How Accurate Is Dexcom G7? MARD and Key Limits

The Dexcom G7 continuous glucose monitor delivers an overall mean absolute relative difference (MARD) of about 8% when worn on the upper arm, meaning its readings are, on average, within roughly 8% of a lab-quality blood glucose measurement. That single-digit figure places it among the more accurate consumer CGM systems available, but MARD is an average, and averages can hide the moments that matter most. The sensor’s real-world performance shifts depending on where you wear it, how long it has been on, whether your glucose is crashing or climbing, and even whether you are lying on it in your sleep.

What an 8% MARD Actually Means Day to Day

In the pivotal adult trial of the Dexcom G7, arm-placed sensors returned an overall MARD of 8.2% and abdomen-placed sensors came in at 9.1%, based on nearly 78,000 matched data pairs from 316 participants. The study also found that about 95% of readings on the arm fell within 20 mg/dL (for low glucose) or 20% (for higher glucose) of the reference value.¹ In concrete terms, if your blood glucose is 150 mg/dL, the G7 is likely to show something between roughly 120 and 180 mg/dL, and most of the time it will be tighter than that. If your glucose is 100 mg/dL, the expected spread is narrower in absolute milligrams but proportionally the same.

A newer, extended-wear version of the G7 designed for 15.5 days of use posted a similar overall MARD of 8.0%, with about 94% of readings meeting the 20/20 agreement standard.² So the longer sensor life does not appear to come at a meaningful cost to accuracy.

The First Day Is the Worst Day

One pattern that catches many users off guard is the warm-up accuracy dip. Day 1 MARD on the arm was about 11.9% in the adult pivotal trial, jumping to a noticeably tighter 8.4% on day 2 and settling around 7.2% by day 4. Abdomen sensors followed the same curve, starting near 12.9% on day 1 and improving to about 7.7% by day 4.³ A study in pregnant women with diabetes showed an even starker break-in effect: only about 79% of readings met the 20/20 standard on day 1, compared with roughly 96% by day 4 and 97% by day 10.⁴

The practical takeaway is straightforward: treat readings during the first several hours after insertion with extra skepticism. If you are making a high-stakes dosing decision on day 1, a confirmatory fingerstick is a smart move. By mid-sensor life, accuracy is at its best, and it tends to hold through the end of the wear period, including the 12-hour grace window after day 10.

Where Accuracy Drops Most: Low Blood Sugar

MARD is a percentage, and percentages behave strangely at low numbers. When glucose is, say, 60 mg/dL, even a modest absolute error of 9 mg/dL works out to a 15% MARD. A real-world pediatric study in hospitalized children found exactly this pattern: MARD was about 8.5% in the hyperglycemic range (above 180 mg/dL), rose to around 11% in the normal range, and peaked at roughly 15% during hypoglycemia.⁵ Paradoxically, the absolute error in milligrams was actually smallest during low glucose, but because the denominator is small, the percentage error balloons.

This matters because hypoglycemia is exactly the moment you need your CGM to be right. The G7 does perform better at catching lows than its predecessors. A head-to-head comparison across Dexcom’s fifth-, sixth-, and seventh-generation systems found that the G7 had a lower rate of missed hypoglycemia detections when the alert threshold was set at 55 mg/dL.⁶ Still, if you feel symptomatic and the CGM says you are fine, trust your body and check with a fingerstick.

The Physiological Lag Between Blood and Sensor

Every CGM reads glucose from interstitial fluid, the thin layer of liquid surrounding your cells beneath the skin, not directly from your bloodstream. Glucose has to travel from blood vessels into that fluid before the sensor can detect it. In healthy adults under fasting conditions, this transport takes about five to six minutes on average.⁷ The G7’s sensor processing adds its own lag on top of the physiological one, though the system’s overall lag has been measured at roughly 3.4 to 3.6 minutes, shorter than older Dexcom models.⁶

In practice, these lags combine so that the number on your screen is always a few minutes behind what is happening in your blood. During stable glucose, the difference is invisible. During a rapid drop or spike, it is not. If you have just finished a meal and your blood glucose is climbing steeply, the CGM will underestimate the peak and then continue to show elevated values after your blood sugar has already started to fall. The same delay applies in reverse during fast drops, which is why CGM trend arrows exist: a steeply falling arrow means the actual blood glucose is probably already lower than the number displayed.

