What Wearing a CGM Is Like, and What the Data Is Worth

Oak
cgm glucose wearables metabolic-health self-measurement
A lab machine lowers a coin-sized sensor with a hair-thin glowing filament into a block of clear gel, where light blooms spread outward.

Wearing a continuous glucose monitor is, for most people, uneventful. A spring-loaded applicator pushes a flexible filament a few millimeters into the subcutaneous fat of the back of the upper arm, the introducer needle retracts immediately, and what stays behind is a soft wire thinner than a human hair under an adhesive patch about the size of two stacked coins. Insertion usually registers as a snap and a brief pinch. After the first hour most people stop noticing the sensor except when a doorframe or a car seat catches it. The discomfort people do report is almost always mechanical (adhesive pulling, pressure on the site while sleeping) rather than the filament itself.

The more interesting question for a technically fluent reader is what the resulting time series is worth. A modern factory-calibrated sensor gives you roughly 1,300 to 2,700 glucose readings over 10 to 14 days, at 1- or 5-minute resolution, from interstitial fluid rather than blood. That is enough to characterize your own postprandial responses, your overnight baseline, and the effect of sleep and exercise on both, provided you understand the sensor’s error structure well enough not to over-read small differences.

What is physically in your arm

The filament is a flexible electrode, typically platinum or carbon with a glucose oxidase enzyme layer and a diffusion-limiting membrane over it. Glucose in interstitial fluid diffuses through the membrane, the enzyme oxidizes it, and the resulting current is proportional to glucose concentration over the working range. This amperometric design has been the backbone of wearable glucose sensing for two decades, and the engineering problems that remain are stability of the enzyme layer, biofouling, and the foreign-body response around the filament 1.

Insertion depth is about 5 mm for the FreeStyle Libre family and roughly 4 to 7 mm depending on device and applicator geometry. That places the tip in subcutaneous adipose tissue, below the dermis and above muscle. There is a needle involved, but only transiently: an introducer needle carries the filament through the skin and withdraws in the same motion, leaving no rigid metal in place. The Dexcom G6 and G7 applicators work the same way. Whether you feel it depends mostly on site vascularity and whether the needle clips a small cutaneous nerve, which happens occasionally and produces a sharp point of soreness that usually fades within a day.

If the sensor hurts when you press on it, the usual cause is that the filament is tethered in tissue that you are compressing, or that a small hematoma formed at insertion. Persistent pain, spreading redness, warmth, or discharge is a different matter and is a reason to remove the sensor and contact a clinician. We are describing measurement here, not giving medical advice.

Accuracy, lag, and the errors you should expect

Interstitial glucose tracks blood glucose with a physiological delay of roughly 5 to 15 minutes, longer when glucose is changing quickly. During a steep postprandial rise, the sensor reports a value that blood reached several minutes earlier. Device algorithms partially compensate with prediction, which introduces its own artifacts: sharp corrections and occasional overshoot at inflection points. For interpreting meal responses this matters mainly for peak timing, not peak magnitude.

Mean absolute relative difference (MARD) for current consumer sensors is generally reported in the 8 to 10 percent range against a laboratory reference. Treat that as an average over a wide glucose range, not a per-reading guarantee. Error is not uniform: it is typically worse in the first 12 to 24 hours after insertion, worse at low glucose, and worse during rapid change. Compression lows, where lying on the sensor restricts local perfusion and produces a false 45 to 60 mg/dL trough that recovers the moment you roll over, are the single most common artifact in overnight data from non-diabetic wearers. They are easy to spot: an abrupt drop with no dietary or exercise antecedent, a flat bottom, and an equally abrupt recovery.

The acetaminophen interference people associate with Dexcom is a chemistry problem, not a software one. Acetaminophen is electroactive at the potential used to detect hydrogen peroxide from the enzyme reaction, so it contributes current that the sensor reads as glucose. Dexcom’s G6 and G7 use a membrane that blocks this at standard doses, though the manufacturer still notes interference at higher intakes. Ascorbic acid at supplement-level doses can do something similar on some platforms. If you are taking either, annotate your timeline so you do not chase a phantom excursion.

How to get a usable dataset out of 14 days

Wear position matters more than most people expect. Use the posterior upper arm, rotate arms between sensors, and avoid sites with thin subcutaneous fat or heavy mechanical loading. Apply to clean, dry, alcohol-wiped skin and let it dry fully before the applicator touches it. If the adhesive lifts, an overpatch is a reasonable fix, but do not cover the small vent hole on the transmitter housing.

Discard the first 12 hours of every sensor. The warm-up period is nominally an hour, but the settling of the tissue response around the filament takes longer, and early data tends to run low with more noise. If you are running two sensors in parallel to estimate your own device-level variance, which we think is the most informative thing a curious wearer can do, expect 10 to 15 mg/dL disagreement between simultaneous sensors on the same person at the same time. That number is the resolution floor for any individual reading, and it should discipline how much you read into a 6 mg/dL difference between two breakfasts.

For export, Dexcom Clarity produces a CSV with timestamps, event type, and glucose in mg/dL or mmol/L. The Libre app exports similarly, with separate rows for historic (15-minute) and scan-derived values on older hardware. Both use local time without offset, so daylight-saving transitions will silently corrupt an unwary join against other time series. Convert to UTC on ingest.

A minimal analysis worth doing in pandas: resample to a uniform 5-minute grid with linear interpolation across gaps shorter than 20 minutes and NaN beyond that, then compute per-day mean, standard deviation, coefficient of variation, and the fraction of time in a range you choose in advance. Compute overnight (00:00 to 06:00) mean separately, since it is the least behaviorally contaminated window and the most stable day-to-day summary you will get. For meals, log start times to the minute and extract a 3-hour window from each, reporting incremental area under the curve above the pre-meal baseline rather than peak value alone, since peak is the statistic most corrupted by lag and noise.

