MENU
Advanced Materials Intelligence
  1. Home
  2. Continuous Glucose Monitoring (CGM) Explained

Continuous Glucose Monitoring (CGM) Explained

← Back to Biosensors & POCT guides

TECHNOLOGY EXPLAINER

Continuous Glucose Monitoring (CGM)
— measuring interstitial fluid, not blood, with an enzyme electrode

A thin sensor inserted under the skin measures glucose in the interstitial fluid that fills the space around cells, by turning an enzyme reaction into an electric current, and it keeps doing so for days. The number it reports is not the glucose in blood itself but the value in a second compartment that follows blood with a delay. This explainer sets out the mechanism, calibration, lag, the accuracy metric MARD, interfering substances and FDA classification of these sensors, using regulatory documents and peer-reviewed papers.

Built from primary sources: FDA regulation text, review documents and 510(k) summaries; manufacturers' labelling; peer-reviewed papers / Last updated October 2026

Hero image: a sensor patch on the upper arm, with a cut-away of the skin
AI-generated concept image. A round patch on the upper arm and, in a cut-away, the thin sensor that enters the tissue beneath it. It does not show the shape or size of any specific product, or the real thickness of the skin layers.
What this article covers
  1. What CGM is, in three points
  2. A tiny enzyme electrode under the skin: what reacts, and how
  3. A materials engineer's view (1): four hurdles — enzyme, sterilisation, biocompatibility and adhesion
  4. Calibration: done at the factory, or by the user
  5. Lag: blood and interstitial fluid are different compartments
  6. The accuracy metric MARD: definition and limits
  7. Our calculation: why the same sensor can show a different MARD
  8. Wear duration: what each product's documents state
  9. Interfering substances: what labelling and review documents say
  10. The FDA's iCGM special controls: what a device must meet
  11. From PMA approval to De Novo, and then over the counter
  12. A materials engineer's view (2): “accurate” compared with what?
  13. Open problems, and what this article could not confirm
  14. Glossary / References / Claim-to-source audit
How claims are labelled in this article

Sourced = stated in an official document, manufacturer labelling or a paper (link given)
Our calculation = a value this article derived from assumptions it states
Not yet confirmed = a plan or target with no confirmed track record
Structural readings and materials-design interpretations are marked separately as Commentary. This article explains technology. It gives no advice on diagnosis, treatment or how to read a measured value, and it does not recommend any product. For how to use a product or handle a reading, follow the product's instructions for use and your healthcare professional.

1. What CGM is, in three points

CGM (continuous glucose monitoring) is a device that measures glucose in the fluid beneath the skin and reports it electronically every few minutes. A fingerstick blood glucose meter (SMBG) captures a single moment; CGM leaves a continuous record along the time axis (our commentary).

  • What it measures: interstitial fluid, not blood. The FDA review document for the Dexcom G6 says the sensor continuously measures glucose concentration in interstitial fluidSourced, and the FDA summary for FreeStyle Libre likewise says a sensor placed under the skin generates a current according to glucose in interstitial fluidSourced
  • How it measures: an enzyme, glucose oxidase, oxidises glucose, and the reaction is converted to an electric current. The FDA documents state that the size of the current is proportional to the amount of glucoseSourced
  • What it is evaluated against: in clinical studies, blood drawn at the same time is measured by a laboratory method and compared. The Libre summary (SSED) says a YSI 2300 analyser was the referenceSourced
The single most important line in this article

A CGM number is not “glucose in blood” itself. It is an estimate: glucose in interstitial fluid, measured by an enzyme electrode and converted by an algorithm. That is why there is lag, why there are calibration schemes, and why accuracy can only be discussed together with the conditions: compared with what, in whom, over which range. Since 2018 the FDA has classified this kind of device as an “iCGM” with special controlsSourced.

2. A tiny enzyme electrode under the skin: what reacts, and how

The heart of the sensor is a thin, thread-like (wire-like) part inserted into subcutaneous tissue. Putting the FDA documents side by side gives the following picture.

Conceptual illustration of continuous glucose monitoring
Fig. 1 Conceptual image (AI-generated).A CGM filament sits in subcutaneous interstitial fluid, not inside a blood vessel, and measures glucose continuously; electronics remain on the skin surface. See the original schematic for electrochemical details. AI-generated concept, not a product structure, scale drawing or micrograph.
View the original SVG schematic, source notes and numerical details
Subcutaneous sensor and the enzyme-electrode reaction (concept diagram) Dimensions and layer thicknesses are schematic. It does not depict the structure of any specific product. Transmitter Skin adhesive layer Epidermis Dermis Subcutaneous tissue (ISF) Electrode at the sensor tip ● Glucose What happens at the tip electrode Glucose enters the enzyme layer Glucose oxidase oxidises the glucose Path A (gen. 1) Oxygen route: H2O2 is oxidised at electrode Path B (gen. 2) A mediator carries electrons to electrode Current → algorithm → estimated glucose
Fig. 1 Concept diagram (vector drawing). On the left, a schematic of skin and a subcutaneous sensor; on the right, the flow of the enzyme-electrode reaction. The reaction pathway follows the Dexcom G6 review document [Reference 4] and the Libre review documents [References 5, 6]; the grouping into generations follows a peer-reviewed review [Reference 14]. Pathways A and B illustrate the difference between generations. Which one a given product uses cannot be told from the public documents checked here (the Dexcom G6 document describes the formation of hydrogen peroxide).
ItemWhat the FDA documents saySource
Enzyme and currentDexcom G6: glucose oxidase catalyses oxidation of glucose and produces hydrogen peroxide, which generates a current “proportionate to the interstitial glucose concentration”De Novo decision summary [Reference 4]
Enzyme and currentFreeStyle Libre 2 and 3: glucose oxidase oxidises glucose and transfers electrons to an electrode, producing a current. The size of the current is proportional to the amount of glucose in the subcutaneous space510(k) summary [Reference 6]
Form of the sensorDexcom G7: a “wire-type” sensing mechanism. Stelo: a “small and flexible wire” inserted into subcutaneous tissue. Libre: the sensor “tail” is placed under the skin guided by a needle, which is then retracted[References 7, 9, 6]
Reportable rangeDexcom G6 and G7: 40 to 400 mg/dL; values below are shown as “LOW” and above as “HIGH”[References 4, 7]
Reporting intervalDexcom G6 and G7: every 5 minutes. Libre 2 and 3: a glucose measurement every minute[References 4, 7, 6]

All Sourced (FDA review documents and 510(k) summaries). Products are updated over time.

