The First Regulatory Clearance of an Open-Source Automated Insulin Delivery Algorithm

Abstract

Open-source Automated Insulin Dosing (OS-AID) algorithms are made publicly accessible so that every facet of their operation can be understood. Currently, commercial AID algorithms are kept proprietary trade secrets, despite the role they take in making life and death decisions for people living with type 1 diabetes. Loop was the second OS-AID algorithm, developed initially by Nate Racklyeft and Pete Schwamb. In 2018, the nonprofit organization Tidepool (Palo Alto, CA) announced the launch of the “Tidepool Loop” initiative with the aim to generate real-world evidence and obtain regulatory clearance. By the end of 2020, the U.S. Food and Drug Administration received Tidepool’s application for an interoperable automated glycemic controller based on Loop. After 2 years, the FDA approved the Tidepool Loop on January 23, 2023.

In 2018, the nonprofit organization Tidepool (Palo Alto, CA) announced the launch of the “Tidepool Loop” initiative with the aim to generate real-world evidence and obtain regulatory clearance for the open-source automated insulin delivery (AID) algorithm “Loop.” ¹ By the end of 2020, the U.S. Food and Drug Administration (FDA) received Tidepool’s application for an interoperable automated glycemic controller (iAGC) based on Loop. After 2 years, the FDA approved the Tidepool Loop iAGC on January 23, 2023. ²

Open-source AID algorithms are made publicly accessible so that every facet of their operation can be understood. Currently, commercial AID algorithms are kept proprietary trade secrets, despite the role they take in making life and death decisions for people living with type 1 diabetes. Open-source AID algorithms have received support from in silico analysis, ³ animal trials, ⁴ human observational data,^{5 -7} and more recently randomized clinical trials (RCTs). ⁸ A significant body of evidence now exists and clearly supports that AID systems that can make hundreds of automated insulin adjustments per day benefit time in range and quality of life. Even prior to regulatory clearance or the publication of RCT findings, an international consensus regarding the use of open-source AID by health care providers was published and endorsed by several professional diabetes organizations. ⁹

Loop was the second open-source AID algorithm, developed initially by Nate Racklyeft and Pete Schwamb. The Loop algorithm predicts future glucose by adding together the effect of carbohydrates on increasing glucose, insulin on decreasing glucose, and then two forms of short-term adaptation dubbed glucose momentum and retrospective correction. Insulin delivery is then modulated to attempt to drive the future glucose toward a desired, customizable target. The aggressiveness of the system can be tuned and personalized, and there is complete transparency in its operation.

Distributing an AID system with the Tidepool Loop algorithm will require an interoperable alternate controller enabled (ACE) pump and integrated continuous glucose monitoring system (iCGM). It is important to understand that alternate controller enabled does not imply alternate controller required. While the insulin pump manufacturers Insulet (Acton, MA) and Tandem (San Diego, CA) are both listed as class II ACE pumps, this designation does not require the manufacturer to offer an ACE pump that is compatible with the Tidepool Loop iAGC. True interoperable Bluetooth pumps do exist, eg, the Dana insulin pump series by SOOIL (Yonggudaero, South Korea) but do not have regulatory clearance in the United States. Hence, Tidepool Loop will require business partnerships with sensor and pump manufacturers to form a complete AID system. Despite this situation, this clearance still sets a very strong precedent and becomes yet another landmark step in the field of diabetes technology. Why might this be? The precedent for iAGC was established with the Control-IQ algorithm, and subsequent cleared iAGCs include the Basal-IQ and OmniPod 5 algorithms. For the first time, a nonprofit organization received clearance for an iAGC—even given the challenges through the COVID pandemic. Second, source code of the Tidepool Loop algorithm, meeting notes and documentation for regulatory clearance are all openly available to the public. Furthermore, this is the first time that real-world clinical evidence, rather than a clinical trial, was used to support regulatory clearance of an iAGC. This effort was driven by more than one thousand open-source AID users willing to donate their data for scientific purposes. ⁶ It not only sets a precedent for patient and stakeholder engagement but also for obtaining iAGC clearance without an ACE pump partner. Tidepool also established standard operating procedures SOP-0016 “Tidepool Loop Connected Device Integration and Validation Process and Plan” to bring on pump and CGM partners and SOP-0018 “Tidepool Loop Regulatory Determination Process” to provide a pathway for future algorithm modification.

SOP-0016 and SOP-0018 spell out very specific guidelines for working with partners and making algorithm changes. SOP-0016 means that a full AID system will require direct collaboration with an iCGM and ACE pump manufacturer, which does not currently happen in the open-source AID community. SOP-0018 spells out when another 510(k) submission would be required for algorithm modification. The 510(k) must show that the device is substantially equivalent (in terms of safety and effectiveness) to a previously approved device. This would likely require additional real-world evidence or clinical trials, which are costly.

It must first be made clear that open-source software has a variety of versions and implementations but only one version of Loop has received clearance. Tidepool’s implementation of Loop will be largely open-source, but the components allowing access to partner devices are likely to be obfuscated at the request of the partners. This version is modified from the implementations that were most used in the Loop Observational study. ⁶ User interface and safety changes were added and many of these have carried forward into the fully open-source version of Loop. Conversely, some features of the open-source iterations of Loop were either not implemented at the time of the observational study (eg, the ability for the user to switch between automated corrections boluses or basal rate changes), or not in use by the majority (eg, override presets) and are therefore not part of the Tidepool Loop initial clearance.

The consideration of real-world evidence was a suggestion by the open-source AID international consensus group, ⁹ and a welcome move by the US FDA. While a clearance does not mean FDA support for the existing open-source AID systems, it is certainly reassuring that a subset of the functionality now available and real-world evidence from Loop passed regulatory scrutiny. It supports the notion that the evidence is out there ¹⁰ and requires the medical system and regulatory officials to keep pace. This step helps bring more equality to iAGC developers, who could not previously receive clearance without iCGM and ACE pump partners. However, the #WeAreNotWaiting diabetes movement continues demanding safety and innovation at a faster pace than industry and regulators can currently offer.

The special controls for class II medical devices are defined by the operating characteristics of a predicate device. Ideally, these controls could be supplanted with rigorous criteria guaranteeing safe plug and play operation that could be tested in silico without the need for expensive clinical trials. Moving ahead, interoperability, with clear and secure information sharing standards, should be a requirement and not a suggestion for CGM and insulin pump manufacturers. This may extend beyond the perceived powers of many regulatory bodies (including the FDA) but should be made law by democratic majority and lawmakers given the wider benefits for people with diabetes, health professionals, and care systems.