Automated insulin delivery (AID) systems have become the standard of care for type 1 diabetes (T1D), offering substantial improvements in glucose management 1, 2. Two main categories of AID algorithms co-exist: closed-source proprietary and open-source AID (OS-AID) 3. All current open-source AID algorithms began as customizable tools/algorithms developed, refined and shared by the diabetes community 2. OS-AID algorithms can be self-built by people with T1D (PwT1D) for use with a variety of glucose sensors and pumps. Our previous study suggested that OS-AID algorithms (mainly Loop) were non-inferior to commercial AIDs based on a full-group analysis that included users of both the Medtronic MiniMed 670/770G and the Tandem Control-IQ systems, in time spent in optimal range (TIR% 3.9–10.0 mmol/L) 4. However, it is essential to evaluate OS-AID directly against newer commercial systems equipped with automatic bolus feature 5, 6, which represent the mainstay of AID therapy. Such comparison is particularly timely as Tidepool Loop, a modified version of the Loop OS-AID algorithm, was cleared by the U.S. Food and Drug Administration (FDA) and is currently being marketed 7. Here, we perform a subanalysis of our previous study 4 to compare Loop with Tandem Control-IQ, a widely-used commercial closed-source AID algorithm among adults with T1D. Our previous study was a prospective, observational, non-inferiority, non-randomized, real-world, comparative study conducted between April 2021 and September 2022 (ISRCTN11362144) 4. In brief, main inclusion criteria included adults with T1D for more than 1 year and using an AID for more than 3 months (any open-source AID, Tandem with Control-IQ or Medtronic Minimed 670/770G AIDs). The duration of the study was 12 weeks, with weeks 0 to 4 of additional blinded continuous glucose monitoring (CGM, Dexcom G6, Dexcom, San Diego, CA, USA) data collection, followed by weeks 4 to 12 of observation. The duration of blinded CGM was chosen based on consensus guidelines deeming that at least 2 weeks of CGM data is sufficient for CGM metric calculation and clinical interpretation 8, and to minimize participant burden associated with wearing an additional study device alongside their insulin pump and CGM. The study included five visits, which were done either in person at the Montreal Clinical Research Institute (IRCM) or using a secured virtual format. The initial study's flowchart is shown in Figure S1. Throughout the study, participants continued using their own AID systems based on their personal CGM, with all settings customized according to personal preference by participants themselves 4. For this subanalysis, the primary outcome was TIR% based on the study CGM's glucose values from weeks 0 to 4. Secondary outcomes included other CGM-derived metrics (singular and composite glucose outcomes as recommended 8) during daytime (06:00–24:00), nighttime (00:00–05:59) and 24 h 8. Safety outcomes included percentage of time spent below range (TBR% 10 mmol/L and > 13.9 mmol/L and CV. However, mean TBR < 3.9 mmol/L was higher among autobolus users compared with non-users (4.2% vs. 1.9%, p = 0.016). Moreover, no difference was observed in the aforementioned CGM variables when comparing non-autobolus Loop-AID participants and Tandem Control-IQ participants. Finally, when comparing CGM metrics of Loop participants using the autobolus feature and Tandem users, we found differences only in TBR < 3.9 mmol/L (4.2% vs. 2.1% respectively, p < 0.001). No severe hypoglycemic or DKA event occurred in either group during the study. A total of 55 adverse device events were reported among 50 participants over the 12 weeks, with similar rates between the Loop OS-AID and Tandem Control-IQ groups (Table S3). Our study was unable to detect statistically significant differences in TIR% or hyperglycemia between Loop OS-AID and Tandem Control-IQ users, even after adjusting for confounding factors. Loop OS-AID users had lower mean glucose levels when compared to Tandem Control-IQ users. On average, both groups met all recommended glucose targets, including TIR%, TAR% and TBR% 8. However, our findings show greater TBR% (< 3.9 mmol/L) for the Loop-AID group compared with the Tandem Control-IQ group (3.1% vs. 2.3%). Reasons may include a lower glucose target leading to more aggressive insulin delivery, a lower level of fear of hypoglycemia as shown in our previous study, and greater knowledge of diabetes management. These could lead to a more lenient approach to glucose management, increasing hypoglycemia risks. Nevertheless, TBR% < 3.9 mmol/L remained below the recommended threshold in both groups, and time spent in critical hypoglycemia (< 3.0 mmol/L) did not differ between the two groups and was within the recommended threshold. Further, both reported no serious adverse events, minimal adverse device effects and high percentages of time spent in automated mode (≥ 95%). Collectively, these findings provide reassuring evidence regarding the effectiveness and safety of both systems. Limited evidence exists on the comparison between OS-AIDs and commercial closed-source AIDs. Existing studies suggested that OS-AIDs achieved better glucose outcomes when compared to both sensor-augmented pump therapy 9 and Medtronic 670G 10 for glucose management. Only two observational studies have used newer AIDs with automatic bolus feature as comparators (both used Tandem Control-IQ and compared to AndroidAPS OS-AID). They found that AndroidAPS OS-AID achieved similar glucose regulation when compared to Tandem Control-IQ 11, 12. These results align well with our findings. Randomized controlled studies with cross-over design comparing different algorithms for the same individual are needed to further assess efficacy and safety. Notably, the versions of Loop OS-AID used in this sub-analysis were all from before September 2022. However, Loop OS-AID still achieved numerically higher TIR% than Tandem Control-IQ. First, it is plausible that, with a larger sample size to increase statistical power and with more advanced Loop algorithms, a statistically significant superiority in TIR% over Control-IQ may be observed. In addition, inherent differences in system functionality and user engagement may contribute