When Bryan Mazlish’s son was diagnosed with Type I diabetes, there were unexpected challenges. Managing diabetes on a day-to-day basis was tough, so he hacked into his son’s insulin pump and continuous glucose monitor to create the world’s first ambulatory real-world artificial pancreas. Now his mission is to make it available to everyone.
Bryan Mazlish: A nice demo that we showed at Google IO earlier this summer, where we showed our use case for one of their forthcoming APIs. We’re really at the vanguard of digital health medical device enterprise software, and it’s incredibly exciting but also challenging place to be. We’re enthusiastic about the prospects for what we can do for a whole lot of people.
Ginette: I’m Ginette.
Curtis: And I’m Curtis.
Ginette: And you are listening to Data Crunch.
Curtis: A podcast about how data and prediction shape our world.
Ginette: A Vault Analytics production.
This episode of Data Crunch is brought to you by Lightpost Analytics, a company helping bridge the last mile of AI: Making data and algorithms understandable and actionable for a non-technical person, like the CEO of your company.
Lightpost Analytics is offering a training academy to teach you Tableau, an industry-leading data visualization software. According to Indeed.com, the average salary for a Tableau Developer is above $50 per hour.
If done well, making data understandable can create breakthroughs in your company and lead to recognition and promotions in your job.
Go to lightpostanalytics.com/datacrunch to learn more and get some freebies.
Curtis: Today we get to speak with a man who, after studying computer science at Harvard, went to start a stock-trading algorithm company on Wall Street until his life experienced a twist. Now he’s the president and co-founder of one of the leading digital health medical device enterprise software companies, which employs machine learning to customize and automate medicine intake, all because of an unexpected challenge that showed up in his life.
Bryan: My name is Bryan Mazlish. I’m one of the founders of Bigfoot biomedical. My background is in quantitative finance. I spent 20 years on Wall Street, first at a large investment bank and then about a decade running a fully automated trading business where we built algorithms to buy and sell stocks completely automated fashion, and it was about 6 or 7 years ago that my path took a change . . .
Ginette: Bryan’s son was diagnosed with Type 1 diabetes, which Bryan says wasn’t entirely unexpected because his wife has the same disease. But what was unexpected was the intensity of managing the disease on a day-to-day basis. He was surprised with how antiquated the insulin management technology was. There wasn’t technology that could anticipate his son’s insulin needs and automatically give him the insulin he needed.
Bryan: You have a need to take insulin to just simply to live. This is something that needs to be delivered on a constant basis, 24 hours a day. You can take this in one of two ways: you can use an insulin pump that delivers this in a continuous basis, and you can also take a once-a-day injection, and the benefit of the pump is that you can vary that at different points in the day. When you take an injection, it lasts for up to 24 hours, and it doesn’t have the same flexibility, but it does have the benefit of not having to wear a device to deliver the insulin. And that’s just the baseline, on top of that you need to take insulin to offset meals, primarily carbohydrates and high glucose levels. So when you’re going to sit down to eat breakfast, lunch, or dinner, or even a snack, you need to estimate the amount of carbohydrate and glucose impact of the meal that you’re about to consume, and then dose that amount of insulin, either through an insulin pump or through an injection at that time.
Ginette: Figuring out how much insulin to give yourself is tough. If your body has too little glucose as a result of your estimate, you can experience hypoglycemia, which is severe cases could lead to a coma or death, and if your body has too much glucose, you could end up with problems in the future, like kidney failure, vision problems, and lack of feeling in your extremities.
Curtis: Injecting the wrong amount of insulin is obviously dangerous either way and getting guidance from doctors, as Bryan’s family experienced, can be tough. Often patients meet with their doctors for about 15 minutes three or four times a year where their doctors try to give them guidance on how to manage their diabetes on a daily basis. But as you can imagine, this advice is limited based on the data doctors receive from patients, which can be skewed and inaccurate.
Bryan: For the vast majority of people who are making injections, there is very little support in capturing any kind of data or enabling any kind of analysis on that data given the lack of it. The standard of care is a handwritten logbook that clinicians ask patients to fill out for a week or two before their appointments. As you can imagine, a lot of that is either not done, or it is done the night before they go to the doctor’s office as best as they can remember.
