1. Top of page
  2. Abstract
  3. Clinical evaluation of a bi-hormonal closed loop
  4. Clinical evaluation of uni-hormonal closed loop
  5. Hypoglycaemia prevention and closed-loop control
  7. References

Closed-loop algorithms can be found in every aspect of everyday modern life. Automation and control are used constantly to provide safety and to improve quality of life. Closed-loop systems and algorithms can be found in home appliances, automobiles, aviation and more. Can one imagine nowadays driving a car without ABS, cruise control or even anti-sliding control? Similar principles of automation and control can be used in the management of diabetes mellitus (DM). The idea of an algorithmic/technological way to control glycaemia is not new and has been researched for more than four decades. However, recent improvements in both glucose-sensing technology and insulin delivery together with advanced control and systems engineering made this dream of an artificial pancreas possible. The artificial pancreas may be the next big step in the treatment of DM since the use of insulin analogues. An artificial pancreas can be described as internal or external devices that use continuous glucose measurements to automatically manage exogenous insulin delivery with or without other hormones in an attempt to restore glucose regulation in individuals with DM using a control algorithm. This device as described can be internal or external; can use different types of control algorithms with bi-hormonal or uni-hormonal design; and can utilise different ways to administer them. The different designs and implementations have transitioned recently from in silico simulations to clinical evaluation stage with practical applications in mind. This may mark the beginning of a new era in diabetes management with the introduction of semi-closed-loop systems that can prevent or minimise nocturnal hypoglycaemia, to hybrid systems that will manage blood glucose (BG) levels with minimal user intervention to finally fully automated systems that will take the user out of the loop. More and more clinical trials will be needed for the artificial pancreas to become a reality but initial encouraging results are proof that we are on the right track. We attempted to select recent publications that will present these current achievements in the quest for the artificial pancreas and that will inspire others to continue to progress this field of research.

Clinical evaluation of a bi-hormonal closed loop

  1. Top of page
  2. Abstract
  3. Clinical evaluation of a bi-hormonal closed loop
  4. Clinical evaluation of uni-hormonal closed loop
  5. Hypoglycaemia prevention and closed-loop control
  7. References

A bi-hormonal closed-loop artificial pancreas for type 1 diabetes

F. H. El-Khatib,1S. J. Russell,2D. M. Nathan,2R. G. Sutherlin,2E. R. Damiano1

1Department of Biomedical Engineering, Boston University, Boston, MA, USA, and2Diabetes Unit and Department of Medicine, Massachusetts General Hospital, and Harvard Medical School, Boston, MA, USA

Sci Transl Med 2010; 2: 1–12

Background: This paper is a report about the performance of a bi-hormonal artificial pancreas which aims to automatically control BG concentrations within the non-diabetic range in people with type 1 diabetes. This closed-loop control system, developed by the authors, delivers both insulin and glucagon to imitate the normal physiology of BG control in the human body.

Methods: The closed-loop control system consists of three components: a venous BG monitor which measures BG concentration every 5 min, infusion pumps to deliver the fast-acting insulin analogue lispro and glucagon subcutaneously, and a computer-based control algorithm that automatically computes insulin and glucagon doses to be delivered based on sampled BG concentrations. Eleven adults with type 1 diabetes and no endogenous insulin secretion were studied in 27-h experiments, which included three carbohydrate-rich meals. The study experiments were repeated with time-to-peak (tmax) plasma lispro concentration setting doubled, based on the results of the initial experiments.

Results: Two patterns of BG control emerged in the first study (lispro tmax set to 33 min). In six subjects, the closed-loop system achieved a mean BG concentration of 140 mg/dl (lower than the target of ≤ 154 mg/dl recommended by the American Diabetes Association) with no instances of hypoglycaemia requiring treatment. The time-to-peak plasma lispro in the six subjects ranged from 56 to 72 min (mean, 64 ± 6 min). On the other hand, the other five subjects had a mean BG concentration of 144 mg/dl and experienced hypoglycaemia that required treatment; however, these subjects were later found to have slower lispro absorption kinetics than the six subjects who did not become hypoglycaemic. The time-to-peak plasma lispro in the five subjects ranged between 71 and 191 min (mean 117 ± 48 min). In the repeated study (lispro tmax set to 65 min), no hypoglycaemia occurred in either group and average aggregate mean BG concentration was controlled to 164 mg/dl.

Conclusion: Automated closed-loop control of BG concentrations to the near-normal range without risk of hypoglycaemia will be feasible with a bi-hormonal artificial endocrine pancreas.

