CGM Guide Series – Step 3
CGM Accuracy – What the Numbers Actually Mean
The headline accuracy number you will see in marketing material is a statistical average. It does not tell you about the 1% of readings that fall well outside any accuracy window – the ones where the clinical consequences of acting on the wrong number could be serious. GNL reports CGM accuracy using ±20/20 and ±40/40 agreement only. This section covers what those numbers mean, and the black swan zone beyond them.
Part of: CGM for T1D – The GNL Framework → Step 3 of 3
Why GNL does not use MARD
The headline accuracy figure most manufacturers lead with is a statistical average of the percentage difference between the CGM and a blood glucose reference. It tells you how the sensor performs on average across a study. It does not tell you how often readings fall far enough from reality to drive an unsafe insulin dose.
An average can look reassuring while hiding the tail. Two sensors with the same average can behave very differently at the edges – and the edges are where dosing decisions go wrong.
GNL therefore reports CGM accuracy using ±20/20 and ±40/40 agreement only. These are the numbers that describe the distribution of readings, including how often a device falls outside the clinically safe band. The DSNFUK CGM comparison chart uses the same measures, and so does the framework in this guide.
The 20/20 and 40/40 agreement rates
Agreement rates tell you about the distribution – how often is the CGM within a clinically meaningful range, and how often is it far outside it?
±20/20 agreement
The percentage of readings where the CGM is within 20 mg/dL (1.1 mmol/L) of the reference when glucose is below 100 mg/dL (5.6 mmol/L), or within 20% of the reference when glucose is above 100 mg/dL. A reading in this band will almost always lead to a correct insulin dosing decision. Current framework-qualified CGMs achieve 93-95% ±20/20 agreement. That means roughly 19 readings in 20 are in the clinically safe zone.
±40/40 agreement
The percentage of readings within 40 mg/dL (2.2 mmol/L) or 40% of the reference. Readings outside this band are far enough from reality to create a meaningful dosing error – either significantly overcorrecting a low glucose, or ignoring a high that is genuinely dangerous. Current framework-qualified CGMs have between 0.5% and 1% of readings outside ±40/40. That is the black swan zone.
The black swan
The 1% that demands a precautionary response → read this before relying entirely on your CGM
Between 0.5% and 1% of CGM readings from framework-qualified devices fall outside the ±40/40 agreement band. That sounds small. But consider what it means in practice.
A person using CGM continuously wears a sensor for 10 to 15 days. At a reading every 5 minutes, that is around 2,000 to 4,000 readings per sensor. At a 1% miss rate, that is 20 to 40 readings per wear period that are far enough from reality to matter clinically.
Those readings do not arrive labelled as errors. They look like every other reading. The CGM is confident. The alarm is silent. The person sees a number and responds to it.
This is not a design flaw unique to any manufacturer. It is a property of interstitial glucose sensing. Interstitial glucose lags behind blood glucose by 5 to 15 minutes. During rapid glucose change, that lag can compound errors. Pressure on the sensor, hydration, exercise, and perfusion can all push readings temporarily outside the accuracy window.
The precautionary principle – not risk-benefit analysis
This is not a risk-benefit question where you weigh the advantages of CGM against a small risk of error. The end consequence of acting on a significantly wrong reading in the wrong direction can be severe hypoglycaemia or DKA. Those are not acceptable outcomes to trade off against convenience.
The precautionary principle applies: when a small percentage of readings can produce a harmful outcome, the appropriate response is not to ignore it – it is to maintain the ability to detect it.
What this means in practice
You do not need to test your finger prick glucose every time you see a reading. CGMs are extraordinary tools – the 99% of readings that fall within the accuracy window represent a level of insight that was not available to people with T1D a decade ago.
But you do need:
- An in-date blood glucose meter, with in-date test strips, accessible
- The habit of reaching for it when something does not feel right – when the CGM says low but you do not feel low, when the CGM says high but you recently gave insulin, when an alarm fires and the clinical picture does not fit
- If you calibrate your CGM – to calibrate from a proper finger prick, not from the CGM reading itself
The lost art of knowing when to do a finger prick is one of the most important safety habits in diabetes management. The DSN Forum UK nurses who collaborated on this guide said it clearly: CGM has pushed us forward a hundred steps. From an education perspective, we might need to take half a step back – and make sure every person using CGM has a working meter and the understanding that occasionally they will need it.
Episodes 35 and 36 of the GNL Podcast cover this in depth – with Professor Othmar Moser and the DSN Forum UK team. The conversation includes the clinical situations where finger prick glucose is most likely to be needed, and the case for calibration where it remains available.
Accuracy and AID systems
When a CGM is integrated with an automated insulin delivery (AID) system, the accuracy of the CGM has a direct effect on insulin delivery decisions made by the algorithm. A reading that is wrong drives an insulin response that is wrong – automatically, without a human decision step.
This is why iCGM (integrated CGM) designation – required in the US for AID integration – demands higher accuracy standards and stricter study design. It is also why the framework in this guide matters most in the context of AID use.
AID compatibility for each device is covered on the individual device pages.
Accuracy and calibration
Some CGMs offer the option to calibrate – to enter a blood glucose reading from a finger prick to refine the sensor’s accuracy in the first hours of wear or when readings seem inconsistent.
Calibration can improve accuracy when done correctly. It can introduce error when done incorrectly – for example, by entering a CGM reading rather than a true blood glucose measurement. If calibration is available and you use it, it should be based on a properly performed finger prick from a clean, dry, uncompressed fingertip.
Devices where calibration is available are noted on each device page.
Evidence backbone
This page is built on the GNL CGM Evidence Pack. John Pemberton is lead or contributing author on every entry below. The Evidence Pack itself sits inside Grace, the GNL clinical knowledge engine; the source papers are linked here for independent verification.
- Pemberton et al 2026 – International clinical opinion on transparency, standardisation, and calibration alignment in CGM. Modified Delphi consensus, 21 international authors. Diabetes, Obesity and Metabolism. DOI: 10.1111/dom.70460
- Pemberton et al 2023 – CGM accuracy: contrasting CE marking with the FDA and TGA. Narrative review. Diabetes, Obesity and Metabolism. DOI: 10.1111/dom.14962
- Pleus et al 2025 – IFCC Working Group clinical assessment guideline for CGM system performance. John Pemberton co-author. Clinica Chimica Acta. DOI: 10.1016/j.cca.2025.120728
- Moser & Pemberton 2024 – Rethinking the safety and efficacy assessment of HCL systems. The CGM-AID safety bridge. Diabetic Medicine. DOI: 10.1111/dme.15305
- DSNFUK CGM Comparison Framework v2 (March 2026) – The practical UK clinical reference, authored by John Pemberton. DSNFUK chart
Overall evidence grade: C (clinical opinion, consensus guidelines, methodology papers). Confidence: HIGH – John is at the centre of the international consensus on CGM standardisation.
When in doubt, check with a finger-prick – and model the outcome
The ±20/20 and ±40/40 agreement framework tells you when to trust the CGM. The Hypo and Hyper Explorer lets you work through what a treatment decision looks like when the number might be at the edge of the accuracy window.