A Model-Based Design Journey from Aerospace to an Artificial Pancreas System

[AUDIO LOGO]

It was early 2013 at Medtronic Diabetes, and I found myself sitting in the office of our senior director of software development. I'd only been at Medtronic for two years, and was trying to build a reputation. But on this day, the senior director was unhappy with me, and I was being reprimanded for it. She had discovered that I was quietly working behind the scenes with some of her team on a technology to automatically generate code directly from MATLAB and Simulink Design models for our advanced therapy control algorithms.

She was displeased for two reasons. First, she thought it was a waste of time to devote resources to a project that was doomed to fail. And second, her whole organization was devoted to creating software code, and people on her team were getting nervous about what automatic code generation would mean for their jobs. I argued back that code generation technology was maturing, and has been successful in other engineering domains. Also far from threatening the jobs of software developers, these new tools would allow them to work more efficiently to develop software much more quickly.

She was unmoved by my arguments and ordered me to stop working on code generation. So, like any good engineer, I kept working on it. Only now, a little more quietly. Fast forward a few months to the fall of that year, and Medtronic Diabetes found itself in a state of panic. We learned that some of our competitors were gaining a lead in the development of automated insulin delivery systems for treating type 1 diabetes. One in particular was scheduled for a first human trial of their closed-loop algorithm the very next summer at a diabetes camp led by a well known Stanford researcher.

The president of Medtronic Diabetes would not allow us to get beat in this race, and ordered that same senior director of software development to find a way to speed the development of our own algorithm. An emergency meeting was called and the senior director was met with cries of disbelief from her own staff. You see, we had just finished developing a much simpler algorithm and the process took years. Now she wanted to develop this next generation algorithm that was 10 times more complex in a matter of months. It wasn't going to happen.

At this point, I raised my hand and proposed an alternative approach. In a private follow up meeting, I informed her that I had kept working on that generation technology, and I already had a version of the embedded code of our next generation algorithm on my laptop. What I did not how to do, though, was how to integrate the code into our insulin pump system. But her team did. At this point, she asked me how confident I was that this would actually work.

On the inside, I was thinking, I'd give it a 50/50 chance. But I told her with as much confidence as I could muster that it was a slam dunk. She then assembled a small team of engineers to work with me very quietly to integrate the code. Within six weeks, the small team demonstrated a working prototype of our closed-loop system. And that summer, we tested it for the first time in humans in that diabetes camp alongside our competitor. The next two years was a whirlwind of accelerated development, clinical trials, and FDA review.

And in 2017, Medtronic Diabetes commercialized the world's first hybrid closed-loop system a full two years ahead of our nearest competitor. We were not only the first to market, but this system achieved a level of therapy that had been previously unimaginable and led to the American Diabetes Association to raise their standard for diabetes outcomes. Allow me to introduce myself. My name is Lou Lintereur, and I'm the chief engineer of automated insulin delivery systems at Medtronic diabetes.

My education is in aerospace engineering. I hold a bachelor's and master's degrees and aerospace engineering from Purdue and MIT. But above all that, I'm a family man. I've been exceedingly blessed with a beautiful wife of 25 years and six wonderful children. I'm in a bittersweet phase of my life where I only have one child still left at home, my sweet baby girl, Marie, who recently turned 18 years old. My family plays a central role not only in my life story, but in this particular story, as you'll see.

Following my graduate studies, I entered the workforce in the mid 1990s with a job at Hughes Space and communication. This company is now part of Boeing Satellites. It was here that I had my first exposure to what might now be referred to as an early form of model-based development. But in those days, we just called it development. We had a sophisticated simulation of satellite dynamics that was built using a language called Attsim.

Control system designers like myself develop the attitude and pointing control laws in this ADSIM simulation framework. And our software development teams were trained on how to interpret these ADSIM models to create the embedded code for the satellite flight computers. In 1998, I was recruited by NASA Dryden Flight Research Center, now known as the Armstrong Flight Research Center, at Edwards Air Force base in Southern California. This amazing job had me developing experimental flight control algorithms for advanced aerospace vehicles. My favorite and most successful program was the one you see pictured here, and it was a partnership with the Boeing Phantom Works to develop an unmanned combat air vehicle, also known as the UCAV X-45.