  • Comment: The authors demonstrated the feasibility of a bi-hormonal system that used both insulin and glucagon to regulate glycaemia. The overall results were extremely encouraging especially when one considers the big carbohydrate load during the study. However, details in the overall system design undermined these results as follows: (a) the logic behind using different control algorithms for insulin and glucagon was not clear; (b) insulin was delivered via a subcutaneous route using two pumps, one with U-100 insulin and a second one with U-10 insulin; (c) glucose measurements were collected via intravenous line which is infeasible in an outpatient system; (d) moreover, measurement delay and sensor errors were not considered in this design. As the authors stated, glucagon was used only to prevent or treat excursions of BG below 100 mg/dl. In terms of prevention, all subjects in both phases of the study reached BG levels below the desired 100 mg/dl level. In terms of glucose control, the results were quite similar to uni-hormonal systems. An important result that this study presented is the need to personalise controller tuning to subjects, as reflected by the change in the PK parameter in the second phase of the trials. The change of the PK parameter was the reason for the prevention of hypoglycaemia between the first and second phase of the experiments. Although glucagon was used during this study, the full benefit of the use of glucagon was not presented. The efficacy of glucagon should be tested compared to an insulin-only system. Glucagon should be integrated into a system that already has a mechanism to prevent overdose of insulin and considered to be activated as a rescue measure.

Clinical evaluation of uni-hormonal closed loop

  1. Top of page
  2. Abstract
  3. Clinical evaluation of a bi-hormonal closed loop
  4. Clinical evaluation of uni-hormonal closed loop
  5. Hypoglycaemia prevention and closed-loop control
  7. References

Manual closed-loop insulin delivery in children and adolescents with type 1 diabetes: a phase 2 randomised crossover trial

R. Hovorka,1,2J. M. Allen,1,2D. Elleri,1,2L. J. Chassin,1,2J. Harris,2D. Xing,3C. Kollman,3T. Hovorka,1A. M. F. Larsen,2M. Nodale,1A. De Palma,1M. E. Wilinska,1,2C. L. Acerini,1,2D. B. Dunger1,2

1Department of Paediatrics and2Institute of Metabolic Science, University of Cambridge, Cambridge, UK, and3Jaeb Center for Health Research, Tampa, FL, USA

Lancet 2010; 375: 743–51

Background: Closed-loop systems that link subcutaneous continuous glucose measurements to subcutaneous insulin delivery are a promising way to better control BG levels in people with diabetes. This study attempts to establish whether closed-loop insulin delivery can control overnight BG in children and adolescents with type 1 diabetes.

Methods: The study consisted of three randomised crossover sub-studies in 19 patients aged 5–18 years with type 1 diabetes of mean duration 6.4 years (SD 4.0). It compared standard continuous subcutaneous insulin infusion and closed-loop delivery (= 13; APCam01); closed-loop delivery after rapidly and slowly absorbed meals (= 7; APCam02); and closed-loop delivery and standard treatment after exercise (= 10; APCam03). Allocation was by computer-generated random code. Participants were masked to plasma and sensor glucose. In APCam01, investigators were also masked to plasma glucose. During closed-loop nights, glucose measurements were fed every 15 min into a control algorithm calculating rate of insulin infusion, and a nurse then adjusted the insulin pump. During control nights, patients’ standard pump settings were used. Primary outcomes were time for which plasma glucose concentration was 3.91–8.00 mmol/l or 3.90 mmol/l or lower, and analysis was per protocol.

Results: Seventeen patients were studied for 33 closed-loop and 21 continuous infusion nights. Primary outcomes did not differ significantly between treatment groups in APCam01 [12 analysed; target range, median 52% (interquartile range 43–83) closed loop vs. 39% (15–51) standard treatment, p = 0.06; ≤ 3.90 mmol/l, 1% (0–7) vs. 2% (0–41), p = 0.13], APCam02 [six analysed; target range, rapidly absorbed meal 53% (48–57) vs. slowly absorbed meal 55% (37–64), p = 0.97; ≤ 3.90 mmol/l, 0% (0–4) vs. 0% (0–0), p = 0.16] and APCam03 [nine analysed; target range 78% (60–92) closed loop vs. 43% (25–65) control, p = 0.0245, not significant at corrected level; ≤ 3.90 mmol/l, 10% (2–15) vs. 6% (0–44), p = 0.27]. A secondary analysis of data documented increased time in the target range [60% (51–88) vs. 40% (18–61); p = 0.0022] and reduced time for which glucose concentrations were 3.90 mmol/l or lower [2.1% (0.0–10.0) vs. 4.1% (0.0–42.0); p = 0.0304]. No events with plasma glucose concentration < 3.0 mmol/l were recorded during closed-loop delivery, compared with nine events during standard treatment.

