Diabetes Forecast

The Artificial Pancreas

Scientists race to make the closed-loop glucose management system a reality

By Erika Gebel, PhD ,

Though he'd spent a day and night in the hospital, Thomas Brobson was perky and grinning at 8:30 the next morning. "I'm feeling great," said Brobson, who has type 1 diabetes. "I haven't stuck my finger in 24 hours."

Even with his salt-and-pepper hair, Brobson, who runs a Christmas tree farm in southwest Virginia, doesn't look his 50 years. He was diagnosed at age 44 and describes his style of blood glucose control as "naturally aggressive." Yet he didn't seem to mind handing over the reins of his diabetes management to a computer.

Thomas Brobson at the hospital, testing an experimental version of the artificial pancreas.

Brobson had just completed a trial run of what's known as an artificial pancreas. Two weeks before, he'd spent an initial 24 hours in the same University of Virginia hospital, taking care of his diabetes by his usual methods. Under those conditions, Brobson's blood glucose dipped below 70 mg/dl—too low—eight times. But the second time around, with the artificial pancreas in control, his blood glucose level hit that threshold only three times.

This experiment, which will be repeated with a handful of other volunteers, is a pilot study for a major global trial of an artificial pancreas, a critical step in the development of a device that could someday, perhaps within a few years, make the lives of people with diabetes safer, healthier, and easier. The challenges of automating delivery of a medication that can cause severe lows are significant, but people with diabetes, doctors, researchers, regulatory agencies, nonprofit organizations, and corporations are working together to make the artificial pancreas a reality as soon as possible.

The artificial pancreas does not, in fact, actually aim to replace the entire biological organ, just its beta cells. In type 1 diabetes, the pancreatic beta cells—the only cells in the body able to produce insulin—are destroyed by a malfunctioning immune system. The beta cells in people with type 2 have also stopped producing an adequate insulin supply. Insulin's job is to move glucose from the blood into the body's cells, where it can be used for energy. A shortage of insulin traps glucose in the blood, leading to the damaging high blood glucose levels that characterize diabetes.

People with type 1 diabetes, who would benefit the most from an artificial pancreas, and some with type 2 need to inject or pump insulin into their bodies to survive. Figuring out just how much insulin is appropriate can be difficult. Research shows that near-normal blood glucose levels can help stave off devastating diabetes complications that ravage the eyes, nerves, kidneys, and heart. But maintaining that tight level of control requires insulin doses that walk the line between enough and too much. To make matters worse, the body's response to insulin can vary widely and unpredictably, depending on factors like food, exercise, stress, and illness. This volatility can leave a person with excess insulin in the body, which may cause severe hypoglycemia (low blood glucose). The danger is particularly grave, and lows are most common, during sleep.

While the artificial pancreas is not a cure for diabetes, it could help people get all the complication-averting benefits of tight control without the serious dangers of hypoglycemia. An artificial pancreas, Brobson says, "would enhance the quality of my life and relieve what is a constant burden."

At least at first, an artificial pancreas will very likely consist of a continuous glucose monitor (CGM) and an insulin pump programmed with a computer algorithm that "closes the loop," calculating insulin doses from blood glucose readings and telling the pump to deliver the medication. In early versions of the artificial pancreas, the CGM and insulin pump will be similar, if not identical, to those already available. The main difference will be that one of these devices will house the algorithm, a set of mathematical steps that takes into account various factors to determine the proper amount of insulin to dispense.

CGMs transmit glucose readings every one to five minutes from an under-the-skin sensor to a handheld receiver, which can be integrated into a pump. That's an enormous amount of information. Some of it is lost on a human, but not on an artificial pancreas, because algorithms love data. Three CGMs are now on the market, each having received Food and Drug Administration (FDA) approval within the past five years. But current CGMs don't tell a pump what to do. While studies show that CGMs help improve blood glucose control, these devices aren't perfect, and some issues need to be resolved before they are ready for automation as part of an artificial pancreas. Unlike blood glucose meters, CGMs measure glucose levels in the interstitial fluid, the liquid that surrounds the body's cells, which is separate from the blood. Interstitial glucose readings typically lag behind actual blood glucose levels by 8 to 10 minutes, which is most significant during periods of rapid change, such as after a meal. That's a key consideration when designing a closed-loop device, since insulin doses will rely on those measurements.

The Artificial Pancreas

In the University of Virginia study, the artificial pancreas consists of a continuous glucose monitor (CGM), an insulin pump, and an algorithm housed in a laptop computer.

1. The CGM sensors wirelessly send glucose readings to receivers in a fanny pack. (Two sensors are used for added safety.) The data are transmitted via cables to the laptop. Future studies will do this step wirelessly. 2. The algorithm—the brains of the artificial pancreas—uses data from the sensors to figure how much insulin is needed to keep the patient's blood glucose in a safe, healthy range. 3. The insulin pump delivers doses as directed by wireless signals from the computer.