The idea of an enzyme electrode is old: in 1967 Updike and Hicks published “The Enzyme Electrode” in NatureSourced. A recent review sorts the routes by which electrons reach the electrode into three generations. In the first, oxygen is the electron acceptor and the hydrogen peroxide produced is oxidised at the electrode to read a current. In the second, an artificial electron acceptor (a mediator) is used. In the third, electrons pass directly from the enzyme to the electrode. The same review states that the current electrochemical CGM sensors are all based on glucose oxidase (GOx)Sourced [Reference 14]. Mediators and enzyme immobilisation are covered in more detail in our explainer on materials for the enzyme–electrode interface.

3. A materials engineer's view (1): four hurdles — enzyme, sterilisation, biocompatibility and adhesion

What to take from this: a disposable enzyme sensor meets several demanding material requirements at once

Reading the FDA documents through a materials lens, a single sensor has to clear several hurdles.

  • 1. The enzyme must keep working for the whole wear period. The review says GOx has better long-term stability than FAD-dependent glucose dehydrogenase (FAD-GDH), which is why FAD-GDH has had limited use in CGMSourced [Reference 14]. Keeping a protein active for 10 to 15 days in a body-temperature fluid is a major materials challenge (our commentary).
  • 2. Activity must survive sterilisation and storage. The Libre sensor pack is described as electron-beam sterilisedSourced [Reference 6]. The Dexcom G6 review document cites a radiation-sterilisation standard (ISO 11137) and says shelf-life testing supported a useful life of up to 3 months for the sensorSourced [Reference 4]. The enzyme has to withstand radiation and storage (commentary).
  • 3. Biocompatibility as a material that stays in the body. The Dexcom G6 sensor was evaluated as an implant-type device in tissue contact for long cumulative duration (treated as exceeding 30 days), with tests including cytotoxicity, sensitisation, irritation, systemic toxicity, subacute/subchronic toxicity, genotoxicity and implantationSourced [Reference 4]. The Libre SSED lists the ISO 10993 tests, including 4-week and 26-week implantationSourced [Reference 5]. Although one wear is 10 to 15 days, evaluation assumes repeated use (commentary).
  • 4. On the skin side, it becomes an adhesion problem. The Libre review document lists a “medical grade adhesive layer” that holds the sensor to the skin, and adhesive irritation among the possible adverse effectsSourced [Reference 5]. The Dexcom G7 510(k) lists an adhesive tape (product code KGX, Class I) as a separate product codeSourced [Reference 7]. The outer parts of the Dexcom G6 transmitter are described as an epoxy and palladium contactsSourced [Reference 4]. The materials design of skin adhesion is covered in our explainer on patch sensors and skin adhesion.

These four are requirements in a different dimension from the measurement principle. Good data in an enzyme-electrode paper is not enough if all of them are not met (our commentary).

4. Calibration: done at the factory, or by the user

To convert current into concentration, a CGM needs calibration. Who calibrates, and when, has changed with product generations.

Product (review document)What it says about calibration
Dexcom G6 (De Novo, 2018)If the calibration code supplied with the sensor is entered, no calibration is needed during the wear period. Without the code, two fingerstick values are entered at start-up and then every 24 hours. Optional manual calibration is possible
FreeStyle Libre (PMA, 2017)Factory calibration at manufacture. It neither requires nor accepts user calibration with blood glucose values
FreeStyle Libre 2 and 3 (510(k), 2024)Factory calibrated; fingerstick calibration is not required
Stelo (510(k), 2024)Factory calibration only, with no calibration input

All Sourced (FDA review documents [References 4, 5, 6, 9]).

Factory calibration is only workable if sensitivity varies little from lot to lot. The FDA's iCGM special controls also require controls during manufacturing and at product release to ensure the released product meets its performance specificationsSourced [Reference 1]. Removing calibration from the user also means moving the burden to manufacturing control (our commentary).

5. Lag: blood and interstitial fluid are different compartments

One reason CGM numbers disagree with a glucose meter is lag. Two causes add up: a physiological delay as glucose moves from blood into interstitial fluid, and a device-side delay from the sensor's response and signal processing. The difference between the two compartments, and the difficulty of measuring interstitial glucose accurately, are reviewed by Cengiz and Tamborlane [Reference 18].

