Review article
The evolution of control algorithms in artificial pancreas: A historical perspective

https://doi.org/10.1016/j.arcontrol.2019.07.004Get rights and content

Abstract

Blood glucose control algorithms have evolved since the beginnings of the artificial pancreas in diabetes treatment. Although the main problem to solve remains as the regulation of blood glucose into the healthy physiological range, the schemes have evolved over time from on-off schemes to the data-based personalized schemes. The evolution has been in accordance with the understanding of glucose metabolism, the theoretical background to model it, and the availability of sensor technology. The algorithms have allowed the calculation of insulin infusion (sometimes glucagon or glucose), by the highly invasive intravenous route, up to schemes based on the minimally invasive subcutaneous route. Solutions have also been proposed to deal with delays in insulin action and glucose measurement, as well as robust schemes to reject disturbances due to meal intake, exercise, non-modeled dynamics, and parametric variations due to inter- and intravariability of metabolism. Other problem that control schemes have solved is the safety in insulin infusion, including the calculation of insulin on board to avoid episodes of hypoglycemia, guaranteeing glucose regulation in normoglycemia, and decreasing the time in hyperglycemia. To summarize the role of control algorithms in the development of the artificial pancreas, this paper presents a historical review of the proposed control algorithms, from the establishment of the paradigm of artificial pancreas to the present date.

Introduction

When Banting and Best obtained extracts from animal pancreas and injected them to diabetic subjects, they made a breakthrough in medical science. It was 1922 and there was no medical treatment for type 1 diabetes mellitus (T1DM) at that time, so pancreatic extracts could be a therapy option. In their report, Bating and Best established a challenge that opened many scientific and technological fields around diabetes treatment: how to determine an adequate and regulated dosage of the pancreatic extracts (Banting, Best, Collip, Cambell, & Fletcher, 1922). The specificity of the chemical components of the pancreatic extracts and the central role of the hormone insulin in the glucose metabolism were clearly demonstrated, and the scientists involved in that project were awarded with the Nobel Prize in Physiology or Medicine in 1923 (Rosenfeld, 2002, Zaccardi, Webb, Yates, Davies, 2015). The availability of insulin as a clinical drug marked the beginning of the insulin therapy, which consists in subcutaneous injections of the hormone. That new therapy spread worldwide and diabetic patients had the opportunity to be saved, prolonging and improving their lives (Owens, 2002).

The early trials of insulin therapy used insulin extracted from animal pancreases (bovine and porcine, for example). Even though the hormone was purified, diabetic subjects presented immune response, and it was one of the first disadvantages of that promising therapy. After years of investigation around chemistry, biology, and physiology of insulin, a new breakthrough in diabetes technology arrived: synthetic insulin became available (Polonsky, 2012). That was possible after the identification of the insulin gene in the human genome and its biosynthesis in the bacterium E. Coli. The procedure was called rDNA technology, and the product was identically to the pancreatic human insulin. The approval to market biosynthetic insulin (also called recombinant human insulin) was released in 1982 in the UK, Germany, and USA. It represented the solution of an increasing demand of insulin (Chance & Frank, 1993).

Once insulin was available, the insulin therapy was established as the golden treatment for diabetic patients. Although the reduction of blood glucose levels was the main promise of the therapy, after years of treatment some clinical disadvantages were identified. These were called the microvascular and neurologic complications (such as retinopathy, nephropathy, and neuropathy). In 1993 a team of researchers in USA reported a study comparing the results of the conventional insulin therapy (one or two daily insulin injections) against a more intensive therapy using an external pump or three or more daily injections. The study was called the diabetes control and complications trial (DCCT), and the main hypothesis was that long-term complications (micro and macrovascular) could be mitigated by achieving near-normal blood glucose control (tight control) (Diabetes, 1993). That therapy was called the intensive insulin therapy, and its clinical implementation offered two options: multiple daily injections (MDI) therapy and insulin-pump therapy (Heller, Kozlovski, & Kurtzhals, 2007).

