Continuous Glucose Monitoring:
Reading the Line, Trusting the Sensor
Everything in one place. Read the plain version with Jude, earn your way into the accuracy evidence with Grace, then the full methodology with John. Stop wherever you have enough.
How we teach: three rules, borrowed from Taleb
You earn each level by showing you understand it, not by scrolling past it. We only teach what we would use on ourselves and the people we love.
Understanding beats memory and luck, so the checks reshuffle every time you retry. A pass means you got it, not that you guessed it. And we teach you to tell a trend (signal) from one reading (noise).
We give you the scaffolding and get out of your way. Roam where your curiosity leads, go as deep as you want, and ask Grace anything. We will not teach a bird how to fly.
Want this for your own sensor, in your units? Ask Grace, then take it to your care team.
One page, three depths
This guide compounds: each layer rests on the one beneath it. Read Jude’s plain version, then pass a short understanding check to open Grace, then another to open John. You can roam freely within a layer; you cannot skip ahead a layer, because the next one would not make sense and you would be standing on a gap.
The whole thing, in plain words
A continuous glucose monitor, or CGM, is a small sensor you wear on your skin. A tiny soft thread sits just under the surface and reads the glucose in the fluid between your cells, every few minutes, day and night. It sends those readings to your phone or a small receiver, so instead of a single number from a finger-prick you get a line that shows where your glucose is and which way it is heading.
That heading is the part that changes everything. A finger-prick tells you the score right now. A CGM shows you the direction of travel: a flat line, a gentle climb, a fast fall. It can also be set to alarm before things go too low or too high, which is why so many people sleep better with one on.
It does most of what a person with diabetes needs and it has become the standard tool offered to children and adults from the start.1 Two honest catches, though. First, the sensor reads the fluid around your cells, not your blood directly, so when glucose is moving fast the sensor lags a little behind; if a reading surprises you, especially a low, a finger-prick still settles it. Second, no two sensors are exactly the same, and a number that looks precise is still an estimate. Read the line for its shape and direction, and treat any single reading kindly.
Does this match the life of the person wearing it? The CGM is a companion, not a judge. If the line ever makes you feel watched or scolded, the tool is being used against you. Use it to spot patterns and to sleep, not to chase every wobble.The Pemberton lens, lived recognisability, one of the four GNL appraisal lenses.
How good is the number, really
The accuracy number you will see quoted
Manufacturers headline a figure called MARD, the mean absolute relative difference between the sensor and a reference blood-glucose method. A lower MARD reads as better. The catch is that MARD depends heavily on how the study was run, so the same sensor produces very different MARDs depending on the comparator method and who funded the trial.2 3 In one independent three-way study, three current-generation sensors all landed close together at about 11.6 to 12.0% against the same venous reference, while their own pivotal trials had quoted figures several points lower.2
The Dexcom G7 reads MARD 8.2% in its own pivotal trial, 13.6% in a competitor-funded head-to-head, and 12.0% in the first manufacturer-independent three-way study; same sensor, same era.3 The choice of reference method alone can shift a MARD by 5 to 10 percentage points.2 4 So a bare “MARD 8.2%” is not a fact you can rank devices on.
Why the better question is “would the error change a decision”
GNL leads on a different metric: Parkes (Consensus) Error Grid agreement, the share of readings whose difference from the truth is small enough that it would not change a treatment decision, and the zone D+E rate, the share that could lead to a harmful decision.4 A sensor reporting around 99.7% Parkes A+B with a 0% zone D+E rate is operating inside the clinical safety bar for essentially every reading. That is the figure that maps to “is it safe to act on”, which a single MARD percentage does not.
Two more things matter more than the headline. Accuracy in the low range (below 3.9 mmol/L, 70 mg/dL), because that is where a wrong reading is most dangerous; and accuracy when glucose is moving fast, because sensors differ in clinically meaningful directions during rises and falls.2 One sensor reads systematically lower during exercise than a finger-prick; another reads more truly while glucose is climbing. The “best” sensor is the one whose strengths match your day.
