Open-source automated insulin delivery in type 1 diabetes—the evidence is out there

The past decade has seen substantial advances in the development of automated insulin delivery (AID) systems, which can improve glycaemic outcomes and burden of treatment for people with diabetes.^1,2 AID systems use algorithms to adjust insulin delivery based in part on continuous glucose monitoring data. The first broadly available AID algorithm, OpenAPS, is open-source and was designed and developed by a community of people with diabetes and their loved ones before any commercial AID systems were available. Although the number of approved commercial AID systems has increased, there are still constraints in their functionality, efficacy, access, and worldwide availability.¹ The development and dissemination of open-source AID systems, behind the hashtag #WeAreNotWaiting, have been held as exemplars of user-led innovation, which have paved the way for an impactful form of treatment for a complex condition.^3,4

The body of evidence supporting the safety and efficacy of open-source AID and the ethical implications for health-care practitioners supporting the right of people with diabetes and their caregivers to make informed decisions on their treatment was highlighted in January, 2022, by an international consensus statement endorsed by several professional diabetes organisations.¹ Despite this endorsement and growing uptake globally, widespread support from health-care systems and academics for the use of open-source AID has been lacking. A potential critique of open-source AID has been the absence of evidence from randomised controlled trials,^5,6 which have been difficult to generate owing to regulatory and funding barriers, because the algorithm is not owned by academic research or for-profit companies. It should be noted that several commercial AID systems have received regulatory approval on the basis of pivotal clinical trials without control groups or randomisation.

In the September, 2022, issue of The New England Journal of Medicine, Burnside and colleagues⁷ reported findings of the CREATE trial, a landmark, multicentre, open-label, randomised controlled trial using a modified version of the open-source AID system, AndroidAPS, with the OpenAPS 0.7.0 algorithm. The 24-week study included 48 children and 49 adults with little experience of using AID who were randomly assigned to either open-source AID or a control group that used sensor-augmented pump therapy. The primary endpoint was time-in-range glucose concentration (3·9–10·0 mmol/L or 70–180 mg/dL) during the final 2 weeks of the trial. Time-in-range was significantly increased in the AID group (mean 71·2% [SD 12·1]) compared with the control group (54·5% [16·0]; p<0·001), with those using open-source AID spending 3 h and 21 min more within range per day and the greatest benefit being noted overnight. No severe adverse events were observed, with pump hardware malfunction being the main burden for participants.

These findings are in keeping with real-world data, and the effect sizes were similar to other studies on commercial AID systems. Burnside and colleagues’ study⁷ provides reassurance on the safety of this system, in line with real-world data⁴ and international consensus¹ from health-care professionals. Preliminary data from the follow-up study of users who continued using the AID system beyond the trial supports these findings.⁸ Nine (18%) of the 49 adult participants were from non-White ethnic groups and 14 (28%) were from the lower two quintiles of New Zealand’s deprivation index. Limitations of the study include lower baseline HbA_1c (mean 58·4 mmol/mol [SD 8·5]) compared with the general population of people with diabetes, although this is similar to all commercial pivotal studies.

So where does the CREATE trial take open-source AID systems? With support from existing robust data,⁴ a widely endorsed international consensus,¹ and now a high-quality randomised controlled trial,⁷ there is ample evidence that all AID, including open-source AID, is beneficial for people with diabetes. However, barriers to adoption of open-source options still exist.

There has been resistance from the academic community in recognising user-led approaches as an established form of treatment for people with diabetes. Previous publications have given critical subjective views that were not aligned with objective evidence or widespread consensus.⁶ Similarly, a 2022 general consensus statement on AID affords very little support or detail for open-source systems.² If support from key opinion leaders in the field takes a different tone to the published data, it begs the question of what might be influencing their recommendations for clinical practice and policy making? Among health-care professionals, there have been concerns regarding understanding the technicalities of open-source systems.⁹ Open-source systems rely on a user-led model of learning that disrupts the traditional hierarchy of doctor–patient education. Open-source systems do not feature an industry supported programme of professional or patient education, which has been integral for the implementation of commercial systems. Until these dissonances are resolved, it will be difficult for patient-led approaches that do not have the backing of independent funding or large capital by the medical device industry to support and execute their ideas, run trials, lobby for their regulatory approval, or deliver professional training programmes to be given the status they deserve.

Open-source systems are an example of rapid innovation, where real-world data often precede expensive and time-consuming pivotal studies. Despite the previous absence of a randomised controlled trial, the OpenAPS algorithm has been used successfully for years. The reproducibility of results from previous real-world studies on open-source AID, as well as the CREATE trial,⁷ reaffirms the views presented in the international consensus on open-source AID,¹ which supports the use of real-world evidence in regulatory decisions. It is worth noting that the real-world data showing evidence of benefit were felt to be sufficient by the consensus group,¹ and that the US Food and Drug Administration is evaluating Loop as an open-source AID system on the basis of these data.¹⁰ There are substantial cost benefits to the health industry from using an open-source algorithm that is free and has gone through exhaustive development and user testing with millions of hours of real-world experience before the point of the trial. Because lowering the environmental and financial impacts of treatments is a priority, using a cost-effective approach to advance science and generate technological solutions transparent to users and providers, and validate their safety and effectiveness by using user-driven approaches and real-world data, needs to be upheld, rather than demoted.

Along with the open-source community, and in line with professional consensus,¹ Burnside and colleagues⁷ show that open-source AID systems have a place in the management of diabetes. Indeed, the evidence is out there. The scientific community, health-care systems, and regulatory bodies are now urgently required to understand the potential of user-led innovation and to re-think future approaches for the benefit of society.