The DCCT provided encouraging outcomes (retinopathy, nephropathy, and neuropathy were reduced in 76%, 34%, and 60%, respectively), and the hypothesis of reducing long-term complications was proved. The insulin infusion strategy reported in the DCCT re-emphasized the premise of Banting and Best in their initial report “... maintaining blood glucose concentrations close to the normal range... ” (Diabetes, 1993). The DCCT gave the basis of a new paradigm, the insulin infusion pattern must improve the glycemic targets of the T1DM treatment (Nathan, 2014). This established the challenge of developing a device able to measure blood glucose concentration, infuse insulin, and compute automatically the insulin dosage. Such device was called the artificial pancreas (AP), and it integrates three elements: a continuous glucose monitoring (CGM) system, a continuous subcutaneous insulin infusion (CSII) system, and a control algorithm (Hovorka, 2011).

The development of the artificial pancreas has evolved since the DCCT established the paradigm of tight control of glycemia. Cobelli, Renard, and Kovatchev (2011) summarized the milestones on AP progress in a timeline from 1920 to 2010. They focused their discussion in the limitations of CGM and CSII systems and they outlined possible solutions towards the clinical acceptance of AP systems. In 2009, Hoshino, Haraguchi, Mizushima, and Sakai (2009) discussed progress in the development of the AP and they pointed out that its use in clinical situation was limited. They presented the status of the three parts of the AP and concluded that the limiting factor was the slow progress in the development of CGM. Shortly after, Weinzimer summarized the first large-scale randomized trial of the sensor-augmented pump therapy, which includes only use of CGM and CSII systems, that is, a control algorithm is not included (Weinzimer, 2012). He concluded that the sensor-augmented pump therapy did not guarantee meet the glycemic goals. Nevertheless, if the basal insulin delivery was suspended the risk of hypoglycemia and diabetic ketoacidosis was reduced. He briefly discussed some small inpatient clinical trials reporting promising results of closed-loop therapy, that is, AP systems including real-time CGM and CSII systems and a control algorithm. Years after, the closed-loop therapy was consolidated in inpatient clinical trials and the outpatient trials arrived.

Thabit and Hovorka (2016) presented a review focused on progress of AP systems in outpatient settings. They discussed that the outpatient trials both transitional and home studies provided encouraging outcomes towards the use of closed-loop therapy in real life. To reach that goal, they discussed that sensor must be smaller, the wearing time must be increased, and calibration must be avoided. Occlusion of the catheter of insulin must be reduced while the absorption and action of released insulin must be accelerated. They finally pointed out that the control algorithm play a crucial role in the closed-loop therapy facilitating the adaptation and personalization of the AP system. In the same year, 2016, Hanazaki et al. (2016) discussed nocturnal hypoglycemia as the major problem of wearable artificial pancreas. They made a review about two possible solutions of this problem. First, CSII systems with low-glucose suspend feature; that is, if the CGM system reports a glucose value less than 70 mg/dL, the CSII system stops insulin delivery for two hours. Second, bihormonal AP systems delivering both insulin and glucagon. Also in 2016, Trevitt et al. (Trevitt, Simpson, & Wood, 2016) made a comprehensive review about the features of AP systems available in market and those in development by research teams worldwide.

In 2017, the Food and Drug Administration (FDA) approved the MiniMed 670G system (by Medtronic MiniMed, Inc.), the first closed-loop AP system available in market. This motivated a growing number of publications about the role of AP systems in diabetes therapy. For example, in 2018, Kovatchev (Kovatchev, 2018) discussed the advances in closed-loop control of diabetes and the challenges to face wearable and home-use artificial pancreas. He reviewed the development of the diabetes technology and the progress of AP systems since 1970. He defined the concept of treatment ecosystem of diabetes which is a combination of interacting processes such as the diabetic patient, technological treatment, behavioral perturbations, and medical intervention, all of them developing in different time scales. In the same year, Bertachi, Ramkissoon, Bondia, and Vehí (2018) reported a comprehensive review of the clinical trials conducted since 2011. They analyzed the types of AP systems, populations, durations and conditions of the clinical trials. Dadlani, Pinsker, Dassau, and Kudva (2018) reviewed the clinical context of the MiniMed 670G release and current studies that the National Institute of Health (NIH) and other international agencies are funding to conduct large-scale outpatient studies. The review reported that AP systems improve time in target to about 70–75% and time below target less than 4%, and just a unique study reporting 88% of time in the target glucose range. These scores were reached despite the latent risk of hypoglycemia. The authors proposed the development of a new generation of systems able to improve the scores while reducing the risk of hypoglycemia and the burden of the therapy.

