Encyclopedia of Systems and Control

Living Edition
| Editors: John Baillieul, Tariq Samad

Automated Insulin Dosing for Type 1 Diabetes

  • B. Wayne BequetteEmail author
Living reference work entry
DOI: https://doi.org/10.1007/978-1-4471-5102-9_100131-1


The development of automated insulin delivery (also known as a closed-loop artificial pancreas) systems has been an active research area since the 1960s, with an intense focus since 2005. In the United States in 2019, there is currently one commercial device available, with others under development. There is also a strong do-it-yourself community of individuals developing and using closed-loop technology. In this chapter we provide an overview of the challenges in developing automated insulin delivery systems and the algorithms that are commonly used to regulate blood glucose.


Biomedical control Drug infusion Fault detection Clinical trials 


The beta and alpha cells of the pancreas secrete insulin and glucagon, respectively, to regulate blood glucose (BG) concentrations; insulin decreases BG, while glucagon increases BG. The pancreas of an individual with type 1 diabetes no longer produces these hormones, so they must take insulin (by injection, or using continuous insulin infusion pumps) to survive. This is in contrast to people with type 2 diabetes, who have a pancreas that produces insulin but not enough to adequately regulate their BG. People with type 2 diabetes (90–95% of the diabetes population) often regulate BG through diet, exercise, and oral medications. The focus of this chapter is on automated insulin dosing (AID) systems for people with type 1 diabetes.

Before the discovery by Banting and Best that insulin was the hormone that regulated blood glucose in 1921, anyone diagnosed with diabetes was doomed to a short life; for an overview of the history of the development and use of insulin, see Hirsch (2004). For many decades someone with type 1 diabetes survived with multiple daily injections of insulin, but without real feedback. By the 1940s urine could be tested for glucose concentration, but easy-to-use urine strips were not available until the 1960s. Blood glucose test meters were not widely available for home use by an individual until the 1980s. Finally, blood glucose control was not recognized to be important until results of the Diabetes Control and Complications Trial (DCCT) were published in 1993. For many people today the state of the art remains multiple daily injections of insulin (one long-acting shot in the morning, followed by injections of rapid-acting insulin at mealtime, or to correct for high blood glucose) and multiple finger-stick measurements of blood glucose using self-monitoring blood glucose (SMBG) meters – ideally before bedtime, when awakening, at each meal, and before potentially dangerous activities such as driving.

The use of insulin pumps became more common in the late 1990s; these pumps are operated “open-loop” with specified rates of insulin delivery, using rapid-acting insulin. Continuous glucose monitors (CGM), which provide a measurement of the interstitial fluid (just underneath the skin) glucose at frequent intervals, typically every 5 min, have been widely available for less than 15 years. These insulin pumps and CGMs are absolutely critical components of a closed-loop automated insulin dosing (artificial pancreas) system. Note that the CGM signal is related to blood glucose (BG) but will often lag the actual BG, may be biased, will have some uncertainty, and could suffer from signal attenuation or dropout. Devices for a closed-loop system are shown in Fig. 1.
Fig. 1

Components of a closed-loop automated insulin delivery (IAD) system: CGM (sensor), pump and infusion set, smartphone or other receiver with a control interface and containing the control algorithm and other logic. The number of devices can be reduced by putting the control algorithm and interface directly on the insulin pump

To begin our discussion, it is helpful to understand common units and orders of magnitude. In the United States, blood glucose is measured in mg/dl. A healthy individual (without diabetes) will typically have a fasting BG level of 80–90 mg/dl, with brief excursions to 125–150 mg/dl due to meals. An individual with type 1 diabetes can drift to over 400 mg/dl (hyperglycemia) if insufficient insulin is given and yet be in danger of going below 70 mg/dl (hypoglycemia) if too much insulin is given. The common quantity of insulin is international units, which we will refer to as units (U) of insulin throughout this chapter. An individual with type 1 diabetes may have a basal (steady-state) insulin infusion rate of 1 U/h, or 24 U/day, but inject another 24 U of insulin throughout the day to compensate for meals. A population range might be 0.5–3 U/h basal (12–72 U/day), with another 12–72 U/day for meals (while a 50-50 split of basal-meal insulin is a rule of thumb, this can vary significantly from person to person). An individual will typically require one U of insulin for each 10–15 g carbohydrates consumed in a meal. Individuals will also make use of a correction or insulin sensitivity factor to decide how much insulin to give to reduce their blood glucose levels. Typically, 1 U of insulin will reduce BG by 40–80 mg/dL; while time is not explicitly mentioned in these factors, it will typically take 2 h to reduce the BG. Also notice that, since a bolus (injection) of insulin is a pulse input, the fact that a new pseudo steady-state in BG occurs is indicative of an integrating process, at least for the short time scale. On a longer time scale, however, there is a one-to-one relationship between the insulin basal rate (U/h) and BG (mg/dl).

The focus of this discussion has been on insulin, because there is no current glucagon formulation that is stable for long periods of time at body temperature. If someone needs to raise their BG, they will consume carbohydrates; currently glucagon is used to rescue someone who passed out or is in a coma and cannot eat – a glucagon solution is quickly mixed and injected when needed. Stable forms of glucagon are under development, and glucagon has been used in some clinical studies as noted below.


