A community of people with type 1 diabetes got a self-built device approved. What can they offer that big companies can’t?
Ten years ago, a tech-savvy group of people with type 1 diabetes (T1D) decided to pursue a DIY approach to their own treatment. They knew that a fairly straightforward piece of software could make their lives much easier, but no companies were developing it quickly enough.
What this software promised was freedom from having to constantly measure and control their blood-glucose levels. In people without T1D, when glucose levels rise, cells in the pancreas release insulin, a hormone that helps tissues to absorb that glucose. In T1D, these cells are killed by the immune system, leaving people with the condition to manage their blood sugar by taking insulin.
“It is almost inhumane,” says Shane O’Donnell, a medical sociologist at University College Dublin, who, like everyone quoted in this article, lives with T1D. “You’re constantly having to think about diabetes in order to survive.”
Members of the nascent DIY community were using the most sophisticated technology available: insulin pumps and wearable devices called constant glucose monitors. But they still had to read the monitor’s data, forecast their diet and exercise and then calculate the appropriate insulin dose.
What they wanted was automation — an algorithm that would analyse glucose data and program the pump itself. Coalescing around this aim in 2013, the community debuted a hashtag: #WeAreNotWaiting.
Then, in February 2015, group member Dana Lewis shared the code for an algorithm that she and two collaborators had developed and tested.
“We didn’t set out do to anything big,” says Lewis, now an independent researcher in Seattle, Washington. But soon, people who had downloaded and used the algorithm shared their personal experiences and gave feedback. When users suggested tweaks and potential improvements, others tried them and reported back.
Katarina Braune, an endocrinologist at Charité – Berlin University Medicine, estimates that around 30,000 people now use open-source technology for automated insulin delivery (AID). Some use Lewis and colleagues’ original OpenAPS system, which requires a minicomputer to control it, whereas others use either AndroidAPS (which evolved from Lewis’ system) or Loop, which are smartphone applications.
The movement has continued to mature. After years of relying on self-reported data, in the past year, two randomized controlled trials1,2 have shown the safety and effectiveness of open-source systems. And this January, the US Food and Drug Administration (FDA) granted regulatory clearance to an AID system based on an open-source algorithm for the first time.
Today, however, the technology landscape for T1D is much more crowded. The first commercial AID system was launched in 2017 and, currently, five companies sell such systems, with more than 750,000 users.
Is this the beginning of the end for the open-source movement in diabetes care? Some diabetologists think so. But many advocates reject that idea, saying that the community is still pushing the technology in new directions that promise more personalization and automation than commercial versions currently provide.
Gaining traction
Sufyan Hussain, an endocrinologist at King’s College London, says that he was initially sceptical about the DIY AID community. But when he started to engage with it in around 2016, he was “shocked at how well engineered the solutions were in terms of the safety and understanding”.
In 2022, Hussain co-authored an international consensus statement3 — signed by more than 40 medical and legal experts and backed by 9 diabetes charities — calling for health-care professionals to support those wanting to use open-source AID.
Results from randomized trials have further elevated the status of DIY technology. A study published this year1 found that an open-source and a commercial AID system both controlled glucose levels similarly well. And a September 2022 study2 demonstrated the efficacy of an algorithm that runs an artificial pancreas system on Android smartphones. By recruiting participants who were new to the technology, the study addressed long-standing criticisms that advocates had previously cherry-picked data from highly motivated, tech-literate members.
Despite their tens of thousands of users, many of whom have chosen not to use available commercial systems, DIY devices for T1D remain relatively niche. O’Donnell says that a welcoming and supportive community guides people with limited technology skills through setting up systems. But most people with T1D — and most doctors — haven’t encountered these systems, says Aaron Kowalski, president and chief executive of JDRF, a non-profit research organization in New York City that focuses on T1D.
The FDA’s approval of the open-source system Tidepool Loop could change things, says Kowalski. The algorithm underlying it was created in 2016 by people living with T1D, and was initially rolled out through online forums, before a version of it was taken through FDA clearance by Tidepool, a non-profit organization in Palo Alto, California.
Tidepool’s goal in getting Loop through regulation is “to make it more accessible to a wider audience”, says spokesperson Saira Khan-Gallo. “Downloading code and building an app on your phone is not for everyone,” she says. But “the algorithm, the novel features and technology should be available to anyone who is interested”.
