Thank you, Dr. Ross. It is an honor to have Dr. Steven Russell here with us today. He is an Associate Professor at Harvard Medical School. And he is a practicing endocrinologist at Mass General. 

Over the years, Dr. Russell has been a world expert in continuous glucose monitoring and has been a pioneer in the development of the bionic pancreas. So Dr. Russell has led a collaboration between at Mass General as well as the Department of Biomedical Engineering at Boston University, where he has with this group developed a validated bionic pancreas, which automatically delivers both insulin as well as micro doses of glucagon, which is pretty cool, in order to be able to regulate blood glucose levels. 

Their research has progressed in recent years. And it's become one of the first in the world to complete an outpatient trial with automated glucose monitoring during both the day and the nighttime. And this is allowed several more outpatient and multi-center home trials and studies to go on. 

We're particularly excited to have Dr. Russell here because this morning he was able to spend some time with our endocrinology department, specifically our diabetes technology team here at UVA. And we are honored to have Dr. Russell here for medicine grand rounds to walk us through the complex endeavor of developing a bionic pancreas and kind of helping us to understand the recent progress and next steps in this field. So without further ado, please help me in welcoming Dr. Russell. 

[APPLAUSE] 

Thank you very much. It's a pleasure to be here. I noticed how pretty this area was when I was flying in to the Charlottesville airport. And the weather was great so I was able to go running last night after I got in. So it's nice to be in someplace a little bit warmer than Boston right now. 

So I'm going to start with disclosures. Obviously, we have to work with a lot of medical device companies to do projects like this. But I'm going to particularly point out that there has now been a company that's been founded to commercialize this technology called Beta Bionics. And I have not joined that company, because that would have prevented me for conflict of interest reasons from continuing to do the research that I do. So I've stayed in the academic realm, but continue to collaborate quite closely with the founders of that company who are actually also academics themselves, Ed Damiano and Faris El-Khatib were at Boston Biomedical Engineering. And then it became clear that in order to bring this to as many patients as possible, they would need to found a company. 

And recently, they just completed a fundraising round. So they raised $50 million in a series B, which they think should get them all the way to approval of the product and start a commercialization. So that was pretty exciting for something that started off as just an academic collaboration. 

So the reason we're doing all of this is because glucose control in patients with type 1 is a really difficult task. They have to maintain average glucose near the normal range in order to avoid complications. An A1C of 7% is the goal for most patients. And that means an average glucose less than 154 milligrams per deciliter. 

And it turns out that that's really tricky to achieve, though, because insulin is absorbed relatively slowly. They don't have normal counter regulatory function. And so as they try and tighten up their glycemic control, they put themselves at a very high risk of hypoglycemia. And those hypoglycemic events can really be scary, unpleasant, and even can lead to accidents and deaths. 

So as a consequence of that, most people are not able to actually meet goals for therapy. So if you look at population-based study of people at specialized diabetes centers-- and I'm not even talking about now people with type 1 who aren't managed at academic centers-- of those people managed at academic centers, only 30% of them are able to meet a goal for therapy with an A1C less than 7%. 

And for adolescents and children, the number is even lower. About 15% of them are able to meet goals for therapy. So this means that they're at risk for complications in the long run that can dramatically, negatively influence their quality of life and also shorten their life and reduce their function in society. 

So the way we ask people to manage diabetes right now really puts it-- it's almost unique in medicine. We give them this drug insulin, which has the highest rate of drug complications of any drug other than chemotherapy, because it has such a narrow therapeutic range, and we give them this drug and we have them determine the dosing multiple times a day. So we ask them to measure their blood glucose. Some of them now have access to this CGM technology. 

I apparently don't know how to use it. Oh, let's see, yeah-- so some of them have access to continuous glucose monitoring technology, which we'll talk about in a little bit. But most of them still check their finger sticks, and they calculate the number of carbs in food, which is actually tricky to get right. And then they have to use all that information to calculate doses of insulin. And that also can be challenging for them. 

Plus, even if they do everything right, there's so many unmeasured factors, like their level of stress, how much sleep they've gotten, for a woman where they are in the menstrual cycle, all those things affect insulin sensitivity. So you can do the same calculation for the same meal and get a completely different results on two different occasions, which is frustrating for patients. 

So they then deliver the dose. And for the reasons I've mentioned, the response can be unpredictable. And what we usually tell patients is they're going to have to go through that cycle probably 10 times a day in order to get decent blood glucose control. That's what studies say. If you want to get an A1C of 7%, it usually takes 10 blood glucose checks a day. 

So what we're trying to do is take advantage of the same technology that I mentioned earlier, this continuous glucose monitoring technology, which provides glucose measurements every five minutes, 288 times a day. That's more information than people are really capable of processing and doing anything with. But it's exactly the kind of information that computers are good at processing. 