Blood glucose versus interstitial glucose: the offset (concept diagram) Curves are schematic, not measured data. Axis values are omitted on purpose. Level Time Lag Blood glucose (lab method) CGM value (ISF + processing) * Reported lag: 5.3 to 6.2 min by microdialysis in eight healthy, overnight-fasted adults [Reference 15]. * Prototype sensor, 37 patients: mean raw-signal lag 9.5 min (SD 3.7); a prediction algorithm shortened it by 4 min [Reference 16]. * The two differ in population and conditions, so the figures cannot be lined up to rank anything.
Fig. 2 Concept diagram (vector drawing). When glucose in blood changes, the interstitial value follows after a delay, more smoothly. The curves are schematic. The numbers in the notes come from peer-reviewed papers [References 15, 16]. A simulation study [Reference 17] points out that interstitial glucose is not just a time-shifted copy of blood glucose but also changes in shape — a distorted mirror.
  • Physiological delay: Basu and colleagues placed microdialysis catheters in the abdominal subcutaneous space of eight healthy, overnight-fasted adults and compared the appearance of intravenously given glucose tracers in plasma and in interstitial fluid. The mean delay in the interstitial space was 5.3 to 6.2 minutesSourced [Reference 15].
  • Delay including the device: Schmelzeisen-Redeker and colleagues used a prototype glucose sensor in 37 patients (108 data sets) and reported the time delay of the raw signal relative to blood glucose as a mean of 9.5 minutes (SD 3.7), median 9 minutes. A prediction algorithm shortened the delay by 4 minutes on averageSourced [Reference 16].
  • Distortion: Cobelli and colleagues used model-based simulation to show that interstitial glucose is not a blood curve merely shifted in time, but has a more complex relationship that depends on both the glucose level and the equilibration timeSourced [Reference 17].
Our calculation: how many mg/dL does a delay correspond to?

Suppose blood glucose changes at a constant rate and the interstitial value is simply delayed in timeOur calculation.

  • Rate of change 1 mg/dL per minute, delay 5.3 minutes → difference 1 × 5.3 = about 5 mg/dL
  • Rate of change 2 mg/dL per minute, delay 9.5 minutes → difference 2 × 9.5 = about 19 mg/dL

Assumptions and limits: a real interstitial value is not just a time shift but also changes in shape [Reference 17]. This is a rough estimate of the order of magnitude, not a prediction of any device's error. The FDA's iCGM special controls also require accuracy of the displayed rate of change, for example that no more than 1% of measurements indicate a positive rate of change greater than 1 mg/dL/min when the true rate of change is more negative than −2 mg/dL/minSourced [Reference 1].

The 2018 review document for the Dexcom G6 included, among the conditions of use, a statement that the device should not be used for treatment decisions while glucose is rising or falling rapidlySourced [Reference 4]. That was the labelling at the time of review, not the current instructions for use. We cite it as an example of how the physical property of lag is reflected in precautions for use.

6. The accuracy metric MARD: definition and limits

The number most often quoted for CGM accuracy is MARD (mean absolute relative difference). The definition is simple: for each pair, take the difference between the CGM value and the reference value, divide by the reference value and take the absolute value, then average over all pairs and express it as a percentage (our commentary).

But this number is not an intrinsic property of a sensor. Two reviews explain why.

  • Heinemann and colleagues write that MARD is the parameter used most often to characterise CGM measurement performance and that comparison looks straightforward, but that interpreting the MARD figures supplied by manufacturers is difficult because of a number of complex aspects. They state that MARD should not be used as the sole parameter, especially for non-adjunctive use (use for treatment decisions)Sourced [Reference 19].
  • Vigersky and McMahon write that MARD is strongly affected by the design of data collection in an accuracy study: participant demographics such as type of diabetes and age, sensor wear site, and the share of data collected in each glycaemic range all influence it. They add that no clinical study has shown that MARD can distinguish a safe sensor from an unsafe one, and that alternatives include error grid analysisSourced [Reference 20].

Review documents show the effect of conditions plainly. The Dexcom G6 document gives MARD against a laboratory method by CGM value range for adults (159 participants) with abdominal wear.

Range of CGM valueNumber of pairsMARD (against the laboratory method)
Below 54 mg/dL38313.8%
54 to 69 mg/dL1,53711.5%
70 to 180 mg/dL9,45310.9%

All Sourced (Dexcom G6 De Novo review document [Reference 4]; adults, 159 participants, abdominal wear, two studies at 11 centres combined; ranges are defined by the CGM value). The same document says the reference was a laboratory method and that, for some young children (ages 2 to 6), a blood glucose meter was used.

The 2017 approval summary (SSED) for FreeStyle Libre also shows how the same device can look different under different conditions. In the first US study, MARD against a laboratory method (a YSI 2300 analyser) was 12.5%; after the device was modified, it was 9.7% in the pivotal studySourced [Reference 5]. Figures from different studies cannot be lined up to rank products.

On standardisation, the CLSI document “POCT05 Performance Metrics for Continuous Interstitial Glucose Monitoring” (2nd edition, published November 2020, reaffirmed July 2025) deals with CGM performance metricsSourced [Reference 21]. Because its full text is paywalled, this article confirms only its title and standing.

7. Our calculation: why the same sensor can show a different MARD

Our calculation: how MARD is computed, and the effect of study design

(1) The calculation: suppose three pairs of reference and CGM values were as followsOur calculation.

  • Reference 100, CGM 110 → |110 − 100| / 100 = 10%
  • Reference 180, CGM 162 → |162 − 180| / 180 = 10%
  • Reference 60, CGM 72 → |72 − 60| / 60 = 20%
  • MARD = (10 + 10 + 20) / 3 = about 13.3%

(2) The effect of study design: suppose a sensor's error is fixed by range at 20% below 60, 10% from 70 to 180, and 8% above 180 (assumed values for illustration). If two studies collect data in different proportions, the overall MARD of the same sensor differsOur calculation.

The same sensor, a different MARD depending on the data mix (our calculation) Errors by range (20% below 60, 10% from 70 to 180, 8% above 180) are assumed values for illustration. Study A: more data in the low range Under 60: 20% 70 to 180: 50% Over 180: 30% → MARD 11.4% Study B: more data in the high range 5% 70 to 180: 50% Over 180: 45% → MARD 9.6% A: 0.20×20 + 0.50×10 + 0.30×8 = 11.4 | B: 0.05×20 + 0.50×10 + 0.45×8 = 9.6 Same sensor error, yet a gap of 1.8 percentage points (our calculation) * Errors by range and the data mix are assumptions showing MARD depends on study design; not real sensor or study values. * Factors (demographics, wear site, share of each range): review [Reference 20]. Real range-by-range data: table in Chapter 6.
Fig. 3 Drawing that includes our calculation (vector drawing). Bar widths show the mix of data; the figures on the right are overall MARD values calculated by this article from the assumed errors. They are not values for any real sensor or study. That MARD depends on study design comes from a review [Reference 20].

