French startup Diabeloop has developed an 'autonomous' medical device with embedded machine learning for treating type 1 diabetes.
Dealing with chronic disease can be a big chore both for affected patients. As Erik Huneker, CEO of French medical device startup company Diabeloop said last week at the Leti Innovation Days in Grenoble, “A person with type 1 diabetes needs to think and make a decision about 40-50 times a day about how much to eat and how much insulin to take.”
This makes management of diabetes a perfect candidate for the utilization of artificial intelligence. That’s why Huneker’s company has developed what he claims is the world’s first autonomous medical device, which uses machine learning in real time to determine the optimum insulin to administer automatically to a patient without them having to even think about it.
Type 1 diabetes affects about 2 million people in Europe and what Diabeloop aims to do with its closed loop medical system is to help people with the condition try and live normal lives by improving their treatment dramatically.
Erik Huneker, CEO Diabeloop
Huneker emphasized they’re not simply developing an app. Its system, the DBLG1, comprises a continuous glucose sensor, a patch insulin pump and a dedicated handset that houses the Diabeloop algorithm (loop mode) and serves as a user interface with the system. The handset has its own custom operating system and connects with the sensor and the insulin pump using secure Bluetooth.
The system senses and monitors the patient’s blood glucose every five minutes, analyzes the data in real time, and can inject the correct does of insulin before the patient even knows he or she needs insulin, since it monitors daily patterns and over time will have learned the optimal parameters for the patient. The continual assessment takes into account current glucose levels, glucose trend, the hypoglycemic risk (short and long term) and special events such as meals, physical activities, and calibration, to provide an optimal recommendation adapted to the needs of the patient.
Huneker said, “For each person the system runs 150 million equations per day for the machine learning.”
The company’s DBLG1 obtained CE marking certification in November 2018, and earlier this year entered into a mutual distribution agreement with Netherlands-based ViCentra, which provides the insulin delivery system used as part of Diabeloop’s product.
The results of a trial with 68 patients was published in the Lancet Digital Health journal. It found that the DBLG1 system improves glucose control compared with sensor-assisted insulin pumps; it concluded that its finding supports the use of closed-loop technology combined with appropriate health care organization in adults with type 1 diabetes.
One of the outcomes of the study was that patients spent 68.5% of their time in the target glucose range, meaning a 10 point increase in the time spent in the target glucose range. There was also a decrease by 50% of time spent in hypoglycaemia (<0.7 g / l), in other words, a gain of more than 30 minutes less per day in hypoglycaemia. A third conclusion was that a reliable closed-loop system that was functional for more than 84% of the time in a home setting for 12 weeks.
While this is focused on adults and the system is already being used by around 30 patients, the company is now also looking to adapt its ‘artificial pancreas project’ to juvenile diabetes and thus improve the quality of life for each child in the short term and the long-term. A study will be conducted in two centers in France and one in Belgium. The goal is to include around 20 children who will be followed for 6 weeks.
We asked Huneker why his company was using a standard handset rather than developing its own smaller integrated device and possibly its own system on chip. He stressed that to do that could have needed years of development and the company really needed to get something to market that could be trialed on patients. This is why they are using standard off-the-shelf components – like the Dexcom G6 continuous glucose monitoring sensor system and the Kaleido insulin pump.
Huneker said that the big challenge right now is to be able to do more on the device. “We need more computing capability.” He suggested they are planning to integrate more sensors and more computing power, and maybe three years from now could even think about developing a more integrated system that could be incorporated in a wearable device.
Diabeloop emerged from a medical research project initiated by diabetologist Dr Guillaume Charpentier in 2011; the company was established in 2015 with support from CEA-Leti in Grenoble to help develop the product. The company now has around 50 people and has raised €13.5 million to date.