So the idea is to take all that information and then delegate the dosing decisions to computer algorithms, which will then decide on the doses, automatically deliver those doses by controlling insulin pumps, and then hopefully, if you have an adaptive algorithm, you'll do a good job of making good decisions about that dosing. But we know that things can change. You decide to make an insulin dosing decision. And then it takes a relatively long time for insulin to be absorbed. So things could have changed by the time the insulin that you gave an hour ago is finally absorbed and is reaching its peak effect. 

But the good thing is is we can invest less in each of those decisions if we're making 288 decisions a day, instead of 10 decisions today. So we don't have to give all the insulin for a meal right up front. We can give some of it and then fill in a little bit more if it looks like we really need it. 

Now, I'm going to tell you mostly about our efforts in this area. But we are certainly not the only people doing this work. And, in fact, people at the local Diabetes Technology Center have done a lot of this work. And an algorithm that they developed has been licensed to type 0 and then onward to tandem diabetes care. And it's currently in a pivotal trial. So an insulin only configuration of an artificial pancreas using that algorithm, we hope will be available to patients next year. 

And, of course, Medtronic, I'll tell you, has developed early generation artificial pancreas that's now available. And they've licensed some additional technology from academics for their next generation of artificial pancreas. 

These are both insulin only systems. And there's multiple other companies. Bigfoot Biomedical, Lilly now has made an investment in this to develop these insulin only systems. There's a smaller number of academic groups mainly, but also some companies, working on bihormonal systems that deliver both insulin and glucagon. 

And the reason that makes sense-- I'm just going to skip this for a moment-- the reason that makes sense is that the normal regulation of glucose by the pancreas doesn't rely on a single hormone. It actually relies on multiple hormones, but the two most important ones are insulin, which lowers blood glucose, and glucagon, which raises it, counters the effect of insulin. 

And we all know that in type 1 diabetes, the beta cells are destroyed by an autoimmune attack, and they can't produce insulin anymore. What is less often talked about is the fact that in the absence of those beta cells, the regulation of the glucagon is also inefficient. It doesn't work properly to counter hypoglycemia anymore. And so you lose this ability for the glucagon to be released in the presence of hypoglycemia. 

Normally, what it would do is it would act on the liver to break down glycogen that's stored in the liver. That glucose is then released into the bloodstream. And you can support the blood glucose. And typically, this happens in normal function in the late period after a meal as the blood glucose is coming back down to the normal range and also during exercise. And those functions are lost in people with type 1 diabetes. So that's why we think that a bihormonal system is important. 

And another example of that, as I mentioned earlier, that the insulin is absorbed relatively slowly, it takes about an hour or more for insulin dose given subcutaneously to reach its peak levels in the blood. So if you have done very well by balancing your insulin delivery with your insulin need and you have a relatively flat blood glucose profile, that's great. You're giving, let's say, a basal dose of insulin. But now you start to exercise. Even moderate exercise can lead a lot of glucose to be taken up into muscle in an insulin independent way. And so you can get a pretty rapid fall in blood glucose. 2 milligrams per deciliter per minute or 3 milligrams per deciliter per minute is not uncommon at all. 

And so that means that, let's say, even you were at 100, you could reach that hypoglycemic threshold in 15 minutes from the time you start exercising. Now, if we stop delivering insulin at the time we see that blood glucose falling, that's going to have virtually no effect on insulin levels 15 minutes later, because it takes an hour. We've already delivered insulin an hour ago that's just peaking now. And insulin 15 minutes ago that's going to deliver-- or 45 minutes ago that's going to peak in 15 minutes from now. So stopping insulin delivery can be helpful to reduce the severity of that hypoglycemic event. But it's not going to be sufficient to totally prevent that hypoglycemic event. 

On the other hand, glucagon is absorbed quite rapidly from the subcutaneous space. The levels come up to detectable in less than 5 minutes. And it peaks in 15 to 20 minutes. So it's much, much faster than insulin. And so that allows you to support the blood glucose with little tiny doses of glucagon until the insulin levels drop, which can take some time, obviously, after you stop delivering the insulin. 

So that's why we use a bihormonal system. But currently the only available artificial pancreas system right now is the 670G, which is from Medtronic. This, like most of them in development, is an insulin only system. It uses only insulin. And so it tries to suspend insulin delivery when it predicts that the blood glucose is going to go into the hypoglycemic range. 

It also has the capability of increasing basal insulin levels if the blood glucose is rising. But it has a fairly limited range in which you can do that. And if the blood glucose is rising and it calculates that the amount of increased basal that it can give isn't enough, it will alarm. And it will ask the user to check a finger stick. And that will enter into the device and that will calculate a correction. 