Conclusions: Closed-loop systems could reduce the risk of nocturnal hypoglycaemia and allow for better BG control in children and adolescents with type 1 diabetes.

  • Comment: Dr. Hovorka and his collaborators have demonstrated the benefit of closing the loop to reduce the risk of nocturnal hypoglycaemia in children and adolescents. This study as indicated by the authors was designed to prove the concept but, technically, the loop was closed manually by a team member. The control algorithm was an adaptive model predictive control (MPC). Manual glucose measurements from continuous glucose monitoring (CGM) were entered to a user interface; the controller then suggested the new infusion rate and these values were manually entered to the insulin pump to close the loop. Three different randomised crossover studies were reported with the aim of demonstrating closed-loop control over several daily challenges such as slowly and rapidly absorbed meals and exercise. The study demonstrated that automatic glucose regulation is feasible and it can reduce the risk of hypoglycaemia; however, one of the exclusion criteria was recurrent severe hypoglycaemia which undermined the aim of the study. The closed-loop system should be investigated by the authors over an extended period of time with volunteers who experience recurrent severe hypoglycaemia to provide the necessary seal of approval. Closed-loop or semi-closed-loop systems that can reduce the likelihood of nocturnal hypoglycaemia in young subjects with type 1 diabetes will most probably be the first to be commercialised.

Closed-loop insulin delivery using a subcutaneous glucose sensor and intraperitoneal insulin delivery: feasibility study testing a new model for the artificial pancreas

E. Renard,1J. Place,1M. Cantwell,2H. Chevassus,3C. C. Palerm2

1Endocrinology Department, Le Centre Hospitalier Universitaire Montpellier and Unité Mixte de Recherche Centre National de la Recherche Scientifique 5232, University of Montpellier, Montpellier, France,2Medtronic Diabetes, Northridge, CA, USA, and3INSERM Clinical Investigation Centre 001, Le Centre Hospitalier Universitaire Montpellier, Montpellier, France

Diabetes Care 2010; 33: 121–7

Background: Attempts to build an artificial pancreas by using subcutaneous insulin delivery from a portable pump guided by a subcutaneous glucose sensor have encountered delays in and variability of insulin absorption. In an effort to improve on the consistency and timeliness of such a closed-loop system, this paper tested closed-loop intraperitoneal (IP) insulin infusion from an implanted pump driven by a subcutaneous glucose sensor via a proportional–integral–derivative (PID) algorithm.

Methods: Two-day closed-loop therapy (except for a 15-min pre-meal manual bolus) was compared with a 1-day control phase with IP open-loop insulin delivery, according to a randomised order, in a hospital setting with eight patients with type 1 diabetes treated with implanted pumps. The primary end-point was the percentage of time spent with BG in the range 4.4–6.6 mmol/l.

Results: During the closed-loop phases, the mean ± SEM percentage of time spent with BG in the 4.4–6.6 mmol/l range was significantly higher (39.1% ± 4.5% vs. 27.7% ± 6.2%, p = 0.05), and overall dispersion of BG values was reduced among patients. Better closed-loop glucose control came from the time periods excluding the two early postprandial hours with a higher percentage of time in the 4.4–6.6 mmol/l range (46.3% ± 5.3% vs. 28.6% ± 7.4%, p = 0.025) and lower mean BG levels (6.9 ± 0.3 vs. 7.9 ± 0.6 mmol/l, p = 0.036). Time spent with BG < 3.3 mmol/l was low and similar for both investigational phases.

Conclusions: The results of this paper demonstrate the feasibility of IP insulin delivery for a closed-loop artificial pancreas and support the need for further study of this form of insulin delivery. The features of the pre-meal bolus still need to be explored in terms of timing and amount.

  • Comment: Dr. Renard and collaborators have demonstrated the use of a hybrid closed-loop system that utilised prandial insulin delivery, subcutaneous glucose measurement and a PID algorithm. Subcutaneous insulin delivery is one of the limitations of a closed-loop system because of delays and modulation of insulin absorption. A closed-loop insulin delivery system that will administer insulin via the IP route should achieve superior glucose control and minimal to no hypoglycaemic episodes due to the lake of subcutaneous depot and fast delivery method. The results presented in the study support the use of IP delivery for the artificial pancreas. Tighter glucose control was obtained in the non-postprandial periods compared to the control group. However, overnight control was as good as the control group and prandial control remained a challenge for the PID algorithm even with a 30% pre-meal bolus.

