Clinical Study Designs and Accuracy Criteria Behind FDA Authorizations of Continuous Glucose Monitors

Clinical study designs and accuracy criteria behind FDA authorizations of continuous glucose monitors

FDA's evidentiary expectations for continuous glucose monitoring (CGM) systems have moved from device-by-device PMA review to a defined Class II framework. Early systems were approved as Class III devices through premarket approval (PMA), and the accuracy evidence was judged case by case. In March 2018, FDA granted the Dexcom G6 De Novo (DEN170088) and created the integrated continuous glucose monitoring system (iCGM) classification. That classification set pre-specified, statistically framed accuracy thresholds as special controls ^110251026. Those special controls now support a steady stream of 510(k) clearances. They also support 510(k) clearances of over-the-counter and implantable sensors ^101410341052. The 2026 dual glucose-ketone De Novo carried iCGM-style glucose criteria into a separate classification with its own special controls ².

This article covers the clinical study architecture common to these authorizations, the performance metrics FDA reviews (MARD and point-accuracy agreement rates), the iCGM special controls, and device-level results from FDA decision summaries and PMA summaries of safety and effectiveness.

Key takeaways

• One study template across pathways. Pivotal CGM studies are prospective, single-arm, multicenter U.S. studies. Sensor readings are paired with venous plasma glucose measured on a YSI laboratory analyzer during in-clinic sessions spread across the wear period ^{8908918201002601833}.
• MARD is reported but is not the iCGM acceptance criterion. The iCGM special controls set thresholds on the lower one-sided 95% confidence bound of agreement rates, stratified by glucose range. They do not set a MARD threshold ^21222327.
• Safety constraints sit alongside accuracy. The special controls bar opposite-extreme readings (iCGM below 70 mg/dL when reference glucose is above 180 mg/dL, and the reverse). They also cap rate-of-change errors and require data availability, secure transmission, interference testing, sensor-life enforcement and usability validation ^{28293031866372374375376}.
• Pediatric data are expected or must be justified. The special controls require similar accuracy in children or an FDA-acceptable justification for omitting pediatric data. Recent iCGM studies have enrolled children down to age 2 ^32622833849.

Regulatory pathways at a glance

Under 21 CFR 862.1355(a), an iCGM automatically measures glucose in bodily fluids continuously or frequently for a specified period. It must reliably and securely transmit readings to digitally connected devices, including automated insulin dosing systems, and may be used alone or with those devices ¹⁰²⁷. This interoperability-centered definition explains why the special controls go beyond point accuracy to cover data availability and transmission.

The common clinical study architecture

Pivotal studies supporting both PMAs and iCGM 510(k)s share a recognizable design.

Design and setting. The Dexcom G4 PLATINUM pivotal study was open-label, non-randomized and single-arm, with 72 adults at four U.S. sites ⁸⁹⁰. The Eversense PRECISE II study was prospective, non-randomized, blinded and single-arm, with 90 adults at eight U.S. sites ⁶⁰¹. The Dexcom G7 pivotal study enrolled 482 adults and children aged 2 years and older at 12 U.S. sites ⁸³³.

Reference method. Reviewers expect a laboratory comparator, typically the YSI 2300 STAT Plus analyzer measuring venous plasma glucose ^8201002836. The iCGM special controls codify this. Comparator specimens must be collected in parallel and measured on an FDA-accepted laboratory method that is precise, accurate and traceable to a higher-order reference ¹⁷. For the youngest children, sponsors have used capillary SMBG meters as the comparator. Examples are ages 2 to 5 in the Dexcom G6 and FreeStyle Libre 2 studies ⁶²⁵⁸⁴⁹.

Clinic-session sampling. Accuracy is captured in extended clinic sessions with frequent venous draws:

• Dexcom G4 PLATINUM: 12 to 13 hour sessions on days 1, 4 and 7, with samples about every 15 ± 5 minutes ⁸⁹¹⁸⁹².
• Eversense PRECISE II: day-long sessions on days 1, 30, 60 and 90, with sampling about every 15 minutes ⁶⁰²⁶⁰³.
• Guardian Sensor 3: frequent-sample testing on sensor days 1, 3 and 7, with YSI reference values every 5 to 15 minutes ⁵²⁸.

Glucose manipulation. To cover the measuring range, sponsors deliberately raise and lower glucose in eligible subjects. In the Dexcom G6 Study 1, glucose was manipulated for participants aged 13 and older. Study 2 did not manipulate glucose ⁶²⁶. The Eversense 365 ENHANCE study included glucose challenges for eligible participants ¹⁰⁵⁰.