But that combined with the fact that it's approved for use with four daily calibrations means that on average we find people are checking fingers sticks 8 to 10 times a day with this system. Well, I mean, that's how many times you're supposed to check if you don't even have an artificial pancreas system. And a lot of people find that it's actually more work to use this than it was to use their previous whatever they were doing. 

In many cases, they were using as a CGM device that doesn't require as frequent calibrations. And there's been a significant amount of frustration with some limitations that have been put on this. For instance, you can't control your target glucose at all. That's something that's set for you by the company and you can't change it. 

So the other thing that's frustrating for me as an academic is that they got this approved without doing a randomized controlled trial with a control group. All they did was they showed that it was safe. So they did a relatively short study. They compared the blood glucose with the artificial pancreas feature turned on to a run-in period. And it turns out there was no actual difference in the mean glucose between the run-in period without the automated features and when they turned the automated features on. Glucose was the same. 

There was a modest reduction in hypoglycemia when they turned those automated features on. But we don't know if that was a study effect or it was actually due to those features. So they're now, after approval, running a randomized controlled study. And so we'll finally understand what the actual benefits of this are. But I would say that this is an early attempt at this sort of technology, and better will be coming along. 

So the system that we've worked on, as I mentioned, is a bihormonal system. But we've also developed an insulin only system because it's going to take some time, as I mentioned later, to get a stable glucagon approved. That's another barrier that we deal with. 

But one of the key things that we focus with both the insulin only in the bihormonal system is we wanted this to be very, very easy for both the provider to start and the user to use, because right now we've got about 30% of the population of type 1 diabetes patients who are using diabetes technology, like pumps and CGMs. The use of pumps has been relatively flat over a long period of time. We don't seem to be able to break out of that 30% number. 

CGM use is going up more rapidly than that. And. That's a good thing. But we're concerned that if we don't do something to make it a lot easier to use, essentially we'll keep making the glycemic control better for the people who already have relatively good control and not doing anything for the 70% of people who are doing relatively poorly. 

So we wanted it to be very easy to start, so there would be no barriers for the doctor. You don't have to do complicated optimization protocols before you put someone on the bionic pancreas. We set it up so that the only information it needs to start is the body weight, which is just used for initial dose scaling. And then it automatically adapts from that point onward. And it continues to adapt over time. For instance, if the person gained weight or lost weight, or if they were a child and they went into adolescence and then they came out of adolescence, all those things dramatically change insulin requirements. So we set up the system to automatically adapt to that over time. 

And also by making it very easy to use, we're hoping that it can actually be started in areas where there's no endocrinologists available, that it can be started by primary care practitioners out there who wouldn't feel comfortable starting someone on an insulin pump and a CGM and playing with basal rates and carb ratios and correction factors and all that. There's no numbers here. This is just diabetes without numbers. The system automatically turns on and starts and runs. 

It doesn't even require announcement of meals, although we do suggest that people do it. But when they announce a meal, they're not announcing a meal by counting the carbs, because we know people aren't very good at that. Instead, it's a qualitative meal announcement. So they just say whether the meal is at the beginning, middle, or end of their day-- breakfast, lunch, dinner-- and how large it is to their typical meal at that time of the day. So it's a typical meal, greater than typical, less than typical. 

The system then learns what they mean by that over time and adapts a meal priming bolus to account for 75% of the predicted insulin requirement for that meal. And then it fills any additional insulin in automatically. But if you forget to do that, it will still regulate that meal well. It will just have a higher postprandial hypoglycemia after that meal because of the slow absorption of insulin. 

So how does this work in practice? This is one of our experiments from very early on. This is before we had the meal announcement-- or the meal priming. So this is totally automated glucose control. And this is a person with type 1 diabetes with no residual C peptide. So they don't make any of their own insulin. 

And so they come in. They are in pretty good control at the time they come in. But they're dropping a little bit. And the system gives them tiny little bits of glucagon. 

So on this scale, this is 10 micrograms of glucagon here. This is 5 micrograms of glucagon. So you can see very tiny doses. A rescue dose of glucagon as 1 milligram, 1,000 micrograms. So we're giving very tiny fractions of that. 

And you can see the blood glucose levels off. Then they come along, and they eat a meal here. This meal, it happens to be pizza, 108 grams of carbohydrates, a big meal. Blood glucose starts rocketing up after they eat the meal. And the system automatically starts dosing insulin. That's what these blue strikes downward are. And you can see the blood glucose turns round, starts coming back down. 

Now, this was early on. We hadn't really optimized our insulin dosing regimen. You can see it looks like we're going to go hypoglycemic. 

But the system then gives glucagon. And this is 15 micrograms of glucagon, 1.5% of a rescue dose, so tiny amount. You can see almost immediately there's this change in the trajectory. And with a few little additional doses of glucagon, we prevent a hypoglycemic event. At the bottom of this green zone is 70 milligrams per deciliter. We overshoot up a little bit to about 120. And then things settle down and we're in this target range of 70 to 120 all night long. And then we give another-- you know, there's a breakfast meal and the process starts over. So it's just automated dosing interplay between insulin and glucagon regulated by these control algorithms. 