There are multiple challenges to individuals regulating BG. Insulin delivered subcutaneously takes a long time to act to decrease the BG. While the pancreas of a healthy individual has a peak in insulin action less than 5 min after a “glucose challenge,” insulin delivered subcutaneously has a peak action of roughly 75 min and continues to act for 4–8 h. Thus, an individual must realize that insulin given 2 h ago may have 50% of the insulin effect remaining. Current insulin pumps allow individuals to keep track of “insulin on board” (IOB), an estimation of the amount of insulin remaining to act. Bequette (2009) reviews the research protocols used to develop pharmacodynamics models and to calculate IOB.

Meals and exercise represent the greatest “disturbances” to regulating BG. The dynamic impact of a meal varies tremendously with meal content – a glass of orange juice can have a rapid effect on BG (and is often taken if an individual is becoming hypoglycemic), while a high-fat meal has an effect that extends over several hours. A meal also requires a significant amount of insulin to compensate for the carbohydrates in the meal. An individual with a basal insulin requirement of 1 U/h and a carb-to-insulin ratio of 10 g/U might need 7.5 U of insulin to cover a 75 g carbohydrate meal. Thus, the amount of insulin given for the meal is equivalent to 7.5 h worth of basal insulin. It would be rare for any manufacturing process to have this type of rangeability in an actuator!

Aerobic exercise can cause a relatively rapid decrease in BG and lead to reduced insulin needs for several hours, while anaerobic activity may result in a short-term increase in BG. An individual with type 1 diabetes walks a tightrope between under-delivery of insulin, which can lead to high blood glucose (hyperglycemia) and over-delivery of insulin, which can lead to low blood glucose (hypoglycemia). The risks of hyperglycemia are generally long-term in that higher mean glucose levels are correlated with micro- and macrovascular diseases and retinopathy. The risks of hypoglycemia, on the other hand, are largely short-term, such as drowsiness and, in a worst-case scenario, a diabetic coma leading possibly to death.

Overnight hypoglycemia is perhaps the greatest fear of a parent of a child with type 1 diabetes; it is common for a parent to check blood glucose at around 1 am or so. This fear often causes people to under-deliver insulin overnight, which can lead to hyperglycemia and long-term consequences. CGMs that provided alarms to awake individuals, or their parents, were not as effective as expected, partially due to a relatively high false positive rate for the devices at that time (the early 2000s); these problems were noted by Buckingham et al. (2005). These challenges motivated the development of initial closed-loop systems, in the form of low-glucose suspend algorithms, which shut off an insulin pump to avoid hypoglycemia, with a focus on overnight applications.

A number of faults can occur with sensors and insulin infusion sets. CGM signals can attenuate due to a person putting pressure on the sensor. Also, Bluetooth or other signals between the devices can drop out. Infusion sets are normally worn for 3 days before replacement, but these sets can fail in less time, sometimes immediately after insertion. Bequette (2014) provides an overview of these problems and possible fault detection techniques.

Control Algorithms and Objectives

Four types of algorithms have typically been used to regulate glucose by manipulating insulin infusion: (i) on-off, (ii) fuzzy logic/expert systems, (iii) proportional-integral-derivative (PID), and (iv) model predictive control (MPC). The control objectives are often either to control to a specific set point or to control to a desired range of glucose; the set point or range can vary with time of day. An overview of the different algorithms is provided by Bequette (2012).

Early closed-loop research focused on overnight control of blood glucose using a simple on-off algorithm, that is, if the glucose reached a hyperglycemic threshold, then the pump was shut off (low-glucose suspend) for a period of time. The next advance was to develop a predictive low-glucose suspend (PLGS) algorithm to shut off the pump if the CGM was predicted to go low during a future prediction horizon (often 30 min) (Buckingham et al. 2010; Cameron et al. 2012a).

A fuzzy logic-based strategy, based on the CGM, its rate of change, and acceleration, is used by Mauseth et al. (2010). The MD-Logic system by Atlas et al. (2010) uses a combination of set point and control-to-range concepts.

The proportional-integral-derivative (PID) controller is ubiquitous in chemical process applications. Steil et al. (2011) proposed a PID controller with a model-based insulin feedback term, making it similar to a cascade control strategy; an analysis of this approach is provided by Palerm (2011). The model-based insulin feedback plays a similar role to IOB since high model-predicted insulin predictions correspond to a high IOB.

Model predictive control (MPC) uses a model to forecast the effect of proposed insulin infusion rates on glucose over a prediction horizon. The majority of proposed AID algorithms are based on MPC. Objectives include tracking a BG set point (which could be changing with time), keeping BG within a desired range (Zone-MPC or control to range) (Kovatchev et al. 2009b; Grosman et al. 2010), or minimizing risk. Cameron et al. (2012b) develop a multiple model probabilistic predictive control (MMPPC) strategy that accounts for uncertainty in predictions and manipulates insulin to minimize hypoglycemic risk.

In a simulation-based study, Cameron et al. (2011) compare basal-bolus, PID, MPC, and an enhanced MPC algorithm based on risk management. The merits of PID and MPC were analyzed by Steil (2013) and Bequette (2013), respectively. An overview of the different algorithms, delivery methods, and other engineering decisions is provided by Doyle et al. (2014). Much of the effort has involved the use of a single hormone, but a number of two-hormone (insulin and glucagon) studies have been performed.

