Children are the most delicate and unpredictable of all patients. As charming as their precociousness is, the unique challenges of treating pediatric diabetes are making it difficult for researchers attempting to create an artificial pancreas for this endearing population.
Twenty years ago, Frank Doyle, MS, PhD, wrote his first research proposal as a young professor for a device that could automate insulin delivery. Now known widely as the “artificial pancreas,” researchers around the world are working to refine this technology and bring various iterations to market. But a large problem remains: The current models do not work well for kids.
Teenagers with diabetes provide one subset of challenges. Their bodies demand higher doses of insulin, which leads to greater risk of low blood sugar hours after meals. Young children face entirely different obstacles.
“Young children are quite insulin sensitive, so very small changes in insulin delivery can have a profound effect on the glucose levels,” explains Stuart Weinzimer, MD, associate professor of pediatrics and lead investigator of the Artificial Pancreas Program at Yale University, New Haven, Conn.
“And, behaviorally, young children are quite unpredictable,” he went on. One day may be spent running around a playground, and the next day may include hours of naps and cartoons. This variability in physical activity has tremendous influence on insulin needs, but diet is also a concern. “Young children are notoriously picky eaters. You might give them their insulin and then they decide that they’re not going to eat.”
Parents of children with diabetes often find themselves in this predicament: struggling to keep their son or daughter’s insulin and blood sugar levels in balance and worrying endlessly about a life-threatening dip or spike. “This is something that can’t be stressed enough,” Weinzimer emphasizes. “Parents sacrifice their lives to this disease. They live in constant fear — constant fear — of very severe hypoglycemic reactions. And if they’re successful in avoiding those, they live in fear of longterm complications.”
Mothers and fathers trying to protect their child from blood sugar lows often then see glucose levels getting too high. If they try to avoid highs, then they run the risk of their child having a seizure. As a result, parents end up stressed, burnt out, and depressed. Weinzimer has witnessed time and time again the devastating toll that diabetes can take not just on children but also on their families. “Parents really do have to be constantly vigilant and hovering over their kids to protect them, and it’s hard to have a normal life when you’re doing that,” he says.
Bespoke Mathematical Suites
In 2006, Eyal Dassau, PhD, joined Doyle’s group in the quest for the artificial pancreas, at a time when the team had begun clinical testing of the control design. In the intervening years, literally hundreds of adults have tested a version of the system operating on various forms of the software developed at the University of California, Santa Barbara (UCSB).
With the help of a $1.8-million grant from the National Institutes of Health (NIH), Dassau — the principal investigator — is working with Doyle and Weinzimer to design a closed-loop system for young children that can offer families some relief. They hope that the pediatric artificial pancreas will allow parents to sleep through the night without worrying about blood sugar dips and even remotely monitor their child’s glucose levels from a mobile device.
Until this study, Dassau and Doyle’s research at UCSB has focused on an artificial pancreas system for adults with type 1 diabetes. They build algorithms that tell the device how much and when to provide insulin.
Doyle claims that the biggest challenge in building the artificial pancreas is uncertainty. “If we knew how the body behaved very precisely, and it was very reproducible hour-after-hour, day-after-day, this would not be a very hard problem,” he explains.
From an engineering perspective, this holds true for any medical control system. Our bodies frequently change, as do our diet and exercise regimens — making insulin demand a moving target. Because of these unpredictable variables, designing an algorithm that will work in all conditions and all people is next to impossible.
So, the UCSB team takes a different approach. Instead of a one-size-fits-all solution, they customize the algorithm to each individual. “We collect data, we inform the algorithm from their medical record, and that is used to initialize the first version of the algorithm for the subject,” Doyle says.
The studies in adults have shown that the algorithm can learn over time, gaining insight into each patient’s habits and activities. The artificial pancreas further adapts to the individual’s needs as a result. Dassau, Doyle, and Weinzimer plan to take the same tailored approach in their NIH project with children. Their team is adjusting the algorithm to each pediatric participant, along with incorporating other effective techniques from the adult system.
A key part of their algorithm is zone model predictive control. Unlike most mathematical problems, there is no single numeric answer for the correct blood sugar level. The artificial pancreas needs to keep glucose within a certain range rather than trying to reach a specific number. The ongoing stream of information from the glucose sensor made this a challenge. “Glucose sensors are noisy, and that noise will drive a traditional algorithm to keep adjusting the pump,” Doyle explains.
“Parents sacrifice their lives to this disease. They live in constant fear — constant fear — of very severe hypoglycemic reactions. And if they’re successful in avoiding those, they live in fear of long-term complications.” — Stuart Weinzimer, MD, associate professor, pediatrics; lead investigator, Artificial Pancreas Program, Yale University, New Haven, Conn.
Zone model predictive control combines two notions of engineering to allow the artificial pancreas to distribute the correct amount of insulin without becoming overactive. The first portion, called model predictive control (MPC), manages complex processes and has been used for a variety of tasks for decades, including oil refinement and chemical processing. Doyle and his colleagues developed the zone innovation, which allows it to keep glucose within a range rather than aiming for a target number. “The blood sugars can roam in that range, and the algorithm won’t fight to try to correct,” he continues.
The models of the artificial pancreas that are closest to reaching the market cannot entirely automate blood sugar management. Rather, they offer a hybrid system where patients still input insulin amounts for meals but can rest easy between meal times and at night.
When glucose does start to fall out of line, the device lets the patient know. Th e researchers are particularly excited to build a wireless tracking system for parents to receive these alerts, even when their child is at school or elsewhere.
“We developed a monitoring system, named E911, that can alert patients and their families on pending hypoglycemia with a GPS location of the event and a link to a Google map,” Dassau says. The warning would be sent to a mobile phone or other remote device.
Weinzimer believes that this is a crucial element of the pediatric artificial pancreas — worth sacrificing a bit of battery life for. “As a pediatric endocrinologist, I’d rather us have remote monitoring capabilities and change the batteries more frequently,” he says.
The scientists are experimenting with different types of technology, such as low-energy Bluetooth, to find a balance between functions and battery duration. They are, for example, trying to find the optimal communication frequency between the computer and the pump. The more often information is transmitted, the more energy the device needs.
Weinzimer has posed a number of additional questions that need answers. “What if you lose connection between your pump and your sensor? Do you have the system suspend, or do you have it fall back into a preset delivery pattern,” he ponders. “How many transmissions do you need to miss before you get kicked out of closedloop and fall back into a manual safety condition?” All of these details still need to be worked out.
Version 2.0 and Beyond
Both Doyle and Weinzimer emphasize that the artificial pancreas is a “staged solution.” Within about three years, artificial pancreas systems for adults will likely become available in the U.S., followed a few years later with systems for young children and then teenagers. It will continue to evolve from there.
“These systems are not going to take away all the uncertainty, but if they allow for a good night’s sleep, it would be transformative in the lives of some parents,” Weinzimer says.
Doyle imagines a fully automated and invisible technology when he proposed the notion of an artificial pancreas two decades ago — something that could perhaps be implanted rather than external. Once in place, it would allow people with diabetes to live life without ever worrying over their blood sugar levels or injecting insulin. Such a device may someday exist, but for now he and his colleagues are excited to make advances toward a pediatric system. “It is really gratifying to see how far we’ve come,” Doyle says.
— Mapes is a Washington D.C.–based freelance writer
and a regular contributor to Endocrine News. She wrote
about advanced business degrees in the April issue.