The long-term goal of a fully automated insulin delivery (AID) system that operates without user input to maintain glucose levels within target range remains a future aspiration and may never be fully realized using insulin alone, but steady progress continues toward developing systems that are “good enough.”
Current AID Systems
An AID system consists of an insulin pump and a continuous glucose monitor (CGM), paired with an algorithm — often housed on a smartphone app — that enables the devices to communicate and automate insulin delivery. While such systems are most commonly used by those with type 1 diabetes (T1D), evidence suggests benefit for people with insulin-treated type 2 diabetes as well.
Currently, all commercially available AIDs are termed “hybrid closed-loop (HCL)” and require varying degrees of user input for operation. This can lead to dosing errors, particularly when managing food and exercise. Current fast-acting insulins administered subcutaneously still aren’t rapid enough to prevent post-meal glucose spikes, while unplanned exercise can cause glucose levels to drop too quickly for the system to correct.
Despite these limitations, current HCL systems achieve better glycemic control than nonconnected devices and are continually advancing. “The main message is to take these technologies and deliver them to people. We shouldn’t wait for perfection. I think we should expect incremental changes,” AID technology Pioneer Roman Hovorka, PhD, professor of metabolic technology research at the University of Cambridge, Cambridge, England, said in a plenary talk at the Advanced Technologies & Treatments for Diabetes (ATTD) 2025 meeting.
David M. Maahs, MD, the Lucile Salter Packard Professor of Pediatrics and Division Chief of Pediatric Endocrinology at Stanford University and the Lucile Packard Children’s Hospital, Palo Alto, California, told Medscape Medical News, “Hybrid closed-loop systems still require you to put in carbs, preferably before you eat, and give that bolus. There are people who don’t give the bolus and the systems don’t do as well as they would if they did. But it’s still better than an old pump without the algorithm in it to give extra insulin if you go high and to cut insulin if you’re predicted to have a low.”
This “gap” in the loop, Maahs noted, is steadily narrowing and will continue to do so in the future, thanks to smarter algorithms powered by artificial intelligence and the development of faster insulins. “I think it will always be the case that if you have some user input, you’ll probably always do a little better. But I think we’ll get to the point where people might say it’s good enough if it’s above 70% time in range. And these systems are going to continue to get better and better and smarter and smarter.”
AID Technology
The field has been advancing since 2013, with the first US approval of Medtronic’s pump-CGM system that could suspend insulin delivery when blood glucose dropped below a certain level. Next came the “predictive low glucose suspend” feature with Medtronic’s 640G, misleadingly labeled the “world’s first artificial pancreas.” More recently, systems such as the Medtronic 680G also automatically correct for high glucose levels. Some systems require only meal “announcements” rather than entering a carb count, as well as exercise notifications.
There are currently seven AID systems commercially available in the United States, each using different algorithms and with differing customization and automation capabilities: Medtronic’s 670G and 680G, Omnipod 5, Tandem Control IQ, Beta Bionics iLet, Tidepool Loop, and Sequel Twiist. The first four are also available in Europe, where two others, the CamAPS FX, and Diabeloop DBLG1, are also approved. All but the Diabeloop are available for pediatric patients of varying ages, most commonly 6 years or older.
There are also open-source do-it-yourself algorithm systems built by users, including Tidepool Loop (approved by the US Food and Drug Administration in 2023), Trio, and Android APS. These systems sometimes outperform the commercial ones due to more aggressive algorithms. While none are “fully closed,” some are being investigated for that use.
Despite any limitations, AIDs are increasingly viewed as standard of care for people with T1D. On March 18, 2025, a “call to action” published in Diabetes Technology & Therapeutics said that AIDs should be offered to all patients at the time of T1D diagnosis or soon thereafter, or the reasons for not giving that choice should be documented in the medical record.
The document also said that the choice of specific device should be based on the individual patient’s circumstances, preferences, and needs, and that national healthcare systems should ensure unfettered access to AID systems. “This is a powerful statement. There is still a long way to go, but the whole field is aiming in this direction,” Hovorka said.
