in range (70–180 mg/dL) among the three study arms. There was less time <70 mg/dL with both the static (1.8%±1.4%, P=0.009) and dynamic set-point (2.7±1.5, P=0.051) arms compared to the usual-care arm (5.5%±4.2%). HF testing showed preliminary user satisfaction and no increased risk of diabetes burden or distress.
Conclusions
The insulin-only configuration of the BP by means of either static or dynamic set-points and initialized only with body weight performed similarly to other published insulin-only systems.
Comment
In this feasibility study another configuration of the closed-loop system was tested for safety using different glucose set-points. The system included the Tandem t:slim infusion pump, G4 Dexcom sensor, and “bionic pancreas” algorithm running on Apple iPhone 4S. This system was initially configured as a bihormonal system, meaning a dual delivery of insulin and glucagon in response to sensor glucose levels. Although no direct comparison was made between the single and dual hormone closed-loop system, several studies showed some advantages over insulin-only mainly in relation to exercise and hypoglycemia (12) while others failed to show any differences (13).
The complexity of dual hormone delivery—especially the need for stable glucagon—concerning the failure to deliver glucagon during closed-loop as well as the costs involved have led to testing the system first with insulin only. The system is also hybrid closed-loop but has a special way to deliver the premeal bolus. The user needs to indicate the size of the meal (i.e., larger than typical, typical, etc.) and the time of the day (i.e., breakfast, lunch, or dinner). Then, the system delivers 75% of the insulin meal-priming bolus based on the size and type of the meal. The insulin-only configuration of the system was tested in the home settings in well-controlled adults with type 1 diabetes. Actually, standard use was compared to two sessions of 5 days of closed-loop, once with fixed set point of 130 mg/dl (only in one patient the set point could be reduced to 115 mg/dl) and second with an automated determined dynamic set point between 110 to 130 mg/dl. Significant reduction in hypoglycemia was observed with no change in time in range but with higher mean glucose for the closed-loop compared to standard care. These outcomes might be the result of the initial well-controlled population included in the study or the relative high glucose set point chosen. The ability to set different glucose set points is important as there will be people with diabetes who will need different levels of glucose control, or different sets of protection from hypoglycemia. The benefits of dynamic set points should be further evaluated.
FASTER INSULIN USED IN CLOSED-LOOP
Faster compared with standard insulin aspart during day-and-night fully closed-loop insulin therapy in type 1 diabetes: a double-blind randomized crossover trial
Dovc K1, Piona C2, Yeşiltepe Mutlu G3, Bratina N1, Jenko Bizjan B1, Lepej D4, Nimri R5, Atlas E6, Muller I6, Kordonouri O7, Biester T7, Danne T7, Phillip M,5,8 Battelino T1,9
1Department of Pediatric Endocrinology, Diabetes and Metabolic Diseases, University Medical Centre-University Children's Hospital, Ljubljana, Slovenia; 2Pediatric Diabetes and Metabolic Disorders Unit, University City Hospital, Verona, Italy; 3Department of Pediatric Endocrinology and Diabetes, Koç University Hospital, İstanbul, Turkey; 4Department of Pulmonology, University Medical Centre-University Children's Hospital, Ljubljana, Slovenia; 5The Jesse Z. and Sara Lea Shafer Institute for Endocrinology and Diabetes, National Center for Childhood Diabetes, Schneider Children's Medical Centre of Israel, Petah Tikva, Israel; 6DreaMed Diabetes Ltd., Petah Tikva, Israel; 7Diabetes Centre for Children and Adolescents, Kinder- und Jugendkrankenhaus Auf der Bult, Hannover, Germany; 8Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel; 9Department of Pediatric Endocrinology, Diabetes and Metabolic Diseases, University Medical Centre-University Children's Hospital, Ljubljana, Slovenia
This manuscript is also discussed in the article on New Insulins, Biosimilars, and Insulin Therapy, page S-46.
Objective
In this study we aimed to determine the safety and efficacy of day-and-night fully closed-loop insulin therapy using faster (Faster-CL) compared with standard insulin aspart (Standard-CL) in young adults with T1D.
Methods
A total of 20 patients with T1D on insulin pump therapy (11 females, aged 21.3±2.3 years, HbA1c 7.5±0.5% [58.5±5.5 mmol/mol]) participated in a double-blind, randomized crossover trial. Participants underwent two 27-hour inpatient periods with unannounced afternoon moderate–vigorous exercise and unannounced/uncovered meals. We compared Faster-CL and Standard-CL in random order. The fuzzy-logic control algorithm DreaMed GlucoSitter was used during both interventions. Glucose sensor data were examined by intention-to-treat principle with the difference (between Faster-CL and Standard-CL) in proportion of time in range 70–180 mg/dL (TIR) over 27 hours as the primary endpoint.