For the same reason, the view that a smaller MARD is simply better is risky. Vigersky and McMahon also say clinically meaningful assessments, such as error grid analysis or clinical studies in people with diabetes, should be used in place of MARD [Reference 20]. The FDA's iCGM special controls likewise set their criteria not as MARD but as lower bounds on agreement rates by range (Chapter 10).

8. Wear duration: what each product's documents state

ProductWear duration and features in the FDA documentDocument and pathway
FreeStyle Libre (first generation)Up to 10 days after an initial 12-hour warm-up. Gives a value only when the user scans the sensorPMA P160030 (approved 2017)
Dexcom G6Up to 10 days. Abdomen only for adults; abdomen or buttock for childrenDe Novo DEN170088 (2018)
Dexcom G7Up to 10 days plus a 12-hour grace period. Sends a value every 5 minutes510(k) K234133 (2024)
Dexcom G7 15 DayUp to 15 days plus a 12-hour grace period. Indicated for ages 18 and over510(k) K243214 (2025)
FreeStyle Libre 3 / 3 PlusUp to 14 / 15 days. Worn on the back of the upper arm. Contains a silver oxide battery510(k) K233537 (2024)
Stelo (OTC)Up to 15 days plus a 12-hour grace period. The document says it uses the same hardware as the G7 and that firmware changes extend wear from 10 to 15 days510(k) K234070 (2024)
Lingo (OTC)Up to 14 days. Based on the FreeStyle Libre 2 sensor510(k) K233655 (2024)

All Sourced (FDA review documents and 510(k) summaries [References 4 to 10]). The year is the year of the decision. How a longer wear affects accuracy and safety was examined in each study, and this article does not compare products.

The Stelo 510(k) summary says that, with the same hardware design, firmware changes extended the wear period from 10 to 15 days while maintaining accuracySourced [Reference 9]. It is one example showing that making a sensor last depends not only on the enzyme and membrane materials but also on signal-processing software (our commentary).

9. Interfering substances: what labelling and review documents say

In an enzyme electrode, substances other than glucose that react at the electrode, or that affect the enzyme reaction, can shift the reading. The FDA's iCGM special controls require clinically acceptable performance in the presence of reasonably expected interfering substances, including endogenous substances, metabolites, foods, supplements and medicationsSourced [Reference 1]. What follows reports what manufacturers' labelling and review documents state; it is not a basis for any individual's decision about use.

SubstanceWhat the documents saySource
AcetaminophenDexcom states that, for G6 and G7, amounts above the standard-to-maximum dose (1 g every 6 hours in adults) may raise readings, and that earlier generations (G4/G5) were affected. In the G6 review document, in a clinical study of 65 adults the mean maximum interference effect after a single 1 g dose was 5.2 mg/dL; in bench testing, about 2 mg/dL (physiological concentration at a standard dose) caused no clinically significant change, while 6.5 mg/dL caused up to 22 mg/dL of bias and 20 mg/dL up to 70 mg/dL[References 12, 4]
HydroxyureaDexcom states that taking it makes sensor readings higher than actual glucose and that the degree depends on the amount in the body[Reference 12]
Ascorbic acid (vitamin C)FreeStyle Libre labelling: above the maximum recommended daily allowance, sensor readings may be falsely raised[References 13, 5]
Salicylic acid (e.g. aspirin)FreeStyle Libre labelling: sensor readings may be slightly lowered[References 13, 5]

All Sourced (manufacturers' pages and FDA review documents). For hydroxyurea, Dexcom's page also recommends against using the system for treatment decisions. Labelling may change with product, version and region, so any actual decision rests on the latest instructions for use and a healthcare professional. The bench-test concentrations are transcribed from the review document.

Substances differ in effect, presumably because what reacts at the electrode depends on the electrode potential, the selectivity of the membrane and whether a mediator is used (our commentary). The review notes that second-generation designs can operate at an applied potential lower than that needed to oxidise hydrogen peroxide directly, which reduces the influence of interferentsSourced [Reference 14]. For acetaminophen, which strongly affected earlier Dexcom generations, the review document says the G6 sensor was designed to reduce the effectSourced [Reference 4]. The material makeup used to achieve this was not described in the public documents checked for this article. The thinking behind selectivity of electrodes and membranes is covered in our explainers on electrochemical sensors and on materials for the enzyme–electrode interface.

10. The FDA's iCGM special controls: what a device must meet

The FDA classifies CGM designed to connect digitally with other devices, such as automated insulin dosing systems, as iCGM (integrated continuous glucose monitoring system), Class II (special controls), under 21 CFR 862.1355Sourced [Reference 1]. The special controls in the regulation text include numerical criteria that must be met in a clinical study in adults. Each is expressed as “the lower one-sided 95 percent confidence bound of the agreement rate must exceed this value”.

iCGM special controls: agreement criteria to be met in an adult clinical study (excerpt) Each row: the lower one-sided 95% bound of agreement with the reference must exceed the bar value (21 CFR 862.1355(b)(1)(v)) 0% 100% (A) Below 70 mg/dL: within ±15 mg/dL > 85% (B) 70 to 180 mg/dL: within ±15% > 70% (C) Above 180 mg/dL: within ±15% > 80% (D) Below 70 mg/dL: within ±40 mg/dL > 98% (E) 70 to 180 mg/dL: within ±40% > 99% (F) Above 180 mg/dL: within ±40% > 99% (G) Whole range: within ±20% > 87% The regulation also requires: no opposite-range disagreements (low vs high), at most 1% rate-of-change direction errors, equivalence in children (or a justification for its absence), and acceptable data gaps.
Fig. 4 Concept diagram (vector drawing). Criteria (A) to (G) of 21 CFR 862.1355(b)(1)(v) drawn as a chart [Reference 1]. Bar length is proportional to the criterion. These are the criteria required for review, not performance values of any product.

