Closing the loop

Abstract

Closed-loop systems have become necessary components of our technologically dependent world. Algorithms are responsible for implementing control in a variety of settings ranging from the temperature of your home to the altitude and speed at which you fly on an airplane. The closed-loop algorithms eliminate the need for constant monitoring and adjustment by an individual. Diabetes mellitus is a disease where the human body’s innate ability to control blood glucose levels fails, making glycaemic control extrinsic and up to the conscious efforts of that individual. However, the ever-increasing amount of variation and unpredictability in people’s lives have led to difficulties with manual ‘open-loop’ control and have ushered in the need to develop a system that can more effectively control blood glucose. An artificial pancreas aims to greatly reduce the difficulty of maintaining euglycaemia and to improve the individual’s quality of life in the process. Although the idea of an artificial pancreas has existed for decades, the feasibility of such a device has only recently come into focus with the combination of user friendly continual glucose monitors and insulin infusion devices. The degree of control that the algorithm possesses as well as the method of administration of one or two hormones are the variables that researchers are currently adjusting. This past year has seen dramatic advances in clinical trials of closed-loop systems, ranging from a multinational study of subcutaneous model predictive closed-loop control to a study of closed-loop insulin delivery during pregnancy complicated by type 1 diabetes (T1D). Trials with overnight control in adults and bi-hormonal infusion have also provided further evidence for the feasibility of a closed-loop system. Advances in controller technology have led to the use of euglycaemic zones in place of discrete set-points and the development of adaptive model predictive controllers based on the utilisation of past behaviour profiles. The inclusion of new safety precautions aims to further reduce the amount of time spent outside the euglycaemic range. Further refinement of controllers in silico brings researchers closer to achieving the ultimate goal of fully automated artificial pancreases. Our goal was to select the most influential and ground-breaking publications that focus on the quest to make an artificial pancreas available to people afflicted with diabetes mellitus around the world. Hovorka R Institute of Metabolic Science, University of Cambridge, Cambridge, UK Nat Rev Endocrinol 2011; 7 : 385–95 The artificial pancreas consists of a continuous glucose monitoring (CGM) device, an insulin pump, and a control algorithm that mediates the delivery of insulin to achieve glycaemic control. Two algorithmic approaches, a model predictive control and a proportional-integral-derivative control, have emerged as the top choices for the controller of the closed-loop method. These algorithms are implemented in a number of different systems that vary in complexity from the more simple suspended insulin delivery to the intricate dual-hormone closed-loop system using both insulin and glucagon. The introduction of artificial pancreas prototypes into clinical practice has already begun with the use of a low-glucose suspend pump in 2009 in parts of Europe. The more complex systems, however, are still being tested either in silico or in vivo and will not see clinical use for at least several years. These closed-loop therapy systems are targeted for use in patients who struggle with hypoglycaemia and children who have difficulties managing their diabetes; however, it is hoped that all patients suffering from T1D will be able to use a fully closed-loop artificial pancreas in the near future. Comment: Hovorka provides a synopsis of the current status of closed-loop artificial pancreas technology as well as a brief yet effective analysis of the controllers that have undergone clinical studies. The inclusion of the suspended insulin delivery system in the review reflects the increasing utilisation and research of semi-closed-loop delivery methods as a stepping stone in the development of a fully automated closed-loop artificial pancreas. The different physiological and lifestyle factors, such as subcutaneous insulin delivery and exercise, serve as the biggest hurdles for closed-loop glycaemic control. Hovorka rightly points out that the future success of an artificial pancreas does not rest merely on the effectiveness of the control algorithm, but also on the establishment of an appropriate infrastructure to educate both healthcare professionals and patients; user-friendliness and portability will factor greatly into the wide acceptance of the artificial pancreas. However, there is also a stress not to delay clinical use of such a medical device because of its phased process of improvement. Small future updates, as with any other