How to Choose a Continuous Glucose Monitor (CGM)

Before starting, it’s worth understanding the CGM Black Swan

Why? The first question should always be:

“Do I understand the risk of using CGM readings for insulin dosing — and is that risk the same for all CGMs?”

For some devices, the risk is well characterised and clearly acceptable. For others, it isn’t. That difference matters, and it should shape how much risk we’re willing to take.

 Ok, risk understood, let’s begin.

What is Continuous Glucose Monitoring (CGM)?

A Continuous Glucose Monitor (CGM) is a wearable device that tracks glucose levels day and night, providing real-time readings, trend arrows, and alerts for high and low glucose. CGMs are now central to diabetes management, giving insight far beyond traditional finger-prick blood glucose monitoring.

The picture below shows how CGM works and how trend arrows let you “see into the future” of your glucose levels.

For the basics, start with CGM Foundations — how continuous glucose monitoring can guide bolus insulin decisions — and then explore the GAME-SET-MATCH framework for dynamic glucose management strategies.

However, not all CGMs are created equal, and choosing the right one requires far more than comparing headline accuracy numbers and app features.

Adjunctive vs non-adjunctive CGM – why it matters

CGM systems fall into two broad categories:

• Adjunctive CGM – requires confirmation with finger-prick blood glucose (SMBG) before insulin dosing and hypoglycaemia treatment.
• Non-adjunctive CGM – approved for insulin dosing decisions and hypoglycaemia treatment without mandatory finger-prick checks.

This guide focuses only on systems with non-adjunctive approval.

When we are using CGM for insulin dosing and hypo treatment, risk assessment comes first — then performance.

Why risk assessment comes first

Before comparing CGMs based on;

• Advertised performance, usually MARD or Mean Absolute Relative Difference,
• The alarms and alerts,
• Size,
• Days it lasts,
• Integrations with pumps,
• Etc, etc, etc

It’s essential to first understand the risk profile of each device.

Think of choosing a CGM like scouting a football player:

If I claimed I had scored as many goals as Messi and Ronaldo, but cost half the price, should you sign me immediately?

Or would you first ask:

• Where did you score those goals? (Top leagues or Sunday league?)
• How competitive were the matches? (Training games or real fixtures?)
• How do you perform at the highest level, under real pressure?

Five key questions for CGM accuracy studies

The answers tell us whether the CGM has been tested across the full glucose range (typically 2.2–22.2 mmol/L or 40–400 mg/dL) and at the rates of change that occur in daily life.

Q1. Has the data been peer-reviewed or reviewed by the FDA?

High-quality CGM evaluations are reviewed either by:

• Reputable diabetes technology journals, and/or
• Regulators such as the FDA.

That scrutiny covers study design, participant selection, test procedures, and outcomes. It allows clinicians, people with diabetes, and regulators to critique the evidence themselves.

We need this level of transparency considering CGM readings are used to make decisions about a drug (insulin) that carries one of the highest risks of any self-administered medicine.

Q2. Is there sufficient data on people with type 1 diabetes?

Data sufficiency exists to ensure that reported CGM accuracy reflects clinical risk. For insulin dosing, evidence must be independent, representative, and collected under conditions where treatment decisions are actually made. This requires coverage across the full glucose range (including hypoglycaemia and marked hyperglycaemia), across the full sensor lifecycle (early, mid, and late wear), and across a sufficiently large number of individuals to capture between-person variability and real-world failure modes. Data volume alone is insufficient if measurements are clustered within too few people or too narrow a set of conditions.

Participant number is therefore the primary anchor, with 50 participants as the baseline and a hard minimum of 40, alongside ≥70% representation of people with type 1 diabetes. To reflect the non-linear loss of independence of data below 50 participants, banded thresholds are used for paired CGM–reference values:

• ≥50 participants require ≥5,000 paired points;
• 46–49 require ≥10,000;
• 40–45 require ≥15,000;
• and <40 is not acceptable regardless of data volume.

Q3. Were meal and insulin challenges performed?

A robust accuracy study should deliberately trigger high and low glucose episodes.

• Food without insulin → rapid glucose spikes.
• Too much insulin without food → fast drops and hypoglycaemia (“rage bolus” territory).

