CGM Guide — Part 2

How Accurate Are CGM Systems?

Understanding what CGM accuracy metrics actually mean, and which systems have the depth of evidence to support insulin dosing decisions.

A reminder before we look at accuracy

For insulin dosing decisions, we only have sufficient data for a small number of CGM systems.

Risk assessment always comes first, with no free passes. For a deeper explanation (not needed for most), see CGM Black Swan.

Therefore, we only discuss CGM accuracy for these systems:

Accu-Chek SmartGuide
Dexcom G6
Dexcom One
Dexcom G7
Dexcom One+
EverSense
FreeStyle Libre 2 and 2 Plus
FreeStyle Libre 3 and 3 Plus
Medtronic Guardian 4
Medtronic Simplera

Performance metrics — why averages can mislead

There are many accuracy metrics, each with limitations. The most common is MARD (Mean Absolute Relative Difference), which gives an overall average error. But an “average” tells you very little about risk at the extremes — where clinical decisions matter most.

Why MARD alone is not enough — and why study design comes first

A low MARD can be produced by a study that never tested CGM under challenging conditions — without meal or insulin challenges, with too few participants, or without peer review. The headline number can look reassuring while the evidence base is thin. Episode 35 of The GNL Podcast with Professor Othmar Moser explains exactly how study design determines what accuracy data actually mean — and why the five questions on the previous page must be answered before any performance comparison. Episode 36 shows how the DSN Forum applied this framework to build their comparison chart, with three specialist nurses explaining how data sufficiency became the first filter, ahead of performance numbers and features.

River depth analogy showing how CGM average error can mask dangerous outliers at the extremes

Agreement rates: 20/20 and 40/40

Agreement rates show the percentage of readings that fall within an acceptable error range:

20/20 — within this zone, no meaningful clinical risk
40/40 — outside this zone, high clinical risk

Diagram explaining the 20/20 and 40/40 CGM agreement rate zones and what they mean for dosing risk

The 20/20 zone

A reading is within the 20/20 zone if:

For CGM values <5.5 mmol/L (100 mg/dL), the comparator is within ±1.1 mmol/L (±20 mg/dL). Example: CGM = 4.5 mmol/L (80 mg/dL) — the “true” value must be between 3.4–5.6 mmol/L (60–100 mg/dL).
For CGM values ≥5.5 mmol/L (100 mg/dL), the comparator is within ±20%. Example: CGM = 10.0 mmol/L (180 mg/dL) — the reference must be 8.0–12.0 mmol/L (144–216 mg/dL).

The higher the percentage of readings within 20/20, the more dependable the CGM is for low-risk insulin dosing.

The 40/40 zone

The 40/40 metric is a wider safety net that helps flag readings with a high chance of leading to incorrect insulin decisions.

For CGM values <5.5 mmol/L, the comparator must be within ±2.2 mmol/L (±40 mg/dL). Example: CGM = 4.5 mmol/L (80 mg/dL) — reference must be 2.3–6.7 mmol/L (20–120 mg/dL).
For CGM values ≥5.5 mmol/L, the comparator must be within ±40%. Example: CGM = 10.0 mmol/L (180 mg/dL) — reference must be 6.0–14.0 mmol/L (108–252 mg/dL).

If a CGM frequently falls outside even the 40/40 window, those readings are very likely to cause incorrect dosing decisions.

Important: because different CGMs are calibrated and tested differently, the same 20/20 or 40/40 percentage does not always mean the same real-world accuracy.

How do the CGM systems stack up?

The DSN Forum UK CGM comparison charts provide the most up-to-date published comparison data.

The first chart shows which CGMs have a study design score of 5, confirming they have been assessed thoroughly.

Overall CGM study design score chart showing which devices meet the five evidence quality criteria

The second chart summarises study design scores and 20/20 and 40/40 agreement rates for CGMs with a study design score of 5.

CGM performance comparison chart showing 20/20 and 40/40 agreement rates for devices with a study design score of 5

If a CGM has a study design score of 5, achieves around 90% of readings within 20/20, and has ≥99% within 40/40, it can support insulin dosing decisions with a low rate of high-risk outliers. All of the CGM systems discussed on this page meet this criterion.

Around 0.5–1% of readings from even well-tested systems may still be problematic. This is why occasional cross-checking with an ISO-standard blood glucose test remains a worthwhile practice, and why understanding the CGM black swan problem is so important.

Those with iCGM approval are the most accurate and have the greatest depth of data. That includes the Abbott, Dexcom, and Eversense CGM systems.

This content is for educational exploration only. It describes average responses and general principles. It is not medical advice and cannot replace individual clinical guidance from your diabetes care team.

Continue the guide

← Part 1: How to Choose a CGM Part 3: CGM Features →

Back to guide overview