software, should be expected. Although hybrid insulin delivery systems are probably the next step, fully automated closed-loop control will hopefully reach clinical use in the next one to two decades. Pickup JC Diabetes Research Group, Division of Diabetes and Nutritional Sciences, King’s College London School of Medicine, Guy’s Hospital, London, UK Diabetes Technol Ther 2011; 13 : 695–8 Although much of the current research involving the artificial pancreas involves closed-loop insulin delivery, the concept of using a hybrid or semi-closed-loop delivery system has existed for over 30 years. Current subcutaneous glucose measurement and insulin infusion in the mealtime setting present large hurdles for the closed-loop method; therefore, the simplicity of the semi-closed-loop system has regained attention. Although many attempts at hybrid insulin delivery have involved long periods of closed loop with short pre-meal boluses, ‘control-to-target’ or ‘control-to-range’ stress the exact opposite approach. The majority of the time is spent in an open-loop setting with short periods of closed-loop management when blood glucose levels become too high or low. An insulin pump with an automatic low-glucose suspend (LGS) (Paradigm® VeoTM, Medtronic Inc., Northridge, CA, USA) has been used clinically in parts of Europe since 2009. The LGS functions by stopping basal insulin delivery when the blood glucose levels fall below a certain set-point. Studies have shown that the LGS pump has effectively reduced the frequency of hypoglycaemic events, especially nocturnal incidences, in both adult and paediatric diabetic patients (1, 2). There is still much to be studied about the LGS pump including the personal thresholds and the best response to an LGS alarm. Future research is aimed at using insulin boosters to eliminate hyperglycaemic events and the use of an algorithm that can predict the trend towards a hyperglycaemic or hypoglycaemic event rather than a reaction once the event has already occurred. Comment: The challenges of a fully automated closed-loop insulin delivery system have made the semi-closed-loop system a likely first step in the development of an artificial pancreas. Pickup explains that the hybrid control offers overall better glycaemic control than current insulin management with minimal user intervention. LGS pumps are already being used in Europe and it seems likely that pumps that both suspend and boost insulin delivery are the next steps. Other semi-closed-loop systems have been developed that focus on a long period of closed-loop management with short open-loop periods. Research is on its way with different strategies to address meal disturbance and to find ways to minimise postprandial hyperglycaemic events. However, the semi-closed-loop model should be seen as an intermediate step in the development process, not the end-product. Even though the hybrid model is currently the most effective controller, insulin infusion without any user intervention is the ultimate goal. Kovatchev B 1,2 , Cobelli C 3 , Renard E 4 , Anderson S 5 , Breton M 1 , Patek S 2 , Clarke W 6 , Bruttomesso D 7 , Maran A 7 , Costa S 7 , Avogaro A 7 , Mann CD 3 , Facchinetti A 3 , Magni L 8 , Nicolao GD 8 , Place J 4 , Farre A 4 1 Department of Psychiatry and Neurobehavioral Sciences, University of Virginia, Charlottesville, VA, USA, 2 Department of Systems and Information Engineering, University of Virginia, Charlottesville, VA, USA, 3 Department of Information Engineering, University of Padova, Padova, Italy, 4 Department of Endocrinology and UMR CNRS, CHU and University of Montpellier, Montpellier, France, 5 Department of Medicine, Section Endocrinology, University of Virginia, Charlottesville, VA, USA, 6 Department of Pediatrics, University of Virginia, Charlottesville, VA, USA, 7 Department of Clinical and Experimental Medicine, University of Padova, Padova, Italy, and 8 Department of System and Informatics, University of Pavia, Pavia, Italy J Diabetes Sci Technol 2010; 4 : 1374–81 Background: Increasing effort has been focused on the development of subcutaneous–subcutaneous (SC-SC) closed-loop glucose control, using CGM coupled with an insulin pump and a control algorithm. This paper summarised data found in a previous publication. Methods: The design of the control algorithm was done entirely in silico. Adults (n = 20) were recruited for the clinical experiments from the USA, Italy and France. All subjects participated in both open-loop and closed-loop sessions, which were scheduled 3–4 weeks apart lasting 22 h each. A CGM and an insulin pump were used. During open-loop control the patient performed insulin dosing under physician supervision, whereas a control algorithm performed insulin dosing during closed-loop control. Results: The in silico design resulted in quick and cost-effective system development, testing and regulatory approvals (which took less than 6 months compared with the years it would