What to aim for, in the standard targets
The international consensus (Battelino 2019) sets population targets for most adults with type 1 diabetes, in the standardised glucose ranges every sensor report now uses.5
| Metric | Range | Target | Why it is set there |
|---|---|---|---|
| Time in Range | 3.9 to 10.0 mmol/L (70 to 180 mg/dL) | over 70% | The protective middle; about 7.0% (53 mmol/mol) HbA1c on average |
| Time Below Range (L1) | under 3.9 mmol/L (70 mg/dL) | under 4% | Limits time spent low |
| Time Below Range (L2) | under 3.0 mmol/L (54 mg/dL) | under 1% | The serious-low ceiling |
| Time Above Range (L1) | over 10.0 mmol/L (180 mg/dL) | under 25% | Caps higher glucose |
| Time Above Range (L2) | over 13.9 mmol/L (250 mg/dL) | under 5% | Caps very-high glucose |
Targets are population goals for most adults with type 1 diabetes; older or higher-risk people, and pregnancy, use different, looser or tighter sets, agreed with the care team.5
A headline accuracy figure is only as good as the method behind it. When you are handed a MARD, ask the two questions that keep everyone honest: against which reference, and during what kind of glucose movement? A number with no method attached is marketing, not evidence.The Goldacre lens, evidence-grade discipline, one of the four GNL appraisal lenses.
The methodology, the zones, the limits
Why MARD is not a league table
MARD is one number summarising thousands of paired sensor-versus-reference readings. It is unstable across three axes, each of which can move it several points.2 4
| Axis | What shifts the MARD | Worked example |
|---|---|---|
| Comparator method | Capillary BG versus venous YSI versus a lab analyser | One FreeStyle Libre 3 sensor: 11.6% vs YSI venous, 9.5% vs Cobas venous, 9.7% vs capillary, same study2 |
| Protocol dynamism | Mean rate of glucose change during the study; slow protocols sit in the easy steady-state window | Pivotal trials around 0.04 mmol/L/min; the IFCC stress protocol demands above 0.06 to 0.07 mmol/L/min6 |
| Sponsorship and outlier handling | Who funded it, and whether early sensor failures were clipped | Dexcom G7: 8.2% (own pivotal), 13.6% (competitor-funded), 12.0% (independent)3 |
A The discordance figures are Grade A to B from named head-to-head studies; the three-axis framing is the GNL canonical position per the MARD framing policy.3 4 The Parkes (Consensus) Error Grid agreement and the zone D+E rate are anchored in the clinical consequence of error, which is why GNL leads on them rather than on MARD.
iCGM, eCGM, and what the categories mean
In the United States the FDA defines an integrated CGM (iCGM) class: a sensor that meets a published statistical accuracy bar across the glucose range and may therefore drive an automated insulin system non-adjunctively, that is, you can dose from it without a confirmatory finger-prick. Sensors that do not clear that bar, or are not cleared for it, are read as eCGM, more cautiously positioned. The iCGM bar is encoded by Parkes-zone agreement far more directly than by a single MARD figure, which is the regulatory reason GNL leads on zones.4 6
The wider problem is transparency. FDA iCGM submissions are publicly accessible with age-specific accuracy data; in Europe the equivalent database (EUDAMED) is not expected to be fully open before 2030, and several systems hold CE marking for non-adjunctive use in children with no publicly available paediatric accuracy data.6 That is an interpretation gap, not a safety verdict: without the protocols and reference data, you cannot tell whether two sensors differ for real or for methodological reasons.
The calibration corridor (the A-P-B model)
Even with full transparency and aligned procedures, sensors can still differ, because their calibration algorithms are trained on different reference datasets (capillary, venous, or arterialised-venous blood). Capillary and venous glucose differ systematically, by around 10% on average in type 1 diabetes, from about 5% at rest to 30% during a rapid post-meal swing, and the gradient can even reverse during a fast fall.6 This defines a physiological corridor, and a sensor sits somewhere relative to it:
| Zone | Where the sensor reads | Clinical reading |
|---|---|---|
| A, above | Above the capillary-venous corridor | May prompt earlier or more frequent corrections; may make hypoglycaemia exposure look smaller than it is |
| P, physiological | Within the venous-to-capillary corridor | The range that underpins the legacy CGM outcome evidence |
| B, below | Below the corridor (below venous) | May generate more conservative dosing and may under-report hyperglycaemia |
An automated insulin delivery algorithm adapts to whatever values it receives; it cannot detect or correct a systematic calibration bias in the sensor feeding it.6 The paper offers an illustrative individual moving from a Zone B sensor to a Zone P sensor on the same interoperable system, with similar Time in Range but a plausible HbA1c shift of around 5 mmol/mol (0.5%); a teaching point about exposure, not an outcome to expect on any one switch.