"CGMs are only approved for tracking and trending" blood glucose, not for determining an insulin dose, says Chip Zimliki, PhD, the chair of the artificial pancreas working groups, the units of the FDA that review research proposals related to the development of the device. "Sometimes these sensors can be off. If you are basing your dosing on that, it can be problematic." That's why, if someone on a CGM sees his or her glucose go out of range, the results must be double-checked with a finger-stick meter before corrective steps are taken.

The next component, an insulin pump, is a beeper-sized device that is often tethered to the body via a flexible tube inserted into the tissue just under the skin. It delivers insulin as directed by the user. Yet unlike insulin that comes from the pancreas, insulin delivered into tissue takes time to get into the bloodstream, where it can lower blood glucose. People with diabetes typically take insulin 15 minutes or so before a meal, giving it time to start working. So another hurdle in developing an artificial pancreas is optimizing the timing of insulin doses to prevent post-meal glucose spikes. There are faster insulins in the pipeline, so it may not be long before developers of the artificial pancreas have more tools to help them achieve automation.

Any of the CGMs and pumps on the market could end up in an artificial pancreas. Even with their limitations, says Zimliki, "we are hopeful that we can do this with existing technology." As of now, researchers have used different combinations of pumps and CGMs in their clinical trials. Researchers at the University of California–Santa Barbara (UCSB) have made testing different CGM, pump, and algorithm combinations easier by creating a computer platform, the Artificial Pancreas System (APS), that has "plug and play" capabilities. "We're the glue of sorts, pulling pieces together," says Francis Doyle III, a professor of chemical engineering at UCSB who has spearheaded the APS. "We've integrated into our platform . . . two different sensors and three different pumps." The APS is now being used in artificial pancreas clinical trials across the globe.

A future artificial pancreas might use sleeker-looking hardware than early incarnations and perhaps employ implantable pumps or blood glucose monitoring that doesn't break the skin. "I think of it like cell phones," says Brobson. "First, cell phones were giant unlikable things. Now, we have the iPhone."

Boris Kovatchev, PhD, developed the algorithm used in the University of Virginia experiment.

The stereotype of the disheveled scientist didn't come from Boris Kovatchev, PhD, associate professor of psychiatry and neurobehavioral sciences at the University of Virginia, whose spotless office, neat goatee, and tidy tweed jacket are as meticulous as his mathematical prowess. The algorithm he developed drives the artificial pancreas in the Virginia study, its dense equations too formidable for all but the most quantitative of minds.

Kovatchev's algorithm isn't alone in the race to close the loop; several other contenders are quietly attempting to capture pancreatic perfection in bytes. The simplest of these is being studied by the company Medtronic Diabetes, a pioneer in testing an artificial pancreas. Its algorithm is somewhat generic and uses just three variables to calculate an insulin dose, according to Stuart Weinzimer, MD, an endocrinologist at Yale University School of Medicine, who has tested the experimental Medtronic devices: "What is the blood sugar? Is it rising or falling? And is the long-term goal higher or lower?"

The more complex approach being championed by Kovatchev and others uses custom-made algorithms designed specifically to model the workings of the human pancreas. These models can incorporate knowledge accrued through decades of diabetes research and account for the imperfections of mechanical systems, like delays related to CGM measurements or insulin delivery.

The two model-based algorithms furthest along in clinical studies are those of Kovatchev and Roman Hovorka, PhD, of the University of Cambridge in England. Both algorithms calculate insulin doses anew as fresh glucose readings flow in. The main difference is that Hovorka's modifies itself, too, while the Kovatchev algorithm, after initialization, doesn't change. "There are patients that don't need adaptations," says Hovorka. "But there are those who change their insulin needs dramatically from day to day." Even so, Hovorka concedes that the simplest algorithm that can do the job will be the best choice.

Roman Hovorka, PhD, is working on a different artificial-pancreas algorithm at the University of Cambridge in England.

The algorithms of Kovatchev and Hovorka will be compared directly in an upcoming clinical trial at six European centers. Though this protocol seems to pit these two scientists against each other, Kovatchev likes to emphasize the congeniality and collaborative spirit that characterizes the worldwide effort to develop an artificial pancreas. He proudly displays a Christmas card signed by members of the Hovorka lab.

In the end, the artificial pancreas may offer plenty of room for different ideas. "I can't believe there's a winner-takes-all solution," says UCSB's Doyle. "Just like today with competing technologies, you have a lot of players on the market. I believe you will see the same thing with algorithms." Different algorithms may cater to varying needs. For example, there could be an algorithm optimized for someone who is extremely physically active, another for pregnant women, and a third for teens who don't manage their diabetes carefully.