So we've now tested this system in a lot of studies. We've done a lot of outpatient studies. And we've now tested it in people from the age of 6 to now 80. That's not quite updated. Body weights of 21 to 128 kilograms, including someone with a BMI of 45, and total daily insulin doses from 11 to 145. And these were all exactly the same algorithm. No difference between the algorithm that was used for a six-year-old to an 80-year-old or somebody who weighed 21 kilos to 128 kilos. All we put in was their body mass. 

And we've tested both the bihormonal and insulin-only configuration. I'll tell you a little bit more about those data. We've done four studies with the insulin-only configuration in type 1-- sorry 2 in type 1 and 1 in type 2. And we recently did a small pilot study in people with cystic fibrosis-related diabetes. About 30% of people with cystic fibrosis wind up developing diabetes because of damage to their islets in the pancreas. 

And so I'm going to show you detailed data from one of these studies that we published in The Lancet in 2016. And to orient you to this graph, on the left side, this was a random order crossover trial. It had 39 people in the study. They did 11 days of their own care, whatever that was. It was all insulin pump users in this trial. But some of them used continuous glucose monitoring, some didn't. 

And you can see that there is a there's a dramatic range of mean glucose. There's all the way from over 240, down to just above 90 in these folks. The size of the circle is proportional to the amount of time they're spending hypoglycemic. So this would be 5% of the time, less than 60. 

And you can see they do pretty well. This line here is the average glucose that would be consistent with an A1C of 7%. And our study population here is doing a lot better than the average of people with type 1 diabetes, not surprisingly, because they're people who volunteer for studies, and they all use insulin pumps. 

So they have an average glucose of 162, which is pretty good. That would be a predicted A1C of 7.3%. And 41% of them would be predicted to have an A1C of less than 7%. 

The amount of hypoglycemia they have, the time less than 60, is also pretty good 1.9%. On the other hand, on the bionic pancreas, you can see that the range of mean glucose is much, much narrower. And that's due to the adaptive nature of the bionic pancreas. If the main glucose is too high on one day, the system will get more aggressive with insulin dosing. If you're having too much hypoglycemia, it will get less aggressive with insulin dosing. 

And so you can see that we lowered the glucose down to 141 here. That would be a predicted A1C of 6.5%. And at the same time, we've cut hypoglycemia to 1/3 of what it was, from 1.9% of the time down to 0.6% of the time. And now instead of 41% of people predicted to have an agency of less than 7%. It's 92% of people would be predicted to have an A1C of less than 7%. 

The amount of glucagon we used in this study was about 1/2 a milligram a day, or about 8 micrograms per kilo per day. Unfortunately, we don't know exactly what would be a normal production of glucagon under these circumstances. But we think there's probably not much more than endogenous production would have been had the body been capable of making glucagon in response to hypoglycemia. 

We told people that meal announcements were optional. But we encouraged them to do them. And interestingly, the average number of meal announcements a day was 2.6. But some people announced 1.3 meals on average. And we looked and we found that they had just as good glycemic control as the people who announced more meals. So that's consistent with the idea that even if you don't announce the meals you can get good glycemic control. 

We've also played a little bit with the target. That study was using our most aggressive target. And so we wanted to know what would happen if we used a less aggressive target. This is the glucose that the bionic pancreas is trying to bring the blood glucose back to if it deviates from it. So if you go high, it tries to bring it down to that number. If you go low, it tries to bring it back up to that number. 

And so that study I was showing you was using this target of 100 milligrams per deciliter. Again, about 90% of people would be predicted to have an A1C less than 7%. As we raise that target to, let's say, 115, now we're down to 80%. And if we raise it to 130, we're at 35% of people would be getting to target. And the average glucose goes up. 

What's interesting is that the hypoglycemia doesn't really change. We have very low levels of hypoglycemia regardless of the target. And in every case, they're better than usual care here, which has about 1.4% of time less than 60 in this study. And only 35% of people predicted to have an A1C less than 7%. 

So what is the cost associated with going to that more aggressive target? It doesn't seem to increase the amount of hypoglycemia. It turns out that the cost is you have to use more glucagon. So this is the glucagon utilization. And you see I told you we had 8 micrograms per kilo approximately at that most aggressive target. As we go to less and less aggressive targets, we use less and less glucagon. 

But it turns out this amount of glucagon is really well tolerated. We did a randomized controlled, double blinded study, where we had people use either glucagon or placebo in addition in the bionic pancreas. And there was no difference in any of the adverse events between the glucagon and placebo arms of the study, even though they used this amount of glucagon in the glucagon days. And in fact, we asked them every day-- they were randomized every day whether they would get glucagon or placebo-- and we ask them everyday, do they think they got glucagon or placebo? And they were right 42% of the time. In other words, they had no idea whether they were getting glucagon or placebo, which, again, suggests that is pretty well tolerated. 

