Tip · Tracking

Continuous glucose monitoring (CGM)

Q: Does the CGM show me my personal trigger foods?

Less reliably than the advertising suggests. In an inpatient study with fully controlled diets, the same meal was served again to the same person a week later; repeatability was at an intraclass correlation of 0.17 to 0.28, and the scatter for identical meals was just as large as for completely different ones. If you want to assess a food, you have to test it several times and average the results.

Q: Why does my sensor show higher values than my blood glucose meter?

Because it does so systematically. In a randomized crossover study with 15 healthy adults, CGM readings were around 0.9 mmol/l above the capillary reference values both fasting and after eating, and the deviation differed from person to person. In addition, the sensor does not measure blood but tissue fluid, with a time lag.

Q: My reading was 150 after breakfast – is that bad?

In 36 healthy young adults, the mean peak after a meal was 143.3 mg/dl with a standard deviation of 23.5 mg/dl. A value of 150 is therefore within the normal range. On top of that, the sensor overestimates: in a crossover study with 15 healthy people, time above 7.8 mmol/l was reported about four times too high uncorrected, and still about twice too high after aligning baseline values.

Q: Does a CGM do anything for me if my HbA1c is normal?

According to the current meta-analysis of 23 studies with 1,074 non-diabetics: in prediabetes, mean blood glucose improved; in normoglycemic people, no appreciable benefit was found. There was no established effect on body weight. Studies on hard endpoints do not exist at all.

Q: Can I detect early diabetes with a CGM?

Not as the sole tool. When 18 CGM experts were asked to assess the same 20 reports from people without diabetes, they agreed only at a Fleiss kappa of 0.36, which counts as poor agreement. Normal values for people without diabetes are simply lacking. HbA1c and fasting glucose remain the diagnostic measures.

Q: What is the most common problem when wearing one?

The skin. In a chemical analysis of 40 extracts from 27 diabetes devices, every single extract contained at least one known allergenic substance, and more than half of the devices contained acrylates. Among real users with type 1 diabetes, 16.2 percent reacted positively in patch testing, and 28.1 percent among those with a rash. Anyone with a known acrylate or colophony allergy should stay away from it.

A sensor on the upper arm, 14 days of glucose readings. Supported in diabetes; in metabolically healthy people there is no study on the benefit.

A CGM is a sensor on the upper arm that estimates the glucose level in subcutaneous fat tissue for two weeks. The tip is here because it has been measured precisely how the sensor deviates from the reference measurement in metabolically healthy people — and because this deviation decides whether a curve merely looks abnormal or actually is.

In short

The sensor does not measure what you think, and not as precisely as you think. Against capillary reference values, CGM readings in 15 healthy adults were 0.9 ± 0.6 mmol/l too high when fasting and 0.9 ± 0.5 mmol/l too high after eating (both P < 0.001); time above 7.8 mmol/l was overestimated about fourfold uncorrected and still about twofold after correction. The same meal produces very different curves in the same person: intraclass correlation 0.28 for the Abbott sensor, 0.17 for the Dexcom. Benefit has been measured in prediabetes, not in normoglycemic people.

What lies behind it

The sensor does not measure blood sugar but glucose in the interstitial fluid of subcutaneous fat tissue. An enzymatic wire generates a current proportional to the concentration, and a factory calibration converts it into a blood glucose value. Two systematic properties follow from this: a time lag relative to the blood and a calibration offset that differs from person to person.

For people with diabetes this is not critical. In metabolically healthy people, whose entire daily range lies between about 90 and 145 mg/dl, an offset of this size is, by contrast, a considerable share of the signal.

What testing against the reference measurement showed

The cleanest work on this comes from Bath: 15 healthy people, 7 laboratory visits in a randomized crossover design with a washout period of at least 48 hours, each with a carbohydrate load and measurement over 120 minutes every 15 minutes by CGM and capillary sampling. The CGM readings were 0.9 mmol/l above the reference both fasting and postprandially, and the fasting deviation differed significantly between participants (P = 0.001).

The bias also depended on the test meal: the glycemic index of a smoothie was 69 by CGM (95 % CI 48–99) and 53 by capillary measurement (40–69; P = 0.05). The authors’ conclusion: for precise quantification of glycemic responses to foods, capillary measurement is preferable.

Why the trigger-food test is so difficult

Under inpatient conditions with fully controlled diets, the same meal was served again to the same person about a week later; 1,189 responses from 30 people without diabetes were analyzed. The correlation between the rounds was r = 0.46 (Abbott) and r = 0.45 (Dexcom), and the intraclass correlation was 0.28 and 0.17.

The mean bias was small (Abbott −0.7 mg/dl, Dexcom 1.3 mg/dl) — on average the sensor measures correctly, but the individual reading is not reproducible, and the scatter for identical meals was about as large as for different ones. A single oatmeal spike is no information about oatmeal.

