Metabolic modelling and the closed-loop insulin delivery problem

Recent advances in the precision control strategy of artificial pancreas

The scientific diagnosis and treatment of patients with diabetes require frequent blood glucose testing and insulin delivery to normoglycemia. Therefore, an artificial pancreas with a continuous blood glucose (BG) monitoring function is an urgent research target in the medical industry. The problem of closed-loop algorithmic control of the BG with a time delay is a key and difficult issue that needs to be overcome in the development of an artificial pancreas. Firstly, the composition, structure, and control characteristics of the artificial pancreas are introduced. Subsequently, the research progress of artificial pancreas control algorithms is reviewed, and the characteristics, advantages, and disadvantages of proportional-integral-differential control, model predictive control, and artificial intelligence control are compared and analyzed to determine whether they are suitable for the practical application of the artificial pancreas. Additionally, key advancements in areas such as blood glucose data monitoring, adaptive models, wearable devices, and fully automated artificial pancreas systems are also reviewed. Finally, this review highlights that meal prediction, control safety, integration, streamlining the optimization of control algorithms, constant temperature preservation of insulin, and dual-hormone artificial pancreas are issues that require further attention in the future.

Continuous glucose monitoring for the prevention of morbidity and mortality in preterm infants

BACKGROUND

Preterm infants are susceptible to hyperglycemia and hypoglycemia, conditions which may lead to adverse neurodevelopment. The use of continuous glucose monitoring devices (CGM) might help keeping glucose levels in the normal range, and reduce the need for blood sampling. However, the use of CGM might be associated with harms in the preterm infant.

OBJECTIVES

Objective one: to assess the benefits and harms of CGM alone versus standard method of glycemic measure in preterm infants. Objective two: to assess the benefits and harms of CGM with automated algorithm versus standard method of glycemic measure in preterm infants. Objective three: to assess the benefits and harms of CGM with automated algorithm versus CGM without automated algorithm in preterm infants.

SEARCH METHODS

We adopted the standard search strategy of Cochrane Neonatal to search the Cochrane Central Register of Controlled Trials (CENTRAL; 2020, Issue 9), in the Cochrane Library; MEDLINE via PubMed (1966 to 25 September 2020); Embase (1980 to 25 September 2020); and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) (1982 to 25 September 2020). We also searched clinical trials databases, conference proceedings, and reference lists of retrieved articles for randomized controlled trials and quasi-randomized trials.

SELECTION CRITERIA

Randomized controlled trials (RCTs) and quasi-RCTs in preterm infants comparing: 1) the use of CGM versus intermittent modalities to measure glycemia (comparison 1); or CGM associated with prespecified interventions to correct hypoglycemia or hyperglycemia versus CGM without such prespecified interventions (comparison 2).

DATA COLLECTION AND ANALYSIS

We assessed the methodological quality of included trials using Cochrane Effective Practice and Organisation of Care Group (EPOC) criteria (assessing randomization, blinding, loss to follow-up, and handling of outcome data). We evaluated treatment effects using a fixed-effect model with risk ratio (RR) for categorical data and mean, standard deviation (SD), and mean difference (MD) for continuous data. We used the GRADE approach to assess the certainty of the evidence.

MAIN RESULTS

Four trials enrolling 138 infants met our inclusion criteria. Investigators in three trials (118 infants) compared the use of CGM to intermittent modalities (comparison one); however one of these trials was analyzed separately because CGM was used as a standalone device, without being coupled to a control algorithm like in the other trials. A fourth trial (20 infants) assessed CGM with an automated algorithm versus CGM with a manual algorithm. None of the four included trials reported the neurodevelopmental outcome, i.e. the primary outcome of this review. Within comparison one, the certainty of the evidence on the use of CGM on mortality during hospitalization is very uncertain (typical RR 3.00, 95% CI 0.13 to 70.30; typical RD 0.04, 95% CI -0.06 to 0.14; 50 participants; 1  study; very low certainty). The number of hypoglycemic episodes was reported in two studies with conflicting data. The number of hyperglycemic episodes was reported in one study (typical MD -1.40, 95% CI -2.84 to 0.04; 50 participants; 1 study). The certainty of the evidence was very low for all outcomes because of limitations in study design, and imprecision of estimates.  Three studies are ongoing.

AUTHORS' CONCLUSIONS

There is insufficient evidence to determine if CGM improves preterm infant mortality or morbidities. Long-term outcomes were not reported. Clinical trials are required to determine the most effective CGM and glycemic management regimens in preterm infants before larger studies can be performed to assess the efficacy of CGM  for reducing mortality, morbidity and long-term neurodevelopmental impairments. The absence of CGM labelled for neonatal use is still a major limit in its use as well as the absence of dedicated neonatal devices.