Wear-period coverage. Accuracy is reported across the claimed wear period, often split into beginning, middle and end segments. This matches the special-control requirement for "consistent analytical and clinical performance throughout the sensor wear period" ¹⁹:

• Dexcom G6 defined days 1–2, 4–5, and 7 and 10 as the beginning, middle and end assessments of a 10-day sensor ³³.
• FreeStyle Libre 2 used clinic sessions on days 1–3, 5–7, 9–11 and 13–15 ⁸⁴⁹.
• Eversense 365 ran accuracy visits through day 365 ¹⁰⁵⁰.

Insertion sites. Results are reported separately for each wear site. In the Dexcom G6 study, adults wore sensors on the abdomen and children could use the abdomen or upper buttocks ⁶²³. Dexcom G7 reported arm results for adults and ages 7 to 17, plus arm and upper-buttock results for ages 2 to 6 ⁸²⁴⁸³⁵. The special controls require labeling to report performance separately for each insertion or use site ³⁷⁷.

Performance metrics FDA reviews

MARD. The mean absolute relative difference against the laboratory reference appears in nearly every summary. Examples include 12.1% for FreeStyle Libre Pro ⁸²¹ and 9.7% for the original FreeStyle Libre ¹⁰⁰⁴. Guardian Sensor 3 MARD ranged from 8.68% to 10.55% depending on insertion site and calibration schedule ⁵³³.

Point-accuracy agreement rates. FDA summaries report the percentage of CGM readings within ±15, ±20, ±30 and ±40 of the reference. The limit is in mg/dL at low glucose and in percent above it. In the PMA era, the mg/dL limits typically applied at reference values of 80 mg/dL or below ^822823543608. The iCGM framework moved the cut point to 70 mg/dL and attached confidence-bound thresholds to each glucose range ²¹²²²³.

Trend (rate-of-change) accuracy. For Dexcom G6, the iCGM rate fell below −2 mg/dL/min while the comparator rate was above +2 mg/dL/min in 0.2% of adult pairs and 0.5% of pediatric pairs ^869870871872.

Calibration dependence. Pre-iCGM summaries often reported accuracy by calibration schedule. For Guardian Sensor 3, overall agreement within 20/20% was 85.7% for abdomen insertion with 12-hour calibration and 91.6% for arm insertion with calibration 3 to 4 times a day ⁵⁴⁴⁵⁴⁷. Dexcom G4 PLATINUM was calibrated about every 12 hours, an average of two fingersticks per day ⁸⁹³.

The iCGM special controls (21 CFR 862.1355)

The special controls in the Dexcom G6 De Novo are the reference point for every iCGM submission.

Design verification and validation: clinical accuracy, special control (1)

Special control (1) requires "robust clinical data demonstrating the accuracy of the device in the intended use population" ¹⁶. The study must represent performance across the intended use population and the measuring range ¹⁸. The adult performance requirements are ²⁰:

Additional safety criteria in special control (1):

• (H) and (I): No corresponding blood glucose above 180 mg/dL when the iCGM reads below 70 mg/dL, and none below 70 mg/dL when the iCGM reads above 180 mg/dL ²⁸²⁹.
• (J) and (K): No more than 1% of iCGM readings may show a rate above +1 mg/dL/min when the true rate is below −2 mg/dL/min, or the reverse ³⁰³¹.
• (vi) Pediatrics: Data showing similar accuracy and rate-of-change performance in children, or a clinical or technical justification acceptable to FDA ³².
• (vii) Data availability: Throughout the claimed sensor life, the device must not allow "clinically significant gaps in sensor data availability that would prevent any digitally connected devices from achieving their intended use" ⁸⁶⁶.

Special controls (2) to (7)

• (2) A strategy for secure, reliable data transmission to provide real-time readings at clinically meaningful intervals to receiving devices ³⁷².
• (3) Manufacturing and release controls that keep released product within the performance specifications of controls (1) and (2) ³⁷³.
• (4) Clinically acceptable performance in the presence of clinically relevant interferents, including endogenous substances, foods, supplements and medications ³⁷⁴.
• (5) Measures preventing sensor use beyond the claimed wear period ³⁷⁵.
• (6) A usability study showing the intended user can use the device safely and obtain the expected accuracy ³⁷⁶.
• (7) Labeling that reports, for each intended use population and each insertion site: accuracy in the ranges <54, 54–70, 70–180, 180–250 and >250 mg/dL; rate-of-change accuracy; frequency and duration of data gaps; true, false, missed and correct alert rates; and observed sensor life ³⁷⁷.

FDA mapped these controls to five identified risks to health. The risks are inaccurate values driving treatment decisions, poorer pediatric performance, loss of data from signal drop-out or lost communication, insecure data transmission, and inadequate performance when the iCGM is part of a system such as automated insulin dosing ^{380381382383384}.

Device-level evidence

Notes on individual authorizations

Dexcom G6 measured against the thresholds it created. In the adult iCGM-range analysis, every reported lower bound cleared its threshold:

Sources for the observed values and lower bounds: ^627628629867. Thresholds: ^{21222324252627}.