Arm Versus Abdomen Placement

Across every Dexcom G7 study, the upper arm consistently outperforms the abdomen by a small but real margin. In adults, the difference was about one percentage point of MARD: 8.2% on the arm versus 9.1% on the abdomen.¹ In children and adolescents, the gap was nearly identical: 8.1% on the arm, 9.0% on the abdomen.⁸

Why the difference? The tissue characteristics of the back of the upper arm, including blood flow and the ratio of fat to muscle, tend to produce a more stable interstitial environment for the sensor. Abdomen tissue is more variable person to person, and movement and compression during daily activities (bending, sitting, wearing a seatbelt) can disturb the sensor more readily. If you have the choice, the arm is the better-performing site. For children aged two to six, who were studied only with abdomen placement, the MARD was about 9.3%, and roughly 92% of readings met the 20/20 standard.⁸

How the G7 Stacks Up Against Earlier Dexcom Generations

Each Dexcom generation has improved on the last in a fairly predictable way: lower MARD, shorter lag, fewer user steps, and a smaller physical profile. The G7 showed higher agreement rates on the 15/15 and 20/20 criteria than either the G5 or the G6 when all were compared with arm placement, and it detected more hypoglycemia episodes than either predecessor.⁶ The insertion process also got simpler: the G7 required fewer decisions and physical steps from the user than either earlier system, which matters for people with limited dexterity or children inserting sensors on their own.

The sensor’s warm-up time dropped significantly with the G7, and its overall physical footprint shrank. The transmitter is thinner, and the skin contact area is smaller, which can reduce irritation and make the sensor easier to conceal under clothing. For users who relied on the G6 and felt it was “good enough,” the G7 represents a meaningful but incremental accuracy improvement, not a revolutionary leap.

Head-to-Head with the FreeStyle Libre 3

The G7’s main competitor is Abbott’s FreeStyle Libre 3, and comparisons between the two have produced some results that Dexcom users might find surprising. One study that directly compared both sensors on the same participants reported a MARD of 13.6% for the G7 versus 8.9% for the Libre 3, with the Libre 3 also showing a higher percentage of readings within the 20/20 agreement window. The difference was most pronounced after the first 12 hours of wear, when the G7’s MARD climbed to about 15.1% while the Libre 3 remained around 10%.⁹

Those G7 numbers are considerably worse than the 8.2% seen in the pivotal trial, which is worth pausing on. Different study designs, reference methods, participant populations, and environmental conditions can all influence MARD results. The pivotal trial used in-clinic glucose sampling under controlled conditions; a head-to-head study in daily life captures more of the messiness that real users experience. Neither number is “wrong,” but they tell different stories about what to expect.

An even more extreme divergence appeared in a study of recreational scuba diving in people with type 1 diabetes: the G7 posted a MARD of 31%, compared with about 14% for the Libre 3. Neither system met the accuracy threshold for making insulin dosing decisions in that setting.¹⁰ Pressure changes, cold water, and altered blood flow to the skin during diving all create conditions that push CGMs far outside their design envelope. The takeaway is not that the G7 is bad underwater (no CGM is good underwater) but that environmental extremes can degrade accuracy far beyond what headline MARD numbers suggest.

Why MARD Is an Imperfect Yardstick

MARD has become the default number people look at when comparing CGM systems, and that is partly by default: it is the single metric manufacturers report most prominently. But researchers who specialize in CGM evaluation have cautioned that MARD alone is not enough to characterize a system’s real-world performance. Complex factors, including when during the sensor life the reference measurements were taken, which glucose ranges were represented, and how many paired data points were included, can all shift the reported MARD without changing how the sensor actually feels to the wearer.¹¹

Two systems could report identical MARDs yet differ in ways that matter far more clinically: one might have tight accuracy during stable glucose but wild swings during rapid change, while the other distributes its errors more evenly. MARD also does not capture the directionality of error. A sensor that consistently reads a little high is much easier to work around than one that swings unpredictably above and below the true value, even if both have the same average deviation. When you see MARD comparisons between systems, treat them as a useful starting point and not as the final word.

Pressure, Sleep, and Physical Interference

If you have ever woken up to a sudden CGM low alarm, checked your blood with a fingerstick, and found your glucose was perfectly normal, you have probably experienced a pressure-induced sensor attenuation. Lying on the sensor compresses the tissue around it, reducing blood flow to the area and temporarily depriving the sensor of glucose-rich interstitial fluid. The result is a false low reading that can last as long as the pressure continues.¹² The signal typically recovers within minutes of rolling off the sensor.