Now what was surprising about that is we actually reduced hypoglycemia by 92% in that study. How could they not notice that? The difference was that these were people who had hypoglycemia unawareness. So they didn't know that they were hypoglycemic all the time. And they didn't know that they weren't hypoglycemic when we prevented it. But that was reassuring to us that the glucagon was well tolerated. 

Now, we've also, I said, tested an insulin-only system. And this is a study where we compared usual care, the bihormonal system that I was describing before, and an insulin-only system. And you can see that the control with the insulin-only system is definitely better than it was with usual care. That mean glucose goes from 165 to 148 with the insulin only system. It's still 10 milligrams per deciliter higher than the bihormonal system, though. And the amount of hypoglycemia, in this particular study at least-- this is time less than 54-- was 0.6% of the time in usual care, 0.6% of time in insulin-only. So we lowered the mean without increasing hypoglycemia. But we didn't decrease hypoglycemia. 

But with the bihormonal system, we're able to lower the mean even more and cut hypoglycemia to 1/3. So what the bihormonal system does, what that glucagon allows us to do is be more aggressive with insulin dosing, get a lower mean glucose, and not only not increase hypoglycemia, further decrease hypoglycemia down to very low levels. 

And when we look at user preferences, users prefer the bihormonal system to the insulin-only system, regardless of the target. So you see they have different targets along the bottom here. And these are the user satisfaction scores on a scale of 0 to 5. And so they have the highest satisfaction with the bihormonal system. Doesn't vary that much with the target, although they like the lower targets better, probably because they get better blood glucose control. 

With the insulin-only system, the satisfaction is still pretty high. But there is a clear peak around 120. And we think that makes sense to us, because that turns out to be probably the best balance between the best mean glucose, but also still having an acceptable amount of hypoglycemia with the insulin-only system. 

Now, I just want to show this. Those were individual studies I showed you. This is a meta analysis that we did where we just combined all our data. And I also compare it to the general population. 

So if you look at data from that longitudinal and cross-sectional study that I described before, you can see that this is the average glucose for people in the young adult, 18 to 25. It's over 200 milligrams per deciliter. And you could see that would mean only 18% would achieve goal for therapy. It's a little better if you look at the 26 to 49-year-old population. Now, you get a mean glucose of 174, 29% achieving goal. It's very similar in the older adult population. 

But in the usual care arms of our studies, they do better than that. And that's probably just a study effect. It's, you know, you're recruiting people who are enthusiastic about the studies. They're all pumpers. They do better. 

But the insulin-only bionic pancreas does even better than that. So it takes about 42% achieving goal to 48% achieving goal and about halving of the amount of hypoglycemia overall. But then when we go to bihormonal system, you can see that we reduce the mean glucose even more. Now, we're at 87% achieving goal for therapy, and hypoglycemia is reduced even further. 

So the insulin-only systems are good. They're better than usual care. But bihormonal system provides additional benefit beyond what you can get with the insulin-only. 

And this is the same sort of data for kids that have a slightly different target here. This is the higher target of an A1C of 7.5%. And you can see that under usual care, they have very high mean glucoses. 

In a camp setting, where we're all over them and we're adjusting their insulin regimen every day, they do much, much better than that. And so they have a mean of 162, 2.4% of hypoglycemia. But with the bionic pancreas, we get it down to a mean all the way down to 140 with 1.3% of time less than 60. And again, this is exact same algorithm that's used in the adults. And you can see the results are very similar to what we get in adults. 

And recently, we completed this little pilot trial in people with cystic fibrosis-related diabetes. And, again, we have our control group, our insulin only, and our bihormonal. This is just three patients. And each of those lines is one of the patients. 

And you can see that two of them do better on insulin only than usual care, and even better with bihormonal. One of them does great in usual care insulin only and bihormonal. And they have very little hypoglycemia. 

So it may be that in this particular group of people, we don't actually have to use the bihormonal system. They seem to do great on insulin only and may not be a reason to add the glucagon. And we have a larger study that we've applied for an R1 grant. We got a great score. So I think we're going to get it funded to test that and do a long-term study in cystic fibrosis related diabetes as well. 

So there are some challenges to this. One is stable glucagon. Currently, glucagon is unstable. So it's provided as a lyophilized pellet that you have to reconstitute before use. And it's only stable for about a day. So we've been having patients reconstituted every day. That's not really very practical for the long run. 