Effect of 6 months hybrid closed-loop insulin delivery in young people with type 1 diabetes: a randomised controlled trial protocol

INTRODUCTION

Automated insulin delivery (also known as closed loop, or artificial pancreas) has shown potential to improve glycaemic control and quality of life in people with type 1 diabetes (T1D). Automated insulin delivery devices incorporate an insulin pump with continuous glucose monitoring(CGM) and an algorithm, and adjust insulin in real time. This study aims to establish the safety and efficacy of a hybrid closed-loop (HCL) system in a long-term outpatient trial in people with T1D aged 12 -<25 years of age, and compare outcomes with standard therapy for T1D as used in the contemporary community.

METHODS AND ANALYSIS

This is an open-label, multicentre, 6-month, randomised controlled home trial to test the MiniMed Medtronic 670G system (HCL) in people with T1D aged 12 -<25 years, and compare it to standard care (multiple daily injections or continuous subcutaneous insulin infusion (CSII), with or without CGM). Following a run-in period including diabetes and carbohydrate counting education, dosage optimisation and baseline glucose control data collection, participants are randomised to either HCL or to continue on their current treatment regimen. The primary aim of the study is to compare the proportion of time spent in target sensor glucose range (3.9-10.0 mmol/L) on HCL versus standard therapy. Secondary aims include a range of glucose control parameters, psychosocial measures, health economic measures, biomarker status, user/technology interactions and healthcare professional expectations. Analysis will be intention to treat. A study in adults with an aligned design is being conducted in parallel to this trial.

ETHICS AND DISSEMINATION

Ethics committee permissions were gained from respective institutional review boards. The findings of the study will provide high-quality evidence on the role of HCL in clinical practice.

Multi-level Supervision and Modification of Artificial Pancreas Control System

Artificial pancreas (AP) systems provide automated regulation of blood glucose concentration (BGC) for people with type 1 diabetes (T1D). An AP includes three components: a continuous glucose monitoring (CGM) sensor, a controller calculating insulin infusion rate based on the CGM signal, and a pump delivering the insulin amount calculated by the controller to the patient. The performance of the AP system depends on successful operation of these three components. Many APs use model predictive controllers that rely on models to predict BGC and to calculate the optimal insulin infusion rate. The performance of model-based controllers depends on the accuracy of the models that is affected by large dynamic changes in glucose-insulin metabolism or equipment performance that may move the operating conditions away from those used in developing the models and designing the control system. Sensor errors and missing signals will cause calculation of erroneous insulin infusion rates. And the performance of the controller may vary at each sampling step and each period (meal, exercise, and sleep), and from day to day. Here we describe a multi-level supervision and controller modification (ML-SCM) module is developed to supervise the performance of the AP system and retune the controller. It supervises AP performance in 3 time windows: sample level, period level, and day level. At sample level, an online controller performance assessment sub-module will generate controller performance indexes to evaluate various components of the AP system and conservatively modify the controller. A sensor error detection and signal reconciliation module will detect sensor error and reconcile the CGM sensor signal at each sample. At period level, the controller performance is evaluated with information collected during a certain time period and the controller is tuned more aggressively. At the day level, the daily CGM ranges are further analyzed to determine the adjustable range of controller parameters used for sample level and period level. Thirty subjects in the UVa/Padova metabolic simulator were used to evaluate the performance of the ML-SCM module and one clinical experiment is used to illustrate its performance in a clinical environment. The results indicate that the AP system with an ML-SCM module has a safer range of glucose concentration distribution and more appropriate insulin infusion rate suggestions than an AP system without the ML-SCM module.

Performance Analysis of Fuzzy-PID Controller for Blood Glucose Regulation in Type-1 Diabetic Patients

This paper presents Fuzzy-PID (FPID) control scheme for a blood glucose control of type 1 diabetic subjects. A new metaheuristic Cuckoo Search Algorithm (CSA) is utilized to optimize the gains of FPID controller. CSA provides fast convergence and is capable of handling global optimization of continuous nonlinear systems. The proposed controller is an amalgamation of fuzzy logic and optimization which may provide an efficient solution for complex problems like blood glucose control. The task is to maintain normal glucose levels in the shortest possible time with minimum insulin dose. The glucose control is achieved by tuning the PID (Proportional Integral Derivative) and FPID controller with the help of Genetic Algorithm and CSA for comparative analysis. The designed controllers are tested on Bergman minimal model to control the blood glucose level in the facets of parameter uncertainties, meal disturbances and sensor noise. The results reveal that the performance of CSA-FPID controller is superior as compared to other designed controllers.