Ginette: So Bryan decided to take matters with his son’s insulin titration into his own hands, and he hacked into the insulin pump and into the continuous glucose monitor and created the first ambulatory real-world artificial pancreas.
Bryan: We had the benefit of a full stack of servers and computers. I had a couple of employees—I owned my own trading business—I had the flexibility to repurpose those resources to start to develop tools for my family at first. We developed improved alarms and communications, methods for managing my son’s glucose, and then that ended up with a . . . the first do-it-yourself automated insulin delivery system that has been quite transformational for my wife and son, and they’ve been using that for more than 5 years now.
Curtis: Once Bryan saw how helpful his innovative automated insulin delivery system was for his family, he decided to work toward making it available to other people. He started Bigfoot Biomedical to further develop his invention into a software product, and he used his experience building stock-trading algorithms to hopefully make life easier for a lot of people.
Bryan: Frankly it’s a bit easier to predict glucose values than it was to predict stock prices.
So the first leg of the problem is getting the data. The ability to capture data through IoT devices, whether it’s a diagnostic data for the glucose monitoring, that’s via episodic finger sticks where you take a small drop of blood and put it onto a test strip, or through a continuous monitor, which we are leveraging through our partner Abbott Diabetes Care for their groundbreaking Libre sensor, where you put a small wire under your skin, and it stays on your body for up to a week or two, and that sends glucose readings every few minutes to a smartphone or another device. This is a key component of of the approach.
In addition to that the ability to understand when the insulin dosing is taking place is the other key component in enclosing this feedback loop. When it’s an insulin pump, that’s an electronic syringe effectively, so we have the ability to capture that data, but connecting it to an IoT web and mobile backend in a secure manner is a key step forward that we’ve made. You’ve got a device that’s on your person that’s connected to the Internet with a drug that has lethal amounts of drug in it. You need to do that in a secure fashion, and we have innovated pretty meaningfully to enable that to happen on commercial off-the-shelf smartphone.
Ginette: As you can imagine, security is hugely important with these IoT devices. As Bryan pointed out, if a device is connected to the Internet and it has lethal amounts of drugs in it, safeguarding the device from some nefarious person trying to hack it is key.
Bryan: The design for the security was one of the first things we did, and the security is designed for the entire system. It’s not something you can bolt onto a device that’s been out in the world. In fact, one of the reasons I was able to do what I was able to do for my wife and son was because of a security hole in some existing medical devices, and thankfully I did it for good, but we have built from the manufacturing of the embedded devices to the mobile systems to the cloud-based software a comprehensive security solution.
Curtis: Bryan’s team has acquired and developed insulin delivery technologies, and they’ve also licensed glucose measurement tools that feed data to their algorithms. For instance, they gather data from Abbott Diabetes Care’s continuous glucose monitor, which is also known as a CGM.
Bryan: That is one of the inputs to our personalization and automation algorithms to our machine learning and our feedback control algorithms. That is essentially the sensor for our system. We have sensors in the system, which are the glucose monitoring systems, and then we have actuators, which are the insulin pump or the insulin pen, and the brain in between the two is where the Bigfoot magic happens, and that’s where we do the analysis and do a lot of what is missing in the current standard of care.
Those CGM technologies are available currently. You can go to CVS and buy them, but they require a prescription. What’s new is the intelligence that we’re putting between those sensors and the delivery devices. Right now it’s primarily a person’s head that has to take the data from these diagnostic devices, figure out how much insulin to dose based on that glucose data and the potentially outdated information from their clinician visit, which maybe months old, and then try to determine how much insulin to give, and then follow-up over the next few hours to see whether or not that was the right amount, because insulin takes up to four or five hours to actually work its course through the body, and you can have events happen far beyond after you dose the insulin. By leveraging these seamless data capture, we’re able to be much smarter relieve the burden of both dosing decision and the subsequent cognitive load to manage that glucose over the time period by which the insulin works.
Ginette: So what else is going on behind the scenes? We’ll get to that right after this break.
This episode of Data Crunch is brought to you by Lightpost Analytics, a company that specializes in teaching you analytics skills that get you promoted.
Part of any good analyst’s toolbox is knowing how to analyze data and when AI can be applied. But, you also must know how to communicate your insights so that they get used. Data visualization with Tableau is one of the best options on the market.
