Glucose Monitors for Non-Diabetics: What You Can Learn and What You Can't

Oak
cgm glucose wearables metabolic-health data-analysis
A glowing grazing creature drinking in a bioluminescent forest, its flank lit in a rising and falling arc echoed on vines behind it.

Yes, you can buy a continuous glucose monitor without diabetes and without a prescription, and yes, it can teach you something useful. Since 2024, over-the-counter CGMs (Dexcom Stelo, Abbott Lingo and Libre Rio) have been cleared in the US for adults not using insulin, so the access question is settled. The harder question is what the data is worth. Our view: a two-week to four-week wear, treated as a designed experiment rather than a continuous anxiety feed, is one of the cheapest high-resolution measurements you can make on your own physiology. Wearing one indefinitely and reacting to every excursion is mostly noise, and interpreting any abnormal pattern you find requires a clinician, because CGM is not a diagnostic instrument for diabetes or prediabetes.

What a CGM measures, and why that matters for interpretation

A CGM does not measure blood glucose. A filament sits in the subcutaneous interstitial fluid and generates current proportional to glucose via immobilized glucose oxidase, and firmware converts that current to a glucose estimate using a factory calibration. Two consequences follow. First, there is a physiological lag between plasma and interstitial glucose, roughly 5 to 15 minutes, largest when glucose is changing fastest, so the peak you see after a meal is later and often blunter than the true plasma peak. Second, the sensor signal drifts over its wear period, and manufacturers compensate with algorithms rather than fingerstick calibration. The engineering constraints here (enzyme stability, oxygen dependence, membrane biofouling, the tradeoff between response time and signal-to-noise) are well documented in the sensor literature and have not fundamentally changed since electrochemical sensors displaced earlier approaches.12

Accuracy is usually reported as MARD, the mean absolute relative difference against a laboratory reference. Modern sensors land around 8 to 10 percent MARD overall, but the error is not uniform: it is worse in the first 12 to 24 hours after insertion, worse at low glucose, and worse during rapid change. For a non-diabetic reader, a 9 percent MARD means a displayed 110 mg/dL is consistent with a true value anywhere from roughly 100 to 120. That is fine for seeing the shape of a response and useless for arguing about whether your fasting glucose is 94 or 99.

The non-invasive alternatives you may have seen advertised do not work yet. Sweat, tear, and saliva glucose sensors have been demonstrated repeatedly in the laboratory, but the correlation with blood glucose is weak and confounded by secretion rate, contamination, and dilution.34 Optical and radar-based “no needle” wrist devices have generated a large patent portfolio without producing a validated product.5 If a device claims to measure glucose without penetrating skin, assume it is estimating it from heart rate and motion.

Why a technically fluent non-diabetic would wear one

The strongest reason is that glucose is one of very few metabolically informative analytes you can sample every five minutes in a free-living person, and time-series density buys you things a single fasting lab draw cannot. You can measure your own postprandial response to specific foods and, more usefully, to specific combinations and orderings. You can quantify the effect of a 15-minute walk after a meal versus sitting. You can see what a night of four hours of sleep does to your morning curve, and what an acute stressor does at 3 pm with no food involved. Individual postprandial responses to identical meals vary widely between people, which is precisely why an n-of-1 measurement adds information that population nutrition advice cannot.

There is also a forward-looking case that glycemic variability itself, separate from mean glucose, carries information about cardiovascular risk and aging biology. This is a hypothesis under active investigation rather than established fact, and the reviews making the case are explicit that outcome trials in people without diabetes do not yet exist.6 Treat variability metrics as interesting personal descriptors, not as a risk score. Similarly, the professional consensus statements on CGM were written for insulin-treated diabetes, where the value of the device is in avoiding hypoglycemia and guiding dosing, and they do not extend to healthy users.7

The reasons not to wear one continuously are real. Sensor noise and compression artifacts (lying on the sensor, which produces false lows) will generate alarming numbers that mean nothing. A reading of 68 mg/dL at 4 am on a sensor with 10 percent MARD in the low range is not evidence of anything. People with a tendency toward food anxiety should be careful here, and if a CGM is making you restrict foods you have no clinical reason to avoid, stop wearing it.

Which sensor to use

For a self-experiment, we would use a Dexcom Stelo or an Abbott Libre 3 series sensor, and the choice comes down to data access. Stelo reports every 15 minutes with 15-day wear and is designed for the non-insulin market. The Libre 3 Plus reports every minute over 15 days, which is a meaningfully denser time series and the better instrument if you care about peak timing. Lingo is the consumer-branded Abbott sensor with a more restrictive app and a heavier emphasis on a proprietary score, which we would avoid.

Whatever you pick, plan your data path before you insert the sensor. All three vendors let you export, but the export granularity differs and some of it is only available through a clinician-oriented portal. Abbott’s LibreView produces a CSV with columns for device timestamp, record type, historic glucose, and scan glucose, at whatever cadence the sensor logged. Dexcom’s Clarity export gives a similar CSV with an “EGV” (estimated glucose value) column plus event rows. Read it in with pandas, parse the timestamp with an explicit format rather than letting the parser guess, set a DatetimeIndex, and resample to a fixed grid with df.resample('5min').mean() before you compute anything, because the raw rows contain duplicated timestamps and gaps around sensor restarts. Drop the first 24 hours of every sensor as burn-in.