have taken with animal trials). Closed-loop control reduced nocturnal hypoglycaemia from 23 to five episodes (p < 0.01). There was a mean of 1.15 hypoglycaemic episodes per subject overnight on open-loop control, which was reduced to 0.25 episodes per subject on closed-loop control. Closed-loop control also increased the amount of time spent overnight within the target range from 64% to 78% (p = 0.03). However, the percentage of time within the blood glucose target range of 3.9–10 mmol/l during postprandial breakfast control decreased from 61% in open-loop control to 52% in closed-loop control. This equates to 9% less time spent below 10 mmol/l on closed-loop after breakfast with 0.4 mmol/l higher average blood glucose. During closed-loop control, an attending physician decided to override the insulin suggested by the closed-loop algorithm on four occasions, resulting in 2.5 h (2%) loss of closed-loop control time in all 20 patients. Conclusions: A system using personalised model predictive control (MPC) to control blood glucose in T1D was developed in silico and then tested in a clinical setting. Subsequent studies have adopted such technology in their clinical approaches. Comment: The practicality of an ‘artificial pancreas’ has been contested by the feasibility of its outpatient use due mainly to the cumbersome technology it requires and the difficulties of creating a closed-loop system that is fully automated. Applications such as minimally invasive subcutaneous CGM, which samples glucose through a minimally invasive sensor implanted in the subcutaneous tissue, have risen to the forefront of artificial pancreas research. This paper by Kovatchev and co-workers made two significant contributions: (1) in silico designs for control algorithms; and (2) multinational testing of SC-SC closed-loop glucose control using fully automated CGM data transfer. The study was a pilot study designed to evaluate a hybrid control methodology. Hence, all subjects participated in both open-loop and closed-loop control under identical conditions in a tightly controlled hospital setting; however, the order of open-loop vs. closed-loop control was not randomised. Typically, randomisation is necessary to avoid learning effects. In this study, data from the open-loop control were used to supply the algorithm with information regarding meals during the closed-loop control. Furthermore, the control cannot be considered fully automated because the algorithm did not automatically control the insulin pump. Future studies should include randomisation, real-life meals as well as full automation as a step toward ambulatory evaluation of the proposed technology. Overall, the study provided strong evidence for the use of closed-loop insulin delivery as it did reduce the number of hypoglycaemic events nearly fivefold and increased the time spent in the target glucose range. But despite the overall success of CGM performance, there were 15 instances across all subjects when the CGM devices suffered from a transient loss of sensitivity and the sensor readings did not correspond to the reference blood glucose. This highlights the challenge of CGM accuracy in the context of a closed-loop system. While the need for more accurate sensors may solve the problem, CGM technology may still face some inaccuracies that could raise a safety issue to a closed-loop system. Therefore, other solutions need to be developed, such as algorithms to detect sensor inaccuracy or drift and combining glucometer data as additional input to the closed-loop system. Hovorka R 1,2 , Kumareswaran K 1,3 , Harris J 1 , Allen JM 1,2 , Elleri D 1,2 , Xing D 4 , Kollman C 4 , Nodale M 1 , Murphy HR 1 , Dunger DB 1,2 , Amiel SA 5 , Heller SR 6 , Wilinska ME 1,2 , Evans ML 1,3 1 Institute of Metabolic Science, University of Cambridge, UK, 2 Department of Paediatrics, University of Cambridge, UK, 3 Department of Medicine, University of Cambridge, UK, 4 Jaeb Center for Health Research, Tampa, FL, USA, 5 Diabetes Research, Weston Education Centre, King’s College London, London, UK, and 6 Diabetes Centre, Clinical Sciences Centre, Northern General Hospital, Sheffield, UK BMJ 2011; 342 : d1855 Background: Overnight closed-loop control can reduce the risk of hypoglycaemia in children and adolescents. This study was done to extend these findings to adults. Methods: This study carried out two sequential, open-label, randomised controlled crossover studies comparing overnight closed-loop delivery of insulin with conventional insulin pump therapy after two meal scenarios: ‘eating in’ and ‘eating out’. Thirteen adults participated in the ‘eating in’ scenario [60 g carbohydrate (CHO) meal] and were randomly assigned overnight treatment with either closed-loop delivery of insulin or conventional insulin pump therapy during two separate study nights, separated by an interval of 1 to 3 weeks. Twelve adults were recruited for the ‘eating out’ scenario and followed