FDA Authorization of Automated Insulin Delivery Systems: iCGM, ACE Pumps, iAGC Controllers and Supporting Clinical Evidence

How FDA authorizes automated insulin delivery: iCGM, ACE pumps, iAGC controllers and the clinical evidence behind them

FDA now regulates most automated insulin delivery (AID) systems as interoperable components rather than single integrated systems. Three De Novo classifications granted between 2018 and 2019 set up that framework. Each one created a Class II device type with special controls: the integrated continuous glucose monitoring system (iCGM), the alternate controller enabled (ACE) infusion pump, and the interoperable automated glycemic controller (iAGC). Each later sensor, pump and algorithm can then reach the market through its own 510(k). Before this framework, FDA approved AID-type systems as Class III premarket approval (PMA) products, and some Medtronic components still have PMA histories. This article covers the regulatory architecture, the special controls, and the clinical evidence FDA relied on at each stage.

From Class III artificial pancreas systems to Class II interoperable components

FDA's 2012 guidance on IDE and PMA content for artificial pancreas device systems (APDS) grouped systems by control strategy. Threshold suspend systems reduce or suspend insulin at or near a low threshold. Control-to-range systems act only when glucose approaches upper or lower limits. Control-to-target systems aim for a set glucose level at all times, with insulin-only, bihormonal and hybrid (patient-administered meal bolus) variants ^{447448449450451}. The guidance listed the APDS product codes as OZO (threshold suspend), OZP (single hormonal control) and OZQ (bihormonal control), all Class III ⁴⁵².

PMA approvals under that model included:

The three De Novos below replaced that single-system model with interoperable Class II device types:

Integrated CGM (iCGM): DEN170088, Dexcom G6

Indication and classification

The Dexcom G6 was authorized for real-time CGM in people aged 2 years and older. Its indication covers replacing fingerstick testing for treatment decisions and use alone or with digitally connected devices, including automated insulin dosing systems ¹⁰³. FDA classified the device type as Class II under 21 CFR 862.1355 and created product code QBJ ¹⁰⁴¹⁰⁵.

Special controls: prespecified accuracy thresholds

The iCGM special controls are unusual because they write numerical accuracy acceptance criteria into the classification. Under special control (1), "Design verification and validation," the adult clinical-study requirements are expressed as lower one-sided 95% confidence bounds ³²³³²⁴:

The control also prohibits large opposite-direction errors. No corresponding blood glucose value may read above 180 mg/dL when the iCGM reads below 70 mg/dL, and none may read below 70 mg/dL when the iCGM reads above 180 mg/dL ³³². Other elements of the review covered interoperability (interface specifications, data exchange and responsibilities for postmarket reporting), cybersecurity, and labeling under 21 CFR Parts 801 and 809 ^4041424344.

Clinical evidence

Two prospective US studies at 11 centers enrolled 380 people aged 2 years and older with type 1 or insulin-using type 2 diabetes, and 324 completed ¹³⁹¹⁴⁰. The pivotal study used a manual applicator. A substudy checked whether the marketed automated applicator performed comparably. Participants wore sensors for up to 10 days, and accuracy was measured against laboratory blood glucose. In participants aged 13 and older, investigators deliberately raised and lowered glucose to cover the 40 to 400 mg/dL range ¹⁴¹¹⁴².

• Overall, 89.9% of 25,101 paired readings fell within 20% of reference, with a reported 95% lower bound of 88.7% ¹⁴³.
• In adults, 88.5% of 1,920 iCGM readings below 70 mg/dL were within 15 mg/dL. In the 70 to 180 mg/dL and above-180 mg/dL ranges, 74.1% and 85.5% were within 15% ¹⁴⁴.
• In pediatric participants, only 48.9% of 90 readings below 54 mg/dL were within 15 mg/dL of comparator ¹⁴⁵¹⁴⁶.
• MARD was 10.7%, 9.2% and 9.6% at the beginning, middle and end of wear ¹⁴⁷.
• 84% of 374 evaluated sensors lasted through day 10, and 9.6% shut off early ¹⁴⁸.
• A separate acetaminophen interference study (70 adults, 1,000 mg dose) found MARD of 7% before dosing and 8% after ^149150151.

FDA concluded that probable benefits outweighed probable risks under general and special controls and classified the device type as Class II ⁴⁵.

ACE pump: DEN180058, t:slim X2 with interoperable technology

Indication

The t:slim X2 insulin pump with interoperable technology is indicated for subcutaneous insulin delivery at set and variable rates in people requiring insulin. It can receive, execute and confirm commands from compatible connected devices, including automated insulin dosing software. Labeling limits use to NovoLog or Humalog U-100, in people aged 6 years and older ¹⁶⁵.