Most algorithms require that a meal be “announced” to provide feedforward control through the associated meal insulin bolus; this requires the individual to estimate the carbohydrates in their meals and provide this information to the “hybrid” controller (the term commonly used for the combination of feedforward and feedback control). The MMPPC approach of Cameron et al. (2012b, 2014) also anticipates and detects meals, reducing the burden on individuals to provide a meal announcement and insulin bolus at mealtime.

Exercise can rapidly decrease BG and increase insulin sensitivity (reducing the insulin required) for many hours; thus it is desirable to provide exercise information as part of an AID strategy. Stenerson et al. (2014) incorporate heart rate and an accelerometer into a PLGS system but find little additional benefit due to the use of heart rate. Turksoy et al. (2015) use energy expenditure and galvanic skin resistance as additional sensor inputs to improve BG control during exercise. Breton et al. (2014) add heart rate to a control-to-range MPC strategy to improve BG regulation during exercise.

Fault Detection

Possible component-related faults include loss of the sensor signal (due to Bluetooth dropouts), pressure-induced sensor attenuation (PISA, due to pressure on the sensor), partial or complete insulin infusion set failure, and loss of controller (smart phone battery depletion). Baysal et al. (2014) present an approach to detect PISAs based on signal characteristics such as rate of change. Howsmon et al. (2017, 2018) present a statistical process monitoring type of approach to detect insulin infusion rate failures. Most systems default to a pre-programmed basal insulin delivery rate upon loss of communication of 20 min or more.

Simulation Models

There are two commonly used models to simulate the response of blood glucose to subcutaneous insulin infusion and meal carbohydrates. Kovatchev et al. (2009a) discuss the use of a simulator that has been accepted by the FDA to demonstrate preliminary results that can be used in investigational device exemption (IDE) applications for clinical trial studies. Wilinska et al. (2010) present a simulation framework based on the model developed by Hovorka et al. (2004).

The Hovorka simulation model is used to illustrate the effects of insulin and carbohydrates on BG. Figure 2 compares the effect of basal insulin rate on the steady-state BG for three different insulin sensitivities (nominal + 20%). Notice that even with a fixed sensitivity a relatively small change in insulin basal rate has a large effect on the steady-state BG. Also, insulin sensitivity varies with time of day and with exercise and illness. In this example, the nominal sensitivity curve with a basal rate of 1.058 U/h yields a steady-state BG of 120 mg/dL. If the actual sensitivity is 20% less, the BG is 190 mg/dL, while if the sensitivity is 20% more, the BG is 75 mg/dL. An individual that struggles with hypoglycemia is likely to err on the side of a lower basal insulin rate. Figure 3 is an open-loop simulation for a 50 g carbohydrate meal, with and without an insulin bolus provided at mealtime, with the desired range of 70–180 mg/dL also plotted; while it is clearly important to provide an insulin bolus at mealtime, it is known that many adolescents fail to do this two or more times a week, leading to higher mean BG levels.
Fig. 2

Steady-state relationship between insulin delivery and BG, for three different insulin sensitivities. The nominal condition used in the dynamic simulations that follow is shown as the “x” (1.058 U/h, 120 mg/dl)

Fig. 3

Nominal sensitivity and open-loop simulations for a 50 g carbohydrate meal. Desired range of 70–180 mg/dl is shown. Illustrates the importance of providing an insulin bolus at mealtime

The closed-loop (using PID) simulation study shown in Fig. 4 is for breakfast, lunch, dinner, and a snack of 75, 50, 75, and 40 g, respectively; a scenario with feedforward/feedback control (insulin bolus at mealtime) is compared with feedback only. Clearly much better results can be achieved when a “meal announcement” is given so that an insulin bolus can be delivered; the artificial pancreas literature refers to this feedforward/feedback strategy as “hybrid control.” Forlenza et al. (2018) have shown clinically the BG control improvement when an insulin bolus is given 20 min ahead of the meal vs. using only feedback control action.
Fig. 4

Closed-loop simulations of PID-based feedback with and without feedforward control. Nominal sensitivity with three meals (75, 50, 75 g) and a snack (40 g); desired range of 70–180 mg/dl is shown. Illustrates the importance of providing an insulin bolus at mealtime (feedforward or meal announcement)

Clinical Trial Results

Much effort is required to conduct clinical trials, including (in the United States) approvals from an Institute Review Board (IRB), an FDA Investigational Device Exemption (IDE), review by a data safety management (DSMB) team, and clinical trials registration; see Bequette et al. (2016, 2018) for overviews of the regulatory process. Initially, simulation-based studies are conducted to validate expected algorithm performance (Patek et al. 2009). Typically, system safety is then verified in a hospital or clinical research center. These are often followed, after DSMB review and perhaps another IDE, by supervised outpatient studies in a hotel or diabetes camp and then by short- and long-term outpatient studies at home. Clinical studies have been inconsistent with performance metrics, so Maahs et al. (2016) proposed clinical trial metrics that include mean CGM as well as %CGM time in range for a large number of ranges. A recent review of clinical trial results is presented by Boughton and Hovorka (2019), for both single (insulin) and multi-hormone (insulin +  glucagon) studies.

