Written and maintained by CASRAI Editorial Board
Last updated
This page describes products and general use; it is not medical advice.
CGM accuracy describes how closely a continuous glucose monitor’s sensor readings match a laboratory reference measurement. It is usually summarised as MARD (mean absolute relative difference, in percent; lower is better), but the US regulator judges an integrated CGM (iCGM) by a different test: the share of readings that fall within set error bands in each glucose range. This guide is for clinic staff, buyers and students who need to read an accuracy claim correctly. It explains both measures, shows why MARD values depend on the reference method, and lists what the FDA record does and does not say about FreeStyle Libre and Dexcom Stelo.
What CGM accuracy means
A CGM estimates glucose in interstitial fluid with a sensor worn on the body. “Accuracy” is the agreement between those estimates and a reference value measured at the same moment. The reference method is part of the claim: a laboratory venous analyser and a fingertip meter are not the same yardstick. Accuracy also depends on glucose range, wear day and rate of change, which is why the regulation described below asks for results by range rather than one number.
For a general explanation of the point-of-care meters often used as a reference, see what a glucometer is. For the regulatory category a CGM belongs to, see what a medical device is.
What is MARD?
MARD is the average of the absolute percentage differences between sensor readings and paired reference readings. A MARD of 10% means that, on average, the sensor differs from the reference by 10% of the reference value, in either direction. Lower is better. Because MARD is a single average, it hides whether errors cluster at low glucose or during fast changes.
MARD is a summary statistic. It is not the criterion in the FDA regulation for iCGM systems, and that regulation sets no MARD cutoff.
How FDA sets iCGM accuracy (21 CFR 862.1355)
An integrated continuous glucose monitoring system is a Class II device under 21 CFR 862.1355, with special controls. The regulation requires accuracy to be shown with clinical data from the intended-use population, across the full measuring range, against an FDA-accepted traceable laboratory glucose method, and to hold throughout the wear period. The adult performance criteria use the lower one-sided 95% confidence bound of the proportion of readings within each zone:
| Reference glucose range | Reading must fall within | Required proportion (lower 95% bound) |
|---|---|---|
| Below 70 mg/dL | ±15 mg/dL | More than 85% |
| 70 to 180 mg/dL | ±15% | More than 70% |
| Above 180 mg/dL | ±15% | More than 80% |
| Below 70 mg/dL | ±40 mg/dL | More than 98% |
| 70 to 180 mg/dL | ±40% | More than 99% |
| Above 180 mg/dL | ±40% | More than 99% |
| Full range | ±20% | More than 87% |
Source: 21 CFR 862.1355, Cornell Legal Information Institute copy, checked 2026-10-30.
The same regulation adds requirements that a single average cannot show: a limit on rate-of-change error, no pairing of a low sensor value with a high reference value or the reverse, and pediatric accuracy comparable to adult accuracy or otherwise justified. Labeling must report performance by population and insertion site, accuracy by glucose range, rate-of-change performance, and the frequency and duration of data gaps.
Why MARD numbers differ: the reference method
A 2024 head-to-head study (Hanson K, Kipnes M, Tran H, Journal of Diabetes Science and Technology) compared Dexcom G7 and FreeStyle Libre 3 sensors in 55 analysed adults with type 1 or type 2 diabetes. Accuracy was assessed against two kinds of reference: a YSI laboratory analyser and capillary blood glucose values. Against the YSI reference, the study reported a MARD of 8.9% for FreeStyle Libre 3 and 13.6% for Dexcom G7.
| Reference method | Dexcom G7 MARD | FreeStyle Libre 3 MARD |
|---|---|---|
| YSI laboratory analyser | 13.6% | 8.9% |
Source: Hanson K, Kipnes M, Tran H. J Diabetes Sci Technol 2024;18(3):598-607, PubMed 38189290, checked 2026-10-30. The reference method is part of every MARD claim, so a MARD quoted without its reference method cannot be compared with another. A single study in one adult sample is not a ruling on either product.
FreeStyle Libre accuracy
In the study above, FreeStyle Libre 3 had a MARD of 8.9% against a YSI reference. That is one study of one model in one adult sample, not a figure for the whole FreeStyle Libre family. For any other Libre model, use the accuracy table in the manufacturer’s instructions for use, which the iCGM regulation requires to be broken out by population, insertion site and glucose range.
Dexcom Stelo accuracy
The FDA record for the Stelo Glucose Biosensor System (Dexcom, Inc.) is 510(k) K234070, decision date 2024-03-05, “Substantially Equivalent”, product code SAF, class 2, regulation 862.1355, described as an “Integrated Continuous Glucose Monitor For Non-Intensive Glucose Monitoring, Over-The-Counter” (openFDA, checked 2026-10-30). That record establishes its regulatory category. It does not state a MARD, and this page does not give one. Accuracy data for Stelo belongs in the device labeling.
How to read a CGM accuracy claim
| Check | Why it matters | Source |
|---|---|---|
| Reference method named? | The 2024 study assessed accuracy against both a YSI laboratory analyser and capillary values; a MARD means little without its reference. | Hanson 2024 |
| Results by glucose range? | The iCGM criteria are set separately for below 70, 70 to 180 and above 180 mg/dL. | 21 CFR 862.1355 |
| Population and insertion site stated? | Labeling must report performance by population and insertion site, accuracy by glucose range, and rate-of-change performance. | 21 CFR 862.1355 |
| Sample size and wear days? | The Hanson study analysed 55 adults; accuracy must hold through the wear period. | Hanson 2024; 21 CFR 862.1355 |
Common mistakes
- Treating a lower MARD from one study as proof that one brand beats another; different references and samples are not interchangeable.
- Assuming FDA clearance means a particular MARD. The regulation uses percent-within-zone criteria, not a MARD threshold.
- Quoting one model’s study figure as the accuracy of every model in a product family.
FAQ
What is MARD?
MARD is the mean absolute relative difference between sensor and reference glucose values, expressed as a percentage. Lower means closer agreement on average.
How accurate is FreeStyle Libre?
In one 2024 head-to-head study of 55 adults, FreeStyle Libre 3 had a MARD of 8.9% against a YSI laboratory reference, compared with 13.6% for Dexcom G7. Other models have their own labeling.
How accurate is Dexcom Stelo?
The FDA 510(k) record K234070 classifies Stelo as an over-the-counter integrated CGM under 21 CFR 862.1355 but gives no MARD. Consult its labeling for accuracy data.
Does FDA require a specific MARD for CGMs?
No. 21 CFR 862.1355 sets percent-within-zone criteria by glucose range, plus other requirements, rather than a MARD cutoff.
Who uses it: clinic staff and buyers comparing monitors, and students learning how device accuracy is reported. Related monitoring supplies are listed in diagnostic, monitoring and imaging. See also the diabetes clinic supply checklist, what the Omnipod is and the general validity vs accuracy comparison.
Sources
- 21 CFR 862.1355, Integrated continuous glucose monitoring system, https://www.law.cornell.edu/cfr/text/21/862.1355 (checked 2026-10-30)
- openFDA 510(k) record K234070, Stelo Glucose Biosensor System, https://api.fda.gov/device/510k.json?search=k_number:K234070 (checked 2026-10-30)
- Hanson K, Kipnes M, Tran H. Comparison of Point Accuracy Between Two Widely Used Continuous Glucose Monitoring Systems. J Diabetes Sci Technol 2024;18(3):598-607, https://pubmed.ncbi.nlm.nih.gov/38189290/ (checked 2026-10-30)