Repeat the same meal at least three times on different days before believing anything about it. Within-person variability in response to an identical meal is substantial, driven by sleep, prior activity, time of day, and the preceding meal.

Is it worth wearing one if you are not diabetic

The strongest evidence for CGM is in insulin-treated diabetes, where sensor use is associated with improved glycemic control and reduced hypoglycemia across age groups, including adolescents and young adults where adherence is historically difficult 23. Qualitative work with type 1 patients describes the trade the technology makes: better information and less finger-pricking against adhesive irritation, alarm fatigue, and the psychological weight of continuous numbers 4. That last cost is real for non-diabetic wearers too, and it is the main reason we suggest fixed wear periods rather than indefinite use.

For a metabolically healthy person, the value is characterization rather than management. Two weeks tells you your fasting baseline, your typical overnight floor, how your particular breakfast behaves, and whether a post-meal walk changes anything measurable for you. Research groups are building models on CGM traces to predict dysglycemia before it appears in standard labs, which suggests the shape of the curve carries information that a single fasting glucose does not 5. What no consumer CGM will do is diagnose anything. Diabetes and prediabetes are defined by fasting plasma glucose, HbA1c, or an oral glucose tolerance test, run in a laboratory and read by a clinician. If your sensor data looks concerning, that is a reason to get venous labs, not a reason to conclude anything.

Questions people also ask

Is wearing a CGM uncomfortable? Rarely, after the first few hours. The common complaints are adhesive itching, a bruise at the insertion site, and snagging the sensor on clothing. If pain persists beyond a day or the site becomes red and warm, remove it and talk to a clinician.

Does the FreeStyle Libre have a needle? Can you feel it? It has an introducer needle that inserts the sensor filament and withdraws immediately. Most people feel a brief snap and pinch at application. Nothing rigid stays in the arm.

How deep does the CGM filament go? About 5 mm for the Libre, roughly 4 to 7 mm across devices, which puts it in subcutaneous fat rather than muscle or a blood vessel.

Why is acetaminophen a problem with Dexcom? Acetaminophen is electrochemically active at the sensor’s operating potential and can be counted as glucose. Current membranes block this at standard doses, but high intake can still shift readings upward.

Why don’t hospitals use CGM routinely? Inpatient use faces workflow and validation barriers rather than sensor-quality ones. Studies of implementation during COVID-19 found nursing workload, unfamiliarity, connectivity, and unclear protocols were the main obstacles to bedside adoption 6. Accuracy in critically ill patients with altered perfusion, edema, or vasopressor use is also less well characterized, though feasibility work in acute pediatric settings has been encouraging 7.

Will there ever be a needle-free version? Tear-fluid and near-infrared approaches have both been demonstrated, and tear glucose correlates with blood glucose well enough to be interesting 8. Optical methods still struggle with skin heterogeneity, temperature, and calibration drift, and none has displaced the enzymatic subcutaneous sensor 9.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Qiuyuan Chen, Yan Zhao, Yunqi Liu. Current development in wearable glucose meters. Chinese Chemical Letters, 2021. https://doi.org/10.1016/j.cclet.2021.05.043 ↩

  2. Matthew Leinung, Shannon Thompson, Elizabeth Nardacci. Benefits of Continuous Glucose Monitor Use in Clinical Practice. Endocrine Practice, 2010. https://doi.org/10.4158/ep09287.or ↩

  3. Myron Andrew Lee, Deborah Jane Holmes‐Walker, Kaye Farrell, et al. Impact of continuous glucose monitoring in youth with type 1 diabetes aged 15–21 years. Internal Medicine Journal, 2022. https://doi.org/10.1111/imj.15347 ↩

  4. Anxela Soto-Rodriguez, Ana Fernández-Conde, Raquel Leirós-Rodríguez, et al. The Experience of Patients with Type 1 Diabetes Mellitus with the Use of Glucose Monitoring Systems: A Qualitative Study. Nursing Reports, 2025. https://doi.org/10.3390/nursrep15080294 ↩

  5. Kaniz FatemaTuz Zahura, Tamanna Akter, Kazi Foyeza Akther, et al. Early Prediction of Type 2 Diabetes Using Deep Learning on Continuous Glucose Monitoring (Cgm) Data - A Systematic Review. Journal of Evidence-Based Medical Research, 2025. https://doi.org/10.66687/jebmr.1.03.2025.13 ↩

  6. Eileen R. Faulds, Laureen Jones, Molly McNett, et al. Facilitators and Barriers to Nursing Implementation of Continuous Glucose Monitoring (CGM) in Critically Ill Patients With COVID-19. Endocrine Practice, 2021. https://doi.org/10.1016/j.eprac.2021.01.011 ↩

  7. Verónica Izquierdo, Nerea Afonso-Bouza, Eva María Montoto-Méndez, et al. Feasibility of continuous glucose monitoring in children with diabetic ketoacidosis: an exploratory observational study. European Journal of Pediatrics, 2025. https://doi.org/10.1007/s00431-025-06368-2 ↩

  8. Alicja E. Kownacka, Dovile Vegelyte, Maurits Joosse, et al. Clinical Evidence for Use of a Noninvasive Biosensor for Tear Glucose as an Alternative to Painful Finger-Prick for Diabetes Management Utilizing a Biopolymer Coating. Biomacromolecules, 2018. https://doi.org/10.1021/acs.biomac.8b01429 ↩

  9. Aminah Hina, Wala Saadeh. Noninvasive Blood Glucose Monitoring Systems Using Near-Infrared Technology—A Review. Sensors, 2022. https://doi.org/10.3390/s22134855 ↩