the same protocol except that their meals were larger (100 g CHO) and accompanied by alcohol. The primary outcome was the time plasma glucose concentrations were in target (3.91–8.0 mmol/l) during closed-loop control. Results: In the “eating in scenario,” overnight closed-loop insulin delivery increased the time plasma glucose concentrations were in target by a median 15% (interquartile range 3%–35%), p = 0.002. In the “eating out” scenario, closed-loop insulin delivery increased the time plasma glucose concentrations were in target by a median 28% (2%–39%), p = 0.01. Analysis of the data showed that with closed-loop delivery the overall time plasma glucose was in target increased by a median 22% (3%–37%), p = 0.001. In conclusion, closed-loop delivery reduced overnight time spent hypoglycaemic (plasma glucose ≤ 3.9 mmol/l) by a median 3% (0%–20%), p = 0.04, and eliminated plasma glucose concentrations below 3.0 mmol/l after midnight. Conclusions: These two crossover trials provided evidence that in adults with T1D, closed-loop delivery of insulin may improve overnight control of glucose concentrations and reduce the risk of nocturnal hypoglycaemia. Comment: According to Hovorka and co-workers, no evidence prior to this study had been shown to support that overnight closed-loop insulin delivery in adults could reduce the number of glucose excursions. Thus, this study was designed to test overnight closed-loop systems in adults with T1D and made two significant contributions to the field: (1) overnight closed-loop delivery of insulin in adults with T1D improves nocturnal glucose control; and (2) overnight closed-loop delivery of insulin can operate safely and effectively across a range of age groups and lifestyles, including scenarios with large meals and alcohol consumption. The success of this study can be attributed to its use of a robust and computationally efficient algorithm, which resulted in a complete avoidance of hypoglycaemia, and the controller's ability to discriminate between both slowly and rapidly absorbed meals, thus coping with the considerable variability in gut absorption. In addition, the study's incorporation of ‘eating in’ and ‘eating out’ scenarios enhanced the real-world applicability of the control system and brings the artificial pancreas one step closer to outpatient use. One main downside of the study was its need for manual inputting of sensor glucose concentrations into the algorithm and manual adjustment of the insulin pump, which rendered the control semi-automated as opposed to fully automated. In addition, the study team delivered the meal bolus at mealtime based on exact estimation of the meal CHO content. While this study presented its limitations with closed-loop control, the research community should strive to refine their algorithms, miniaturise their devices, include real-life challenges as well as variation in meal content and time and then follow up with more advanced solutions for glucose control during both day and night. Murphy HR 1 , Elleri D 1,2 , Allen JM 1 , Harris J 1 , Simmons D 3 , Rayman G 4 , Temple R 5 , Dunger DB 2 , Haidar A 1 , Nodale M 1 , Wilinska ME 1,2 , Hovorka R 1,2 1 Institute of Metabolic Science, Metabolic Research Laboratories, University of Cambridge, Cambridge, UK, 2 Department of Paediatrics, University of Cambridge, Cambridge, UK, 3 Institute of Metabolic Science, Cambridge University Hospitals NHS Foundation Trust, Cambridge, UK, 4 Diabetes Centre, Ipswich Hospital NHS Trust, Ipswich, UK, and 5 Elsie Bertram Diabetes Centre, Norfolk and Norwich University Hospital NHS Trust, Norwich, UK Diabetes Care 2011; 34 : 406–11 Background: There has been evidence to support the use of insulin pump therapy, continuous glucose measurements (CGM) and sensor-augmented pump (SAP) therapy in pregnancy; however, the benefits have not been well established, particularly during late pregnancy. This study aimed to evaluate closed-loop insulin delivery with an MPC algorithm during early (12–16 weeks) and late (28–32 weeks) gestation in pregnant women with T1D. Methods: Ten women with T1D were studied over 24 h during early and late gestation. Basal insulin infusion rates were calculated using CGM sensor glucose values and the MPC algorithm every 15 min. A nurse adjusted the basal insulin infusion rate before each insulin delivery as a safety precaution. At 18:00 h, women ate a standardised evening meal. At 07:00 h, they ate a standardised morning meal. Prandial insulin boluses were calculated according to the women’s insulin–carbohydrate ratio and capillary fingerstick glucose concentrations. The study ended at 12:00 h. Mean glucose and time spent in target (63–140 mg/dl), hyperglycaemic (>140 to ≥180 mg/dl) and hypoglycaemic (<63 to ≤ 50 mg/dl) ranges were calculated using plasma and sensor glucose measurements. At the end of the study, models were used to compare glucose control during early and late gestation. Results: Throughout