IP insulin delivery may be one of the better methods to regulate glycaemia and to minimise hypoglycaemia using implantable or external pumps. The research of IP delivery as a way for faster insulin action is undergoing and should be considered as a substitute for subcutaneous delivery. The use of an external pump with IP delivery using U-100 insulin will be an improvement on the implantable system that uses U-400 and requires frequent maintenance. Future studies are needed; these studies should explore better and more sophisticated control strategies than the PID algorithm. Furthermore, reducing the occurrence of nocturnal hypoglycaemia and improving prandial glucose control without a pre-meal bolus given by the patient should also be investigated.

MD-Logic Artificial Pancreas system: a pilot study in adults with type 1 diabetes

E. Atlas, R. Nimri, S. Miller, E. A. Grunberg, M. Phillip

The Jesse Z. and Sara Lea Shafer Institute for Endocrinology and Diabetes, National Center for Childhood Diabetes, Schneider Children’s Medical Center of Israel, Petah Tikva, Israel

Diabetes Care 2010; 33: 1072–6

Background: The authors describe the principles and clinical performance of the MD-Logic Artificial Pancreas (MDLAP) system for controlling BG in adults with type 1 diabetes.

Methods: This system uses fuzzy logic theory to imitate the decision-making processes of experts in diabetes care. Seven adults with type 1 diabetes (aged 19–30 years, mean diabetes duration 10 ± 4 years, mean A1c 6.6% ± 0.7%) underwent 14 full closed-loop control sessions of 8 h (12 total) and 24 h (two total). During the 8-h sessions, subjects arrived at the clinic after an overnight fast, were connected to the MDLAP system and were monitored for 8 h in the fasted state or after a mixed meal consisting of 40–60 g carbohydrate. In the 24-h sessions, subjects consumed three mixed meals and slept approximately 8 h while connected to the MDLAP system.

Results: Peak postprandial BG excursion was limited to 224 ± 22 mg/dl, returned to < 180 mg/dl within 2.6 ± 0.6 h and was kept stable in the normal range for at least 1 h in all cases. In the 24-h closed-loop control, BG levels were kept in the normal range (between 70 and 180 mg/dl) 73% of the time, were > 180 mg/dl 27% of the time and never fell below 70 mg/dl. Symptomatic hypoglycaemia was not noted during any of the trials.

Conclusions: This pilot study shows the MDLAP system to be a promising tool for individualised glucose control in patients with type 1 diabetes. Postprandial glucose highs were minimised and hypoglycaemia was prevented.

  • Comment: This paper presents a fuzzy logic controller which is a different control strategy from the common modern control strategies such as MPC, PID with insulin feedback that make use of insulin–glucose dynamics. The method relies on a fuzzy model/set that is based on expert knowledge and relies on the subject’s clinical parameters. As reported by the authors, this is the first clinical evaluation of such a control strategy. Encouraging results were presented with good glucose control and no hypoglycaemia episodes. Prandial glucose peaks were reasonable, although glucose values at mealtime were missing. The time to maintain near normal glycaemia was good as well. However, these results are no different from other closed-loop systems. The advantage of fuzzy logic over other methods currently used was not proven and will require comparative studies in different daily life scenarios. As indicated by the authors the controller design is based on traditional treatment/individual subject treatment plan. This design may limit the system since there is a need to tune the patient’s basal rate and correction factor in order to reach good control. The system should therefore be further examined with moderately to poorly controlled patients. Closed-loop artificial pancreas systems will be based on different control strategies and it is unlikely that only one strategy will be used. Researchers should be encouraged to seek different good control methods such as fuzzy logic to overcome the glucose regulation problem of type 1 DM.

Closed-loop artificial pancreas using subcutaneous glucose sensing and insulin delivery and a model predictive control algorithm: preliminary studies in Padova and Montpellier

D. Bruttomesso,1A. Farret,2S. Costa,1M. C. Marescotti,1M. Vettore,1A. Avogaro,1A. Tiengo,1C. Dalla Man,3J. Place,2A. Facchinetti,3S. Guerra,3L. Magni,4De Nicolao G4, C. Cobelli,3E. Renard,2A. Maran1

1Department of Clinical and Experimental Medicine, Division of Metabolic Diseases, University of Padova, Padova, Italy,2Department of Endocrinology, University Hospital Centre, University of Montpellier, Montpellier, France,3Department of Information Engineering, University of Padova, Padova, Italy, and4Dipartimento di Informatica e Sistemistica, University of Pavia, Pavia, Italy

J Diabetes Sci Technol 2009; 3: 1014–21

Background: MPC algorithms use data from an individual’s daily life events to build personalised, predictive models to optimise BG control. The authors describe the preliminary clinical experience using an MPC algorithm for closed-loop control of BG in patients with type 1 diabetes.