A protocol worth running

Run two weeks, not forever, and structure it. Week one is observation: eat and live normally, log meals with timestamps to the minute, and do not change behavior. This gives you a baseline distribution and identifies your natural excursions. Week two is intervention, and the design matters more than the device. Pick two or three standardized test meals you can reproduce exactly (mass out the ingredients), and repeat each one at least three times on different days at the same time of day, because within-person variance for the same meal is large enough that a single trial tells you almost nothing. Then vary one factor: a 20-minute walk starting 15 minutes after eating, or the same carbohydrate eaten after a protein and fat course rather than first.

For each meal, compute incremental area under the curve over 120 minutes using the pre-meal 15-minute mean as baseline, plus peak height and time to peak, and compare replicate means rather than eyeballing overlaid curves. Over the whole wear, compute mean glucose, standard deviation, coefficient of variation, and percentage of time between 70 and 140 mg/dL. In people without diabetes wearing research-grade sensors, mean glucose typically sits in the high 90s to low 100s mg/dL, coefficient of variation usually under about 20 percent, and the large majority of the day between 70 and 140. Use those as orientation, not as thresholds you have passed or failed.

The interpretive limit is worth stating plainly. If your fasting values are consistently above roughly 100 mg/dL, or your post-meal values routinely exceed 180, the correct next step is a venous fasting glucose, an HbA1c, and a conversation with a physician, not a purchase decision about breakfast. CGM output is not a diagnostic criterion for any glucose disorder, and sensor error in that range is large enough to matter. A CGM also responds to acute physiological stress unrelated to diet or metabolic health, which is one reason isolated excursions are hard to interpret in people without diabetes.8

Questions people also ask

Should non-diabetics use a glucose monitor? For a bounded self-experiment, we think yes, if you will design it and analyze the export rather than watch the app. Continuous indefinite wear without a clinical indication mostly produces noise and worry.

Can I monitor my blood sugar without pricking my finger? Yes, with a CGM, but the filament still penetrates skin. Devices that claim to read glucose optically or through a wristband with no skin penetration are not validated, and the non-invasive sweat and tear approaches remain research-stage.34

What is a normal blood sugar level for a non-diabetic? Fasting venous glucose below 100 mg/dL and a two-hour post-load value below 140 mg/dL are the conventional laboratory reference points. On a CGM, most non-diabetic adults spend the large majority of the day between 70 and 140 with a mean near 100, though sensor error of 8 to 10 percent means you should not read single values precisely.

Can anybody wear a glucose monitor? Anyone over 18 not taking insulin can buy an OTC sensor in the US. If you take insulin or a sulfonylurea, you need a prescription sensor and clinical guidance, because the accuracy characteristics at low glucose matter for dosing decisions.7

Why would a non-diabetic wear one? To measure personal postprandial responses, to quantify the effect of exercise timing and sleep loss on your own curves, and to generate a high-density time series you can integrate with other molecular data.6

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. T. Koschinsky, L. Heinemann. Sensors for glucose monitoring: technical and clinical aspects. Diabetes/Metabolism Research and Reviews, 2001. https://doi.org/10.1002/dmrr.188 ↩

  2. Martina Vettoretti, Giacomo Cappon, Giada Acciaroli, et al. Continuous Glucose Monitoring: Current Use in Diabetes Management and Possible Future Applications. Journal of Diabetes Science and Technology, 2018. https://doi.org/10.1177/1932296818774078 ↩

  3. Jayoung Kim, Alan S. Campbell, Joseph Wang. Wearable non-invasive epidermal glucose sensors: A review. Talanta, 2018. https://doi.org/10.1016/j.talanta.2017.08.077 ↩ ↩2

  4. Jacquelyn Yazdani, Kamal Mafatia, Md. Harun-Or-Rashid, et al. Tear-based glucose monitoring: A non-invasive approach to diabetes control in resource-limited settings. Biosensors and Bioelectronics, 2026. https://doi.org/10.1016/j.bios.2025.118209 ↩ ↩2

  5. Olena Litvinova, Magdalena Eitenberger, Aylin Bilir, et al. Patent analysis of digital sensors for continuous glucose monitoring. Frontiers in Public Health, 2023. https://doi.org/10.3389/fpubh.2023.1205903 ↩

  6. Cristina Văcărescu, Dragos Cozma. Beyond Diabetes: Continuous Glucose Monitoring as a Candidate Precision Tool for Cardiovascular Prevention and Healthy Longevity—A Hypothesis-Generating Narrative Review. Medicina, 2026. https://doi.org/10.3390/medicina62081513 ↩ ↩2

  7. Thomas C. Blevins, Bruce W. Bode, Satish K. Garg, et al. Statement by the American Association of Clinical Endocrinologists Consensus Panel on Continuous Glucose Monitoring. Endocrine Practice, 2010. https://doi.org/10.4158/ep.16.5.730 ↩ ↩2

  8. Adnan Babar, Shafi Ullah Khan, Syed Junaid Hussain Bukhari, et al. Blood Glucose Concentration in Non-Diabetic Individuals Undergoing Dental Extraction: Comparison of Pre vs Post Extraction levels. Pakistan Armed Forces Medical Journal, 2026. https://doi.org/10.51253/pafmj.v76isuppl-8.12258 ↩