Beyond the numerical criteria, the special controls set out items such as the followingSourced [Reference 1].

  • Comparison with simultaneously collected blood values measured by a laboratory method that the FDA accepts and that is traceable to a higher-order reference
  • Consistent performance throughout the wear period, and no clinically significant gaps in data
  • A mechanism preventing a disposable sensor from being used beyond its claimed wear period
  • A usability study, a strategy for secure and reliable data transmission, and manufacturing and release controls
  • Labelling that reports accuracy by glucose range (below 54, 54 to below 70, 70 to 180, above 180 to 250, above 250 mg/dL), accuracy of rate of change, frequency of data gaps, alert rates, performance by insertion site, and the observed sensor life

11. From PMA approval to De Novo, and then over the counter

The regulatory position of CGM has shifted considerably in under ten years. In FDA terms, approval (PMA), a grant (De Novo) and clearance (510(k)) are separate procedures. Fig. 5 sorts them out so they are not confused.

Key FDA milestones for CGM (this article's selection) Purple: PMA approval / orange: De Novo grant / grey: rule published / green: 510(k) clearance 27 Sep 2017 PMA approval P160030 FreeStyle Libre Class III 27 Mar 2018 De Novo grant DEN170088 Dexcom G6 iCGM, Class II 18 Feb 2022 Rule published 21 CFR 862.1355 87 FR 9237 iCGM classified 5 Mar 2024 510(k) clearance K234070 Stelo OTC, age 18+ 29 May 2024 510(k) clearance K233655 Lingo OTC, age 18+ 7 Jun 2024 510(k) clearance K233861 Libre Rio OTC, age 18+ * Dates: FDA PMA review document, De Novo and 510(k) databases, Federal Register [References 1-3, 5, 9-11]. * The three OTC products are indicated for people not using insulin. Stelo is described as OTC for ages 18 and over, not on insulin. * The milestones are this article's selection and do not list every approval or clearance.
Fig. 5 Concept diagram (vector drawing). PMA (approval), De Novo (grant of a new classification) and 510(k) (clearance) are different procedures. None of them is a judgment on anything but a specific intended use, and none means “safe and accurate in any use”. The choice of milestones is this article's.
Category (FDA product code)Description in the FDA classification database
QBJIntegrated CGM, factory calibrated. An iCGM designed to connect with digitally connected devices, including automated insulin dosing systems
QLGIntegrated CGM, factory calibrated, not for use with automated insulin delivery systems (the user manually controls therapy actions)
SAF / SBHiCGM for non-intensive glucose monitoring / management, over the counter. For persons not on insulin and without a significant risk of problematic hypoglycaemia. Devices that control insulin delivery are excluded from connection

All Sourced (FDA product classification database; the Stelo 510(k) record [Reference 9] shows the classification name and product code SAF). The Libre 3 510(k) lists product codes QBJ and QLG [Reference 6].

Stelo's indication is for people 18 and over who are not on insulin, and it states that the user is not intended to take medical action based on the device output without consultation with a qualified healthcare professionalSourced [Reference 9]. Being sold over the counter and being usable for treatment decisions are separate matters, and the FDA classes reflect that difference (our commentary). The classification of medical devices and home-use devices is covered in our explainer on device classification and home use.

12. A materials engineer's view (2): “accurate” compared with what?

Fig. 6: a bench evaluation with several sensor tips dipped in solution
Fig. 6 Concept image (AI-generated). A laboratory scene in which sensor tips are dipped in a temperature-controlled buffer to examine their response, an in vitro evaluation. It does not show any specific apparatus, product or test condition.
What to take from this: accuracy is a function of the reference, the population and the site, not only of the sensor material
  • The reference differs. The FDA's special controls call for a laboratory method that the FDA accepts and that is traceable to a higher-order reference [Reference 1]. In the Dexcom G6 studies the reference was a laboratory method for adults, and a blood glucose meter (SMBG) for some young childrenSourced [Reference 4]. With a different reference, the same “agreement rate” means something different (commentary).
  • The site differs. The G6 review document says adult performance was evaluated at the abdomen only and that performance at other sites was not evaluated and may differSourced [Reference 4]. The regulation also requires performance by insertion site in the labelling [Reference 1]. If the state of subcutaneous tissue differs by site, the same sensor can give different results (commentary).
  • Sensors vary even against each other. In the G6 review document, 67 participants wore two devices at the same site at the same time. The mean absolute relative difference between the two devices was 9.6% and the mean coefficient of variation was 8.4%Sourced [Reference 4]. This figure shows the sensor's own reproducibility, without the reference-method problem (commentary).
  • The test tube is not the body. Even if a response is good in vitro, in the body there are additional effects: loss of enzyme activity, protein adsorption, reaction of surrounding tissue, and movement at the wear site. This is one reason review documents call for clinical studies (our commentary). The reaction that sensors meet inside the body is covered in our explainer on implantable sensors.

13. Open problems, and what this article could not confirm

(1) Membrane, enzyme immobilisation and mediator makeup by product

The FDA documents checked for this article (review documents and 510(k) summaries) do not give the composition of each product's diffusion-limiting membrane, the way the enzyme is immobilised, or whether a mediator is used. The Libre SSED says only that glucose oxidase transfers electrons to an electrode, and the use of a mediator could not be confirmed. This article therefore does not describe product-by-product material differences.

