Fraser Cameron, University of Texas at El Paso, United States
The Artificial Pancreas: Algorithms, Results, and Considerations in Controlling Blood Glucose Levels in People with Type 1 Diabetes Mellitus
Type 1 Diabetes Mellitus (T1DM) destroys the body’s capacity to produce insulin. For people with T1DM this imposes the burden of injecting insulin to regulate their blood glucose concentration. For good control, an individual must check their glucose concentration and adjust insulin delivery 4-6 times per day. Poor control incurs the risk of chronic organ damage from excessive glucose levels (hyperglycemia), or the acute effects of depriving the brain of fuel, e.g. low glucose levels (hypoglycemia). The artificial pancreas (AP) project aims to reduce the burden of glucose regulation. In this talk, I will focus on the algorithms that calculate how much insulin to inject via an insulin pump in response to the signal(s) from a continuous glucose monitor (CGM). Three things complicate AP algorithm design: the relative severity of low vs. high glucose levels, the relative slowness of insulin vs. meal and exercise disturbances, and the weak capacity for raising the glucose levels by suspending insulin delivery. Different algorithms lighten different parts of the burden. For example, a recent 2,000-night outpatient trial tested a predictive pump shut-off algorithm, which suspends insulin delivery for brief periods to mitigate hypoglycemia overnight. An extension of this trial will also add insulin to mitigate predicted hyperglycemia overnight. During the day, meals complicate control, prompting several trials to use meal announcements to handle them. Finally, my algorithm and others, provide full glucose control, responding to unannounced meals as they happen. A practical, profitable product requires many safety features. It needs to guard against device failure and the variety of people and everyday life. Accelerometers, for instance, can measure activity and torso orientation, which can help improve exercise response, meal detection, and removal of orientation-dependent sensor anomalies. Likewise, analytic and hardware redundancy can help adapt models and detect poor calibration and pump failures early. I will illustrate the key aspects of the AP problem, and present clinical and simulation results for various personal and published AP implementations. Lastly, I will discuss safety algorithms and the future of the AP project.
Hailing from Winnipeg, Fraser Cameron graduated with distinction from the Systems Design Engineering Program at UW in 2002. He then worked as an electrical engineer in Winnipeg for a year before joining the Department of Aeronautics and Astronautics at Stanford University. He graduated in 2010 with a Ph.D. thesis titled "Explicitly Minimizing Clinical Risk through Closed Loop Control of Blood Glucose in Patients with Type 1 Diabetes Mellitus". He continued his research into closed loop control of blood glucose levels in people with Type 1 Diabetes throughout a postdoc at Rensselaer Polytechnic Institute and now as a faculty member at the University of Texas at El Paso. His research has been tested in over 4,000 nights of outpatient trial and roughly 10 clinical trials total. He has been involved in many control theory aspects of diabetes control including full closed loop control, dead-band control, actuator failure detection, sensor failure detection, calibration, and event detection. This research has resulted directly in 16 refereed publications and over 36 conference presentations.
Invited by the Department of Electrical and Computer Engineering