Cell-matrix interaction during strain-dependent remodelling of simulated collagen networks

The importance of tissue remodelling is widely accepted, but the mechanism by which the remodelling process occurs remains poorly understood. At the tissue scale, the concept of tensional homeostasis, in which there exists a target stress for a cell and remodelling functions to move the cell stress towards that target, is an important foundation for much theoretical work. We present here a theoretical model of a cell in parallel with a network to study what factors of the remodelling process help the cell move towards mechanical stability. The cell-network system was deformed and kept at constant stress. Remodelling was modelled by simulating strain-dependent degradation of collagen fibres and four different cases of collagen addition. The model did not lead to complete tensional homeostasis in the range of conditions studied, but it showed how different expressions for deposition and removal of collagen in a fibre network can interact to modulate the cell's ability to shield itself from an imposed stress by remodelling the surroundings. This study also showed how delicate the balance between deposition and removal rates is and how sensitive the remodelling process is to small changes in the remodelling rules.

A bio-inspired glucose controller based on pancreatic β-cell physiology

INTRODUCTION

Control algorithms for closed-loop insulin delivery in type 1 diabetes have been mainly based on control engineering or artificial intelligence techniques. These, however, are not based on the physiology of the pancreas but seek to implement engineering solutions to biology. Developments in mathematical models of the β-cell physiology of the pancreas have described the glucose-induced insulin release from pancreatic β cells at a molecular level. This has facilitated development of a new class of bio-inspired glucose control algorithms that replicate the functionality of the biological pancreas. However, technologies for sensing glucose levels and delivering insulin use the subcutaneous route, which is nonphysiological and introduces some challenges. In this article, a novel glucose controller is presented as part of a bio-inspired artificial pancreas.

METHODS

A mathematical model of β-cell physiology was used as the core of the proposed controller. In order to deal with delays and lack of accuracy introduced by the subcutaneous route, insulin feedback and a gain scheduling strategy were employed. A United States Food and Drug Administration-accepted type 1 diabetes mellitus virtual population was used to validate the presented controller.

RESULTS

Premeal and postmeal mean ± standard deviation blood glucose levels for the adult and adolescent populations were well within the target range set for the controller [(70, 180) mg/dl], with a percent time in range of 92.8 ± 7.3% for the adults and 83.5 ± 14% for the adolescents.

CONCLUSIONS

This article shows for the first time very good glucose control in a virtual population with type 1 diabetes mellitus using a controller based on a subcellular β-cell model.

Evolution of devices in diabetes management

Recent technological advancements in insulin administration and glucose monitoring have allowed patients with diabetes to become increasingly involved in their own care. Devices replacing the traditional vial and syringe, such as insulin pens, are gaining popularity and offer simple and convenient insulin administration. Pen devices are associated with improved dose accuracy, reducing the risk of hypo- or hyperglycemia, and are continually being updated with new safety features in order to optimize their performance. In patients for whom glucose variability remains a problem, continuous subcutaneous insulin infusion via an implanted canula or continuous intraperitoneal insulin infusion via an implanted pump is safe and effective when used correctly, although cost can be a limitation. More accurate retrospective and real-time continuous monitoring devices, which can better detect blood glucose excursions, have become standard components of modern-day diabetes management. The most recent devices have sensor-signaling capabilities with wireless data transmission, leading to reduced time delay and more accurate alerts. Ultimately, though, while self-management remains a critical factor in improving glycemic control at present, human error may undermine even the most accurate treatment interventions. A key long-term goal in diabetes management is, therefore, to develop an automated and accurate closed-loop system for blood glucose monitoring and insulin delivery to better reflect the physiological mechanisms of glucose homeostasis and remove the "human" element. This "artificial pancreas" would offer the most innovative intervention for diabetes management and has the potential to considerably reduce the patient's burden of self-care.

A new general glucose homeostatic model using a proportional-integral-derivative controller

The glucose-insulin system is a challenging process to model due to the feedback mechanisms present, hence the implementation of a model-based approach to the system is an on-going and challenging research area. A new approach is proposed here which provides an effective way of characterising glycaemic regulation. The resulting model is built on the premise that there are three phases of insulin secretion, similar to those seen in a proportional-integral-derivative (PID) type controller used in engineering control problems. The model relates these three phases to a biological understanding of the system, as well as the logical premise that the homeostatic mechanisms will maintain very tight control of the system. It includes states for insulin, glucose, insulin action and a state to simulate an integral function of glucose. Structural identifiability analysis was performed on the model to determine whether a unique set of parameter values could be identified from the available observations, which should permit meaningful conclusions to be drawn from parameter estimation. Although two parameters--glucose production rate and the proportional control coefficient--were found to be unidentifiable, the former is not a concern as this is known to be impossible to measure without a tracer experiment, and the latter can be easily estimated from other means. Subsequent parameter estimation using Intravenous Glucose Tolerance Test (IVGTT) and hyperglycaemic clamp data was performed and subsequent model simulations have shown good agreement with respect to these real data.