Evidence: engineering rather than clinical

The retrieved decision summary pages describe no clinical study for the ACE pump ¹⁷⁰¹⁵². Authorization instead rested on technical evidence. FDA found the communication protocols and failsafe features adequate and the cybersecurity information acceptable ¹⁸¹. Risk mitigations included validation of delivery accuracy, reliability testing, occlusion-detection validation and drug-compatibility testing ¹⁹⁰.

Special controls

The ACE pump special controls require:

• Infusion-accuracy verification across foreseeable conditions and programmable rates and volumes, using a statistically valid number of devices.
• Clinically relevant hazard-detection testing, drug compatibility and stability testing, and biocompatibility assessment ¹⁷⁹.
• Validated communication specifications, failsafes for conflicting commands, and data logging to trace dosing errors ¹⁸⁰.
• Labeling that discloses pump performance and wireless quality of service ¹⁸²¹⁸³.

FDA explained that labeling performance parameters lets developers judge whether a given pump meets the needs of a proposed connected system ¹⁸⁴. This is the core idea behind the interoperable model.

iAGC: DEN190034, Control-IQ Technology

Indication

Control-IQ is software-only. It works with compatible iCGMs and ACE pumps to increase, decrease or suspend basal insulin based on current and predicted glucose. It can also deliver correction boluses when glucose is predicted to exceed a threshold. The original indication was type 1 diabetes in people aged 14 and older, with NovoLog or Humalog U-100 ¹²³¹²⁴.

Pivotal randomized trial

The pivotal study was a prospective, multicenter randomized trial. It compared Control-IQ closed-loop control (CLC) with a sensor-augmented pump (SAP) without automation, randomized 2:1 after a run-in. The randomized phase lasted 6 months, and an extension allowed up to 15 months of follow-up ⁷⁹⁸⁰. A total of 168 participants completed the trial (112 CLC, 56 SAP) at seven US sites. Ages ranged from 14 to 71 years, and 31 participants were aged 14 to 18 ⁸¹⁸².

On safety, neither arm had severe hypoglycemia. One DKA event in the CLC arm was attributed to infusion-set failure. Reportable hyperglycemia/ketosis events numbered 13 with CLC and 2 with SAP ⁸⁸⁸⁹. Two unanticipated adverse device effects involved software defects. Both led to temporary study suspension and were corrected with patches. One of them caused inappropriate insulin suspension in about 45 subjects, with one ketosis case and no DKA ⁹¹. FDA noted that a 168-person, 6-month study could not detect differences in rare events such as DKA or severe hypoglycemia. It also noted that the hyperglycemia/ketosis difference might reflect differences in reporting ⁹²⁹³.

Special controls and benefit-risk

The iAGC special controls require:

• Clinical evidence representative of intended users and use scenarios, for each indicated therapeutic agent or with an FDA-acceptable justification.
• Human factors validation and documented design inputs and outputs, with traceability from hazards to risk controls. FDA may also require postmarket data collection ¹¹²¹¹³.
• Interoperability verification: secure, reliable transmission; eligibility and performance specifications for connected devices; safe behavior when communication is lost or restored; validated pairing and data exchange; handling of out-of-bounds data; and safe software updates ¹¹⁴¹¹⁵.
• Logging of critical commands, malfunctions, alerts and connectivity events, plus restrictions on which glucose-input devices and pumps the controller may connect to ¹¹⁴¹¹⁵.
• A training plan for users and healthcare providers, and labeling of clinical performance, compatible configurations and unsafe-use populations ¹¹⁸¹¹⁹.

FDA reviewed a software bill of materials as part of its cybersecurity assessment ¹¹⁶¹¹⁷. It concluded that benefits outweighed risks. Because of uncertainty about rare events, it planned additional postmarket data to confirm real-world safety ¹²⁰.

The 510(k) wave after the De Novos

Once the three device types existed, FDA cleared later products through 510(k). In FDA's 510(k) database, QJI (iAGC) lists 24 clearances after DEN190034, QFG (ACE pump) lists 29 after DEN180058, and QBJ (iCGM) lists 17 after DEN170088. Examples include:

• ACE pumps: Omnipod DASH (K191679, 2019) ²⁴⁴, Omnipod 5 ACE Pump (K203768) ²⁴¹, iLet ACE Pump (K223846) ²³⁷, Tandem Mobi (K223213) ²³⁸, Accu-Chek Solo (K213134) ²⁴⁰, twiist system (K250930) ²²⁵, MiniMed 780G insulin pump (K251032, 2025) ²²⁴ and Mint ACE Pump (K263422, October 2026) ²¹⁶.
• iCGMs: Dexcom G7 (K213919) ²⁵⁷ and FreeStyle Libre 2 and Libre 3 (K222447) ²⁵⁵.
• iAGC algorithms: Tandem, Insulet, Tidepool, Beta Bionics, CamDiab, DEKA, Diabeloop and Medtronic have all received clearances ^{192193194195196197198199200201202203204205206207208209210211212213214215}.

