Closed-Loop Insulin Delivery: We're “Virtually” There

T he introduction of real-time continuous glucose monitoring (RT-CGM) devices provided the missing link that made closed-loop insulin delivery using external sensors and external pumps a practical reality. With RT-CGM, standard or more advanced computer algorithms could be used to run insulin pumps based on real-time sensor glucose levels. At the dawn of the closed-loop era, regulatory requirements demanded that the algorithms for use in closed-loop systems be tested first in laboratory animals. The pioneering work of Gary Steil, Kerstin Rebrin, and their colleagues at the Medtronic MiniMed Laboratories that demonstrated the effectiveness of the Proportional-Integrative-Derivative algorithm in dogs ¹ set the stage for the for “first in man” studies involving adults and children with type 1 diabetes (T1D). ^2,3 More recently, at the University of Virginia, Boris Kovatchev and associates developed computer models of glucose metabolism in people with type 1 diabetes that could be used for in silico testing of closed-loop controller algorithms. ⁴ When the University of Virginia in silico model was accepted by the U.S. Food and Drug Administration as a substitute for animal studies, the number of investigator groups and projects devoted to artificial pancreas projects and algorithm development virtually exploded. The importance of the article by Schmidt et al. ⁵ in this issue of Diabetes Technology & Therapeutics is that it adds to the underlying database upon which to build more comprehensive models of glucose regulation that can be used for the development of new algorithms.

While these studies are clearly of great value, you will have to excuse the skepticism of an old clinical investigator regarding how well we can extrapolate and predict glucose regulation in real T1D patients based on data that are obtained under highly artificial research center conditions. Unlike the study protocol described in their methods section, in real life, “system inputs” are not introduced in a predetermined and organized order. Moreover, we will never be able to accurately record or measure all the disparate influences that impact on glucose control in individuals with T1D, especially in children and adolescents. That's why I like to call the algorithm that we currently use in our Children's Diabetes Clinic the “MPC Algorithm”: Confusing to Model, Impossible to Predict, and Difficult to Control. Rather than facing all the rigors of everyday life, subjects in the study by Schmidt et al. ⁵ were allowed to rest for 2 h prior to the first study procedure so that they could recover from the effects of transportation to the hospital. In addition, between study procedures, the subjects spent the day reclining in bed. The investigators indicate that they plan to incorporate the data that they are acquiring into models of glucose metabolism that will be used to develop a virtual clinic. I can assure you that patients in the virtual clinic will be a heck of a lot easier to deal with than the often-annoying adolescents who we care for every day in our real clinics.

Approval of algorithms based on in silico modeling has paved the way for a variety of short-term, clinical research center (CRC)–based pilot and feasibility studies. Led by vision of Dr. Aaron Kowalski, the Juvenile Diabetes Research Foundation (JDRF) was at the forefront of support for studies being carried out by a rapidly increasing number of investigator teams in the United Sates, Europe, and Israel; the National Institute of Diabetes and Digestive and Kidney Diseases has also made a major commitment to this area of research. Closed-loop, CRC studies have examined the performance of the systems for overnight control and during and after exercise, whether a dual-hormone system that administers glucagon as well as insulin can reduce the risk of hypoglycemia, whether the tendency to postmeal hyperglycemia can be blunted by manual premeal priming boluses and by premeal injections of pramlintide, and to test the efficacy of new and modified controller algorithms. The JDRF Ultra-Fast Acting Insulin project is a complementary research program directed at improving closed-loop performance by developing methods to accelerate absorption and time–action profiles of subcutaneously injected rapid-acting insulin analogs.

The overarching goal of everyone involved in artificial pancreas development is to move these systems out of the hospital and into outpatient treatment as soon as possible. There is no question that the first generation of external closed-loop systems has to be better than what we are doing now. Moreover, from the patients' perspective, the introduction of an artificial pancreas for outpatient treatment would represent the first advance in diabetes technology that would improve metabolic control while simultaneously markedly reducing the burdens of caring for their disease (Big Gain/Less Pain). The major obstacle to future outpatient use is to ensure the safety of the closed-loop system. Specifically, the scientists and engineers who are developing these systems must ensure that every possible safeguard is in place to prevent excessive administration of insulin due to a system malfunction. As illustrated by the really bad sensor performance in the study by Schmidt et al. ⁵ (mean absolute relative difference of 21.6%), we need more accurate and easier-to-use sensors. Dual sensors, improved algorithms to detect failing sensors, limitations of maximum delivery doses, and ensuring the integrity of radiofrequency transmissions are some of the other ways that system safety can be enhanced. Considerable work also remains to enhance the device–patient interface to make it as easy to use as possible and to limit the risk of overdelivery because of user error (e.g., permitting only blood glucose values transmitted from a linked meter for sensor calibrations to prevent use of made-up, manually entered values).

Although overdelivery of insulin is the major concern when closed-loop systems are turned on to prevent hyperglycemia, turning off a pump for a brief period of time to prevent or reduce the magnitude, duration, and consequences of hypoglycemia, especially during sleep at night, appears to be of little or no risk and of potentially great benefit (Big Gain/No Pain). The Medtronic Veo™ Low Glucose Suspend (LGS) system will automatically suspend a pump's basal insulin infusion for 2 h if sensor glucose levels fall to the preprogrammed alarm level and the patient does not respond to the alarm. A previous study from our laboratory showed that such a 2-h interruption of a pump's basal rate is followed by a modest rise in blood glucose levels and no clinically significant increases in blood ketone levels. ⁶ Moreover, Buckingham et al. ⁷ reported a case study involving T1D patients who were wearing a sensor on nights in which there was a hypoglycemic seizure; in each case, the seizure was preceded by several hours of hypoglycemia, providing a window of opportunity where shutting off the pump might have prevented the seizure. The next step will be an LGS system that will temporarily shut off a pump based on a projected rather than an actual low sensor value. Our data suggest that this approach may be able to prevent more than 75% of hypoglycemic events, as well as more than 50% of hypoglycemic alarms. ⁸

Although it is rare to get groups of diabetologists to agree on anything, there has been an almost universal clamor by clinicians in the United States to make the Veo LGS system available to our patients with T1D. We would like to join our friends in Australia, Canada, and Europe whose regulatory authorities have already approved this potentially lifesaving device.