And so we knew that if we justified an interest, that companies would come out with a stable glucagon for us. And, in fact, they have. So a pharmaceutical company in Denmark called Zealand has produced a glucagon analog, which has 7 amino acid changes. It's stable for more than a month at 40 degrees with agitation. So it's much more stable than insulin. And so it has a specific activity comparable to glucagon. Its side effect profile is identical to glucagon. 

And so we have tested it in the bionic pancreas. And we compared it to freshly reconstituted glucagon that we use now. And we set up a very challenging regimen, where we gave extra insulin with the bionic pancreas didn't know about. And we made them exercise. And we did everything we could do to force hypoglycemia. And that would mean that we give plenty of glucagon, or dasiglucagon. 

And what you're able to see here is that this is the blood glucose profile through BET experiment with glucagon in black and dasiglucagon in blue. And you can see that the dasiglucagon was just as effective at raising the blood glucose when necessary as the glucagon was. And so if you look at key glycemic outcomes, there were no statistically significant differences. But there was actually a trend for the dasiglucagon to be better. So we are now moving forward to bring that into use in the bionic pancreas in outpatient studies. And that's what's going to be used in our pivotal trial. 

Another one is sensor reliability and accuracy. Obviously, we're basing everything on the accuracy of these CGMs. But, fortunately, CGM accuracy has gotten very, very good recently. I used to show this thing, we're saying that the AP depends on-- artificial pancreas depends on CGM, which depends on blood glucose monitoring. And now, we even have no calibration CGMs. So it may be that we don't even need blood glucose monitoring, except as a sanity check in the future. 

And there are lots of different continuous glucose monitors on the market. We've tested them all as time has gone on. And we have also been concerned about certain problems that they have. For instance, many of the early versions were vulnerable to acetaminophen. If you took Tylenol, it would make your glucose look higher than it really was. That's a major problem, because a lot of things contain Tylenol. Even things people don't necessarily realize have Tylenol in them. 

But fortunately, now, multiple sensors have come out that are more accurate, are not vulnerable to acetaminophen. The most recent probably best two sensors are the Dexcom G6, which doesn't require calibrations and has relatively low errors. It has a average error of 8.8%. And it's actually approved to release replace finger sticks in the US, meaning that you can use the information from the CGM to decide on insulin dosing without doing an affirmatory finger stick. 

Another system that has come out recently is an implantable CGM. So you have this little what looks like, I don't know, two grains of rice put together or something. It gets implanted under the skin of the upper arm in about a 5 to 10 minute surgical procedure. And then you wear a transmitter over it. And it communicates with it with RFID technology. And what's great about that is if the transmitter falls off, you can just put it right back on and you're back in business immediately. Whereas if a Dexcom falls off, you have to put on a new sensor and you have to wait for two hours to warm up and so forth. 

The other thing that's cool about this one is that the little transmitter has a vibrating motor in it. So you have alarms right on your body. And it will buzz you if you're dropping low or going high, even if you don't have your phone with you, which is how people normally get this data. 

But we have done independent tests of all of these sensors. And, for instance, we directly compared this implantable sensor to the Dexcom G5 and found that they had comparable accuracy. These numbers are a little higher for the average errors, because these are true in use. These aren't in hospital studies that the 8% and 9% errors come from. These are people using them at home and comparing them to finger stick devices, which are, of course, less accurate than laboratory studies. 

But the point is that this new CGM we think is accurate enough to drive closed loop control. And we just completed a study. All our previous studies had used the Dexcom as the input. We just completed a study using this implantable Eversense system as the input, and we got comparable results. So now we can configure our system with either the implantable sensor or the non-implantable sensor as the input. And it can be up to patients which they like better. 

Another challenge is the slow insulin absorption I mentioned. And I'll just very briefly tell you that when we first started these studies, we discovered that some people had extremely slow insulin absorption. And that caused big problems with the artificial pancreas system, because if the blood glucose was still high, it would assume it had to give more insulin. But in reality, the insulin that it thought had been absorbed hadn't really been absorbed yet. 

And so here's an example of an experiment that went well back in 2008 and '09 when we were first doing these. And here's an experiment that went really badly. 

Is this working? 

I think it's working. 

And now sense what we found is that we were causing a lot of hypoglycemia in these people. And what we had to change the pharmacokinetics settings in our algorithm to solve the problem. And what we did is we changed the settings. We brought those same people back and found that once we changed the settings, we could take somebody like this who was having a lot of hypoglycemia. With new settings, no hypoglycemia. 

But it raised a point that by assuming that insulin was absorbed slowly, we were giving up a degree of control, because we weren't as aggressive as we would like to be. And what we found is that when we did that, people who could have used the more aggressive settings, we lost about 20 milligrams, 15 to 20 milligrams per deciliter of control in those people by going to a less aggressive setting. So wouldn't it be great if we had more rapid acting insulins? 