to this trend. Specifically, this may be due to the lower glucose targets chosen by Loop users and the system's highly customizable settings, which allow tailoring to individual preferences and lifestyles 4, 13. Further, the Loop OS-AID algorithm delivers automatic boluses more frequently than Tandem Control-IQ (every 5 min vs. every hour) 1, allowing a more aggressive response to hyperglycemia. Yet this may have contributed to the slightly higher hypoglycemia observed in the Loop OS-AID group. However, no difference in TIR% was observed between autobolus and non–autobolus Loop users, nor between non–autobolus Loop users and Tandem Control-IQ users. These findings should be interpreted cautiously, as the analyses are limited by a small sample size. More recent evidence 14 suggests that PwT1D can achieve TIR consensus goals without meal announcement on an OS-AID, something not yet possible with current commercial AIDs. With the rapid iteration among AID algorithms (particularly open-source), studies will need to be conducted on a regular basis to provide the most up-to-date evidence. This is the first study to compare Loop OS-AID (Loop) with a commercially available AID that also allows for automatic bolus feature (Tandem Control-IQ). Of note, a newer version of Loop has been recently cleared by the FDA. Strengths of our study include its prospective design, along with nationwide recruitment and the use of a blinded standardized CGM for 4 weeks. Several limitations must be acknowledged. First, although prospective, the design remains observational and causal inference is not possible. Users selected their own AID system, which conferred a selection bias. Moreover, OS-AID users may have a higher technology literacy as well as health engagement that could account for the present results and may limit the generalizability of our findings to the broader population. Secondly, the sample size was not powered to detect the difference in TIR% between Loop OS-AID and Tandem Control-IQ. However, on average, both groups successfully met recommended glucose targets, reinforcing that both systems are effective in glucose regulation. Third, we did not collect information on the frequency of system settings adjustment by healthcare professionals, which could potentially affect study outcomes. In conclusion, our study did not identify statistically significant differences in major CGM metrics between Loop OS-AID and Tandem Control-IQ. Both systems can help users achieve TIR consensus targets and demonstrate safety in real-life conditions. Healthcare professionals are encouraged to support the use of all AID systems among PwT1D who show interest. Z.W. and M.L. share co-first authorship. Z.W. designed the study, collected the data, conducted the statistical analysis, interpreted the results and wrote the manuscript. M.L. interpreted the results and contributed to data collection. A.B. contributed to data collection. R.R.-L. supervised the project. All authors critically reviewed/edited the manuscript. All authors read and approved the final version of the manuscript, and accepted responsibility for the decision to submit for publication. Authors would like to thank Josee Gagnon at the Montreal Clinical Research Institute for monitoring the data and Danijela Bovan at the Montreal Clinical Research Institute for coordinating the study. Authors are grateful to study volunteers for their participation, and to patient-partners from the BETTER patient-partner team for helping authors refine study scope. The authors would also like to thank both Canadian Institutes of Health Research for the Foundation Grant (#148464) and Steinberg Foundation for their donation to conduct this study. This work was supported by a Foundation Grant from the Canadian Institutes of Health Research (#148464) and a private donation from Steinberg foundation, both held by R.R.L. Z.W. received grant funding from Canadian Institutes of Health Research, Breakthrough T1D, Diabetes Canada and Société Francophone du Diabète. M.L. received a speaker fee from Sanofi. R.L. has received consulting fees from Abbott Diabetes Care, Adaptyx Biosciences, Biolinq, Capillary Biomedical, Deep Valley Labs, Gluroo, PhysioLogic Devices, Portal Insulin, Sanofi and Tidepool. He has served on advisory boards for ProventionBio, Lilly and Rezolute. He receives research support from his institution from Insulet, Medtronic, Sinocare and Tandem. V.M. received purchase fees: E Lilly (automated insulin delivery system). A.-S.B. is a research scholar from Fonds de recherche du Québec en Santé. She has received financial support from Canadian Institutes of Health Research and Breakthrough T1D Canada. RRL: Research grants: Diabetes Canada, E Lilly, Cystic Fibrosis Canada, Canadian Institutes of Health Research, Fondation Francophone pour la Recherche sur le Diabète, Janssen, Breakthrough T1D, Merck, NIH, NovoNordisk, Societe Francophone du Diabete, Sanofi-Aventis, Vertex Pharmaceutical. Consulting/advisory panel: Abbott, Astra-Zeneca, Bayer, Boehringer I, Dexcom, E Lilly, HLS therapeutics, INESSS, Insulet, Janssen, Medtronic, Merck, Novo-Nordisk, Pfizer and Sanofi-Aventis. Honoraria for conferences: Abbott, Astra-Zeneca, Boehringer I, CPD Network, Dexcom, CMS Canadian Medical&Surgical Knowledge Translation Research group, E Lilly, Janssen, Medtronic, Merck, Novo-Nordisk, Sanofi-Aventis, Tandem and Vertex Pharmaceutical. Consumable gift (in Kind): E Lilly, Medtronic. Unrestricted grants for clinical and educational activities: Abbott, E Lilly, Medtronic, Merck, Novo Nordisk and Sanofi-Aventis. Patent: T2D risk biomarkers, catheter life. Purchase fees: E Lilly (automated insulin delivery system). The remaining authors have no conflicts of interest to declare. The anonymized data that support the findings of this study are available from the corresponding author upon reasonable request. The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/dom.70705. Figure S1: Study flowchart. Table S1: Glucose target and automatic bolus feature used by OS-AID participants. Table S2: Comparison of glucose outcomes (daytime and nighttime). Table S3: Adverse events and effects. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.
Wu et al. (Mon,) studied this question.