(2) Each product's current stated MARD and latest instructions for use

Manufacturers' stated MARD varies with reference method, population, site and date and is updated over time, so this article does not use it. The MARD figures given here are limited to values for specific studies stated in FDA review documents.

(3) Long-term clinical outcomes, and product-specific lag values

Long-term clinical effects of CGM use and product-specific lag values could not be confirmed in the official documents and papers checked, so they are not given. The lag figures here are limited to values from papers on specific populations and devices [References 15, 16].

Summary of this article
  • CGM is an estimate of interstitial glucose measured by an enzyme electrode. The FDA documents say the current produced by the glucose oxidase reaction is proportional to glucose in interstitial fluidSourced
  • Lag has a physiological part and a device part. Reports include 5.3 to 6.2 minutes in healthy adults by microdialysis, and a mean of 9.5 minutes for the raw signal of a prototype sensorSourced
  • MARD depends on study design. It varies by glucose range and study, and reviews say it should not be the sole metricSourced
  • The FDA's iCGM criteria are set as lower bounds of agreement rates by range. For adults, for example, more than 85% within ±15 mg/dL below 70 mg/dLSourced
  • The pathway moved from PMA (approval) to De Novo (grant) to 510(k) (clearance), and to over the counter. The three are separate proceduresSourced
  • From the materials side, enzyme stability, sterilisation, biocompatibility and skin adhesion are all required at once. This is our commentary

14. Glossary

CGM
Continuous glucose monitoring. A device that measures glucose in subcutaneous interstitial fluid continuously, every few minutes.
Interstitial fluid (ISF)
The fluid that fills the space around cells. Glucose reaches it from blood through capillaries.
Glucose oxidase (GOx)
An enzyme that oxidises glucose. Widely used in the sensors of current electrochemical CGM.
Mediator
An artificial electron acceptor that carries electrons between enzyme and electrode. Used in second-generation sensors.
MARD
Mean absolute relative difference. The mean of the absolute difference from the reference divided by the reference (%).
iCGM
Integrated CGM. An FDA Class II (special controls) category for CGM designed to work with other connected devices.
OTC
Over the counter. Available without a prescription.
PMA
Premarket approval. The premarket route for Class III devices.
De Novo
A route to classify a novel low-to-moderate-risk device. A grant creates a new device classification.
510(k)
A premarket notification showing substantial equivalence to a legally marketed device. The result is called clearance.
YSI
A product name for a laboratory glucose analyser. Commonly used as the reference in CGM studies.
Lag
The time difference between a change in blood glucose and its appearance in the CGM value.
Traceability
The property that a measured value links to a higher-order reference (a reference material or method).
Special controls
Device-specific requirements imposed on Class II devices in addition to general controls.

15. References (primary sources)

  1. U.S. Food and Drug Administration (FDA; eCFR) “21 CFR 862.1355 Integrated continuous glucose monitoring system” — www.ecfr.gov
  2. FDA (Federal Register) “Medical Devices; Clinical Chemistry and Clinical Toxicology Devices; Classification of the Integrated Continuous Glucose Monitoring System”, 87 FR 9237, 18 February 2022 — www.federalregister.gov
  3. FDA De Novo database record DEN170088 (Dexcom G6): decision granted 27 March 2018; regulation 862.1355; product code QBJ — www.accessdata.fda.gov
  4. FDA “DEN170088 Evaluation of Automatic Class III Designation for Dexcom G6 Continuous Glucose Monitoring System: Decision Summary” — www.accessdata.fda.gov
  5. FDA “PMA P160030 FDA Summary of Safety and Effectiveness Data (FreeStyle Libre Flash Glucose Monitoring System)”, approved 27 September 2017 — www.accessdata.fda.gov
  6. FDA “K233537 510(k) summary and clearance letter (FreeStyle Libre 2 / Libre 3)”, 23 April 2024 — www.accessdata.fda.gov
  7. FDA “K234133 510(k) summary and clearance letter (Dexcom G7)”, 26 February 2024 — www.accessdata.fda.gov
  8. FDA “K243214 510(k) summary and clearance letter (Dexcom G7 15 Day)”, 9 April 2025 — www.accessdata.fda.gov
  9. FDA “K234070 510(k) summary and clearance letter (Stelo Glucose Biosensor System)”, 5 March 2024 — www.accessdata.fda.gov
  10. FDA “K233655 510(k) summary and clearance letter (Lingo Glucose System)”, 29 May 2024 — www.accessdata.fda.gov
  11. FDA “K233861 510(k) summary and clearance letter (Libre Rio Continuous Glucose Monitoring System)”, 7 June 2024 — www.accessdata.fda.gov
  12. Dexcom “Interfering substances and risks” (labelling page for Dexcom G6 / G7) — www.dexcom.com
  13. Abbott “What substances could interfere with the results generated by the FreeStyle Libre 2 Plus system?” (support page for Saudi Arabia) — www.freestyle.abbott
  14. Hatada M, Okuda-Shimazaki J, Sode K. “Glucose Oxidoreductases for Glucose Monitoring: A Bird's Eye View of Glucose-Sensing Enzymes II”, J Diabetes Sci Technol (2026) (PMC) — pmc.ncbi.nlm.nih.gov
  15. Basu A et al. “Time lag of glucose from intravascular to interstitial compartment in humans”, Diabetes 62:4083 (2013) — doi.org
  16. Schmelzeisen-Redeker G et al. “Time Delay of CGM Sensors: Relevance, Causes, and Countermeasures”, J Diabetes Sci Technol 9:1006 (2015) — doi.org
  17. Cobelli C et al. “Interstitial Fluid Glucose Is Not Just a Shifted-in-Time but a Distorted Mirror of Blood Glucose: Insight from an In Silico Study”, Diabetes Technol Ther 18:505 (2016) — doi.org
  18. Cengiz E, Tamborlane WV. “A tale of two compartments: interstitial versus blood glucose monitoring”, Diabetes Technol Ther 11 Suppl 1:S-11 (2009) — doi.org
  19. Heinemann L et al. “Benefits and Limitations of MARD as a Performance Parameter for Continuous Glucose Monitoring in the Interstitial Space”, J Diabetes Sci Technol 14:135 (2020) — doi.org
  20. Vigersky RA, McMahon C. “The Myth of MARD (Mean Absolute Relative Difference): Limitations of MARD in the Clinical Assessment of Continuous Glucose Monitoring Data”, Diabetes Technol Ther 26:38 (2024) — doi.org
  21. CLSI “POCT05 Performance Metrics for Continuous Interstitial Glucose Monitoring”, 2nd edition (published November 2020; reaffirmed July 2025): product page — clsi.org
  22. Updike SJ, Hicks GP. “The Enzyme Electrode”, Nature 214:986 (1967) — doi.org