Lightpost Analytics will teach you how to find insights and share them effectively, creating improvements for your company and greater success for you.
Go to lightpostanalytics.com/datacrunch to learn more and get some freebies.
Bryan: There’s a couple of different feedback loops that are going on. There’s a the near-term feedback loop, which is you know, this dose that I just gave two hours ago, was it too much or too little and do what I need to adjust for it? And then it is on a longer cadence. The estimates that I’m making for a dose, for example for a bowl of oatmeal in the morning, is that generally riding too much or too little and do we need to either increase or decrease it? And then the ability to interact with the clinician . . . . system is a significant step forward as well because all of the data will be captured.
One of our key differentiators is to deliver a comprehensive solution to both the patients, the providers, and the payers. The vast majority of diabetes technology are built in silos where you have the different aspects that I’ve been talking about delivered by different companies with different prescriptions and they’re not combined in a holistic way. . . . Really bring it all together and deliver a single service, and that’s part of our solution to deliver it all as a Saas-like monthly service.
Ginette: Additional inputs that their algorithms take into account are based on the individual’s fluctuating sensitivity to insulin and carbohydrates.
Bryan: The understanding of the person’s physiology is a key input. Generally their sensitivity to insulin and to carbohydrates, and this varies both over time, longitudinally and diurnally over the course of a day. What our systems take currently is a rough estimate and then based on the data that it captures of system use, both from insulin dosing and glucose variation, it dials in and personalizes it’s underlying model of that person’s physiology, so once it has that and whatever model it’s using, it leverages the historical insulin dosing and glucose to formulate a prediction going out into the future to estimate where their glucose is going to end up over the next few hours. This enables the system to either increase or decrease, or in the case of a person taking injections, provide relevant feedback to the person to help them to manage their glucose. And in doing so, it significantly relieves much of what the person is doing on their own in a regular life.
It also relieves a great deal of the burden for the clinicians, which are typically asked to do a very challenging thing, which is to titrate someone’s insulin model. They are asked to do that with very little data, and with the risk of if it if they intensify too much, they may not have a chance to change that again for up to six months until the next time the person comes in. Our systems can work on a continuous basis, and they can on a near daily basis evaluate what’s happening with the person and adjust their doses either up or down to try to improve their outcomes, and at the end of the day, reduce the burdens on the patients, improve their outcomes and overall footprint of the disease on the healthcare system by reducing the number of hospital admissions.
Curtis: So how has this been received in the medical community?
Bryan: The enthusiasm has been extremely high across both the community, the clinical community, and the investment community. We are still in development phase, so we run a clinical trial. We have a couple more slated online next year. We have a long road because these are class three medical devices. They require the highest scrutiny by the FDA. And we’re not, we’re not marketing it right now, although everyone certainly who founded this company and many of those who have joined want this technology yesterday, and we’ve got a great deal of enthusiasm from some healthcare networks that are looking to do pilots with us when we’re first able to bring something out, probably sometime next year.
Our patients and the first patients will be our children, those of us who founded this company. Their lives are depending on this, and and it’s not just this, but any FDA device requires a very rigorous risk analysis that looks at all the different failure modes that can happen with the type of system, no matter how accurate your sensors or your actuaries are, there’s always failure modes, and it’s not a question of eliminate all your failure modes but making sure that the failure modes are safe and they manage appropriately.
Ginette: Here are some final thoughts from Bryan if you’d like to follow their company’s progress as it works toward getting its products on the market.
Bryan: I’d encourage folks to check our LinkedIn page. We’ve got a whole bunch of openings. We’re hiring in a bunch of different competencies. Follow us on Facebook. We’ve got a pretty robust web presence and pretty enthusiastic following there, and feel free to reach out. We’ve got to contact link on our web page Bigfoot biomedical.com.
Ginette: A huge thank you to Bryan from Bigfoot Biomedical for chatting with us.
And as mentioned in our podcast, if you want some freebies from Lightpost Analytics, the sponsor of today’s episode, go to lightpostanalytics.com/datacrunch. They’ll teach you how to find insights and share them effectively, creating improvements for your company and greater success for you.
You can find that link and all other links to attributions in our show notes at datacrunchpodcast.com
“Cold Funk” by Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License