This proved to be a very successful technology demonstration program. And the X-45 is now displayed in the Smithsonian Air and Space Museum, which you see pictured in the upper right corner there. But this program also introduced me to true model-based development. You can see in this NASA tech brief that I authored back in 1999, how the Boeing team was using an engineering software package called Matrix-X, not only to simulate the flight characteristics of this autonomous vehicle, but also to automatically generate the embedded code for the flight control computers.

This approach dramatically sped the development of our very advanced flight control laws, and did so without compromising the quality of the embedded code. I really loved my work at NASA, and might have stayed there for my entire career. But in 1999, I also got married. And in 2000, we welcomed the first of our six children into the world. And in 2005, an event occurred at home while I was away at work that threatened the safety of my children. This event caused me to reevaluate a career that had me commuting so far away from home every day.

So I made the hard decision to choose to leave NASA and look for work closer to home. A neighbor of mine worked for a medical technology company called Advanced Bionics. And while they did not need engineers with my specific specialty, they did need engineers to do general work and quality and design assurance. It was not necessarily what I was looking for in terms of work, but it was very close to home. And this neighbor helped me gain a position at AB, an entry into the world of medical technology.

Advanced Bionics creates miraculous cochlear implant technology that uses neuro stimulation of the auditory nerve to restore the hearing of people with severe hearing loss. Children who are unable to hear and receive a cochlear implant can learn to interpret speech and be mainstreamed through school, and ultimately work without the need for any special accommodations. A little backstory on this technology. AB was founded by a brilliant entrepreneur engineer named Al Mann. He had a vision to create technologies to reduce the burden of cochlear implants. Excuse me. Of common ailments.

First, he looked at hearing loss and started Advanced Bionics to create cochlear implant technology. He also looked at blindness, and started a company called Second Sight, which uses similar neurostimulation technology to restore some vision. He also looked at diabetes and started a company called MiniMed with the goal of creating an artificial pancreas system as a prosthetic cure for type 1 diabetes. MiniMed was acquired by Medtronic in 2001. I was happy with my work at Advanced Bionics. But in 2011, I was contacted by a former colleague from AB who is now at Medtronic diabetes.

He recalled that I had a background in flight control system development, and informed me that Medtronic was looking for engineers with my specialty to help develop insulin control algorithms for an artificial pancreas system. They were having a difficult time filling the role, though, because biomedical engineers did not typically design feedback control systems. But he knew that I had this experience, and he encouraged me to apply. I viewed this, of course, as an amazing opportunity to apply my skill set in control design to a biomedical application and still be close to home. I landed at Medtronic in August of 2011.

A little primer on diabetes. Diabetes is a metabolic disorder in which the body either does not produce insulin, which is called type 1 diabetes, or does not properly use insulin, or type 2 diabetes, which leads to uncontrolled blood sugar. Insulin is a hormone produced by the pancreas that the body uses to convert the blood sugar into energy needed to sustain life. In effect, glucose in the blood needs to be transported into cells. And you can think of insulin as the key that is needed to open the cells to receive glucose for energy.

An automated insulin delivery system is a way to automatically deliver insulin into the body of a person with type 1 diabetes who does not produce insulin on their own. Before automated insulin delivery systems, like the one pictured here, people with diabetes would need to stick their fingers several times a day to get a drop of blood and check their blood sugar. And then estimate how much insulin to take, which would then need to be injected manually using a syringe. Many people still manage diabetes this way, but automated insulin delivery systems provide a better way to manage blood glucose levels much more effectively.

It starts with a blood glucose sensor that periodically measures glucose just under the skin. A transmitter connected to the sensor then transmits the glucose measurement to an insulin pump. Control software and the insulin pump uses hybrid closed-loop technology to calculate the best dose of insulin to deliver to bring the person's blood glucose back to a healthy level. And then the insulin from the pump is then delivered to the body through an infusion set. And this loop is then repeated every few minutes to control blood sugar.

Over time, the therapy might look like this. You see here a graph of rising and falling glucose over time. And the vertical bars represent the insulin dose adjustments that are made approximately every five minutes by the hybrid closed-loop algorithm. The goal of Medtronic's technology is to treat glucose to a specific target as low as 100 milligrams per deciliter, which is close to normal fasting glucose. The therapy itself is adapted to the person's individual insulin needs.