closed-loop insulin delivery, median plasma glucose concentrations were 117 mg/dl in early gestation and 126 mg/dl in late gestation (p = 0.72). The overnight mean plasma glucose time in target was 84% and 100% in early and late pregnancy, respectively (p = 0.09). Overnight mean time spent hyperglycaemic (>140 mg/dl) was 7% in early and 0% in late pregnancy (p = 0.25) and hypoglycaemic (<63 mg/dl) was 0% and 0%, respectively (p = 0.18). Postprandial glucose control, glucose variability, insulin infusion rates and CGM sensor accuracy were no different in early or late pregnancy. However, it must be noted that time spent with plasma glucose within the target range after breakfast was 59% in early and 47% in late pregnancy, and time spent hyperglycaemic after breakfast was 28% in early and 44% in late pregnancy. Sensor accuracy was reported as the mean absolute relative difference between sensor glucose and paired plasma glucose divided by plasma glucose and was 13.3% (14.7% in early vs. 11.9% in late pregnancy; p = 0.15). Conclusions: This study demonstrates that closed-loop control could be performed in women with T1D during pregnancy; however, much more work is needed to achieve optimal glycaemic control. Comment: Pregnant women suffering from T1D are faced with physiological and hormonal changes during pregnancy, which can contribute to poor glycaemic control and progressive insulin resistance in late gestation. Prior to this study, the effectiveness of a closed-loop system in pregnant women with T1D had not been reported. The study was unique in that it was the first to use plasma glucose measurements during pregnancy, which allowed for the evaluation of sensor accuracy. Unfortunately, the ability of the glucose sensors and algorithm to achieve tight glycaemic control was less than optimal. In this proof of concept study, the algorithm was not modified to distinguish between preprandial and postprandial glucose targets, and postprandial glycaemia was shown to be a major challenge in the closed-loop system. In addition, the closed-loop system was not fully automated. During closed-loop insulin delivery, a nurse had to manually adjust the insulin rate based on the values given by the CGM, which were fed into the MPC algorithm. By way of current CGM technology and tighter glycaemic targets to treat pregnant women with T1D, it can be inferred that the risk for hypoglycaemic events increases because of lower glycaemic setpoint. To confirm clinical effectiveness of closed-loop insulin delivery in women with T1D, a larger randomised study comparing closed-loop with SAP therapy will be needed. Steil GM 1 , Palerm CC 2 , Kurtz N 2 , Voskanyan G 2 , Roy A 2 , Paz S 3 , Kandeel FR 3 1 Children’s Hospital Boston, Boston, MA, USA, 2 Medtronic MiniMed, Northridge, CA, USA, and 3 City of Hope National Medical Center, Duarte, CA, USA J Clin Endocrinol Metab 2011; 96 : 1402–8 Background: Initial studies of closed-loop proportional integral derivative (PID) control in individuals with T1D have shown good overnight performance; however, breakfast meals are routinely more difficult to control and require supplemental carbohydrates to prevent hypoglycaemia. This study assessed the ability of insulin feedback (IFB) to improve the breakfast meal profile. Methods: Subjects with previously diagnosed with T1D (n = 8) were recruited for participation with closed-loop control over approximately 30 h at an inpatient clinical research facility. Participants completed both an open-loop CGM procedure and a closed-loop PID-with-IFB procedure. During open-loop monitoring, adjustments were made to normalise blood glucose to between 90 and 120. An intravenous catheter was inserted for collecting blood samples. Meals were served at 07:00 (44.5 g) [breakfast on day 1 (B1)], 12:00 (62.5 g) [lunch (L)] and 18:00 h (59.5 g) [dinner (D)], with a snack given at 21:00 h and breakfast the next day (B2) at 07:00 h (45 g). A manual 2-U meal bolus was delivered before every meal. If plasma glucose fell below 50 mg/dl, 15 g of supplemental carbohydrate was provided. Outcome measures were plasma insulin concentration, model-predicted plasma insulin concentration, 2-h postprandial and 3- to 4-h glucose rate of change following breakfast after 1 day of closed-loop control. There was also a measure of the need for supplemental carbohydrate in response to nadir hypoglycaemia. Results: Plasma concentrations during closed-loop were well correlated with model predictions. Two-hour postprandial glucose concentration values were 138 ± 24, 158 ± 17, 138 ± 9 and 132 ± 16 mg/dl (B1 L, D and B2, respectively). No hypoglycaemia was observed overnight. Sensors tracked plasma glucose with a mean absolute relative difference of 11.9% (7.3%–20.6%). Glucose concentration during the 3–4 h period after B2 was stable (rate of change of glucose, –0.03792 ± 0.0884 mg/dl/min, not different from zero; p = 0.68) and at daytime target (97 ± 6 mg/dl, not