Medtronic: Designing a Closed Loop System

This past weekend, I traveled a couple hundred miles north to Los Angeles to attend the second annual Medtronic Diabetes Advocate Forum. Medtronic entered the diabetes space as a manufacturer of insulin pumps, and has since expanded its role to offer continuous glucose monitors. Much of the day was spent talking about progress they’ve made with those two product lines, and they discussed their efforts to get FDA approval of their Enlite continuous glucose monitor (CGM) sensor, and their recent release of an external display device for patients and caregivers, the mySentry Remote Glucose Monitor.

I of course always love the opportunity to hang out with other diabetics and to complain to companies about their products without proffering any better solutions, so the Forum was a fun experience. Others have written about their impressions of the event (see here and here), though, so I would like to share my thoughts about what Medtronic presented to us regarding their efforts towards creating a closed loop system—that is, an insulin delivery device capable of determining and administering required insulin doses without human intervention.

 Lane Desborough, Product Strategist at Medtronic, led a discussion of the closed loop system, and what his team faces in designing one. First he explained the problem: we want to be able to control glycemic variation, and we need to do this by transferring variance from one area—blood sugar levels—to another—insulin levels. The pancreas does this naturally; when a non-diabetic takes in sugar in the form of food, his pancreas rapidly increases the rate at which it releases insulin such that normal blood sugar levels are maintained. Conversely, when a non-diabetic exercises, the pancreas quickly decreases the amount of insulin being released, and the liver begins to release stored glucose—and normoglycemia is again maintained. The goal of a closed loop system (often referred to more broadly as an artificial pancreas, though the latter implies fewer hedged bets and a more complete, perfected product), then, is to achieve this same control of blood glucose values by regulation of insulin levels externally.

To build such a system, Medtronic focuses on three components: a glucose sensor, an insulin delivery device, and an algorithm to govern their behavior. Each component in this trinity plays a crucial role in making the system work—the glucose sensor monitors current blood sugar levels in order to determine where the patient is and where she is moving; the insulin delivery system dispenses insulin at varying rates to change blood glucose levels; and the control algorithm makes the decisions, calculating how much insulin is needed based on the current glucose level inputs from the sensor.

Each component in the trinity, however, presents a unique challenge, and Medtronic will have to address these challenges before a practical closed loop system can be used effectively by patients in their daily lives. The first challenge comes with the glucose sensor: as any user of the currently available continuous glucose monitors (CGMs) knows, the existing sensor systems are a work-in-progress. I personally wear the Medtronic CGM, and it is not unusual to have whole days where the sensor is off, showing wildly fluctuating glucose levels when I am more or less stable, or vice versa. This is fine as long as the CGM is acting as a supplement to blood glucose meter readings.  But if it serves as the primary input to an automated system that is determining insulin dosing, it needs to be extremely accurate and reliable to be safe.

Medtronic is addressing this problem of CGM reliability in several ways. First, they are currently working to improve their existing sensor technology, making it easier to use and more consistent. Their Enlite sensor, available in Europe, is allegedly already leaps and bounds above the model currently available in the U.S. in these regards, and Medtronic promises the Enlite 2 that is under development is even better.

Medtronic, though, is interested in taking accuracy and reliability one step further. Greg Meehan, Vice President of the CGM business at Medtronic, described ongoing efforts to design a fully redundant sensor system. The current CGM measures the electric current created when glucose reacts with the enzyme glucose oxidase, and rather than relying on this system alone, Medtronic aims to couple a microelectric sensor with an optical sensor that relies on a different means of detecting glucose. In 2009, Medtronic bought the Danish company PreciSense, which is working on a short-term use sensor that is injected into the upper layer of the skin and measures the amount of glucose binding to fluorescent receptors within the sensor. When glucose binds to the receptors, the receptors give off light that can be measured, quantified, and reported back in terms of glucose concentration. This optical means of quantifying glucose levels in the tissue would act as a second check for the existing sensor, and the two together could be read and correlated by the closed loop system to ensure accurate glucose inputs were being used by downstream algorithms. The optical sensor system is still under development, but PreciSense units have already been tested in pigs and humans. Notably, though, the exact implementation Medtronic chooses for the second sensor matters less than the fact that they are building a redundant, reliable system with distributed risk; as a potential wearer, this is very important to me.

Even assuming perfect glucose sensing capabilities, though, Medtronic faces another hurdle: insulin delivery must be fast and efficient enough to prevent excursions once the closed loop system detects that more or less insulin is needed. As it turns out, this issue of speedy delivery is no small one; the JDRF has identified faster-acting insulin as one of the major funding areas in the next few years. Several companies—including Biodel Inc. and Halozyme Therapeutics—are making progress towards chemically faster insulin, while others are looking into ways to deliver insulin directly into the bloodstream so that it doesn’t have to first diffuse through tissue.