Every person has different needs depending on their age, body weight, diet, level of activity, medications, other health conditions, et cetera, et cetera. And insulin needs can do, can and do change over time. So the algorithm must constantly adapt itself to these changing needs. Finally, you've heard me describe these algorithms as hybrid closed-loop. The term hybrid means that the person still must participate in the therapy. They do this mainly by estimating the insulin needs for meals and delivering this bolus of insulin manually using the pump.

However, people can often forget to bolus for a meal and do not bolus enough for insulin. In these cases, the Medtronic system can automatically deliver correction boluses to help compensate for these forgotten boluses. Back in 2013, a race had started to develop the first commercialized automated insulin delivery system. As I related at the beginning, it was Medtronic's use of model based development techniques that got us across the line first. But over time, several other technologies have emerged.

Medtronic launched the world's first hybrid closed-loop system in 2017. Since that time, Tandem Diabetes launched their control IQ system. This was followed by Insulet and their Omnipod 5 patch pump, which is a tubeless pump design. More recently, Beta Bionics launched their system, the iLet system, which targets an even greater ease of use. And more are coming. So the pressure to remain the leader in automated insulin delivery is fierce.

In Medtronic's journey to full model-based development, overcoming resistance to code generation was one of two hurdles we needed to overcome. The second hurdle was gaining confidence in simulation. In my aerospace days, we would never dream of designing a flight control system without extensive use of high-fidelity simulation. This was not so in medical technology. Engineers in this domain were accustomed to developing algorithms using best guesses based on clinical practice, and then iterating through clinical studies.

When our closed-loop development team started advocating for design techniques that relied more heavily on simulation, we were met with much skepticism and resistance. I recall one leader challenging me that I did not understand the risks involved. While simulation may work in aerospace applications, we were now working in medical technology, and the risks in medical technology are very different. At this point, I thought to myself, in my aerospace days, we use flight simulation to design flight control systems for a killer robot airplane that was designed to carry out surgical strikes in populated areas.

This leader was correct after all. The risks are very different in medical technology. Needless to say, we were ultimately successful in developing what, in my estimation, is a best in class diabetes simulator that contains thousands of virtual patient models derived from real human data. This simulation was used heavily in the development of Medtronic's latest 780G automated insulin delivery system, and now in our next generation system development.

Medtronic's model-based development journey started with some very basic simulation tools that were used to develop glucose prediction and insulin suspend algorithms in our previous 640G product. Later, we added automatic code generation for the entire software development of our first hybrid closed-loop system, the 670G. Next, we tightly coupled our advanced simulation platform with our code generation process, and used full model-based development in our latest 780G system. And now we are working on fully validating our simulation platform so it can be used to support formal regulatory submissions, and possibly even reduce the scope of clinical studies.

Medtronic is a mission-driven company. Our mission is to contribute to human welfare by the application of biomedical engineering to alleviate pain, restore health, and extend life. We at Medtronic take this mission very seriously and apply it in everything we do to create technology that can be considered life changing. Here is an image of a biostater starter device that was developed in the 1970s as a first attempt and an automated insulin delivery system. Here is another image of a girl with diabetes who is using this biostater system. Very few of us would call this life-changing technology.

A couple of months ago, I was approached by a woman with diabetes at an event who wanted to personally thank Me for the technology that Medtronic has created for her. She described it as life changing. And on the right, you can see a photo of that woman who was once that bedridden little girl using the biostater. At a different event last year, I was approached by this woman who also wanted to thank me for the incredible work that Medtronic is doing. She contracted diabetes as a young girl, and at that time, her doctor told her that if she wanted to live a long life, she could not eat cake and other sweets.

She since developed a fear of sweets, and then told me that in her whole life she had never eaten a piece of cake. My mind at that time immediately went to all the birthday parties, other celebrations, and so forth that she could not fully participate in growing up. And then she told me something else. She told me that she now feels so confident with the therapy that she gets from the Medtronic 780G system that on this very evening she would have her first piece of cake. And I watched with delight as she ate that cake.

That right there is life changing. This is the type of technology that is being developed at Medtronic, and that would not be possible without the amazing tools like MATLAB and Simulink, and the support from the MathWorks team. Thank you.

[AUDIO LOGO]