Buckingham and co-workers, in a series of articles, studied 160 subjects over 5722 subject nights using a predictive low-glucose suspend (PLGS) algorithm and showed hypoglycemia reduction in all age groups. A modification to the algorithm also bolused insulin if the glucose was predicted to be high, using a predictive hypo-/hyper-glycemic mitigation (PHHM) algorithm; 58 subjects over 2436 subject nights were studied in this phase. The PLGS and PHHM studies are summarized in Bequette et al. (2018); PHHM can be viewed as a control-to-range algorithm focused on overnight care.

Pinsker et al. (2016) report clinical trial results of a head-to-head comparison of MPC and PID in a study involving 30 participants and concluded that MPC had better performance with more time in range and lower mean glucose. A Zone-MPC study by Forlenza et al. (2017) was designed for prolonged infusion set wear, specifically to increase the probability of infusion set failures, in a 2-week study involving 19 subjects. While most studies use “meal announcement” (feedforward control), Cameron et al. (2017) present results on a multiple model probabilistic predictive control (MMPPC) system that anticipates and detects meals and does not require meal announcement. An MMPPC-based study by Forlenza et al. (2018) demonstrates the improvement and control that can be achieved if meal insulin boluses are given 20 min in advance of the meal. Dassau et al. (2017) report a comprehensive study involving 29 subjects over 12 weeks. In the longest study to date, Musolino et al. (2019) propose a protocol for a 6-month study involving 130 participants; the nonlinear MPC algorithm developed by Hovorka et al. (2004) will be used in this study.

A number of studies use both insulin and glucagon. El-Khatib et al. (2010) report in-patient clinical studies using a PD controller that is active under certain glucose concentrations to manipulate glucagon. Insulin is administered based on an adaptive, model predictive control strategy with a very short prediction horizon, making it similar to a PID controller. Russell et al. (2014) present outpatient results for a 5-day study with 20 adults and 32 adolescents. Blauw et al. (2016) stress the advantages of using a single integrated device (rather than separate smartphones and pumps) to manipulate both insulin and glucagon in a 4-day study involving 10 subjects. El-Khatib et al. (2017) study 39 subjects in a dual-arm at-home study of 11 days in closed-loop and 11 days in conventional therapy.

Commercial Devices

The first threshold-based low-glucose suspend system approved by the US FDA (in 2013) was the Medtronic 530G, while the Medtronic 640G uses a predictive low-glucose suspend algorithm. The first closed-loop system approved by the US FDA (in 2017) was the Medtronic 670G, which is commonly called a “hybrid closed-loop” system because meal announcement (an estimated of the amount of carbohydrates in the meal, with the associated insulin bolus) is required. Garg et al. (2017) report results on adolescents and adults in a 3-month at-home study using the 670G.

As of July 2019, a number of potential commercial devices have gone through various stages of clinical trials. Buckingham et al. (2018) report results for an OmniPod system based on the OmniPod patch pump, a Dexcom CGM, and a MPC algorithm developed by Doyle and co-workers at UCSB and Harvard. Brown et al. (2018) report a study involving five subjects and the use of a T:slim Pump, Dexcom CGM, and a “ControlIQ” algorithm developed at the University of Virginia by Kovatchev and co-workers. The iLet system by Damiano at Boston University is a device that delivers both insulin and glucagon; they plan to first commercialize the insulin-only system. Bigfoot Biomedical also has a closed-loop system under development, based on a model predictive control algorithm.

Do-It-Yourself (DIY) Movement

Frustrated by the slow development in commercial devices for automated insulin delivery, a large do-it-yourself community has started an open-source movement. Initial commercial CGM manufacturers did not provide a way of sharing CGM data in real time, leading to a community called Nightscout that shared ways of “hacking” CGMs to share data in real time in 2013 (Lee et al. 2017). Soon thereafter (2014) the open APS movement began, allowing DIY “Loopers” to implement their own closed-loop systems (Lewis 2018). The general approach that was developed is similar to a model predictive control algorithm, using models to predict the effect of meals and insulin on future CGM values (Diamond et al. 2019). The algorithm calculates the current insulin bolus (when spread over a 30-min interval) that will yield the desired CGM value at the end of the prediction horizon (the insulin action time). This involves an algebraic calculation rather than an optimization problem that is used in traditional MPC; the calculated insulin is not delivered if any CGM is predicted to be below a hypoglycemic threshold during the prediction horizon. Barnard et al. (2018) discuss challenges and potential benefits of collaborations between DIY individuals, device manufacturers, regulatory agencies, and caregivers.

Summary and Future Directions

Automated insulin delivery systems are far more than control algorithms to regulate BG by manipulating insulin (and perhaps glucagon). It is important to have easy-to-calibrate (or calibration-free) sensors (CGM), insulin infusion pumps, and infusion sets that have a low failure rate and easy-to-use interfaces to make it clear whether the system is under manual (open) or automatic (closed) loop. System faults should result in defaulting to a safe mode, which again should be clear to the user.