Early this year, Boughton and Hovorka (2019) proposed six stages in the development of the artificial pancreas, and they grouped the stages in three generations of devices. The first generation includes the three first stages related to the improvements of the therapy based only in CGM and CSII systems. Stage 1: The CSII system shuts off when user does not respond to the low glucose alarm of the CGM system. Stage 2: Reduction or cessation of insulin delivery of the CSII system before blood glucose gets low. Stage 3: Same that the second stage but including insulin dosing above high threshold. The first generation is called open-loop therapy because the improvements only include the CGM and CSII systems without intervention of a control algorithm. The second generation includes the concept of automatic insulin infusion, it is called the closed-loop therapy, the improvements are in the next two stages. Stage 4: The artificial pancreas is formed by a hybrid closed-loop system with automatic infusion at basal stage and meal-time manual-assist bolusing. Stage 5: The manual meal-time bolus was eliminated and it was replaced by the full automated closed-loop system. The third generation corresponds to full automated multi-hormone (insulin and glucagon) closed-loop systems, this is the Stage 6.

As Kovatchev mentioned in Kovatchev (2018), the current status of AP systems is due to almost a century of research in the three elements, and in the integration of them, as well as in the clinical trials to test the impact in T1DM therapy. After the DCCT, the role of the control algorithm in AP systems was clearly identified as the element providing full automation of the insulin release, that is, the closed-loop therapy. Once the performance of CGM (Heinemann et al., 2018) and CSII (Vigersky et al., 2018) was proved in a wide variety of inpatient, transitional and outpatient trials, the conditions were given to test the full automation of AP systems. The review papers discussed in Section 1.2 account for the achievements of AP systems, and they commonly includes a brief description of the related control algorithms. For example, Cobelli et al. (2011), and Hoshino et al. (2009), and Weinzimer (2012) mentioned that proportional-derivative (PD), proportional-integral-derivative (PID), and model predictive controllers (MPC) are the most used control approaches to resolve the full automation of AP systems. Trevitt et al. (2016) briefly mentioned that there are four types of control algorithms used in AP systems: MPC, PID, fuzzy logic (FL) algorithms, and bio-inspired algorithms based on mathematical models. Kovatchev (2018) dedicated a paragraph to the evolution of the PID and MPC algorithms. Although these authors mentioned explicitly the control algorithms in their review papers, they only provided few references. Moreover, the rest of the review articles about AP systems (e.g. Bertachi, Ramkissoon, Bondia, Vehí, 2018, Hanazaki, Munekage, Kitagawa, Yatabe, Munekage, Shiga, et al., 2016, Thabit, Hovorka, 2016, and Boughton & Hovorka, 2019) did not explicitly mention the control algorithms and no references were provided.

To support the analysis of the current progress of AP systems, this document focuses on the evolution of the control algorithms. The main approaches and technical problems that the automation and control systems community has investigated in the last decades are discussed here. Previous reviews of this topic were reported in the papers by Bellazzi, Nucci, and Cobelli (2001) and Parker, Doyle, and Peppas (2001), where the control algorithms published before 2000 were summarized. Doyle III et al. focused on algorithms to control blood glucose via the intravenous route. Whereas Cobelli et al. discussed algorithms used in glucose control via subcutaneous route. Although many control approaches have been discussed in the last two decades, the interest to summarize the wide variety of contributions has begun recently. In 2013, Lunze et al. (Lunze, Singh, Walter, Brendel, & Leonhardt, 2013) presented the first paper devoted to discussed the control algorithms for blood glucose control. That paper also included a review of mathematical models of glucose metabolism. In 2014, Turksoy et al. (Turksoy & Cinar, 2014) represented a review only considering algorithms based in adaptive control. In 2016, Oviedo et al. (Oviedo, Vehí, Calm, & Armengol, 2017) presented a review in the same line of Lunze et al. (2013) regarding the discussion of both mathematical models of glucose metabolism and blood glucose control algorithms. With this background, the aim of this paper is to offer a historical perspective of how the control algorithms have evolved along time. For this end, the study is divided in three eras. The first era covers the closed-loop control algorithms reported between 1963 and 1981 (Section 2). The second era includes the model-based control algorithms covering from 1900s to the present date (Section 3). The third era includes current control algorithms based on data (Section 4). Finally, a discussion and conclusions are presented Section 5.