16. Claim-to-source audit

Claim in the textBasisType
That CGM measures glucose in interstitial fluid. For the Dexcom G6: the sensor continuously measures interstitial glucose; glucose oxidase oxidises glucose and produces hydrogen peroxide, generating a current proportional to interstitial glucose; reporting every 5 minutes; a reportable range of 40 to 400 mg/dL; up to 10 days; calibration by calibration code or two fingerstick entries; a laboratory method (including YSI) as the clinical reference, with a blood glucose meter for some young children; the study size (380 participants, 11 centres); evaluation at the adult abdomen only; paired same-site devices with MARD 9.6% and CV 8.4% (67 participants); sensor shelf-life testing supporting up to 3 months; the types of biocompatibility test; transmitter exterior (epoxy and palladium contacts); reference to a radiation-sterilisation standardDEN170088 review document
Reference 4 https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN170088.pdf		Sourced
MARD for adults (159 participants) by CGM value range: below 54, 13.8% (383 pairs); 54 to 69, 11.5% (1,537); 70 to 180, 10.9% (9,453). 89.9% within 20% overall. Acetaminophen: mean maximum interference effect 5.2 mg/dL after a single 1 g dose in 65 adults; bench results (no significant change at about 2 mg/dL; up to 22 mg/dL bias at 6.5; up to 70 mg/dL at 20); effects in G4/G5; the condition of use at review (not for treatment decisions while glucose is rising or falling rapidly)DEN170088 review document
Reference 4 https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN170088.pdf		Sourced
The definition of iCGM and Class II (special controls). The adult criteria (A) to (K): lower one-sided 95% bound above 85% for within ±15 mg/dL below 70 mg/dL; above 70% within ±15% from 70 to 180; above 80% within ±15% above 180; 98 to 99% for the ±40 criteria; above 87% within ±20% overall; no opposite-range disagreements; no more than 1% rate-of-change direction errors. Comparison with a laboratory method traceable to a higher-order reference; consistency across the wear period; acceptable performance with interferents; prevention of use beyond wear period; usability study; manufacturing and release controls; labelling content (accuracy by glucose range and insertion site, rate of change, gaps, alerts, sensor life)Regulation text (eCFR)
Reference 1 https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-862/subpart-B/section-862.1355		Sourced
The date the FDA classified the iCGM as Class II with special controls in the Federal Register (18 February 2022; 87 FR 9237)Reference 2 https://www.federalregister.gov/documents/2022/02/18/2022-03504/medical-devices-clinical-chemistry-and-clinical-toxicology-devices-classification-of-the-integratedSourced
That the De Novo for DEN170088 (Dexcom G6) was granted on 27 March 2018, with regulation number 862.1355 and product code QBJReference 3 https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/denovo.cfm?id=DEN170088Sourced
That the PMA for FreeStyle Libre (P160030) was approved on 27 September 2017. That glucose oxidase transfers electrons to a metal electrode and the current is proportional to the amount of glucose. Factory calibration with no user calibration required or accepted. Up to 10 days (after a 12-hour warm-up); a value only on scan. The medical grade adhesive layer and adhesive irritation among adverse effects. MARD of 9.7% in the pivotal study (12.5% in the first US study) against a YSI 2300. Labelling on ascorbic acid and salicylic acid. The ISO 10993 tests (including 4-week and 26-week implantation)PMA P160030 SSED
Reference 5 https://www.accessdata.fda.gov/cdrh_docs/pdf16/P160030B.pdf		Sourced
The FreeStyle Libre 2 and 3 510(k) (23 April 2024): the glucose oxidase principle, electron-beam sterilisation, measurement every minute, 14 / 15 days of wear, the back of the upper arm, factory calibration, a silver oxide battery, age indications, and product codes QBJ, QLG and NBWReference 6 https://www.accessdata.fda.gov/cdrh_docs/pdf23/K233537.pdfSourced
The Dexcom G7 510(k) (26 February 2024): wire-type sensing, 5-minute reporting, 40 to 400 mg/dL, up to 10 days plus a 12-hour grace period, and an adhesive tape (product code KGX, Class I) listed alongsideReference 7 https://www.accessdata.fda.gov/cdrh_docs/pdf23/K234133.pdfSourced
The Dexcom G7 15 Day 510(k) (9 April 2025): up to 15 days plus a 12-hour grace period; indicated for ages 18 and overReference 8 https://www.accessdata.fda.gov/cdrh_docs/pdf24/K243214.pdfSourced
The Stelo 510(k) (5 March 2024): an OTC iCGM for people 18 and over not on insulin; a small, flexible wire; the same hardware design as the G7 with firmware changes extending wear from 10 to 15 days; factory calibration only; the statement about consulting a healthcare professional before medical action; product code SAFReference 9 https://www.accessdata.fda.gov/cdrh_docs/pdf23/K234070.pdfSourced
The Lingo 510(k) (29 May 2024): an OTC iCGM for people 18 and over not on insulin, based on the FreeStyle Libre 2 sensor, up to 14 daysReference 10 https://www.accessdata.fda.gov/cdrh_docs/pdf23/K233655.pdfSourced
The Libre Rio 510(k) (7 June 2024): an OTC iCGM for non-insulin-using people 18 and over, up to 15 days, factory calibratedReference 11 https://www.accessdata.fda.gov/cdrh_docs/pdf23/K233861.pdfSourced
Dexcom's labelling: acetaminophen (G6 and G7 usable at standard to maximum doses; excess amounts may raise readings) and hydroxyurea (raises readings, depending on the amount in the body, with a recommendation against using it for treatment decisions)Reference 12 https://www.dexcom.com/en-us/interferenceSourced