Methods: A total of six patients with type 1 diabetes participated in two pilot studies (Montpellier and Padova, three in each study). Patients were aged 36 ± 8 and 48 ± 6 years, duration of diabetes 12 ± 8 and 29 ± 4 years, haemoglobin A1c 7.4% ± 0.1% and 7.3% ± 0.3%, body mass index 23.2 ± 0.3 and 28.4 ± 2.2 kg/m2, respectively, and were all currently using continuous subcutaneous insulin infusion. Each subject was studied during two separate 22-h overnight hospital admissions 2–4 weeks apart. In both trials, patients used a Freestyle Navigator® continuous glucose monitor and an OmniPod® (Insulet, Bedford, MA) insulin pump. During admission 1 patients followed their regular insulin dosing regimen (‘open loop’) while closed-loop control with the group’s MPC algorithm was used in admission 2.

Results: In Montpellier, two out of three subjects had superior BG control in the closed-loop system compared to the open-loop procedure. In Padova, mean overnight plasma glucose was 134 mg/dl in open-loop vs. 111 mg/dl in closed-loop. Percentage of time spent above 140 mg/dl was 45% vs. 12% and post-breakfast mean plasma glucose was 165 vs. 156 mg/dl during open-loop versus closed-loop trials. The closed-loop system provided a clear advantage in avoiding nocturnal hypoglycaemia in both studies.

Conclusions: These studies demonstrate that closed-loop control using currently available continuous glucose monitors and insulin pumps along with an MPC algorithm is feasible and can be used to achieve improved BG control in patients with type 1 diabetes. However, the algorithm should be improved to offer better postprandial BG control.

Closed-loop artificial pancreas using subcutaneous glucose sensing and insulin delivery and a model predictive control algorithm: the Virginia experience

W. L. Clarke,1S. Anderson,2M. Breton,3S. Patek,4L. Kashmer,1B. Kovatchev3

1Division of Pediatric Endocrinology, Department of Pediatrics, University of Virginia Health Sciences Center, Charlottesville, VA, USA,2Department of Internal Medicine, University of Virginia, Charlottesville, VA, USA,3Diabetes Technology Program, University of Virginia, Charlottesville, VA, USA, and4Department of Systems and Information Engineering, University of Virginia, Charlottesville, VA, USA

J Diabetes Sci Technol 2009; 3: 1031–8

Background: The authors test the ability of a closed-loop system consisting of a personalised MPC algorithm in combination with a continuous glucose monitor and insulin pump to keep BG levels between 70 and 140 mg/dl overnight and to control postprandial BG levels.

Methods: Eight adults with type 1 diabetes aged 27–51 years, average diabetes duration 3–26 years, all using continuous subcutaneous insulin infusion, participated in this pilot study. Each subject was studied overnight and for 4 h following a standardised meal, once using open-loop control and once using a closed-loop system controlled by the MPC algorithm. The authors compared the two trials for average BG levels, percentage of time with BG between 70 and 140 mg/dl, number of hypoglycaemic episodes and postprandial BG excursions.

Results: The closed-loop system was effective in maintaining nocturnal BG levels between 70 and 140 mg/dl in seven out of eight subjects, with one subject’s BG falling to 65 mg/dl, which signalled the insulin pump to shut off. Closed-loop control resulted in significantly fewer nocturnal hypoglycaemic events compared with open-loop control (p = 0.001). Postprandial glucose levels were similar in the two groups.

Conclusions: A closed-loop system using MPC is an effective means to control BG overnight and following a standard breakfast meal. This system clearly outperformed open-loop control in preventing overnight hypoglycaemia, but did not offer a significant advantage following a morning meal.

