In the world of emerging automated insulin delivery systems for diabetes, French company Diabeloop stands out for its design work. This is because the company works on their designs with input from the people who will actually be using their products.
Company leaders have been researching and building this technology for a decade. They’ve worked directly with people with diabetes (PWDs) in the DIY #WeAreNotWaiting community to create a more customizable, practical system instead of the usual “engineering-driven” devices we’re used to.
Evidence of that can be seen in things like the Diabeloop “Zen Mode” feature, which lets users temporarily raise glucose targets slightly to avoid annoying alerts that would otherwise beep after every meal. There’s also a “Privacy Mode” that lets them shut off sharing features in case they need a break.
“With Diabeloop, I am most excited about how the technology performs, really making it possible to maximize time in range,” said Zoe Heineman, a longtime type 1 diabetes (T1D) advocate in New York who joined Diabeloop as Senior VP of North America in February 2020. “The fact it’s designed with us in mind, and it has features that really reflect what we live with, is hugely important to me.”
While it hasn’t yet hit the market, Diabeloop expects to launch its first-generation system in Europe during 2021, as well as a version specifically designed for “highly unstable” T1D. From there, the plan is for a U.S. launch before long.
Diabeloop’s key technology is a smart algorithm that functions as the brains of its new automated insulin delivery system, preliminarily known as DBLG1 (for “Generation 1,” to be followed by DBLG2).
Unlike the two commercial systems currently on the market in the U.S. from Medtronic and Tandem that have their algorithms built directly into the devices, this Diabeloop smart algorithm works on a dedicated handheld receiver or smartphone app. It then communicates with a connected insulin pump and continuous glucose monitor (CGM).
Diabeloop’s artificial intelligence (AI) analyzes glucose data, calculates insulin amounts, and automates the pump delivery based on the CGM readings.
In many ways, what Diabeloop is doing is similar to the work by open-source data nonprofit Tidepool, which filed its Tidepool Loop mobile app with the FDA in late 2020 — except with different hardware devices.
Diabeloop’s first generation works with the colorful hybrid patch-tubed Kaleido pump available outside the U.S., but the company hopes to work with other in-development devices down the road.
Diabeloop has already inked deals with these additional partners:
- Roche Diabetes Care, which is behind the Accu-Chek brand of fingerstick meters and strips, as well as the Accu-Chek brand of insulin pump that was discontinued in the U.S. as of 2017. They’ll be integrating the Accu-Chek Insight pump with Diabeloop’s AID. Roche also owns the mySugr data platform as well as the Solo micropump acquired in 2010 but has only brought it to market in select countries outside the U.S.
- SFC Fluidics, which is developing an open-protocols patch pump closed loop system dubbed “Panda.”
- Japanese company Terumo, who makes the MEDISAFE WITH patch pump overseas.
- Dana pump from South Korean company SOOIL Development, which is working on a next-generation version of its tubed pump tech, the “Dana-i” pump. This will be a two-way communication device that will also use its “AnyDana” mobile app for smartphone control. A clinical study on the use of the Dana-i pump with Diabeloop was launched in early 2020.
Diabeloop secured European regulatory approval in November 2018 for its first generation tech, making it the first early closed loop system to follow the widely-publicized debut of the Medtronic Minimed 670G approved in the U.S. just 2 years earlier.
With that approval, Diabeloop became the first-ever system OK’d by regulators to use the Dexcom CGM — even though Dexcom had been used widely in DIY closed loop systems worldwide.
Clinical trial results look promising. One 2021 study followed 25 patients using the DBLG1, and found that the system was able to “significantly improve glycemic control in real-world conditions, without serious adverse events.”
DiabetesMine spoke with Diabeloop CEO Erik Huneker, to get a better understanding of what’s exciting about this new system.
How exactly does your system work?
DGLB1 is a
DBLG1 artificial intelligence analyzes data in real time and takes into account the patient’s physiology parameters, history and data entries (meals or exercise) to determine the correct dose of insulin to send to the pump.
What’s the deal with the Kaleido patch pump?
Developed by a Dutch startup ViCentra, the Kaleido pump is a hybrid device, meaning it’s both a patch pump and has infusion set tubing.