Future systems will better integrate knowledge about lifestyle (predictable exercise, eating and sleep times) and will include the use of calendar information (e.g., lunch meeting) and additional sensors (such as accelerometers and gyroscope from a smart watch). Indeed, Navarathna et al. (2018) report preliminary results for detecting meal motions using a smart watch, enabling an advisory feedforward action for meal announcement.

It is desirable to reduce the number of devices that must be placed on the body for current AID systems, which include a CGM (with transmitter), insulin infusion catheter attached to an insulin pump, and a smart phone or other device used for the controller. There is some activity toward incorporating the CGM and infusion set into the same device. Indeed, if these were incorporated into a patch pump that also contained the interface and algorithm, only a single device would need to be placed on the body. It is likely, however, that a smart phone would still be used to communicate with the patch pump which may be worn under clothing.

Recommended Reading

Doyle et al. (2014) review the many different engineering decisions that must be made when developing AID systems. Castle et al. (2017) provide a comprehensive appraisal of the future of AID. Ramkissoon et al. (2017) present a detailed assessment of safety hazards associated with AID. A special issue on Artificial Pancreas systems was published in the February 2018 issue of IEEE Control Systems Magazine. Cinar (2018) provides an overview of the papers in the issue. Huyett et al. (2018) presented the effect of glucose sensor (CGM) dynamics. Bondia et al. (2018) review the estimation of insulin pharmacokinetics and pharmacodynamics. El Fathi et al. (2018) focus on meal control, while Messori et al. (2018) perform a simulation-based study of an MPC strategy using individualized parameters. Turksoy et al. (2018) present an adaptive control based procedure to include activity monitor sensors in addition to CGM signals. Bequette et al. (2018) provide an overview of a 5000 subject night study to reduce hypoglycemic risk overnight.