  • Comment: The papers by Bruttomesso et al. and Clarke et al. present preliminary results from a pilot study conducted in three centres that compared an MPC-based artificial pancreas. The study protocol included overnight and breakfast control and lasted 4–5 h post breakfast. The presented closed-loop system succeeded in preventing nocturnal hypoglycaemia in most cases and provided superior glucose control compared to open-loop. However, in terms of other glucose control measures, glucose control during the night under closed-loop control was as good as under open-loop control. The authors used the open-loop session to personalise the MPC algorithm, i.e. the algorithm learned patient meal routine and had the ability to infuse insulin prior to the consumption of the meal, even though post breakfast glucose control under closed-loop was not better than the open-loop admission. These studies include two identical experimental sessions: in the first one the subjects followed their standard treatment (basal profile and bolus calculations) and the second one was a closed-loop study. It should be noted that on the one hand this protocol setting may help in reducing unknown factors such as meal size and mealtimes, but on the other hand the ‘white coat syndrome’ may affect the subject’s decision-making during the first admission and the study results from the first admission. It may be useful to compare the closed-loop performance with ambulatory data from the same individuals to demonstrate better the benefit of closed-loop control.

The control algorithm was tailored to the subject via an aggressiveness parameter q that was designed using clinical parameters. Bruttomesso et al. reported in two cases that the value of the controller’s aggressiveness was insufficient and affected the system’s ability to improve glucose control. The overall results demonstrated the feasibility of closed-loop control and overnight glycaemic control facing sensor dropouts and model mismatch. The authors reported results suggested that the controller managed to overcome a 50 g liquid breakfast meal; however, the trials ended, per protocol, prior to the demonstration of glucose stabilisation which undermined this conclusion. Further evaluation should consider sufficient time postprandial to ensure glucose stabilisation.

Hypoglycaemia prevention and closed-loop control

  1. Top of page
  2. Abstract
  3. Clinical evaluation of a bi-hormonal closed loop
  4. Clinical evaluation of uni-hormonal closed loop
  5. Hypoglycaemia prevention and closed-loop control
  7. References

Real-time hypoglycaemia prediction suite using continuous glucose monitoring: a safety net for the artificial pancreas

E. Dassau,1,2F. Cameron,3H. Lee,4B. W. Bequette,4H. Zisser,1,2L. Jovanovič,1,2H. P. Chase,5D. M. Wilson,6B. A. Buckingham,6F. J. Doyle III1,2

1Department of Chemical Engineering, University of California Santa Barbara, Santa Barbara, CA, USA,2Sansum Diabetes Research Institute, Santa Barbara, CA, USA,3Department of Aeronautics and Astronautics, Stanford University, Palo Alto, CA, USA,4Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA,5Barbara Davis Center for Childhood Diabetes, University of Colorado Health Sciences Center, Aurora, CO, USA, and6Department of Pediatrics, Division of Pediatric Endocrinology, Stanford Medical Center, Palo Alto, CA, USA

Diabetes Care 2010; 33: 1249–54

Background: Although tight glycaemic control brings benefits to diabetic patients, it also increases the risk of hypoglycaemia, especially during sleep. The present study was targeted to develop an advanced algorithm that detects pending hypoglycaemia and then suspends basal insulin delivery. This approach can provide a solution to the problem of nocturnal hypoglycaemia.

Methods: This real-time hypoglycaemic prediction algorithm (HPA) combines five individual algorithms: linear projection, Kalman filtering, hybrid infinite impulse–response filter, statistical prediction and numerical logical algorithm. All are based on CGM 1-min data. An alarm is issued if either a certain number (three, four or five) of the algorithms all predict hypoglycaemia, or the sensor interstitial glucose value is below the hypoglycaemic threshold (70, 80 or 90 mg/dl, respectively). Each combination was assigned three different prediction horizons (V) of 35, 45 and 55. The HPA system was developed using data derived from 21 Navigator studies that assessed Navigator function over 24 h in children with type 1 diabetes. The function of HPA was confirmed using a separate dataset from 22 admissions of subjects with T1DM. During these admissions, hypoglycaemia was induced by gradually increasing the basal insulin infusion rate up to 1.8 times the subject’s own baseline infusion rate.

Results: Using a prediction horizon of 35 min, a glucose threshold of 80 mg/dl, and a voting threshold of three out of five algorithms to predict hypoglycaemia (defined as FreeStyle plasma glucose readings < 60 mg/dl), the HPA predicted 91% of the hypoglycaemic events. When the voting threshold was changed to four of five algorithms, 82% of the events were predicted.

Conclusions: The use of HPA will enable triggering of a warning alarm or automated insulin pump suspension in response to a pending hypoglycaemic event that has been detected prior to severe immediate complications.