The design is a little rectangle that sticks onto your body, similar to how OmniPod works but smaller, measuring just 50 mm by 35 mm and only 12mm in thickness, and weighing in at a very light 19 grams. There are many different colors to choose from.
It can deliver small doses (0.05 unit increments) with incredible accuracy. It actually has short traditional tubing that connects to an infusion set, just like a traditional pump. It’s also stuck on your body where the insulin goes in, so the user ends up wearing two separate adhesive sites on their skin for the pump.
Instead of using controls on the Kaleido pump as you normally would, the pump is connected to our Diabeloop system on a locked-down smartphone (Sony Xperia Z1) that also feeds in data from the Dexcom G6 CGM, and that’s where the algorithm lives to make insulin dosing decisions.
How did the #WeAreNotWaiting community influence development?
We looked at their work, and discussed quite a bit with the DIY community. There was a lot to learn, in particular on what users were looking for in a closed loop. We also compared our algorithm with the DIY systems, and contributed a significant amount of code to Tidepool, for example.
Following feedback with patients, we developed a “Zen Mode” feature. It’s an option that lets you slightly raise your glycemic targets for a short period of time to avoid hypoglycemia and avoid unwanted alerts/alarms. This is particularly convenient during a movie, for example.
There’s also a “Privacy Mode,” in which no data is uploaded to the server at all, for when you want it to stay confidential.
Can users choose their own target range for blood sugars?
Yes, users can adjust the algorithm target between 100 and 130 mg/dL. They can also adjust preferences between a lower mean (but higher probability of hypos) or higher mean, but with much lower hypo risk.
The hypoglycemia threshold is also customizable. We know this is important to people, because not being able to reset these targets has been a point of frustration with some systems.
How do carb-counting and meal announcements work in this system?
The user still has to announce meals. Of course, the more precise they are with carb counting, the more accurate the algorithm will be.
Also, the system allows you to save personalized preferences for each meal (like a typical small breakfast of 20g of carbs, a medium breakfast of 40g of carbs, or a large breakfast of 80g of carbs). So you can choose this option when announcing a meal. The user can decide the level of precision they want to play with. In any case, the algorithm will learn to adapt and correct.
Diabeloop has also started an “unannounced meal” clinical study for the U.S., expected to wrap up in early 2022.
How exactly is AI employed in DBLG1?
Our system is developed using machine learning methods for a given population.
When the system starts, it will then learn the person’s physiology and/or lifestyle and will adapt over time to individual requirements. However, this adaptive machine learning system will follow first and foremost the preferences of the user (glycemic targets, etc.)
What about phone control, or at least mobile app data display?
That isn’t planned for the first commercial release, but we’re definitely looking into mobile app data display, especially for the kids version. Closer integration with the phone will surely follow.
Will the U.S. version be the same as what’s developed for those internationally?
Yes, it’ll be an update to the existing CE Mark. There’s no plan at the moment to use different components in the U.S.
What is the actual launch plan?
Gradual launch is planned to start in 2021 in Europe. Beyond France, which is our “home country,” we’ll be focusing primarily on countries where there’s an appetite for innovative medical devices and fluid reimbursement mechanisms.
We’ve also finished a second financing round in order to support further R&D and commercial interests such as European launch, pediatric approval, our FDA submission and preliminary market introduction in the U.S.
And of course, we’ve started working on our next-generation “DBLG2” already, where we should be able to integrate some additional items to make life even easier for T1D users. We have a few scenarios, but we don’t have a timeline yet.
Would people buy from Diabeloop directly?
We’re currently fine-tuning distribution and commercial agreements in our priority markets, so details are still being worked out.
Finally, can you fill us in on what makes Diabeloop unique?
Founded in 2015, this small company of about 50 employees is based in Grenoble, France, and is a unique collaboration between scientists, healthcare clinicians and patients.
A few years ago, Diabeloop teamed up with CEA-Leti, a major research collaboration in France made up of 12 university hospitals, clinicians and patients, to create a lab where they’d work on this new closed loop system collaboratively. We think that’s pretty unique.
Diabeloop does believe in disruptive innovations, and we think ours is one of those.