  1. Atlas E, Nimri R, Miller S, Grunberg EA, Phillip M (2010) MD-logic artificial pancreas systems. Diabetes Care 33(5):1072–1076CrossRefGoogle Scholar
  2. Barnard KD, Ziegler R, Klonoff DC, Braune K, Petersen B, Rendschmidt T, Finan D, Kowalski A, Heinemann L (2018). Open source closed-loop insulin delivery systems: a clash of cultures or merging of diverse approaches? J Diabetes Sci Technol 12(6):1223–1226CrossRefGoogle Scholar
  3. Baysal N, Cameron F, Buckingham BA, Wilson DM, Chase HP, Maahs DM, Bequette BW (2014) A novel method to detect pressure-induced sensor attenuations (PISA) in an artificial pancreas. J Diabetes Sci Technol 8(6):1091–1096CrossRefGoogle Scholar
  4. Bequette BW (2009) Glucose clamp algorithms and insulin time-action profiles. J Diabetes Sci Technol 3(5):1005–1013CrossRefGoogle Scholar
  5. Bequette BW (2012) Challenges and progress in the development of a closed-loop artificial pancreas. Annu Rev Control 36:255–266CrossRefGoogle Scholar
  6. Bequette BW (2013) Algorithms for a closed-loop artificial pancreas: the case for model predictive control (MPC). J Diabetes Sci Technol 7(6):1632–1643CrossRefGoogle Scholar
  7. Bequette BW (2014) Fault detection and safety in closed-loop artificial pancreas systems. J Diabetes Sci Technol 8(6):1204–1214CrossRefGoogle Scholar
  8. Bequette BW, Cameron F, Baysal N, Howsmon DP, Buckingham BA, Maahs DM, Levy CJ (2016) Algorithms for a single hormone closed-loop artificial pancreas: challenges pertinent to chemical process operations and control. Processes 4(4):39. https://doi.org/10.3390/pr4040039CrossRefGoogle Scholar
  9. Bequette BW, Cameron F, Buckingham BA, Maahs DM, Lum J (2018) Overnight hypoglycemia and hyperglycemia mitigation for individuals with type 1 diabetes. How risks can be reduced. IEEE Control Syst 38(1):125–134.  https://doi.org/10.1109/MCS.2017.2767119MathSciNetCrossRefGoogle Scholar
  10. Blauw H, van Bon AC, Koops R, DeVries JH (2016) Performance and safety of an integrated artificial pancreas for fully automated glucose control at home. Diabetes Obes Metab 18:671–677CrossRefGoogle Scholar
  11. Bondia J, Romero-Vivo S, Ricaret B, Diez JL (2018) Insulin estimation and prediction. IEEE Control Syst Mag 38(1):47–66CrossRefGoogle Scholar
  12. Boughton CK, Hovorka R (2019) Advances in artificial pancreas systems. Sci Transl Med 11:484. eaaw4949.  https://doi.org/10.1126/scitranslmed.aaw4949
  13. Breton MD, Brown SA, Karvetski CH, Kollar L, Topchyan KA, Anderson SM, Kovatchev BP (2014) Adding heart rate signal to a control-to-range artificial pancreas system improves the protection against hypoglycemia during exercise in type 1 diabetes. Diabetes Technol Ther 16(8):506–11.  https://doi.org/10.1089/dia.2013.0333CrossRefGoogle Scholar
  14. Brown S, Raghinaru D, Emory E, Kovatchev B (2018) First look at control-IQ: a new-generation automated insulin delivery system. Diabetes Care 41(12):2634–2636. https://doi.org/10.2337/dc18-1249CrossRefGoogle Scholar
  15. Buckingham BA, Block J, Burdick J, Kalajian A, Kollman C, Choy M et al (2005) Response to nocturnal alarms using a real-time glucose sensor. Diabetes Technol Ther 7:440–447CrossRefGoogle Scholar
  16. Buckingham B, Chase HP, Dassau E, Cobry E, Clinton P, Gage V, Caswell K, Wilkinson J, Cameron F, Lee H, Bequette BW, Doyle FJ III (2010) Prevention of nocturnal hypoglycemia using predictive alarm algorithms and insulin pump suspension. Diabetes Care 33(5):1013–1018CrossRefGoogle Scholar
  17. Buckingham BA, Forlenza GP, Pinsker JE, Christiansen MP, Wadwa RP, Schneider J, Peyser TA, Dassau E, Lee JB, O’Connor J, Layne JE, Ly TT (2018) Safety and feasibility of the OmniPod hybrid closed-system in adult, adolescent, and pediatric patients with type 1 diabetes using a personalized model predictive control algorithm. Diabetes Technol Ther 20(4):257–262CrossRefGoogle Scholar
  18. Cameron F, Bequette BW, Wilson DM, Buckingham BA, Lee H, Niemeyer G (2011) A closed-loop artificial pancreas based on risk management. J Diabetes Sci Technol 5(2):368–379CrossRefGoogle Scholar
  19. Cameron F, Wilson DM, Buckingham BA, Arzumanyan H, Clinton P, Chase HP, Lum J, Maahs DM, Calhoun PM, Bequette BW (2012a) In-patient studies of a Kalman filter based predictive pump shut-off algorithm. J Diabetes Sci Technol 6(5):1142–1147CrossRefGoogle Scholar
  20. Cameron F, Niemeyer G, Bequette BW (2012b) Extended multiple model prediction with application to blood glucose regulation. J Process Control 22(7):1422–1432CrossRefGoogle Scholar
  21. Cameron F, Niemeyer G, Wilson DM, Bequette BW, Benassi KS, Clinton P, Buckingham BA (2014) Inpatient trial of an artificial pancreas based on multiple model probabilistic predictive control (MMPPC) with repeated large unannounced meals. Diabetes Technol Ther 16(11):728–734CrossRefGoogle Scholar
  22. Cameron FM, Ly TT, Buckingham BA, Maahs DM, Forlenza GP, Levy CJ, Lam D, Clinton P, Messer LH, Westfall E, Levister C, Xie YY, Baysal N, Howsmon D, Patek SD, Bequette BW (2017) Closed-loop control without meal announcement in type 1 diabetes. Diabetes Technol Ther 19(9):527–532.  https://doi.org/10.1089/dia.2017.0078CrossRefGoogle Scholar
