Closed-Loop Systems (Artificial Pancreas)

Introduction to Closed-Loop Systems

A closed-loop system, also called an automated insulin delivery (AID) system, joins a continuous glucose monitor (CGM) to an insulin pump through an algorithm that decides how much insulin the pump delivers [1]. The CGM reads glucose in the fluid under the skin, and the pump infuses rapid-acting insulin under the skin [1]. "Artificial pancreas" is the popular name, though it oversells the systems on the market. They are hybrids: the algorithm handles background insulin on its own, but the user still tells it about meals [1][6]. In the Control-IQ trial the FDA reviewed, users spent 71% of the time in the target glucose range during the randomized phase, although the FDA noted that time in range is not a validated stand-in for long-term outcomes [9].

Evolution of Algorithms

The algorithm story is messier than a clean handoff from one design to the next. A 2021 review in Diabetologia describes three families of controller [1]. Model predictive control (MPC) uses a mathematical model to predict glucose over a set time window, then picks the insulin delivery that keeps the prediction closest to target. Proportional-integral-derivative (PID) control looks at how far glucose is from target, how long it has been off target and how fast it is changing. Fuzzy logic follows approximate rules that encode what experienced clinicians would do [1].

Here's what surprised us: the market never fully switched from PID to MPC. The same review lists the MiniMed 670G and its successor, the 780G, as PID systems with insulin feedback [1]. The 780G is still being expanded, with an FDA approval for adults with type 2 diabetes in August 2025 [10]. Meanwhile, Insulet's early feasibility studies of its Omnipod hybrid closed-loop system used a personalized MPC algorithm [2], and Tandem's Control-IQ predicts glucose 30 minutes ahead to adjust basal insulin every 5 minutes [9]. Prediction matters because insulin infused under the skin takes time to be absorbed, so a controller that only reacts is always a step behind [1]. For a side-by-side look, see our article on PID vs. MPC algorithms.

Key Innovations

Several developments shaped the systems sold as of October 2026:

• FDA device classifications for interoperable parts: The FDA created a separate device type for each piece of the loop: the integrated CGM (iCGM) [3], the alternate controller enabled (ACE) pump [14] and the interoperable automated glycemic controller (iAGC), which is the dosing software [15]. These are classifications with special controls, not interoperability standards. Each one is defined as built to communicate reliably and securely with digitally connected devices, and that requirement is what lets one company's algorithm run with another company's sensor [3][14][15]. Control-IQ's De Novo request is the record that classified the iAGC device type [9].
• Tandem's Control-IQ+: Tandem's pumps now run Control-IQ+, the successor to the earlier Control-IQ [4]. The FDA cleared Control-IQ+ in April 2026 for type 1 diabetes from age 2 and type 2 diabetes from age 18 [11]. It raises, lowers or suspends basal insulin from CGM readings and predicted glucose values, and it can deliver automatic correction boluses [11].
• Insulet's Omnipod 5: Its pivotal trial called it the first tubeless, on-body AID system, with the algorithm running inside the adhesive Pod itself [16]. The Omnipod 5's SmartAdjust algorithm adjusts insulin from current and predicted glucose values. Since August 2024 it has been cleared for type 1 diabetes from age 2 and for type 2 diabetes in adults [5].
• MiniMed 780G: MiniMed's PID-based system [1] added an indication in August 2025 for adults with type 2 diabetes who need insulin [10].
• Beta Bionics' iLet: The iLet software needs only the user's body weight to start. It doesn't ask for basal rates, carb ratios or correction factors, and it keeps adapting to the user's insulin needs [6]. It is cleared for type 1 diabetes from age 6 [6].
• twiist: The twiist AID System is an ACE pump with built-in dosing software. The FDA cleared its current version in June 2026 for people ages 6 and up [12]. Its maker lists the FreeStyle Libre 3 Plus and Eversense 365 as compatible sensors and describes it as a system for type 1 diabetes [13].

Frontier of Technology

Dual-hormone systems add glucagon, the hormone that raises glucose, so the system can push back against a low instead of only cutting insulin [1]. In a home-use trial of a bihormonal bionic pancreas, 43 adults with type 1 diabetes spent 78% of the time in range over 11 days, compared with 62% on pump therapy alone [1][7]. Glucagon is the hard part. In solution it quickly forms amyloid fibrils, even within the first day, and the gels that result can block pump catheters; it also degrades chemically over time [8]. When the 2021 review was written, no room-temperature-stable glucagon was approved for continuous delivery under the skin, and no dual-hormone system was on the market [1]. The iLet the FDA cleared in 2023 doses insulin only [6]. Our article on dual-hormone AID systems follows that research further.

Pitfalls and Challenges

Several limits remain, and the FDA's review of Control-IQ names many of them [9]:

• PK/PD Mismatch: Pharmacokinetics (how insulin is absorbed) and pharmacodynamics (how fast it acts) don't keep pace with the sensor. Insulin given under the skin is absorbed with a delay, which is why glucose after meals is still hard for closed-loop systems to control [1]. The FDA also lists, as a risk, a controller's inability to handle insulins with different PK/PD profiles, and addresses it partly through drug compatibility information in the labeling [9].
• Connectivity Failures: The FDA names loss of communication with connected devices as a risk. It notes that connectivity problems between interoperable devices can cause a temporary loss of Control-IQ use [9]. Insecure data transmission and wireless interference are on the same list [9].
• Meal Announcements: The 2021 review found that accurate carb counting and pre-meal boluses are still needed for the best results; attempts to simplify meal boluses or go fully closed-loop gave worse glucose control [1]. Even the iLet, which drops carb counting, requires the user to announce each meal and describe it as "Usual," "More" or "Less" than usual for that meal [6]. How a system is set up around meals is a decision to make with your care team.

Conclusion

As of October 2026, US closed-loop systems include MiniMed's 780G, Tandem's Control-IQ+, Insulet's Omnipod 5, Beta Bionics' iLet and twiist [5][6][10][11][12]. At least three of them now carry type 2 indications for adults [5][10][11]. All five dose insulin alone [5][6][10][11][12], and none removes the need to tell the system about meals [1][6]. Which system fits a given person, and how it's configured, is a decision to make with your care team.