While the artificial pancreas is intended to completely take over insulin dosing and keep blood glucose in the target range, there are already some partial "half-loop" solutions becoming available. A Medtronic device on the market in Europe lets an algorithm have some control over an insulin pump, though it's not a true artificial pancreas. That device, the Paradigm Veo, combines a CGM, a pump, and an algorithm, which can shut off insulin for two hours if blood glucose dips below a threshold. Medtronic sells a similar system, the Paradigm Revel, in the United States—minus the shutoff feature. "The FDA won't let us activate that algorithm until we complete [a] study in the U.S.," says Francine Kaufman, MD, the chief medical officer at Medtronic Diabetes, adding that the trial is about to begin. A recent study by Doyle in Diabetes Care tested another shutoff algorithm that lessens insulin flow if low blood glucose is predicted. The algorithm prevented 84 percent of overnight episodes of hypoglycemia. These shutoff functions do not, technically, close the loop, but they would be necessary in any artificial pancreas to provide a layer of safety. Aaron Kowalski, PhD, who oversees the Artificial Pancreas Project at the nonprofit Juvenile Diabetes Research Foundation (JDRF), is eager to see shutoff technology approved in the United States. "It makes me crazy that insulin pumps today continue to pump insulin into somebody [who is] severely low."

To truly close the loop, a device must actually deliver insulin. A step in that direction, on the path to a fully automated system, would be an algorithm that keeps the user in a safe range, not too high and not too low. This so-called treat-to-range approach gives insulin only if a person's blood glucose level crosses over a particular threshold or if it's predicted to go too high. The user is still responsible for any treatment decisions within the safe range. Several of the algorithms currently in clinical trials, including those of Kovatchev and Hovorka, are treat-to-range technologies.

Indeed, making an artificial pancreas that can safely bring blood glucose under tight control is a formidable challenge. The ideal target is "right next to hypoglycemia," says Kowalski, "so there's very little buffer." To start with, better or multiple CGM sensors are probably necessary to avoid insulin dosing based on faulty readings. An additional degree of safety could be realized by creating an artificial pancreas that provides two hormones. These devices would offer a second hormone to counteract insulin, such as glucagon, and bring blood glucose levels up should they fall too low.

Meals—because their timing, size, content, and effect can be unpredictable—are a particular challenge for an artificial pancreas. Even the fastest insulins currently on the market are taken just before a meal to keep blood glucose from going too high after eating. An artificial pancreas has the disadvantage of not knowing when someone will eat; it bases insulin doses solely on blood glucose levels, which take time to increase after a meal. One way around this problem would be a hybrid artificial pancreas, which is also in development. The device would be fully automated save for the user needing to press an "I'm about to eat" button to increase insulin flow. Ultra-fast insulin, more timely glucose readings, or both, could make such a button unnecessary.

The likely outcome of these efforts will be a progressive rollout of artificial-pancreas versions, the first of which could arrive in less than five years, each with more capabilities than the last. Of course, there's always the possibility that a fully automated system will surprise everyone with quick and dazzling results. The FDA "encourages parallel investigation," says Zimliki. "Whatever comes first and has sufficient data, we will entertain."

What will it take to get an artificial pancreas approved for the market? "That's the million-dollar question," says Zimliki. One thing is certain: It will require a lot of research. Through each successive round of clinical trials, scientists are making life harder for the artificial pancreas, testing its limits and using its failures to make improvements, and incorporating real-world scenarios like exercise. In a recent study, Hovorka's artificial pancreas was able to handle the effects of a glass of wine on blood glucose better than the average adult with type 1 can.

After an artificial pancreas proves itself in the lab, the next step will be to test it in the field. The European Union has recently launched a four-year project called AP@Home. Hovorka, a member of the consortium running the European endeavor, says he hopes its goal of conducting studies at home will be reached later this year. The project will test devices that use existing components as well as develop new hardware that can measure blood glucose and deliver insulin at the same site. Home studies will be needed in the United States, too, and Zimliki says the FDA is working with companies and scientists to design them.

Other companies besides Medtronic Diabetes are embracing the artificial pancreas. In January, Animas Corp., a Johnson & Johnson company that makes the OneTouch Ping insulin pump, announced that it was partnering with JDRF to develop an artificial pancreas. Animas has since joined forces with researchers at UCSB to make their insulin-delivery devices compatible with the Artificial Pancreas System. From all indications, the artificial pancreas is getting closer to becoming a reality in the marketplace, not just on the lab bench.

Even with its potential to improve and save lives, the artificial pancreas isn't a cure for diabetes. A true cure would restore the body's own ability to produce enough insulin and perfectly regulate blood glucose levels. That absolute victory, though vigorously sought, is probably still far off. And while the development of an artificial pancreas is an extraordinary goal, it is one that, after a flurry of activity in recent years, seems suddenly, tantalizingly, within reach. Researchers are attacking the problem with a tangible sense of destiny, and the FDA is facilitating the work. The future of diabetes may rest in a small electronic tool that combines cutting-edge science with a few simple hopes. "An artificial pancreas is there all the time," says Kovatchev. "It can make decisions while a person sleeps." In a not too distant future, people with diabetes may finally be able to rest easy.



Take the Type 2
Diabetes Risk Test