  23. Castle JR, DeVries JH, Kovatchev B (2017) Future of automated insulin delivery. Diabetes Technol Ther 19(S3):S-67–S-72Google Scholar
  24. Cinar A (2018) Artificial pancreas systems. IEEE Control Syst Mag 38(1):26–29CrossRefGoogle Scholar
  25. Dassau E, Pinsker JE, Kudva YC, Brown SA, Gondhalekar R, Dalla Man C, Patek S, Schiavon M, Dadlani V, Dasanayake I, Church MM, Carter RE, Bevier WC, Huyett LM, Hughes J, Anderson S, Lv D, Schertz E, Emory E, McCrady-Spitzer SK, Jean T, Bradley PK, Hinshaw L, Sanz AJL, Basu A, Kovatchev B, Cobelli C, Doyle FJ III (2017) Twelve-week 24/7 ambulatory artificial pancreas with weekly adaptation of insulin delivery settings: effect on hemoglobin A1c an hypoglycemia. Diabetes Care 40:1719–1726CrossRefGoogle Scholar
  26. Diabetes Control and Complications Trial (DCCT) Research Group (1993) 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 329:977–986CrossRefGoogle Scholar
  27. Diamond T, Bequette BW, Cameron F (2019) A control systems analysis of “DIY looping.” Presented at the 2019 Diabetes Technology Meeting, BethesdaGoogle Scholar
  28. Doyle FJ III, Huyett LM, Lee JB, Zisser HC, Dassau E (2014) Closed-loop artificial pancreas systems: engineering the algorithms. Diabetes Care 37:1191–1197CrossRefGoogle Scholar
  29. El-Khatib FH, Russell SJ, Nathan DM, Sutherlin RG, Damiano ER (2010) A bihormonal closed-loop artificial pancreas for type 1 diabetes. Sci Transl Med 2:27ra27Google Scholar
  30. El-Khatib F, Balliro C, Hillar MA, Magyar KL, Ekhlaspour L, Sinha M, Mondesir D, Esmaeili A, Hartigan C, Thompson MJ, Malkani S, Lock JP, Harlan DM, Clinton P, Frank E, Wilson DM, DeSalvo D, Norlander L, Ly T, Buckingham BA, Diner J, Dezube M, Young LA, Goley A, Kirkman MS, Buse JB, Zheng H, Selagamsetty RR, Damiano ER, Russell SJ (2017) Home use of bihormonal bionic pancreas versus insulin pump therapy in adults with type 1 diabetes: a multicentre randomised crossover trial. Lancet 389:369–380CrossRefGoogle Scholar
  31. El Fathi A, Smaqui MR, Gingras V, Bolout B, Haidar A (2018) The artificial pancreas and meal control. IEEE Control Syst Mag 38(1):67–85CrossRefGoogle Scholar
  32. Forlenza GP, Deshpande S, Ly TT, Howsmon DP, Cameron F, Baysal N, Mauritzen E, Marcal T, Towers L, Bequette BW, Huyett LM, Pinsker JE, Gondhalekar R, Doyle FJ III, Maahs DM, Buckingham BA, Dassau E (2017) Application of zone model predictive control artificial pancreas during extended use of infusion set and sensor: a randomized crossover-controlled home-use trial. Diabetes Care 40:1096–1102. https://doi.org/10.2337/dc17-0500CrossRefGoogle Scholar
  33. Forlenza GP, Cameron FM, Ly TT, Lam D, Howsmon DP, Baysal N, Kulina G, Messer L, Clinton P, Levister C, Patek SD, Levy CJ, Wadwa RP, Maahs DM, Bequette BW, Buckingham BA (2018) Fully closed-loop multiple model predictive controller (MMPPC) artificial pancreas (AP) performance in adolescents and adults in a supervised hotel setting. Diabetes Technol Ther 20(5):335–343CrossRefGoogle Scholar
  34. Garg SK, Weinzimer SA, Tamborlane WV, Buckingham BA, Bode BW, Bailey TS, Brazg RL, Ilany J, Slover RH, Anderson SM, Bergentaal RM, Grosman B, Roy A, Cordero TL, Shin J, Lee SW, Kaufman FR (2017) Glucose outcomes with the in-hoe use of a hybrid closed-loop insulin delivery system in adolescents and adults with type 1 diabetes. Diabetes Technol Ther 19(3):156–163CrossRefGoogle Scholar
  35. Grosman B, Dassau E, Zisser HC, Jovanovic L, Doyle FJ III (2010) Zone model predictive control: a strategy to minimize hyper- and hypoglycemic events. J Diabetes Sci Technol 4:961–975CrossRefGoogle Scholar
  36. Hirsch IB (2004) Treatment of patients with severe insulin deficiency: what we have learned over the past 2 years. Am J Med 116(3A):17S–22SCrossRefGoogle Scholar
  37. Hovorka R, Canonico V, Chassin LJ, Haueter U, Massi-Benedetti M, Fedrici MO et al (2004) Nonlinear model predictive control of glucose concentration in subjects with type 1 diabetes. Physiol Meas 25(4):905–920CrossRefGoogle Scholar
  38. Howsmon DP, Cameron F, Baysal N, Ly TT, Forlenza GP, Maahs DM, Buckingham BA, Hahn J, Bequette BW (2017) Continuous glucose monitoring enables the detection of losses in infusion set actuation (LISAs). Sensors 17:161. https://doi.org/10.3390/s17010161CrossRefGoogle Scholar
  39. Howsmon DP, Baysal N, Buckingham BA, Forlenza GP, Ly TT, Maahs DM, Marcal T, Towers L, Mauritzen E, Deshpande S, Huyett LM, Pinsker JE, Gondhalekar R, Doyle FJ III, Dassau E, Hahn J, Bequette BW (2018) Real-time detection of infusion site failures in a closed-loop artificial pancreas. J Diabetes Sci Technol 12(3):599–607. https://doi.org/10.1177/19322968187551CrossRefGoogle Scholar
  40. Huyett LM, Dassau E, Zisser HC, Doyle FJ III (2018) Glucose sensor dynamics and the artificial pancreas. IEEE Control Syst Mag 39(1):30–46Google Scholar
  41. Kovatchev BP, Breton M, Dalla Man C, Cobelli C (2009a) In silico preclinical trials: a proof of concept in closed-loop control of type 1 diabetes. J Diabetes Sci Technol 3(1):44–55CrossRefGoogle Scholar