What does this hurdle mean to Medtronic’s closed loop system? According to Desborough, “I think any technology/pharmacology to decrease the elapsed time to get the insulin to where it needs to be…is the key factor in quality of control.” In other words, figuring out how to correct for the total delay inherent in the system right now—from blood sugar change, to measurements of that change via a CGM, to determination of correct insulin dose, to delivery of that dose, to action of that insulin within the body, to lower blood sugar—is critical to the correct operation of a closed loop system. That said, it is not practical for a company like Medtronic to develop a chemically faster insulin at this stage; they are a medical device company, and it makes more sense to leave the pharmacology to the pharmacologists. That doesn’t mean Medtronic is stuck, though—it just means that, as Desborough says, “control engineers have to make the best of the hand we are dealt.” The alternative to reducing the delay in response time is engineering around the delay. This sort of engineering, luckily, is Desborough’s forte—his graduate work in the early nineties and his subsequent career at Honeywell focused on implementations of algorithms designed to minimize the difference between a variable—say, blood glucose level—and a target value, assuming that we know the delay in time between the execution of the algorithm and the output variable we are trying to make near the target value. What this means practically is that there are ways in software to predict what the system needs to do now to reach a target blood sugar a known amount of time in the future—which is crucial to making a closed loop system work in the absence of instantaneous insulin.

This algorithmic approach (called Minimum Variance Control, as it aims to minimize the variance between a variable and a target value) has a severe limitation, however: it can’t read your mind. In other words, it can handle a known time delay and even some stochasticity, but it can’t predict how much insulin I will need in twenty minutes if it doesn’t know I’m going to eat a piece of cake in ten. Instantaneous delivery of insulin here would be helpful, but even that is not enough; the system would have to wait until blood glucose values were rising before it could respond with an influx of insulin. In theory, it wouldn’t have to wait to the point of hyperglycemia, but one of the beauties of the human pancreas is that it can read our minds, at least in the sense that it is responding not just reactively to blood glucose, but proactively to the myriad bodily and hormonal changes that comprise the concert of human metabolism. And this brings me to a third major challenge of the closed loop system as it is currently being implemented: any algorithm that is developed has to work with a single biometric input—blood glucose values. Building an algorithm that takes glucose values and outputs insulin dosages such that normoglycemia is maintained is feasible, but it’s not easy and it’s not what the body is doing. It is theoretically preferable, then, to have more inputs into the algorithm, so that the system is not limited to decision-making based on a single piece of the body’s puzzle.

So what else should be an input into the algorithms running the closed loop system? As always, there are trade-offs. Dr. Francine Kaufman, the Chief Medical Officer of Medtronic, notes that they tried accelerometers, but there are diminishing returns—the additional complexity only adds minimal signal to the system. When I asked Desborough what was on his data wishlist—what he would measure if feasibility were not an object—and he said, “I believe the biggest bang for the buck is to put a three axis accelerometer directly on the body, i.e. integrated with the glucose sensor transmitter.  You’d want to do quite a bit of signal processing to convert it into a five minute, single valued energy consumption metric.”

An accelerometer is the most obvious next step—as any diabetic can tell you, exercise and movement drives down blood sugar levels remarkably quickly—but I wonder what else is out there that we’re missing. What makes the pancreas so good at what it does? What does it know that we don’t?

If feasibility were no object, I would try to get some measure of good intake. Could we watch salivation? What about molecules expressed or released in the gut that anticipate the food we’re eating? Or maybe even measure the food itself—implant miniature calorimeters in our mouths, or small carbohydrate-binding sensors. And while we’re at it, let’s monitor hormone levels to get an idea of stress and insulin sensitivity—adrenaline, norepinephrine, estrogen. And leptin to estimate the likelihood that we’ll want to be eating any time in the near future.

Of course, feasibility is always an object, even if money were not. We don’t have any means of measuring these things now, and, even if we did, we would still have to figure out in what way these varying signals feed in to the body’s decision-making process. This exercise, though, does drive home how wonderfully complex our bodies are—when they work right. And when they don’t? Well, hopefully Medtronic and those they work with will have a next-best solution in the coming years. I don’t need it to be perfect; I appreciate each incremental step that takes even a little of the burden of being a pancreas off of my own shoulders. In the meantime, I’ll keep taking it one blood glucose value at a time.

Karmel Allison is science editor of ASweetLife.  She writes the blog Where is My Robot Pancreas?

*Disclosure – Karmel Allison was invited to the Second Annual Diabetes Advocate Forum by Medtronic. Medtronic paid her expenses.

Karmel Allison

Karmel was born in Southern California, diagnosed with Type 1 Diabetes at the age of nine, and educated at UC Berkeley. Karmel now lives in San Diego with her husband, where she is loving the sunshine, working in computational biology at the University of California, San Diego, and learning to use the active voice when talking about her diabetes.

Karmel Allison

Karmel was born in Southern California, diagnosed with Type 1 Diabetes at the age of nine, and educated at UC Berkeley. Karmel now lives in San Diego with her husband, where she is loving the sunshine, working in computational biology at the University of California, San Diego, and learning to use the active voice when talking about her diabetes.