Results
The proportion of TIR was similar for both arms: 53.3% (83% overnight) in Faster-CL and 57.9% (88% overnight) in Standard-CL (P=0.170). The proportion of time in hypoglycemia <70 mg/dL was 0.0% for both groups. Baseline-adjusted interstitial prandial glucose increments 1 hour after meals were greater in Faster-CL compared with Standard-CL (P=0.017). The gaps between measured plasma insulin and estimated insulin-on-board levels at the beginning, the end, and 2 hours after the exercise were smaller in the Standard-CL group (P=0.029, P=0.003, and P=0.004, respectively). No severe adverse events occurred.
Conclusions
Fully closed-loop insulin delivery using either faster or standard insulin aspart was safe and efficient in accomplishing near-normal glucose concentrations outside postprandial periods. The closed-loop algorithm was better adjusted to the standard insulin aspart.
Comment
All current closed-loop configurations are limited in their ability to handle meal and exercise, mainly due to the delayed onset and long-lasting action of the existing available insulins. It takes time for insulin action to kick in during meals causing postprandial hyperglycemia and takes time to be wiped out during exercise causing hypoglycemia. Therefore, most closed-loop systems are “hybrid,” meaning the patient still needs to bolus for meals and notify the system of exercise. New and upcoming faster-acting insulins could provide important benefits and may facilitate full CL operation. This study in Diabetes Care examined, for the first time, full closed-loop in young people with T1D using a new faster-acting insulin compared with the regular aspart. No advantage was found for the faster compared to the standard aspart in terms of post-prandial and exercise glucose control or in overall glycemic control. Effective glycemic control especially outside the post-prandial periods was achieved with the fuzzy-logic algorithm in full closed-loop operation in either types of insulin. These findings are consistent with previous studies that showed a marginal or no significant difference in glycemic control between the two insulins used with pump therapy (14) and also in recent studies of hybrid closed-loop with the commercial Medtronic 670G (15) or short-term use of full closed-loop for people with type 2 diabetes (16). It might be that the present closed-loop algorithms may require adaptations to adjust to the faster insulin aspart profile or that the small difference between the two insulin profiles is not sufficient to produce clinically significant change. Additional and longer clinical trials are needed to answer these questions, as there is certainly still a need for insulin with faster action as well as earlier weaning in order to enable a fully automated CL system that will provide a comparable glucose control to closed-loop systems with manual premeal bolus.
SUMMARY
This has been an exciting year as studies with more advanced closed-loop systems, including correction boluses, mobile systems, and connected systems, were published. This year another closed-loop system (Tandem Control-IQ) was approved for clinical use in the United States and several others in Europe (Medtronic 780G, Diabeloop, and Cambridge CAMAPS), expanding the individual options. Different people will need or prefer different system configurations. It will be interesting and important to compare head-to-head these different systems. Effort is being invested to explore the feasibility of full closed-loop systems with no announcements for meal or exercise, adding adjunctive therapies such as Pramlintide. New populations are being tested for CL use, such as adults with T1D and gastroparesis and cystic fibrosis, and preterm infants. Closed-loop systems can be regarded as automated decision support systems in real time. Nevertheless, not all people treated with insulin will use these systems and for those a wide range of decision support systems are being tested.
This year was characterized by a significant increase in the amount of clinical studies evaluating DSS. The COVID-19 pandemic has highlighted the need, which had previously existed, to utilize data gathered from various diabetes technologies remotely to provide automated diabetes-related treatment recommendations, when in-person visits have become limited.
Footnotes
Author Disclosure Statement
RN has received device support for clinical studies from Medtronic, Dexcom, Abbot, and Insulet Corporation. RN received honoraria for participation on the speaker's bureau of Novo Nordisk, Pfizer, Eli Lilly, and Sanofi. RN owns DreaMed stock and reports three patent applications.
BK has received research support handled by the University of Virginia from Dexcom and Tandem; patent royalties handled by UVA from Dexcom, Johnson&Johnson, and Sanofi; consultant/advisory board fees from Sanofi and Dexcom; and speaker honoraria from Dexcom, Tandem, and Sanofi.
MP has received honoraria as well as consultation fees from Sanofi, Novo Nordisk, Eli Lilly, Medtronic, Pfizer, ESP Systems, Qulab Medical, AstraZeneca, and Insulet. The institute headed by MP received research support from Medtronic, Novo Nordisk, Eli Lilly, Roche, Dexcom, Sanofi, Insulet, OPKO Health, DreaMed Diabetes, Bristol-Myers Squibb, and Merck. MP is a stockholder/shareholder of DreaMed Diabetes, NG Solutions, and Nutriteen Professionals and reports two patent applications.