What is well supported

Robust above all is how the sensor behaves: systematic overestimation by 0.9 ± 0.6 mmol/l fasting and 0.9 ± 0.5 mmol/l postprandially against capillary reference, with an overestimation of time above 7.8 mmol/l by about fourfold, and by about twofold after aligning baseline values. In addition, the low repeatability of identical meals, with an ICC of 0.17 to 0.28 and a small mean bias. And a described normal profile: in 36 people without diabetes, median glucose in everyday life was 98.1 mg/dl (interquartile range 93.7–100.8), the mean value before meals was 92.8 ± 9.4 mg/dl and the mean peak afterwards was 143.3 ± 23.5 mg/dl. On the benefit side, one finding holds: in prediabetes, CGM feedback improved mean blood glucose (SMD = −0.54; 95 % CI −1.02 to −0.07; P = 0.03).

What the studies show

Hutchins 2025, Am J Clin Nutr

The first study to test specifically whether CGM accuracy depends on the test meal. 15 healthy adults, 7 laboratory visits in randomized order, measurement over 120 minutes by CGM and capillary sampling. CGM was 0.9 ± 0.6 mmol/l higher fasting and 0.9 ± 0.5 mmol/l higher postprandially (both P < 0.001). Uncorrected, time above 7.8 mmol/l was overestimated about fourfold, and about twofold after correction.

Hengist 2025, Am J Clin Nutr

Directly challenges the sales argument of individual trigger foods. Data from two inpatient feeding studies with 30 participants without diabetes, 1,189 responses to identical meals. Intraclass correlation 0.28 (Abbott) and 0.17 (Dexcom), limits of agreement −29.8 to 28.4 mg/dl and −29.4 to 32.1 mg/dl. The scatter for identical meals was about as large as for different ones. Conclusion: personalization based on CGM requires aggregated repeat measurements.

Freckmann 2024, J Diabetes Sci Technol

The reference study for what is normal in healthy people. 36 people without diabetes (23.7 ± 5.7 years) wore a CGM for up to 14 days: after 3 run-in days, 4 days with fixed meals and 7 days of everyday life. Median glucose 95.0 mg/dl with fixed meals and 98.1 mg/dl in everyday life, mean peak after a meal 143.3 ± 23.5 mg/dl.

Liao 2026, Eur J Med Res

The most recent systematic review for non-diabetics: 23 studies with 1,074 participants from 11 countries, including 7 randomized studies. Mean blood glucose improved (SMD = −0.54; 95 % CI −1.02 to −0.07; P = 0.03), body mass index did not (SMD = −0.25; P = 0.19). The benefit occurred in prediabetes; in normoglycemic people, no appreciable benefit could be demonstrated.

What the measurement does not show

There is not a single hard endpoint. Neither the meta-analysis of 1,074 non-diabetics nor the PREDICT analysis of 3,634 participants reports heart attack, stroke, diabetes incidence or mortality, and the PREDICT authors explicitly label their analyses as non-prespecified exploratory analyses. Interpretation is also uncertain: when 18 CGM experts assessed 20 reports from people without diabetes, agreement was at a Fleiss kappa of 0.36. The reference study on the normal profile also includes only 36 people. For the recommendation to measure twice a year, there is no study.

How to do it

14 days is the wear time of common sensors and corresponds to the study standard, although the reference study discarded the first 3 days as a run-in. Determine your personal fasting offset by measuring capillary blood in parallel on several mornings — the correction halves the overestimation of time above 7.8 mmol/l from about fourfold to about twofold. Test a food at least three times before you judge it. As a guide: peaks around 143.3 ± 23.5 mg/dl after eating are normal in healthy people. The only intervention with a measured CGM effect is exercise after eating: across 23 studies with 408 participants, the peak value fell by 0.63 mmol/l and time in range rose by 6.22 percentage points.

Safety

Skin reactions are the real physical risk. An analysis of 40 extracts from 27 diabetes devices identified 284 individual chemicals, and all 40 extracts contained at least one substance previously described as an allergen; 14 of the 27 devices (52 %) contained acrylates. In a user sample of 204 adults with type 1 diabetes, 16.2 % reacted positively in patch testing, and 28.1 % among those with a rash. Anyone with a known contact allergy to acrylates or colophony should avoid CGM. The psychological burden is measurable: among 56 adults not on insulin therapy, 68 % reported feeling fear of type 2 diabetes when seeing high readings, while 89 % reported positive changes in diet or exercise. CGM is unsuitable for people with an eating disorder.

BK-Score Long used, barely studied

Human evidence		3
Mechanism		7
Safety data		8
Hype gap		3
Track record of use		7

In diabetes, the benefit is established. In metabolically healthy people there is no study showing that avoiding glucose spikes leads to better outcomes – but there are reports of anxiety and unnecessary restriction.

What is rated is the state of knowledge, not the effect. “Safety data 9” means well studied – not harmless.
Subjective assessment by Biohacking Kompakt based on published scoring rules – not a scientific rating and not a medical recommendation. Rules and all ratings (German)

Frequently asked questions about continuous glucose monitoring (CGM)

Does the CGM show me my personal trigger foods?