The standard of care, and where to settle
ISPAD treats CGM as the standard monitoring tool for children and adolescents with type 1 diabetes from diagnosis, with reductions in ketoacidosis, severe hypoglycaemia, and improved quality of life, and stresses that data-interpretation skill, not just device access, drives the benefit.1 The target ranges (Battelino 2019) are population goals; your own sensor, your wear site, your rate of change, and your care team set the real numbers.5
It is the rare, large miss that does the damage, not the average reading. A sensor that is right on a flat Tuesday but reads 21 points low during exercise is dangerous exactly when it matters. Judge a device by its worst case in the situations you cannot afford, not by its tidy headline.The Taleb lens, robustness to outliers, one of the four GNL appraisal lenses.
A metric is only as honest as the protocol behind it. MARD without a named comparator and a stated rate-of-change distribution is a number with no provenance. Name the reference, name the dynamism, name who paid; an accuracy claim that hides its method is not an accuracy claim, it is a brochure.The Hayes lens, technical and methodological rigour, one of the four GNL appraisal lenses.
The whole guide, summarised
Glucose never lies; the sensor only estimates it, every few minutes, day and night. Read the line for its shape, and check a surprising low with a finger-prick.
This page is the taster. The full journey, three modules and their 30 questions, with your progress saved, lives in Learn with Grace. Glucose never lies; come and learn to read the line.
References
Evidence grades A (strongest) to D (editorial or working analysis).
- Tauschmann M, Cardona-Hernandez R, DeSalvo DJ, et al. ISPAD Clinical Practice Consensus Guidelines 2024: Diabetes Technologies, Glucose Monitoring. Horm Res Paediatr. 2025. DOI 10.1159/000543156. CGM as standard of care from diagnosis; benefit depends on data-interpretation skill. C
- Eichenlaub M, Waldenmaier D, Wehrstedt S, et al. Manufacturer-independent prospective head-to-head accuracy of three current-generation CGM systems. J Diabetes Sci Technol. 2025. Triple comparator (YSI venous, Cobas venous, Contour Next capillary); FSL3, G7, Medtronic Simplera. A
- Garg SK, Kipnes M, Castorino K, et al. Dexcom G7 pivotal accuracy. Diabetes Technol Ther. 2022;24(6):373-380 (MARD 8.2%); with Hanson K, Kipnes M, Tran H. Head-to-head G7 vs FSL3. J Diabetes Sci Technol. 2024;18(3):598-607 (G7 MARD 13.6%). A / B
- GNL MARD Framing Policy (locked 1 May 2026), drawing on Pleus S, et al. Comparator-method bias in CGM accuracy. J Diabetes Sci Technol. 2022, and the Parkes (Consensus) Error Grid. Lead on Parkes A+B and zone D+E; MARD secondary, comparator named. D
- Battelino T, Danne T, Bergenstal RM, et al. Clinical Targets for Continuous Glucose Monitoring Data Interpretation: International Consensus on Time in Range. Diabetes Care. 2019;42(8):1593-1603. DOI 10.2337/dci19-0028. PMID 31177185. A
- Pemberton JS, Andrews RC, Barnard-Kelly K, et al. International clinical opinion on transparency, standardisation, and calibration alignment in CGM performance evaluation. Diabetes Obes Metab. 2026;28(4):2551-2565. DOI 10.1111/dom.70460. The A-P-B calibration model, IFCC procedure, transparency asymmetry. C
One page, three voices: Jude, Grace, John. Population-average, not personalised.