Abbott's labelling (FreeStyle Libre 2 Plus): ascorbic acid above the maximum recommended daily allowance may falsely raise readings, and salicylic acid may slightly lower themReference 13 https://www.freestyle.abbott/en-sa/support/what-substances-could-interfere-with-the-results-generated-by-the-freestyle-libre-2-plus-system-1.htmlSourced
The three generations of enzyme electrodes (oxygen and hydrogen peroxide / mediator / direct electron transfer). That current electrochemical CGM sensors are all based on GOx. That GOx is more stable over the long term than FAD-GDH, which has limited FAD-GDH use in CGM. That second-generation designs work at a lower applied potential and reduce interferenceReference 14 https://pmc.ncbi.nlm.nih.gov/articles/PMC13518614/Sourced
The physiological delay of glucose from blood to interstitial fluid (eight healthy adults, overnight fasted, microdialysis; mean 5.3 to 6.2 minutes)Reference 15 https://doi.org/10.2337/db13-1132Sourced
The mean delay of the raw signal against blood glucose for a prototype sensor in 37 patients (108 data sets): 9.5 minutes (SD 3.7), median 9 minutes; 4 minutes shorter on average with a prediction algorithmReference 16 https://doi.org/10.1177/1932296815590154Sourced
That interstitial glucose is not simply a time-shifted copy of blood glucose but has a distorted relationship that depends on equilibration time and glucose levelReference 17 https://doi.org/10.1089/dia.2016.0112Sourced
A review of the difference between the blood and interstitial compartments and the difficulty of measuring interstitial glucose accuratelyReference 18 https://doi.org/10.1089/dia.2009.0002Sourced
That MARD is the parameter most often used for CGM performance, that manufacturers' figures are hard to interpret, and that it should not be the sole parameter for use in treatment decisionsReference 19 https://doi.org/10.1177/1932296819855670Sourced
That MARD depends on study design (participant demographics, wear site, share of data in each glycaemic range), that no clinical study shows MARD separates safe from unsafe sensors, and the alternatives such as error grid analysisReference 20 https://doi.org/10.1089/dia.2023.0435Sourced
The title, edition and publication date (9 November 2020) of CLSI POCT05, and its reaffirmation in July 2025Only the title and standing were confirmed, because the full text is paywalled
Reference 21 https://clsi.org/standards/products/point-of-care-testing/documents/poct05/		Sourced
That Updike and Hicks published “The Enzyme Electrode” in Nature in 1967Reference 22 https://doi.org/10.1038/214986a0Sourced
The way MARD is computed (about 13.3% for the three-pair example); the overall MARD of 11.4% and 9.6% for studies A and B; the conversion of delays of 1 mg/dL/min × 5.3 min and 2 mg/dL/min × 9.5 min into about 5 and about 19 mg/dLOur calculation. The errors by glucose range (20%, 10%, 8%), the mix of data and the “pure time shift” are assumptions set by this article. They are not values for any real sensor or studyOur calculation
The grouping that three OTC products were cleared by 510(k) in 2024 (Stelo, Lingo, Libre Rio), and the choice of milestonesThis article’s own framing. Dates and numbers were checked against each 510(k) document [References 9 to 11]Commentary
Composition of each product's diffusion-limiting membrane, enzyme immobilisation method and whether a mediator is used; each product's latest stated MARD; product-specific lag values; long-term clinical outcomesNot stated in the official documents and papers checked for this article, so not given hereCommentary
That the size of lag may vary with the reference, the site and the individual; that factory calibration moves the burden to manufacturing control; the framing of four hurdles (enzyme, sterilisation, biocompatibility, adhesion); the difference between the test tube and the body; the reading that being sold over the counter differs from being usable for treatment decisionsThis article’s own framing and commentary based on published content. Not views expressed by the authors or regulators of those documentsCommentary
That Figs. 1, 2 and 5 are concept diagrams (vector drawings), Fig. 3 is a drawing that includes our calculation, Fig. 4 charts the numbers in the regulation, and that the hero image and Fig. 6 are AI-generated imagesA note by this articleCommentary

Last updated 3 October 2026. Sources are limited to primary material (FDA regulation text, review documents and 510(k) summaries; manufacturers' labelling pages; peer-reviewed papers and reviews). No market-size or market-share estimates are used. Because the article includes structural readings and materials-design interpretations, those are marked as Commentary and kept separate from sourced fact. Membrane makeup, enzyme immobilisation and mediator use by product, each product's latest stated MARD and lag, and long-term clinical outcomes are not covered, because they could not be confirmed in published primary sources. Figs. 1, 2 and 5 are concept diagrams, Fig. 3 includes our calculation, Fig. 4 charts the numbers in the regulation, and the hero image and Fig. 6 are AI-generated images; none of them shows a real product, cross-section or measured data. This article explains technology. It gives no advice on diagnosis, treatment or how to read a measured value, and it does not recommend any product.

🌐 Japanese