Section snippets

The first era: initial closed-loop control algorithms

In the 1960s, the automation of insulin infusion was clearly identified as a closed-loop control problem where the controlled process is the glucose metabolism of a diabetic patient, the regulated variable is the blood glucose concentration, the sensor is a CGM system, the control signal is the infused insulin, and the actuator is a CSII system. The control problem depends on the glycemic target defined by the physician, and the controller is an algorithm able to compute the insulin to be

The second era: model-based control algorithms

At the beginning of the 1980s, control community paid attention on the control problem underlying the closed-loop therapy and the analysis from control systems point of view began. The new challenge was to develop control algorithms based on physiological information of the glucose metabolism and the control signal should have physiological meaning. That establish a new paradigm encouraged two research areas: mathematical modeling of glucose metabolism and the synthesis of model-based control

The third era: data-based control of glucose in diabetes

Since the DCCT study established the paradigm of the artificial pancreas, its three elements (CGM system, CSII system, and the control algorithm) evolved considerably, as it was discussed in the previous sections. Currently, many clinical trials have proved the usefulness of AP systems as an outpatient therapy. Moreover, such as Trevitt et al. have discussed, there are already some AP systems available in market (Trevitt et al., 2016). The individual development of the three elements has

Conclusion

As it has been demonstrated along the last years, the full automation of the insulin infusion and the solution of the underlying closed-loop control problem are determinant in the success of a diabetes treatment based on the artificial pancreas paradigm. The main problems about available information (mathematical models and/or CGM data) of the current status of the metabolism of a patient have been established, but the solutions are on the road. In the case of the control algorithms, there is

Funding

The author thanks the National Council of Science and Technology in México for the financial support under grant 220187.

Declarations of interest

None.

Griselda Quiroz (PhD 08) is a professor at the Graduated School of Electrical Engineering in Universidad Autónoma de Nuevo León, México. Her research is about mathematical modeling and control of physiological systems.

References (101)

  • L. Heinemann et al.

    Real-time continuous glucose monitoring in adults with type 1 diabetes and impaired hypoglycaemia awareness or severe hypoglycaemia treated with multiple daily insulin injections (hypoDE): Amulticentre

    randomised controlled trial

    (2018)
  • S. Heller et al.

    Insulin’s 85th anniversary-An enduring medical miracle

    Diabetes Research and Clinical Practice

    (2007)
  • H. Lee et al.

    A closed-loop artificial pancreas based on model predictive control: Human-friendly identification and automatic meal disturbance rejection

    Biomedical Signal Processing Control

    (2009)
  • E.D. Lehmann et al.

    A physiological model of glucose-insulin interaction in type 1 diabetes mellitus

    Journal of Biomedical Engineering

    (1992)
  • B.S. Leon et al.

    Inverse optimal neural control of blood glucose level for type 1 diabetes mellitus patients

    Journal of the Franklin Institute

    (2012)
  • K. Lunze et al.

    Blood glucose control algorithms for type 1 diabetic patients: A methodological review

    Biomedical Signal Processing and Control

    (2013)
  • L. Magni et al.

    Model predictive control of glucose concentration in type i diabetic patients: An in silico trial

    Biomedical Signal Processing Control

    (2009)
  • A. Nath et al.

    Observer based nonlinear control design for glucose regulation in type 1 diabetic patients: An LMI approach

    Biomed. Signal Process. Control

    (2019)
  • P. Palumbo et al.

    Mathematical modeling of the glucose-insulin system: A review

    Mathematical Biosciences

    (2013)
  • A.K. Patra et al.

    Adaptive sliding mode Gaussian controller for artificial pancreas in TIDM patient

    Journal of Process Control

    (2017)
  • G. Quiroz et al.

    Theoretical blood glucose control in hyper-and hypoglycemic and exercise scenarios by means of an H algorithm

    Journal of Theoretical Biology

    (2010)
  • E. Ruiz-Velazquez et al.

    Parameter estimation of a meal glucose-insulin model for TIDM patients from therapy historical data

    IET systems biology

    (2018)
  • R. Srinivassan et al.

    A mathematical model for the control mechanism of free fatty acid-glucose metabolism in normal humans

    Computational and Biomedical Research

    (1970)
  • G. Toffolo et al.

