CLOSE IT: open-source AID without meal announcement vs hybrid closed loop

Every automated insulin delivery system sold today is a hybrid closed loop: the algorithm handles the background, but you still have to tell it about food. Count the carbs, enter them, take the bolus. It is the single biggest chore left in the day, and the single biggest source of error — a mis-estimated plate of pasta shows up as a spike two hours later. The obvious question is whether the algorithm could just deal with meals itself. CLOSE IT is the first randomized trial to test that on an open-source system.¹

What was tested. Participants used "Lotus" — a locked build of the open-source oref1 algorithm, the same OpenAPS code that also powers AndroidAPS — paired with a customized Ypsomed pump and a Dexcom G6 sensor. Everyone spent 12 weeks on it as a normal hybrid closed loop, announcing meals as usual. Then they were randomized: half carried on announcing meals, half switched the announcements off completely for 12 weeks and let the algorithm's aggressive micro-bolusing (super micro-boluses, or SMB, plus unannounced-meal detection) handle food on its own.¹

Who it was for. 73 adults with type 1 diabetes, aged 18 to 70, at sites in Australia and New Zealand — all of them already competent, established open-source AID users before randomization. 36 went to the no-announcement arm, 37 stayed on hybrid closed loop.¹

Design. Open-label, parallel-arm, and — importantly — a non-inferiority trial, not a superiority one. The question was never "is skipping meal announcements better?" It was "how much do you give up?" The investigators set the acceptable loss in advance at 7 percentage points of time in range, and measured the final 14 days of the trial phase against the final 14 days of run-in.¹

Key results. Time in range (70-180 mg/dL; 3.9-10.0 mmol/L) fell from 69 ± 11% to 66 ± 8% in the no-announcement arm, while the hybrid closed-loop arm went from 70 ± 9% to 69 ± 13%. The adjusted difference was -2.2 percentage points (95% CI -6.2 to 1.7) — roughly half an hour a day, and the confidence interval stayed inside the 7-point margin, so non-inferiority was met. The authors conclude the system without meal announcement achieved a similar time in range to the same system used as a hybrid closed loop.¹

How to read this honestly. Three caveats matter.

First, this was non-inferiority, not superiority. The point estimate still favours announcing meals. What the trial shows is that the cost of stopping is small and, on the evidence, acceptable — not that you gain anything glycemically. What you gain is the hours of your life spent counting carbs.

Second, the tested system was not the app you would download. Lotus is a locked oref1 build, functionally equivalent to AndroidAPS but deliberately excluding the newer dynamic-ISF variants that many people run today. It is direct evidence for the OpenAPS/AndroidAPS unannounced-meal approach, and only indirect evidence for Trio and iAPS, which commonly ship dynamic-ISF derivatives this trial never tested. It also used a customized pump, not a stock consumer one.

Third, the published abstract reports no safety outcomes. No severe hypoglycemia or DKA figures are given, so this record makes no claim about them. Removing meal announcements means the algorithm must chase a rise it did not see coming, and hypoglycemia risk is exactly the thing a reader would want quantified. Until the full results are read in detail, treat safety as unreported rather than reassuring.

Why it matters anyway. Every gap on the artificial-pancreas roadmap that meal announcement papers over — the algorithm not knowing what you ate (missing context), and the algorithm having to be smart enough to react without being told (algorithm intelligence) — is exactly what a no-announcement system has to solve. CLOSE IT says a free, community-written algorithm running on faster insulin than the analogues of a decade ago can very nearly get there today. The commercial fully-closed-loop systems in development are chasing the same target with far more money behind them; the open-source community got a randomized answer first.

References

1. Wilkinson T, Nanayakkara N, Burren D, Jenkins A, Williman J, Keesing C, Boyle E, Elghattis Y, Bennett A, Gunn T, de Bock M, Cohen ND. Glycemic Outcomes During Outpatient Use of Automated Insulin Delivery Without Meal Announcement in Adults with Type 1 Diabetes: Results from the CLOSE IT Randomized Controlled Trial. Diabetes Technology & Therapeutics (2026), epub 12 Feb 2026. https://pubmed.ncbi.nlm.nih.gov/41681122/ ↩ ↩² ↩³ ↩⁴ ↩⁵