For users on automated insulin delivery systems (often called closed-loop or hybrid closed-loop pumps), these false lows are more than an annoyance. The algorithm may reduce or suspend insulin delivery in response to the artificially low reading, potentially leading to a genuine high later. Placing the sensor in a spot you are unlikely to sleep on, such as the back of the upper arm rather than the abdomen, reduces the frequency of these events.

Acetaminophen and Other Chemical Interferences

Older CGM sensors were notoriously vulnerable to acetaminophen (Tylenol). A pilot study using an earlier-generation Dexcom sensor showed that plasma glucose stayed steady at around 90 mg/dL while the CGM reading shot up to as high as 400 mg/dL after a dose of acetaminophen, purely because the drug’s metabolites are electrochemically active on the same sensor surface that detects glucose.¹³

Starting with the G6, Dexcom introduced a redesigned sensor chemistry specifically to address this. Testing of the G6 found that the average interference from a standard 1-gram acetaminophen dose was only about 3 mg/dL, well below what would be considered clinically meaningful.¹⁴ The G7 uses the same core sensor chemistry as the G6, so it carries the same resistance. That said, Dexcom’s labeling still notes that high-dose vitamin C (ascorbic acid) and hydroxyurea can affect readings. If you take either regularly, it is worth being aware that your numbers may drift in one direction.

What iCGM Clearance Tells You (and What It Does Not)

The Dexcom G7 is classified by the FDA as an integrated continuous glucose monitor (iCGM). This designation means the device met specific performance benchmarks that allow it to be used for insulin dosing decisions without mandatory fingerstick confirmation, and it is cleared to work with compatible automated insulin delivery systems. Not all CGMs carry this designation, and non-iCGM devices should not be assumed to meet the same accuracy or reliability bar.¹⁵

However, iCGM clearance is a threshold, not a guarantee. It confirms the device passed standardized testing under controlled conditions. It does not mean every reading you get in daily life will meet those performance goals. As the studies above illustrate, day 1 readings, extreme glucose levels, physical compression, and unusual environments can all push accuracy below what the FDA-cleared benchmarks suggest. The iCGM label is a floor you can generally rely on, not a ceiling you should expect to hit at every moment.

Situations That Push the G7 Outside Its Comfort Zone

Beyond the factors already discussed, a few additional conditions can degrade accuracy in ways that are not obvious from the spec sheet:

Dehydration: Reduced fluid volume in the interstitial space can alter the concentration gradient between blood and tissue, exaggerating the physiological lag and potentially skewing readings.
Rapid temperature changes: Moving from a heated room into cold outdoor air, or stepping into a hot shower, can temporarily change blood flow to the skin and the sensor site, causing short-lived reading drifts.
Scar tissue and lipohypertrophy: Repeated sensor placement in the same small area can lead to tissue changes that impair glucose diffusion to the sensor. Rotating sites is important for long-term accuracy.
Altitude and pressure extremes: The scuba diving study is an extreme example, but even activities at high altitude with significant barometric pressure changes can affect readings.

None of these mean you should stop trusting your G7. They mean you should recognize that a CGM is a trend tool with very good accuracy most of the time, rather than a perfect laboratory instrument. When the stakes are highest, such as treating a suspected severe low or making a large correction dose, a fingerstick remains the gold standard verification.

Why Pivotal Trial Numbers and Real-World Numbers Often Disagree

If you read Dexcom’s promotional materials, you will see the pivotal trial MARD front and center. Those numbers come from carefully run studies where sensors are applied by trained staff, reference samples are drawn on schedule, and participants are monitored in clinic. Real-world studies tend to return somewhat higher MARD values, because they capture the full range of conditions users actually encounter: forgotten calibrations on systems that need them, sensors bumped during exercise, readings taken during illness or stress, and insertion technique that varies from person to person.

The head-to-head Libre 3 comparison, where the G7 posted a MARD of about 14% rather than 8%, is a good illustration of this gap.⁹ The hospitalized pediatric study, where overall MARD came in around 11% for readings in the normal glucose range, is another.⁵ Neither study invalidates the pivotal data. They simply reflect a different, messier truth. When someone asks you how accurate the G7 is, the honest answer is “around 8% MARD under favorable conditions, and somewhat higher in everyday use depending on the situation.” That range is still good enough to make informed dosing decisions most of the time, which is ultimately what the device is designed to do.