Integration with Automated Insulin Delivery (AID) Systems

Introduction to Automated Insulin Delivery (AID) Systems

An automated insulin delivery (AID) system, which the NIH also calls an artificial pancreas, has three parts: a continuous glucose monitor (CGM), an insulin pump, and a program on the pump or a smartphone that calculates and delivers insulin [1]. Linking the CGM to that program is what turns the sensor from a display into the input that drives dosing.

The first step in that direction was modest. In a 2013 randomized trial of 247 people with type 1 diabetes, a pump that paused insulin when the sensor reached a preset low threshold had 31.8% fewer nighttime low-glucose events than a standard sensor-augmented pump, with no rise in HbA1c (a lab measure of average glucose over about three months) [2]. Current systems do much more. Control-IQ, the earlier version of the algorithm Tandem now sells as Control-IQ+, predicts glucose 30 minutes ahead, can raise, lower or pause basal (background) insulin every 5 minutes, and can give automatic correction boluses [3][6].

Key Market Players and Technologies

Most AID systems are hybrid closed-loop (HCL) systems: the software handles background insulin, but the user still acts at meals [1]. The control algorithms differ. Model predictive control (MPC) forecasts glucose over a time window and chooses the delivery that keeps the forecast closest to target. A proportional-integral-derivative (PID) controller reacts to how far glucose is from target, how long it has been off, and how fast it is changing [4]. A 2021 review lists Medtronic's 670G and 780G as PID systems with insulin feedback and the Cambridge algorithm behind CamAPS FX as an adaptive MPC; we compare the two approaches in PID vs. MPC Algorithms in Automated Insulin Delivery [4].

What lets a sensor from one company drive a pump from another is an FDA device class. An integrated CGM (iCGM) is designed to "reliably and securely transmit glucose measurement data to digitally connected devices, including automated insulin dosing systems" [5]. As of October 2026, the main pairings look like this:

• Tandem & Insulet use both Dexcom and Abbott sensors. Tandem's t:slim X2 pump works with Dexcom sensors and the FreeStyle Libre 3 Plus; the Tandem Mobi works with Dexcom sensors only [6]. Insulet's Omnipod 5 works with the Dexcom G7, Dexcom G7 15 Day, FreeStyle Libre 3 Plus and FreeStyle Libre 2 Plus [7]. It also still supports the Dexcom G6, which Dexcom stopped manufacturing after July 1, 2026 [7][8].
• MiniMed is the name Medtronic's diabetes business now uses, and it trades on Nasdaq as MMED [9][10]. Its ecosystem is no longer all in-house. The MiniMed 780G works with MiniMed's own Simplera Sync and Guardian 4 sensors and with the Instinct sensor, which Abbott makes and which is designed exclusively for MiniMed systems [10][11].
• Beta Bionics takes a different approach with the iLet Bionic Pancreas. It doesn't use carb counting or carb-to-insulin ratios and starts from the user's body weight. It still isn't hands-off: the user announces each meal (breakfast, lunch or dinner) and rates it as "Usual", "More" or "Less" than normal for that meal [12].

Effectiveness and Benefits

The strongest evidence comes from randomized trials, and they show gains in Time in Range (TIR), the share of the day spent between 70 and 180 mg/dL. In Control-IQ's six-month trial of 168 people with type 1 diabetes aged 14 to 71, TIR rose from 61% to 71% with the closed-loop system and stayed at 59% with a sensor-augmented pump [13]. Time below 70 mg/dL and time above 180 mg/dL both improved compared with the control group, and HbA1c fell 0.33 percentage points more [13].

Here's what surprised us about the iLet trial: the system was set up with nothing but each person's body weight [14]. Across 326 people aged 6 to 79, HbA1c fell from 7.9% to 7.3% over 13 weeks with the bionic pancreas and stayed at 7.7% with standard care [14]. Severe hypoglycemia ran 17.7 events per 100 participant-years with the bionic pancreas and 10.8 with standard care, a difference that wasn't statistically significant [14].

Burden matters too. In the FDA's review of Control-IQ, users reported less diabetes-related distress and less fear of hypoglycemia, and the agency noted the system's ability to suspend insulin overnight to lower the risk of nighttime lows [3].

Critical Issues and Pitfalls

An AID algorithm can only act on what the sensor reports. The FDA's Control-IQ review names the failure paths: erroneous CGM data can cause inappropriate insulin delivery or a lack of delivery, and connection problems between devices can temporarily stop automation [3]. Three issues come up again and again:

• Interstitial lag. A CGM reads glucose in interstitial fluid, the fluid around cells under the skin, not in blood. In a Mayo Clinic study of eight healthy adults after an overnight fast, glucose took about 5 to 6 minutes to move from blood into that fluid [15]. The study didn't test fast-changing conditions such as meals or exercise, so it doesn't tell us how large the gap gets then. Predictive algorithms try to look ahead; Control-IQ, for example, forecasts 30 minutes out [3].
• Compression lows. Lying on a sensor can make it read falsely low. Healthy volunteers wearing four sensors each overnight showed sudden aberrant drops that were strongly tied to lying on the sensor, which the authors attributed to reduced local blood flow from tissue compression [16]. The algorithm acts on the reading it receives. The FDA lists hyperglycemia and diabetic ketoacidosis (DKA) among AID risks and names erroneous CGM data that leads to a lack of insulin delivery as one cause [3].
• Bluetooth dropouts. On the MiniMed 780G, the sensor's transmitter sends a reading to the pump over Bluetooth Low Energy (BLE) every five minutes [17]. When the sensor data aren't good enough for Auto Mode, the pump delivers a patient-specific "Safe Basal" rate for up to 90 minutes and switches to Manual Mode if the problem doesn't clear [17]. Systems handle the gap differently: the iLet lets the user enter fingerstick readings to keep dosing for a limited time [12]. The FDA's Control-IQ review also counts data transmission interference and electromagnetic disturbance among the identified risks [3].

How to respond to any of these alerts is a decision to make with your care team.

Conclusion

As of October 2026, sensor choice has widened: Tandem, Insulet and MiniMed pumps each work with at least one Abbott-made sensor [6][7][10]. The weak point hasn't moved, though. Every algorithm doses on what the CGM reports. The direct measurement of interstitial lag we cite was made in fasting, healthy adults [15], so how much that lag matters during meals is still an open question.