Prevention of nocturnal hypoglycaemia using predictive alarm algorithms and insulin pump suspension

B. Buckingham,1H. P. Chase,2E. Dassau,3E. Cobry,2P. Clinton,1V. Gage,2K. Caswell,1J. Wilkinson,2F. Cameron,4H. Lee,5B. W. Bequette,5F. J. Doyle III3

1Department of Pediatric Endocrinology, Stanford University, Stanford, CA, USA,2Department of Pediatrics, University of Colorado, Aurora, CO, USA,3Department of Chemical Engineering, University of California, Santa Barbara, CA, USA,4Department of Aeronautics and Astronautics, Stanford University, Stanford, CA, USA, and5Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA

Diabetes Care 2010; 33: 1013–17

Background: Preventing nocturnal hypoglycaemia remains one of the most challenging goals in the treatment of diabetes. The aim of the present study was to develop a partial closed-loop system to safely prevent nocturnal hypoglycaemia in type 1 diabetes patients by suspending insulin delivery when hypoglycaemia was predicted.

Methods: A total of 40 subjects with type 1 diabetes (age 12–39 years) were studied overnight in the hospital. The control group (= 14) and two experimental groups (= 10 and = 16) were all induced with hypoglycaemia by gradually increasing the basal insulin infusion rate. No pump-shutoff algorithms were used in the control group; in the experimental groups, pump shutoff occurred when either three of five (= 10) or two of five (= 16) algorithms predicted hypoglycaemia based on the glucose levels measured with the FreeStyle Navigator (Abbott Diabetes Care). The insulin pump was suspended manually for 90 min for each shutoff. The five algorithms are a modified linear prediction alarm, Kalman filtering, adaptive hybrid infinite impulse–response filter, statistical prediction and a numerical logical algorithm.

Results: Hypoglycaemia occurred on 13 (93%) of the 14 nights in the control group. In the experimental group that required three algorithms to trigger insulin pump suspension, nocturnal hypoglycaemia was prevented for six (60%) out of 10 nights. In the experimental group that required only two algorithms to trigger insulin pump suspension, hypoglycaemia was prevented for 12 (75%) of 16 nights. There were 25 predictions of hypoglycaemia in this group because some subjects had multiple hypoglycaemia predictions during one night, and hypoglycaemia was prevented for 84% of these events.

Conclusions: Using algorithms to suspend the insulin pump when hypoglycaemia is predicted, it is possible to prevent hypoglycaemia on 75% of nights (84% of events) when it would otherwise be predicted to occur.

  • Comment: One of the major obstacles in intensive insulin management is a higher tendency of hypoglycaemia. The striking fact is that most severe hypoglycaemic events occur overnight (1). Thus, an automatic system that will track BG concentrations and act accordingly may prevent the occurrence of nocturnal hypoglycaemia and improve night-time glucose control of people with type 1 DM.

The HPA is an improved method for the prevention of nocturnal hypoglycaemia which is based on five different individual algorithms as presented by Dassau et al. and Buckingham et al. Each algorithm analyses the available glucose data, separately, and concludes whether a hypoglycaemic event is anticipated or not. The algorithms can be tuned using three parameters: threshold glucose concentration, prediction horizon (in min) and the required number of concurrent algorithms needed to issue an alarm. Using a voting algorithm the system decides whether to suspend insulin delivery for a predefined period of 90 min. This algorithm was initially designed and evaluated on historical clinical data and further evaluated on the University of Virginia/University of Padova metabolic simulator (2) (accepted by the US Food and Drug Administration) with promising results (3).

Buckingham et al. reported a prospective clinical study aimed at evaluating the hypoglycaemia prevention system. A cohort of 26 patients was investigated during the clinical phase where with the first 10 patients a voting threshold of three methods out of five was used; with the last 16 patients this was modified to two out of five in order to improve the performance of the method. Hypoglycaemia was induced by increasing the basal rate of the patient 1.8-fold. In this study, the investigators used 80 mg/dl as a threshold level and a prediction horizon of 35 min. When a hypoglycaemic event was announced by the algorithm insulin was suspended for 90 min and was resumed to default basal rate. In the last 11 patients basal rate was resumed earlier if certain conditions were met. The ability of the algorithm to detect and prevent hypoglycaemic events was improved from the first part of the trial (71% of events were detected) to the second, in which 84% of events were prevented. Although this system presents good results it still cannot guarantee 100% prevention of an event. The use of an automated alert system (4) as well as automated glucagon administration should be investigated. One of the limits of the study is the clinical protocol to promote a high tendency for hypoglycaemia during the study nights, as discussed by the authors. In order to achieve a high rate of hypoglycaemia the investigators increased nocturnal basal rate until there was a > 80% risk of hypoglycaemia. This method may not be similar to the dynamics behind hypoglycaemia events in real life. In order to further examine the performance of this algorithm, a randomised control study evaluating the performance of the HPA vs. control in a group of patients with a tendency to hypoglycaemia should be performed.