And, indeed, now, the pharmaceutical companies have come to the rescue again. Now Novartis has created a new formulation of their rapid acting insulin Novolog that's called faster aspert, or Fiasp. And it's 26 minutes faster to peak. And it has more than double the effect in the first 30 minutes. 

And so we also just completed a study testing this more rapid acting insulin in the bionic pancreas. And I don't know the answer for you yet. But in theory, it should allow us to get better glycemic control with a shorter control group. We don't have to wait as long to know whether the insulin we gave was enough before we give more. So that's exciting. 

And then the final challenge is you have to create hardware. You have to somehow get this to patients. A lot of our earlier studies were done with the system, the algorithm residing on an iPhone, and controlling insulin pumps remotely via Bluetooth. 

And I can tell you I've learned to hate Bluetooth because, you know, Bluetooth drops out and it's a hassle. And patients have to carry the iPhone and two pumps, since it's a bihormonal system. So the company that was started to develop this technology and commercialize it is now building this system that they're calling the Gen4 iLet. 

To give you a sense, that's about the size of a credit card and just a little under 1 centimeter thick. So it's a very small device that has a cartridge for insulin and a cartridge for glucagon. And it has the algorithm. And it will receive the signal from either the Dexcom G6. Or the sensing Onyx Eversense sensor, so patients can have a choice. And there's a dual infusion set. 

And it has a few other little bells and whistles. We believe it will be waterproof. And it will also have inductive charging. So you just put it on a charging mat like some of the latest phones. And, of course, that helps make it waterproof, because you don't have to have any place to plug anything into it. You just use inductive charging. 

So that's the system we're going to take into pivotal trials next year. And I said it uses these cartridges. So we have insulin in this little 1.6 mL cartridge. So you just drop the new cartridge into the system when you're ready for a new one. And the glucagon comes in a 1 mL pre-filled cartridge. And because it's very concentrated, we think that 1 mL will be enough for a week, six days to seven days of glucagon in a single 1 mL cartridge. So that allows you to carry your glucose metabolism in this tiny little package that you could slip into your pocket. 

And it has a very easy user interface. It shows you the glucose. It shows you the insulin and glucagon dosing. And this is how you announce a meal. You just say if you're in the beginning, middle, or end of your day. Is it a tiny, small typical for me or large meal? And you hit that, and that's all you do. 

And we think that that will be something that anyone could manage, perhaps including people who wouldn't be able to manage a sensor augmented pump with having to do a lot of math with carb ratios and carb counting and carb ratios and insulin sensitivity factors and so forth. So we hope this can democratize good glucose control to everybody. 

And we even have a mobile app. So if you have an iPhone, you can see how your glucose is doing on your Apple Watch. 

Currently, we're not to that system yet. We've been using this Gen 3.2 system, which has all the same functionality, but it's a little bit larger. And that system is the one that we've used in this just recently completed study. And we had very good glycemic control of that system. 

Now, the final challenge is we actually have to get this approved by the FDA. And so we've been really engaged with them from the beginning, trying to understand what it would take to get this system approved for use. And so that's an ongoing discussion. They've been really, really helpful at talking with us at every step of the way, helping us understand what was necessary. 

And so we have this whole set of studies laid out. This bridging study, it says it's ongoing. Actually it's done now. We completed it already. That was the first using the new iLet system, although the larger version, not the smaller version. 

We're then going to move into a bihormonal bridging study. So instead of insulin only, it'll be bihormonal, and use the new dasiglucagon, the stable glucagon. And then we're going to start our pivotal study in the second half of next year with that Gen4 iLet. And that will be a random order-- sorry, a randomized parallel design study, 3 months, with another 3 month extension. And we believe that will be enough for us to get approval in 2020 to sell the device. 

And then at the same time, we're going to start a bihormonal study, roughly the same time. But that's going to be a lot longer, because the FDA wants us to have a lot longer period of exposure to this new drug, the glucagon and this new micro dose glucagon, over a long period of time. Chronic micro dose glucagon has never been tested before. And so we have to show that that's safe. And so we think we'll probably look at FDA approval in 2020, two years later. 

But an important point is that we've built the device such that we're going to sell the device for insulin only. That'll be enabled. But the glucagon could be slotted in any time the FDA approves it. 

So it's the same device. Not a different device. We could just send out a code that would allow the glucagon functionality to be unlocked. And then people who already bought the insulin-only device could then just buy glucagon and use it in the bihormonal mode. So they won't have to get a new device. 

And that's important because I don't know for those of you in the diabetes world, there's a four-year cycle. The insurance will only pay for a new insulin pump every four years. So you don't want to lock people in for four years if in two years you're going to be able to have a new configuration. 

And this study is NIH funded, interestingly enough, even though NIH has now stepped up and said, we want to fund studies that will lead to approval and will actually lead to commercial products, because there's this really big unmet need. And the tandem device that uses the UVA algorithm is also being funded by that. So the NIH is actually stepping in and trying to pay for studies that will lead to approval of commercial products. 