Several margins were narrow. The 70–180 mg/dL ±15% bound (71.4% against 70%) and the <70 mg/dL ±15 mg/dL bound (85.4% against 85%) show how tightly the thresholds track the performance of the device that established them. An acetaminophen interference study with 70 subjects at four sites was run separately from the accuracy study ⁶²⁰. It supports special control (4).

Dexcom G7 and range-stratified reporting. The G7 summary reports agreement both by iCGM glucose range (the special-control framing) and by comparator glucose range. At 70–180 mg/dL in the iCGM-range analysis, 85.4% of adult readings and 86.0% of readings in ages 7 to 17 were within 15% ⁸²⁷. Study 2, with 60 subjects aged 12 and older at two sites, separately assessed real-time data-transmission reliability ⁸³³⁸³⁴. This addresses special control (2).

Extending wear. The FreeStyle Libre 2 study had subjects wear sensors for up to 16 days but analyzed only the first-applied sensor ⁸⁴⁹. Accuracy in ages 6 to 17 fell to a MARD of 10.4% on days 13–15 ⁸⁶⁰. The Dexcom G7 15 Day clearance was supported by an adult study (18 and older) against venous YSI through 15 days plus a 12-hour grace period ⁶³³. The available summary rows did not report its MARD or agreement rates ⁶³¹.

Leveraging prior clinical data. Several 510(k)s reuse an earlier pivotal study rather than running a new one:

• The 2023 FreeStyle Libre 2 and 3 clearance (K223435) leveraged the K222447 studies because the systems were physically identical except for an added reader ⁵²⁰.
• Libre Rio, an over-the-counter iCGM for non-insulin-using adults aged 18 and older, cites clinical data established in K222447 ^10571059.
• Lingo relied on a reanalysis of FreeStyle Libre 2 adult data. Its summary states that accuracy met the iCGM special controls ^10351036. For iCGM readings of 55–200 mg/dL, 89.9% of 11,086 pairs were within 20% (lower confidence limit 88.0%) ¹⁰³⁹.

Implantable sensors. For the original Eversense PMA, the PRECISE II safety data covered 9,773 sensor-use days. They included one serious adverse event, a sensor placed too deeply that required surgical removal under general anesthesia ⁶⁰⁵⁶¹⁸. The later Eversense 365 510(k) applied the iCGM special controls (1)(v)(A) to (F) to a one-year sensor ^10461047. Its 95% lower bounds exceeded each threshold ^10521053.

Replacing fingersticks for treatment decisions. The Dexcom G5 indication under product code PQF states that the device is designed to replace fingerstick blood glucose testing for treatment decisions. Users interpret results using trends and several sequential readings ¹⁰⁰⁹. PMA supplement records for the G5 under product code PQF date from December 2016 ⁴¹⁹. The supplement summary rows retrieved here did not describe the evidence behind that indication change ¹⁰⁰⁸.

Beyond glucose. The Libre Duo De Novo (DEN250049) carries the iCGM-style glucose criteria into a combined glucose-ketone classification ⁴⁷³⁹. It adds ketone safeguards: a sensor reading below 0.6 mmol/L must not correspond to blood ketones above 3.0 mmol/L, and a reading above 3.0 mmol/L must not correspond to blood ketones below 0.6 mmol/L ⁴⁸. Labeling must report glucose and ketone accuracy separately by population, analyte and use site ⁴⁹⁴⁰.

Practical implications for sponsors

• Power the study on lower confidence bounds, not point estimates. The iCGM criteria are tested on lower one-sided 95% bounds within each glucose range ²¹²²²³. Paired-point counts in the <70 mg/dL and >180 mg/dL strata therefore drive sample size, and glucose manipulation protocols are standard ^6261050.
• Plan the reference method early. The comparator must be traceable to a higher-order reference ¹⁷. Using an SMBG comparator for very young children has been accepted alongside YSI data for older cohorts ⁶²⁵⁸⁴⁹.
• Report by population, site and wear segment. Special control (7) requires this granularity in labeling ³⁷⁷. Recent summaries report beginning, middle and end-of-wear accuracy separately ^353637837.
• Account for the non-accuracy controls. Data availability, secure transmission, interference, sensor-life enforcement and usability validation are each independent special controls. Each maps to specific risks to health ^{866372374375376382383}.

Related FDA guidance on CGM-derived endpoints in drug trials, Submitting Continuous Glucose Monitoring Data in Clinical Trials, was finalized on May 6, 2026 ⁸⁹⁶. Topics a reader may want to explore further include iCGM interoperability with automated insulin dosing, the design of over-the-counter CGM studies in non-insulin users, and how FDA has handled sub-threshold pediatric strata under special control (1)(vi).