Tidepool and others hoping to roll out open-source algorithms face a big challenge: their products do not stand alone. The algorithms require compatibility with continuous glucose monitors and pumps — made by other companies — thereby demanding a cooperative relationship between manufacturers. Tidepool is yet to announce which device company it will collaborate with to launch Loop.
Interoperability between different products and algorithms could shake up a marketplace in which individual manufacturers have usually developed proprietary, exclusive software. This could have ramifications beyond T1D treatment, affecting any computerized medical hardware that software developers might attempt to improve.
Kowalski points to examples in other industries, such as aerospace, in which companies use engines made by others. “They’re plugging [in] different components from different manufacturers to get the best performance,” he says. “People with diabetes should have the opportunity to use the best tools that work the best for them.”
Several device manufacturers, however, told Nature that they are wary of relying on third parties to provide crucial pieces of AID systems. For instance, Tandem, a pump manufacturer in San Diego, California, says that if an algorithm runs on a smartphone app and not on the pump itself, phone damage or connectivity problems could be a risk to therapy. And Medtronic, a medical technology company in Watford, UK, has decided to prioritize designing their own complete system rather than interoperable components.
Nonetheless, Tidepool’s regulatory approval will hopefully ease the way for future standalone algorithms seeking clearance, says Khan-Gallo. She hopes that having more options will incentivize companies to make their devices cross-compatible.
And where algorithms have led, open-source hardware might follow, says O’Donnell. A team at the University of Otago in New Zealand has run a successful early-stage clinical trial of an open-source insulin pump4. The goal is to provide free-of-charge design plans to qualified manufacturers to build pumps for a fraction of the cost of current commercial ones.
Algorithms ablaze
When asked whether the open-source community of users still has a part to play, Kowalski says: “The DIY community is always going to be there. I think they’re the ultimate testing ground for what people want and need.”
Hussain agrees. Participants’ lived experiences of T1D constantly generate ideas for novel features, he says. But more importantly, the network of online forums and highly motivated members has created a rapid and powerful way to test algorithm functions.
Problematic updates are quickly weeded out, and beneficial innovations self-propagate and achieve widespread use in just months. Currently, Hussain says, “The commercial systems don’t have advanced features that the open-source systems allow.”
Some of these features mean that certain open-source systems come close to regulating blood sugar completely autonomously. Although commercial systems are advancing, none is as near to solving this important problem; they currently use hybrid algorithms that manage insulin dosing most of the time, but they require a lot of manual input. For instance, users must program in meals to ensure that the devices deliver large corrective insulin doses.
By contrast, many open-source systems do not require meal announcements and come close to being a fully closed loop5.
Several advances in technology made this possible, says Lewis. The apps can analyse long-term changes in glucose control and insulin sensitivity resulting from, say, hormonal changes or illness. They do this by retrospectively comparing glucose levels with insulin doses over days or weeks, to adjust upcoming dosing according to changes in insulin sensitivity. They can also deal with perturbations to blood glucose “without knowing whether it was a meal, adrenaline, stress, excitement — whatever, it doesn’t matter”, says Lewis.
Whether open-source systems are better than commercial ones “is a big ‘it depends’ question”, says Rayhan Lal, an endocrinologist at Stanford University in California, and Tidepool’s chief medical adviser, who says that he has started more than 3,000 people on open-source AID systems, and uses one himself.
What matters, Lal says, is that individuals can find what works best for them, in terms of the control they gain or the effort they want to make. Some people might want to tailor their devices; others might prefer the simplicity of a package from a commercial manufacturer.
The DIY community and industry are not in opposition, says Lewis. She is delighted that a safety feature that she wrote and freely shared was incorporated into a commercial device. The open-source community will stay relevant, she says, as long as it offers users choice. “The vision of where I’d love to get to — whether it’s commercial or DIY — is really, really about the person with diabetes and our safety and our quality of life.”
Nature 620, 940-941 (2023)
doi: https://doi.org/10.1038/d41586-023-02648-9
References
Wu, Z. et al. Diabetes 72 (Suppl. 1), 274-OR (2023).
Burnside, M. J. et al. N. Engl. J. Med. 387, 869–881 (2022).
Braune, K. et al. Lancet Diabetes Endocrinol. 10, 58–74 (2022).
Payne, M. et al. J. Diabetes Sci. Technol. https://doi.org/10.1177/19322968221142316 (2022).
Petruzelkova, L. et al. Diabetes Technol. Ther. 25, 315–323 (2023).