different from 90; p = 0.28). During the 30-h closed-loop period, supplemental carbohydrate was given on eight occasions and three subjects received supplemental carbohydrate during B2. Overall, six instances of hypoglycaemia requiring supplemental carbohydrate were observed, and two subjects did not require any interventions with closed-loop insulin delivery. Conclusions: PID control with insulin feedback can achieve a desired breakfast response but more studies will be necessary to achieve optimal control. Comment: Steil and co-workers have been conducting research and clinical trials on the closed-loop system for years, but this study was the most successful in obtaining optimal results. A previous study in 2006 showed peak breakfast glucose concentrations on day 2 reported as 231 ± 12 mg/dl, which were reduced in a study done in 2008 to 204 ± 17 mg/dl, and in this study to 175 ± 8 mg/dl with the use of insulin feedback (3, 4). The overall results of the study are good with excellent glucose control. However, the tuning of the PID algorithm remains a problem with what seems like excessive delivery due to integral error for some of the subjects. Although this study highlights the need for an improved system, the effect of insulin feedback on closed-loop glucose control was shown to improve overnight control. This study reported that it was unclear whether the need for supplemental carbohydrate, eight times during a nine-patient study, could be eliminated by changes in the PID algorithm parameters or whether the algorithm needed further modification. During this trial, pre-meal boluses were also administered, which renders the system semi-automated. Future IFB and PID closed-loop algorithms for the artificial pancreas will require better control tuning as well as additional algorithms to alert against sensor errors and impending hypoglycaemia. Takahashi G 1 , Sato N 1 , Matsumoto N 1 , Shozushima T 1 , Hoshikawa K 1 , Akitomi S 1 , Kikkawa T 1 , Onodera C 1 , Kojika M 1 , Inoue Y 1 , Suzuki K 2 , Wakabayashi G 3 , Endo S 1 1 Department of Critical Care Medicine, 2 Department of Anesthesiology, and 3 Department of Surgery, Keio University School of Medicine, Tokyo, Japan Eur Surg Res 2011; 47 : 32–8 Background: Glucose control has been a topic of serious debate within the Surviving Sepsis Campaign Guidelines (SSCG) over the past years due to arguments both for and against its use in postoperative sepsis patients. Current SSCG state that following initial stabilisation of patients with severe sepsis, blood glucose should be maintained < 150 mg/dl. This study sought to evaluate the feasibility of an artificial pancreas in such patients. Methods: Patients who had surgery for sepsis caused by infections (n = 8) underwent tight glucose control using an artificial pancreas continuously for 7 days after surgery. Patients were sedated and mechanically ventilated. Primary outcomes included blood glucose concentrations over time, insulin dose received, and occurrence of hypoglycaemia. At the end of the study, the patients were divided into two groups (HG, higher insulin requirement; LG, lower insulin requirement) and analysed on the basis of total insulin dose they received over the 7 days. Results: The blood glucose concentration before glucose control was 203.3 ± 9.9 mg/dl, but was successfully brought down to < 150 mg/dl with the use of the artificial pancreas with no events of hypoglycaemia. Patients in the HG group required a mean of 21,824 ± 6,030.4 mU/kg of insulin, and patients in the LG group required 6,254.5 ± 3,402.3 mU/kg. Conclusions: Targeted glucose concentrations could be achieved safely and feasibly with the use of an artificial pancreas in postoperative sepsis patients. Comment: Notwithstanding studies in the past which present data both for and against glucose control in postoperative patients, Takahashi’s study is unique in that it demonstrates the feasibility of a closed-loop control system in postoperative sepsis patients. While it is understood that this was a preliminary study, more research is needed to present the need for tight glucose control in postoperative patients. This study did not have a control group where patients did not undergo glucose control with the artificial pancreas. Takahashi et al. mention, however, that they do plan to assess a larger number of patients and conduct a randomised controlled trial to make their results more conclusive. There seems to be uncertainty on how much insulin requirement is actually necessary for postoperative patients in intensive care in order to avoid the complications that can potentially arise from hypoglycaemia due to the glucose control and hyperglycaemia due to no glucose control. This study purports that, because their artificial pancreas achieved glucose concentrations within a target range of 80–150 mg/dl, the artificial pancreas is capable of acute control; however, if the study were designed to measure