Without these challenges, and without 70% having type 1 diabetes, the CGM may never be tested in the very scenarios where accuracy matters most.

Q4. What percentage of comparison readings were < 4.4 mmol/L (80 mg/dL)?

At least 8% of paired CGM vs comparator readings should be below 4.4 mmol/L (80 mg/dL), as per Performance Metrics for Continuous Interstitial Glucose Monitoring (POCT05).

No one is aiming for 8% hypoglycaemia in real life. This threshold simply ensures enough data in the low range to meaningfully assess performance.

Q5. What percentage of comparison readings were > 16.7 mmol/L (300 mg/dL)?

At least 5% of paired readings should exceed 16.7 mmol/L (300 mg/dL), to adequately test performance in the very high range.

If these extremes are not well represented, the CGM’s ability to reliably detect dangerous highs and lows is unknown. That uncertainty is not acceptable when we are using readings to treat hypos or to deliver large correction doses.

Which CGMs with insulin dosing approval have been tested robustly?

I have co-developed CGM comparison charts with the Diabetes Specialist Nurse (DSN) Forum UK. The most up-to-date version is hosted on their website.

The first chart uses the five study-design questions above to identify which CGM devices have been tested using methods that meet basic international standards.

It provides an overview of CGM systems currently available in the UK and Europe and gives each device a study design score out of 5.

Any CGM system with a study design score less than 5 does not provide enough evidence to understand the risk when using for insulin dosing. Specifically, the risk of using readings in the low and very high glucose ranges — exactly where clinical and insulin dosing decisions are made.

This remains true even if a CGM system has a CE mark for non-adjunctive use across a wide age range.

More recently, CE marking for CGMs used to drive automated insulin delivery (AID), without peer-reviewed published data, has triggered serious concerns. These have been highlighted by paediatric endocrine societies and adult diabetes technology networks.

This does not mean those CGMs and AIDs are unsafe; it means we do not know the risk.

From this point forward, we only discuss CGM systems with a study design score of 5. They are;

• 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

Do all CGMs show you the same glucose level?

Simply, no.

Each CGM shows, on average, glucose values that can sit above, within, or below the blood glucose your body is actually exposed to. This is pronounced when the glucose levels are rising and go above 10.0 mmol/L (>180 mg/dL).

That matters when you’re thinking about how much time in range (3.9–10.0 mmol/L / 70–180 mg/dL) you should aim for.

The same person using different CGMs — can have the same biology but end up with different reported time in range, by up to 10%.

In other words, time in range is only meaningful if you understand which Zone your CGM reads in, especially when above 10.0 mmol/L (180 mg/dL).

Where current CGM systems sit

We don’t think there any Zone A systems on the market, but, with the lack of transparency in data, it is possible. All established CGM systems fall into Zone P or Zone B.

CGM systems reading in Zone P

On average, these systems read in line with physiological glucose exposure, particularly at glucose concentrations above 10.0 mmol/L (180 mg/dL).

For Zone P CGM systems, a 70% Time in Range (TIR) is the reference target. The table below shows the percentage by which each system reads higher than venous glucose when glucose exceeds 10.0 mmol/L (180 mg/dL).

This must be interpreted in context: capillary glucose is typically at 5-10% higher than venous glucose above 10.0 mmol/L. As a result, these systems provide a clear and physiologically meaningful representation of hyperglycaemic exposure, aligning closely with the glucose levels to which tissues are actually exposed.

CGM systems reading in Zone B

On average, these systems read below physiological glucose exposure, particularly at glucose concentrations above 10.0 mmol/L (180 mg/dL).

For Zone B CGM systems, 75-80% Time in Range (TIR) is the reference target. The table below shows the percentage by which each system under-reads relative to venous glucose when glucose exceeds 10.0 mmol/L (180 mg/dL).

This must be interpreted in context: capillary glucose is typically at least 5-10% higher than venous glucose above 10.0 mmol/L. As a result, these systems under-represent hyperglycaemic exposure, and thats why they require 5-10% more time-in-range compared to Zone P CGM systems.

To understand this fully, read the deep dive.

Let’s move forward to assessing accuracy of the CGM systems where their risk is understood and Zone is known.