And we powered-- overpowered the study to show superiority both in lowering of A1C and lowering of hypoglycemia. And these are the clinical sites. We have 16 clinical sites around the country that are going to participate in this study, which will give us a very diverse patient population. 

And so that's it for me. I'd just like to acknowledge the volunteers, hundreds of volunteers, who participated in these studies, and all our group members, my collaborators at Boston University and now at Beta Bionics, Ed and Faris and their whole team. And if you want to know more about what we're doing, we have a website, bionicpancreas.org. And now Beta Bionics has a website, betabionics.org. Thanks very much. 

[APPLAUSE] 

Wow. Can't help but be blown away by all the work you're doing. It's just absolutely amazing. Talking about democratizing, could you tell us a little bit about projected costs for both buying the machine and then for use for a year? 

Yes. So the machine itself, all of the financial projections that Beta Bionics has done are made on the basis of selling it for the same as a current insulin pump, like a tandem insulin pump. They're not going to do any kind of premium on price. 

They have worked really hard to make the device not only very durable, but also manufacturable at low cost. The company is a public benefit corporation. So they have a written right into their charter that their goal is to maximize the availability of the system to patients. 

And they just did this big funding round. And their lead investor was a double impact fund that tries to fund companies that are not only concerned with profit, but are also trying to facilitate the common good. And so they're really focused on trying to make sure it's affordable for everybody. So that's one thing. 

For the insulin-only configuration, basically the ongoing costs would be the same as any insulin pump. Once you add glucagon, though, the glucagon will be an additional expense. And the companies that are making these glucagons have told us that they're modeling their projections on pricing the glucagon essentially at the same on a per day basis as insulin. So it is going to increase the cost in that regard. 

We would just make the argument that the dramatic reduction in hypoglycemia and the further reduction in glucose will justify that. But we're going to have to have those discussions with insurers. Certainly, patients like it a lot better because they can be more spontaneous. They can go exercise and know the glucagon will kick in and prevent a low, rather than having to carry Gatorade and sip on the Gatorade as they're exercising as people with type 1 diabetes do now. 

So our hope is that it's not going to-- it may modestly increase the cost in the short run if you're using the bihormonal system. But by dramatically lowering the mean glucose, we should reduce long-term complications. And by reducing severe hyper and hypoglycemia we'll also reduce the number of emergency room visits associated with those. 

So even in a relatively short-time span, we think it would be better than revenue neutral for insurance companies should save money. That's the hope. Yes, sir. 

With the glucagon as a modified neopeptide, why wouldn't you get an immune response [INAUDIBLE] 

Well, that's always a risk when you have a new peptide? First of all, it's pretty small. And so you generally have a harder time generating immune response to smaller peptides. They have done lots of studies, and they've only seen one episode of someone developing drug antibodies. 

And what was interesting about that, that person developed antibodies to both glucagon and dasiglucagon. So they also had an antibody to the endogenous peptide in addition to the dasiglucagon. They've never seen an instance where somebody had an antibody to dasiglucagon that didn't have an antibody to glucagon itself. 

So it seems to be low risk. But it's possible, certainly. That can happen. We know people develop antibodies against the insulin analogs. But they still seem to work OK. They aren't neutralizing in the sense that they stop them from working. Yes. 

Does the average daily amount of insulin used [AUDIO OUT] or not using one of these [AUDIO OUT] if so do you think there'll be an effect on weight? 

So we've looked at that in a bunch of our studies. And in most of our studies, overall there's no change in the average insulin total daily dose. In one study, we found that there was, and we were curious to why that was. It was the study I showed you with the ball and stick diagram. It turned out that all of the increase in insulin utilization in that study was attributable to the three people who had the highest mean glucoses under usual care. And they had a dramatic reduction in their glucose on the bionic pancreas. 

So in that case, our rationale for that is that they'd simply needed more insulin. And the system gave them more insulin. But for people who are already in reasonable control, the system doesn't use anymore insulin. The average use of insulin is identical. 

In terms of implications for weight, one thing we have noticed, though, is that people use a lot less what we call medicinal carbohydrates. So people, if they don't have glucagon, they have to take sugar. There's these sugar tabs or they eat candy or something every time they run low. And it turns out that's a substantial amount of carbohydrate, 30 to 45 grams a day of carbohydrate taken to treat hypoglycemia. That's dramatically reduced with our system. 

So if anything, my prediction would be we would have a net negative effect on weight, because people would be taking less of this sugar just to keep themselves from going low. Also, they say they feel more empowered to exercise with the system, because they're not as worried about going low, and they don't have to take carbs during exercise. So my hope is that it would have a beneficial effect. But we haven't shown that yet. 

One more question. 

All right, thank you very much. It's a pleasure to be here.