For pumps and sensors, the retrieved summaries emphasize bench and analytical evidence. Omnipod DASH (predicate DEN180058) relied on performance testing against device specifications and updated ISO 14971 risk assessments, and the retrieved pages describe no clinical study ^469470471. Dexcom G7 (predicate G6, K201328) was supported by a pivotal study at 12 US sites enrolling 482 participants aged 2 and older. Accuracy was measured against a YSI laboratory analyzer, and design verification was performed against the iCGM special controls ^479480481482.

Clinical evidence for iAGC 510(k)s

Algorithm clearances show the widest range of evidence types: single-arm pivotal studies, randomized trials, real-world observational data and in silico simulation.

These clearances illustrate several patterns:

• Real-world evidence is accepted for algorithms. Tidepool Loop relied on an observational study of the DIY Loop algorithm instead of a controlled trial of the marketed app ³⁷⁰³⁷³. CamAPS FX and DBLG2 used real-world data to support configurable settings and modes that the clinical trials had not tested ⁵⁶⁶⁵⁴⁸.
• Predicate chains run through the original De Novo. Control-IQ K200467 is the predicate for Omnipod 5, Tidepool Loop and iLet ^493369590. Tidepool Loop, in turn, is the predicate for CamAPS FX and DBLG2 ⁵⁵⁸⁵⁴⁰.
• Population expansions need targeted clinical data. Lowering the age to 2 years required a randomized trial in very young children ⁴⁰³. Type 2 indications for SmartAdjust and Control-IQ+ each rested on a dedicated type 2 study ³⁹⁰⁶¹³. Human factors validation was repeated in the new type 2 population ³⁹¹⁵²⁶.
• Legacy PMA data can carry over. Medtronic's SmartGuard algorithm moved into the QJI framework by leveraging prior PMA pivotal data, without new clinical studies ⁵¹³.

Legacy PMA evidence: MiniMed 670G

The MiniMed 670G (P160017) was the first hybrid closed-loop system approved in this lineage. It was indicated for type 1 diabetes in people aged 14 and older, and users still had to calculate and give meal boluses ^573574575. The pivotal study was a multicenter, single-arm home-and-hotel study of 126 pump users aged 14 to 75 at 10 centers. It included a 2-week run-in, 3 months of home use and a 5-day/6-night hotel phase ^576577578. HbA1c fell from 7.4% to 6.9%, and sensor time in range rose from 66.7% to 72.2% ⁶³⁸⁶⁴⁰. Time at or below 70 mg/dL fell from 5.9% to 3.3% ⁶⁴¹. There were no severe hypoglycemia or DKA events ⁶⁶⁷.

Medtronic components show both pathways at work. The Simplera Sync sensor (P250012, approved August 6, 2025) is a PMA product indicated as an input to compatible Medtronic AID systems ³⁴⁹³⁵⁵. That approval relied on clinical studies from P160017/S118, with no new clinical study ³⁵². The MiniMed 780G pump, by contrast, was cleared as an ACE pump in 2025 ²²⁴.

Next in the ecosystem: glucose-ketone sensing

On August 25, 2026, FDA granted De Novo DEN250049 for the Abbott Libre Duo 10 Day Continuous Dual Glucose Ketone Monitoring System. It created a new Class II regulation, 21 CFR 862.1354 (integrated continuous glucose ketone monitoring system), under product code SJD ¹⁸⁸¹⁸⁹. The device type is designed to send data to digitally connected devices, including AID systems ⁵⁴. Its special controls require clinical accuracy data for each analyte against laboratory glucose and ketone measurements, with adult glucose thresholds modeled on the iCGM framework ⁵⁰⁵¹. The retrieved decision documents do not report the underlying clinical study results ⁵⁹⁶¹.

Takeaways for regulatory strategy

1. Choose the component, then the pathway. Sensors, pumps and dosing algorithms are each regulated under their own Class II regulation (862.1355, 880.5730, 862.1356) and are cleared by 510(k) against an existing device of the same type.
2. iCGM accuracy is a pass/fail gate. The confidence-bound thresholds in the special controls apply to every new sensor, so pivotal accuracy studies need to be powered to meet each range-specific bound ^{325326327328329330331}.
3. Algorithm clearances turn on clinical evidence. The iAGC special controls require clinical data representative of intended users and therapeutic agents ¹¹². New ages, diabetes types or insulins have each been supported by targeted studies, real-world data, or both.
4. Rare-event uncertainty is handled postmarket. FDA has repeatedly said premarket AID studies are underpowered for severe hypoglycemia and DKA and has relied on postmarket data collection instead ^92120416.
5. Interoperability and cybersecurity are reviewed in every submission. Communication failsafes, data logging and cybersecurity documentation appear in the special controls and in each De Novo review. FDA's Control-IQ review also examined a software bill of materials ^{114115116180181}.