  42. Kovatchev B, Patek S, Dassau E, Doyle FJ III, Magni L, De Nicolao G et al (2009b) The juvenile diabetes research foundation artificial pancreas consortium. Control to range for diabetes: functionality and modular architecture. J Diabetes Sci Technol 3(5):1058–1065CrossRefGoogle Scholar
  43. Lee JM, Newman MW, Gebremariam A, Choi P, Lewis D, Nordgren W, Costik J, Wedding J, West B, Gilby N, Benovich C, Pasek J, Garrity A, Hirschfeld E (2017) Real-world use and self-reported health outcomes of a patient-designed do-it-yourself mobile technology system for diabetes: lessons for mobile health. Diabetes Technol Ther 19(4):209219CrossRefGoogle Scholar
  44. Lewis D (2018) History and perspective on DIY closed looping. J Diabetes Sci Technol 13(4):790–793CrossRefGoogle Scholar
  45. Maahs DM, Buckingham BA, Castle JR, Cinar A, Damiano ER, Dassau E, DeVries JH, DoyleIII FJ, Griffen SC, Haidar A et al (2016) Outcome measures for artificial pancreas clinical trials: a consensus report. Diabetes Care 39:1175–1179CrossRefGoogle Scholar
  46. Mauseth R, Wang Y, Dassau E, Kircher R, Matheson D, Zisser H et al (2010) Proposed clinical application for tuning fuzzy logic controller of artificial pancreas utilizing a personalization factor. J Diabetes Sci Technol 4:913–922CrossRefGoogle Scholar
  47. Messori M, Incremona CP, Cobelli C, Magni (2018) Individualized model predictive control for the artificial pancreas. IEEE Control Syst Mag 38(1):86–104Google Scholar
  48. Musolino G, Allen JM, Hartnell S, Wilinska ME, Tauschmann M, Boughton C, Campbell F, Denvir L, Trevelyan N, Wadwa P, DiMeglio L, Buckingham BA, Weinzimer S, Acerini CL, Hood K, Fox S, Kollman C, Sibayan J, Borgman S, Cheng P, Hovorka R (2019) Assessing the efficacy, safety and utility of 6-month day-and-night automated closed-loop insulin delivery under free-living conditions compared with insulin pump therapy in children and adolescents with type 1 diabetes: an open-label, multicenter, multinational, single-period, randomized, parallel group study protocol. BMJ Open 9:e027856.  https://doi.org/10.1136/bmjopen-2018-027856CrossRefGoogle Scholar
  49. Navarathna P, Bequette BW, Cameron F (2018) Device based activity recognition and prediction for improved feedforward control. 2018 American control conference, Milwaukee, pp 3571–3576.  https://doi.org/10.23919/ACC.2018.8430775
  50. Palerm CC (2011) Physiologic insulin delivery with insulin feedback: a control systems perspective. Comput Methods Prog Biomed 102(2):130–137MathSciNetCrossRefGoogle Scholar
  51. Patek SD, Bequette BW, Breton M, Buckingham BA, Dassau E, Doyle FJ III, Lum J, Magni L, Zisser H (2009) In silico preclinical trials: methodology and engineering guide to closed-loop control. J Diabetes Sci Technol 3(2):269–282CrossRefGoogle Scholar
  52. Pinsker JE, Lee JB, Dassau E, Seborg DE, Bradley PK, Gondhalekar R, Bevier WC, Huyett L, Zisser HC, Doyle FJ 3rd (2016) Randomized crossover comparison of personalized MPC and PID control algorithms for the artificial pancreas. Diabetes Care 39(7):1135–1142. https://doi.org/10.2337/dc15-2344. PubMed PMID: 27289127; PMCID: PMC4915560
  53. Ramkissoon C, Aufderheide B, Bequette BW, Vehi J (2017) Safety and hazards associated with the artificial pancreas. IEEE Rev Biomed Eng 10:44–62.  https://doi.org/10.1109/RBME.2017.2749038CrossRefGoogle Scholar
  54. Russell SJ, El-Khatib F, Manasi S, Magyar KL, McKeon K, Goergen LG, …Damiano ER (2014) Outpatient glycemic control with a bionic pancreas in type 1 diabetes. N Engl J Med 371(4):313–325.  https://doi.org/10.1056/NEJMoa1314474
  55. Steil GM (2013) Algorithms for a closed-loop artificial pancreas: the case for proportional-integral-derivative control. J Diabetes Sci Technol 7(6):1621–1631CrossRefGoogle Scholar
  56. Steil GM, Palerm CC, Kurtz N, Voskanyan G, Roy A, Paz S, Kandeel FR (2011) The effect of insulin feedback on closed loop glucose control. J Clin Endocrinol Metab 96(5):1402–1408CrossRefGoogle Scholar
  57. Stenerson M, Cameron F, Wilson DM, Harris B, Payne S, Bequette BW, Buckingham BA (2014) The impact of accelerometer and heart rate data on hypoglycemia mitigation in type 1 diabetes. J Diabetes Sci Technol 8(1):64–69CrossRefGoogle Scholar
  58. Turksoy K, Paulino TM, Zaharieva DP, Yavelberg L, Jamnik V, Riddell MC, Cinar A (2015) Classification of physical activity: information to artificial pancreas control systems in real time. J Diabetes Sci Technol 9(6):1200–1207. https://doi.org/10.1177/1932296815609369CrossRefGoogle Scholar
  59. Turksoy T, Littlejohn E, Cinar A (2018) Multimodule, multivariable artificial pancreas for patients with type 1 diabetes. IEEE Control Syst Mag 38(1):105–124CrossRefGoogle Scholar
  60. Wilinska ME, Chassin LJ, Acerini CL, Allen JM, Dunger DB, Hovorka R (2010) Simulation environment to evaluate closed-loop insulin delivery systems in type 1 diabetes. J Diabetes Sci Technol 4:132–144CrossRefGoogle Scholar

Copyright information

© Springer-Verlag London Ltd., part of Springer Nature 2020

Authors and Affiliations

  1. 1.Chemical and Biological EngineeringRensselaer Polytechnic InstituteTroyUSA

Section editors and affiliations

  • B. Wayne Bequette
    • 1
  1. 1.Chemical & Biological EngineeringRensselaer Polytechnic InstituteTroyUSA