Optimal control of blood glucose: the diabetic patient or the machine?

In this issue of Science Translational Medicine, El-Khatib et al. describe a "closed-loop" bihormonal artificial pancreas, designed to avert episodes of low blood sugar in patients with insulin-dependent diabetes. We discuss the benefits and challenges of therapy directed at tight control of blood glucose and ask whether this and similar technological breakthroughs can address as yet unanswered questions in the biology of diabetes.

Acute response in vivo of a fiber-optic sensor for continuous glucose monitoring from canine studies on point accuracy

The objective of this study was to evaluate the acute response of Sencil(™), a fiber-optic sensor, in point accuracy for glucose monitoring in vivo on healthy dogs under anesthesia. A total of four dogs with clinically normal glycemia were implanted with one sensor each in the chest region to measure the interstitial glucose concentration during the ovariohysterectomy procedure. The data was acquired every 10 seconds after initiation, and was compared to the concentration of venous plasma glucose sampled during the surgery procedures for accuracy of agreement analysis. In the four trials with a range of 71-297 mg/dL plasma glucose, the collected 21 pairs of ISF readings from the Sencil™ and the plasma reference showed superior dispersion of residue values than the conventional system, and a linear correlation (the Pearson correlation coefficient is 0.9288 and the y-intercept is 14.22 mg/dL). The MAD (17.6 mg/dL) and RMAD (16.16%) of Sencil™ measurements were in the comparable range of the conventional system. The Clarke error grid analysis indicated that 100% of the paired points were in the clinically acceptable zone A (61.9%) and B (38.1%).

In silico evaluation of a control system and algorithm for automated insulin infusion in the ICU setting

Background

It is known that tight control of glucose in the Intensive Care Unit reduces morbidity and mortality not only in diabetic patients but also in those non-diabetics who become transiently hyperglycemic. Taking advantage of a recently marketed subcutaneous glucose sensor we designed an Automatic Insulin Infusion System (AIIS) for inpatient treatment, and tested its stability under simulated clinical conditions.

Methods

The system included: reference glucose, glucose sensor, insulin and glucose infusion controllers and emergency infusion logic. We carried out computer simulations using Matlab/Simulink^®, in both common and worst-case conditions.

Results

The system was capable of controlling glucose levels without entering in a phase of catastrophic instability, even under severe simulated challenges. Care was taken to include in all simulations the 5-10 minute delay of the subcutaneous glucose signal when compared to the real-time serum glucose signal, a well-known characteristic of all subcutaneous glucose sensors.

Conclusions

When tested in-Silico, a commercially available subcutaneous glucose sensor allowed the stable functioning of a proportional-derivative Automatic Insulin Infusion System, which was able to maintain glucose within acceptable limits when using a well-established glucose response model simulating a patient. Testing of the system in vivo using animal models is now warranted.

Update on mathematical modeling research to support the development of automated insulin delivery systems

One year after its initial meeting, the Glycemia Modeling Working Group reconvened during the 2009 Diabetes Technology Meeting in San Francisco, CA. The discussion, involving 39 scientists, again focused on the need for individual investigators to have access to the clinical data required to develop and refine models of glucose metabolism, the need to understand the differences among the distinct models and control algorithms, and the significance of day-to-day subject variability. The key conclusion was that model-based comparisons of different control algorithms, or the models themselves, are limited by the inability to access individual model-patient parameters. It was widely agreed that these parameters, as opposed to the average parameters that are typically reported, are necessary to perform such comparisons. However, the prevailing view was that, if investigators were to make the parameters available, it would limit their ability (and that of their institution) to benefit from the invested work in developing their models. A general agreement was reached regarding the importance of each model having an insulin pharmacokinetic/pharmacodynamic profile that is not different from profiles reported in the literature (88% of the respondents agreed that the model should have similar curves or be analyzed separately) and the importance of capturing intraday variance in insulin sensitivity (91% of the respondents indicated that this could result in changes in fasting glucose of >or=15%, with 52% of the respondents believing that the variability could effect changes of >or=30%). Seventy-six percent of the participants indicated that high-fat meals were thought to effect changes in other model parameters in addition to gastric emptying. There was also widespread consensus as to how a closed-loop controller should respond to day-to-day changes in model parameters (with 76% of the participants indicating that fasting glucose should be within 15% of target, with 30% of the participants believing that it should be at target). The group was evenly divided as to whether the glucose sensor per se continues to be the major obstacle in achieving closed-loop control. Finally, virtually all participants agreed that a future two-day workshop should be organized to compare, contrast, and understand the differences among the different models and control algorithms.