  1. Top of page
  2. Abstract
  3. Clinical evaluation of a bi-hormonal closed loop
  4. Clinical evaluation of uni-hormonal closed loop
  5. Hypoglycaemia prevention and closed-loop control
  7. References

Closed-loop control of artificial pancreatic beta-cell in type 1 diabetes mellitus using model predictive iterative learning control

Y. Wang, E. Dassau, F. J. Doyle III

Department of Chemical Engineering and Biomolecular Science and Engineering Program, University of California, Santa Barbara, CA, USA, and Sansum Diabetes Research Institute, Santa Barbara, CA, USA

IEEE Trans Biomed Eng 2010; 57: 211–19

Background: This study proposes a combination of iterative learning control and MPC, called model predictive iterative learning control (MPILC), as a way of improving glycaemic control in patients with type 1 diabetes.

Methods: MPILC combines frequent glucose readings via CGM with the repetitive nature of glucose–meal–insulin dynamics in a 24-h cycle. The proposed algorithm can learn from an individual’s lifestyle, allowing BG control to be improved from day to day.

Results: After less than 10 days, BG concentrations can be kept within a range of 90–170 mg/dl. In general, BG control under MPILC is better than that under MPC. The proposed methodology is also robust to random variations in meal timings within ± 60 min or meal amounts within ± 75% of the nominal value, which further validates MPILC’s superior robustness compared to run-to-run control.

Conclusions: MPILC can improve BG control in people with type 1 diabetes. The algorithm can both improve control from day to day and take advantage of frequent glucose measurement to design the basal and bolus insulin simultaneously; it is also robust to meal variations and subject variability. Furthermore, the use of an adaptive set-point guarantees safe and stable convergence of the algorithm even when subjects are not well controlled and have high glucose levels. This algorithm may be especially useful in treatment of children and adolescents because it does not necessitate user intervention.

  • Comment: Inter- and intra-variability in insulin sensitivity is a well-known phenomenon among diabetes caregivers and patients. Insulin sensitivity tends to vary during the course of the day, from day to day and as a result of specific events such as physical activity, illness and stress. An artificial pancreas system, which relies on patient-specific parameters, should incorporate algorithms that will allow it to learn and overcome the inherent variability in insulin sensitivity.

The authors describe a novel methodology to implement a learning iterative control algorithm within a closed-loop system which is based on model predictive control (MPILC). This interesting combination does not rely on the subject’s input, and thus is very suitable for a closed-loop system. The MPILC is validated using in silico simulations for 50 days on a group of 10 adults. On the first day the in silico subjects were under open-loop control and on the other days they were under the control of the MPILC. Four different simulation settings were chosen in order to evaluate the robustness of the proposed system: (i) simulation day with a fixed amount of consumed carbohydrates at fixed times, (ii) variation of meal size, (iii) variation of mealtime and (iv) combination of the latter two. Although this method has great potential, performance results of the MPILC should be considered with caution, since diurnal as well as day to day insulin variability was not considered in this initial work. Extensive clinical trials are required to validate this method and its ability to regulate glucose control.


  1. Top of page
  2. Abstract
  3. Clinical evaluation of a bi-hormonal closed loop
  4. Clinical evaluation of uni-hormonal closed loop
  5. Hypoglycaemia prevention and closed-loop control
  7. References
  • 1
    The Diabetes Control and Complications Trial Research Group. the effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med 1993; 329: 97786.
  • 2
    Kovatchev BP, Breton M, Dalla Man C, Cobelli C. In silico preclinical trials: a proof of concept in closed-loop control of type 1 diabetes. J Diabetes Sci Technol 2009; 3: 4455.
  • 3
    Dassau E, Cameron F, Lee H, Bequette BW, Zisser H, Jovanovič L, Chase HP, Wilson DM, Buckingham BA, Doye III FJ: Real-time hypoglycemia Prediction suite using continuous glucose monitoring: a safety net for the artificial pancreas. Diabetes Care 2010; 33: 124954.
  • 4
    Dassau E, Jovanovič L, Doyle FJ III, Zisser H. Enhanced 911/GPS Wizard: a telemedicine application for the prevention of severe hypoglycaemia – monitor, alert and locate. J Diabetes Sci Technol 2009; 3: 15016.