    Quantitative estimation of beta cell sensitivity to glucose in the intact organism: a minimal model of insulin kinetics in the dog

    Diabetes

    (1980)
  • K. Turksoy et al.

    Adaptive control of artificial pancreas systems: a review

    Journal of Healthcare Engineering

    (2014)
  • Y. Wang et al.

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

    IEEE Transactions on Biomedical Engineering

    (2010)
  • A. Abu-Rmileh et al.

    A gain-scheduling model predictive controller for blood glucose control in type 1 diabetes

    IEEE Transactions on Biomedical Engineering

    (2010)
  • E. Ackerman et al.

    Model studies of blood glucose regulation

    Bulletin of Mathematical Biophysics

    (1965)
  • I. Ajmera et al.

    The impact of mathematical modeling on the understanding of diabetes and related complications

    Pharmacometrics & Systems Pharmacology

    (2013)
  • A.M. Albisser et al.

    An artificial endocrine pancreas

    Diabetes

    (1974)
  • F.G. Banting et al.

    Pancreatic extracts in the treatment of diabetes mellitus

    The Canadian Medical Association Journal

    (1922)
  • R. Bellazzi et al.

    The subcutaneous route to insulin-dependent diabetes therapy

    IEEE Engineering in Medicine and Biology

    (2001)
  • A. Beneyeto et al.

    Postprandial fuzzy adaptive strategy for a hybrid proportional derivative controller for the artificial pancreas

    Medical & Biological Engineering & Computing

    (2018)
  • R.N. Bergman

    Minimal model: perspective from 2005

    Hormone Research

    (2005)
  • R.N. Bergman et al.

    Quantitative estimation of insulin sensitivity

    American Journal of Physiology

    (1979)
  • R.N. Bergman et al.

    Physiologic evaluation of factors controlling glucose tolerance in man: Measurement of insulin sensitivity and betacell glucose sensitivity from the response to intravenous glucose

    Journal of Clinical Investigation

    (1981)
  • C.K. Botz

    An improved control algorithm for an artificial beta-cell

    IEEE Transactions on Biomedical Engineering

    (1974)
  • C.K. Boughton et al.

    Is an artificial pancreas (closed-loop system) for type 1 diabetes affective?

    Diabetic Medicine

    (2019)
  • H.M. Broekhuyse et al.

    Comparison of algorithms for the closed-loop control of blood glucose using the artificial beta cell

    IEEE Transactions on Biomedical Engineering

    (1981)
  • G.F. Cahill et al.

    Practical developments in diabetes research

    Diabetes

    (1972)
  • J.F. de Canete et al.

    Artificial neural networks for closed loop control of in silico and ad hoc type 1 diabetes

    Computer Methods and Programs in Biomedicine

    (2012)
  • Z. Cao et al.

    Extremum seeking control for personalized zone adaptation in model predictive control for type 1 diabetes

    IEEE Transactions on Biomedical Engineering

    (2018)
  • A. Chakrabarty et al.

    Event-triggered model predictive control for embedded artificial pancreas systems

    IEEE Transactions on Biomedical Engineering

    (2018)
  • R.E. Chance et al.

    Research, development, production, safety of biosynthetic human insulin

    Diabetes Care

    (1993)
  • Charette, W. P. (1968). Control systems theory applied to metabolic homeostatic systems and the derivation and...
  • C. Cobelli et al.

    Artificial pancreas: Past, present, future

    Diabetes

    (2011)
  • V. Dadlani et al.

    Advances in closed-loop insulin delivery systems in patients with type 1 diabetes

    Current Diabetes Reports

    (2018)
  • T. Diabetes

    Control and complications trial research group, the effect of intensive treatment of diabetes on the development and progression of long-term complications of insulin-dependent diabetes mellitus

    The New England Journal of Medicine

    (1993)
  • P. Dua et al.

    Multi-objective blood glucose control for type 1 diabetes

    Medical and Biological Engineering and Computing

    (2009)
  • R. Femat et al.

    Weighting restriction for intravenous insulin delivery on t1DM patient via H control

    IEEE Transactions on Automation Science and Engineering

    (2009)
  • Cited by (30)

    View all citing articles on Scopus

    Griselda Quiroz (PhD 08) is a professor at the Graduated School of Electrical Engineering in Universidad Autónoma de Nuevo León, México. Her research is about mathematical modeling and control of physiological systems.

    View full text