New Data From ATTD
Neale D. Cohen, MBBS, head of the Clinical Diabetes Research group at the Baker Heart and Diabetes Institute, Melbourne, Victoria, Australia, presented results from the CLOSE-IT (Closed Loop Open SourcE In Type 1 diabetes) trial, for which the protocol was published in 2024. The study used an open-source algorithm called “oref” on an Android phone, along with a Dexcom G6 CGM and Ypsomed insulin pump in 75 adults with T1D.
After a run-in period, patients were randomized to either the FCL group or to the control HCL group. The FCL protocol required no manual meal bolusing unless blood glucose levels exceeded 270 mg/dL for more than an hour. The control HCL group used the same system but required carbohydrate counting and manual bolusing. Both groups consumed similar amounts of carbohydrates with no restrictions.
The primary outcome, time in range (70-180 mg/dL on days 155-168), was 69% for HCL group and 66% for FCL group, meeting the noninferiority threshold (P = .009). A secondary endpoint, A1c, showed no significant difference between HCL and FCL groups (6.8% vs 6.9%, respectively). There were no significant differences in time in tight range (70-140 mg/dL) and no excess levels of hypoglycemia in either group.
“I think it’s remarkable that we were able to show that it was equally as good. It comes down to the algorithm. It’s a very aggressive algorithm and you can make it more intense at certain times of the day,” Cohen told Medscape Medical News.
Results from a two-center, randomized crossover phase 2 open-label study of the CamAPS HS algorithm, using faster-acting aspart (Fiasp) insulin, were presented by Nithya Kadiyala, MBBS, a clinical research associate at the University of Cambridge. The study involved 24 adolescents with T1D and a baseline A1c > 7.5% (58 mmol/mol).
Two 8-week periods of FCL with no manual meal announcements or bolusing resulted in a 45.2% time in range (70-180 mg/dL, 3.9-10.0 mmol/mol) compared with 32.3% time in range with standard insulin pump plus CGM, a significant difference (P < .001). Time above 180 mg/dL was significantly lower (P < .001) while time in hypoglycemia range didn’t differ.
“Perhaps you’re thinking 45% time in range achieved with a fully closed-loop [FCL] is well below the recommended guideline target of 70%. However, it’s still a significant improvement in a group that struggles to meet glycemic targets. And of course, we have to remember that this was achieved whilst removing the practical and psychological burdens associated with having to bolus for meals, greatly improving participants’ quality of life,” Kadiyala pointed out.
Also under study and discussed at the meeting were various methods of enhancing the systems by adding hormones such as glucagon to prevent or reverse hypoglycemia, and/or pramlintide (an amylin analog) to minimize postmeal highs, or using adjunctive medications such as glucagon-like peptide 1 receptor agonists to reduce carbohydrate intake, or adding the more rapid-acting inhaled insulin. “I think the area of trying to help the fully closed-loop system by co-administering something is still very much resonating,” Hovorka said.
Beyond Devices to ‘Ecosystem’
Hovorka described closed-loop technology as an “ecosystem” extending beyond the devices and algorithms to include data portals, clinical support, remote monitoring, supply management, reimbursement, procurement, training materials, onboarding process, and customer support.
“These are all the aspects which contribute to a closed-loop system to be successful. If something falls out of this, it just fails. It is a complex issue which needs to be addressed by the manufacturers and the community as well. We want people with type 1 diabetes to have the systems they want and which are the best for them. I think this is what we are all striving for.”
Hovorka received research support from, is on the speaker’s bureau for, and/or received license fees from Abbott Diabetes Care, Dexcom, Ypsomed, and B. Braun, He is the director of CamDiab, and holds patents in the field.
Maahs had received research support from and/or has consulted for NIH, JDRF, NSF, the Helmsley Charitable Trust, Abbott, Aditxt, Lifescan, Mannkind, Sanofi, Novo Nordisk, Eli Lilly, Medtronic, Insulet, Dompe, Biospex, Provention Bio, and Bayer.
Cohen is on the advisory board for, is a speaker for, and/or receives research support from Novo Nordisk, Lilly, Boehringer Ingelheim, Abbott, Medtronic, AstraZeneca, Roche, Novartis, Sanofi, Endogenex, NHMRC, and Telematics Trust.
Boughton is a consultant for CamDiab and has received speaker fees and/or research support from Ypsomed, ABCD Ltd., Dexcom, Abbott Diabetes Care, and CamDiab. She is an associate editor of Diabetologia.
Kadiyala had no disclosures.
