Do CGMs measure ketones? One sensor now does both

Oak
cgm ketones biosensors time-series self-measurement
A feathered reptilian creature on a studio pedestal against black, with two fine glowing quills on its shoulder pulsing amber and blue.

No, a standard CGM does not measure ketones. A Dexcom G7, a FreeStyle Libre 2 or 3, or a Stelo reports one analyte: interstitial glucose, from a glucose oxidase or glucose dehydrogenase electrode. Ketones require a second enzyme and a second working electrode. The exception is Abbott’s Libre Duo 10 Day, a single filament carrying both a glucose electrode and a beta-hydroxybutyrate electrode, which received a CE Mark and then FDA authorization as the first wearable that continuously measures both. If you are on a Libre 3, a Libre 2, a G6, or a G7 today, your ketone data comes from a fingerstick meter, a breath device, or nothing.

What the two electrodes measure

Both analytes are read the same way: an enzyme immobilized on a wire electrode in subcutaneous interstitial fluid oxidizes the target molecule, a mediator shuttles electrons to the electrode, and the resulting current (nanoamps) is proportional to concentration. Amperometric enzyme electrodes are the chemistry behind essentially every commercial glucose sensor, with the design constraints well documented in the biosensor literature 1.

Glucose uses glucose oxidase or a glucose dehydrogenase variant. Ketone sensing uses 3-hydroxybutyrate dehydrogenase, which oxidizes D-beta-hydroxybutyrate (BHB) to acetoacetate with NAD+ as cofactor. That reaction is harder to run on a wearable than the glucose reaction: the cofactor has to be tethered or regenerated so it does not wash out over a 10 to 15 day wear, and the physiological concentration range is roughly two orders of magnitude lower than glucose. Sub-millimolar BHB against a 5 mmol/L glucose background on the same filament means cross-talk between electrodes is the main engineering problem, and the published description of the dual sensor addresses exactly that 2.

Note what is not measured. BHB is one of three ketone bodies. Urine strips measure acetoacetate, which falls as you become keto-adapted even while BHB stays elevated, which is why urine strips read low in people who have been in ketosis for weeks. Breath devices measure acetone. A continuous sensor measuring interstitial BHB is closest to the blood BHB number that clinical thresholds are written against, but it is a different fluid compartment with its own lag.

Which combinations are available

  • Libre Duo 10 Day: glucose and ketone, one sensor, minute-resolution glucose with background ketone monitoring. Availability varies by country and it is positioned for people with diabetes, particularly those on insulin pumps or SGLT2 inhibitors where ketone risk is real.
  • Every other CGM: glucose only.
  • Fingerstick dual meters (Abbott Precision Xtra, Keto-Mojo GK+): separate glucose and BHB strips, same meter body. Cheap, accurate enough, and discrete in time. If you want six ketone points a day, this works.
  • Breath acetone (Biosense): non-invasive, correlates with BHB loosely, poor for absolute values.
  • Non-invasive optical glucose: nothing validated. The review literature on optical and micro-system glucose sensing is a long catalog of promising benchtop results that have not reached accuracy suitable for dosing decisions 3. Treat any smartwatch claiming cuffless glucose as fiction.

The measurement artifacts you inherit

If you are going to fit models to this data, know what is in it before the biology.

Interstitial lag. CGM tracks blood glucose with a 5 to 15 minute delay, part physiological diffusion and part smoothing filter. During a rapid rise (an OGTT, a high-GI meal) your CGM peak arrives late and reads lower than the true plasma peak. Do not compare a CGM peak to a lab plasma value drawn at the same minute and conclude the sensor is broken.

Warmup drift. The first 12 to 24 hours after insertion are the least accurate part of the wear, and many sensors also drift on the last day. If you are running a controlled experiment, put the sensor on two days early.

Compression lows. Sleeping on the sensor produces a 40 to 60 mg/dL trough that recovers in 20 minutes with a sharp, symmetric shape. Flag any nocturnal excursion with a recovery slope steeper than the fall and treat it as an artifact, not hypoglycemia.

Sensor-to-sensor offset. Two sensors on the same person on the same day routinely differ by 10 to 15 mg/dL in mean. Never compare absolute means across a sensor change. Compare within-sensor contrasts.

For ketones, the analogous open questions are what constitutes clinically meaningful accuracy and what thresholds should trigger action. A consensus panel worked through those questions for continuous ketone monitoring, covering the measurement targets and the use cases where continuous ketone data changes management 4. The clinical case, particularly for insulin pump users and people on SGLT2 inhibitors where ketoacidosis can develop at near-normal glucose, is argued in the dual-monitoring literature 5.

Getting the data out

LibreView exports CSV. The columns you care about:

Device, Serial Number, Device Timestamp, Record Type,
Historic Glucose mg/dL, Scan Glucose mg/dL,
Non-numeric Rapid-Acting Insulin, Rapid-Acting Insulin (units),
Non-numeric Food, Carbohydrates (grams), ...
Ketone mmol/L, ...

Record Type is the field people get wrong. 0 is the historic 15-minute series, 1 is a manual scan, and the remaining codes are insulin, food, and notes entries. Filter to Record Type == 0 for a regular grid, then merge scans separately if you want them. Timestamps are local, naive, and will break at DST transitions. Dexcom’s Clarity export is similar in spirit with an Event Type column. Dexcom’s developer API returns data delayed by hours, so for real-time work people run xDrip+ or a Nightscout instance and pull from the Mongo collection instead.

Minimum viable pipeline:

import pandas as pd

df = pd.read_csv("libre.csv", skiprows=1)
g = df[df["Record Type"] == 0].copy()
g["t"] = pd.to_datetime(g["Device Timestamp"], format="%m-%d-%Y %I:%M %p")
g = g.set_index("t")["Historic Glucose mg/dL"].astype(float)
g = g.resample("5min").mean().interpolate(limit=3)  # cap gap filling

Cap the interpolation. A three-hour sensor dropout silently filled with a straight line will wreck any variability metric you compute afterward.

What to compute

Mean and coefficient of variation (SD/mean) over 14-day windows, not daily. CV is the variability metric that survives sensor offset better than SD alone. Time in range is a clinical metric designed for insulin dosing, and for a non-diabetic person it saturates near 100% and tells you almost nothing.

The useful signal for a healthy person is the meal response: align to timestamped meals, take a baseline as the mean of the 30 minutes before, and compute incremental AUC over 0 to 120 minutes with the trapezoid rule, plus peak delta and time to return to baseline. Repeat the same meal three times on different days. The within-person, within-meal standard deviation of iAUC is large, so a single test of a single food is noise.

If you have both analytes, the glucose-ketone index (glucose in mmol/L divided by BHB in mmol/L) is the ratio people track. Divide mg/dL by 18.0 to get mmol/L. The conventional nutritional ketosis band is 0.5 to 3.0 mmol/L BHB, a convention rather than a validated physiological boundary.

If your ketones are rising while you feel unwell, or you use insulin or an SGLT2 inhibitor, that is a clinical situation and you need a clinician, not a dashboard. Ketoacidosis can occur with normal glucose readings, which is the whole argument for continuous ketone data in the first place 5. Nothing in this page is a basis for changing a medication.

Questions people also ask

Does FreeStyle Libre 3 measure ketones? No. Libre 3 and Libre 3 Plus measure glucose only. The Libre Duo 10 Day is the Abbott product with a ketone electrode.

Can Libre 2 check ketones? No. Some Libre reader devices accept Abbott’s blood ketone test strips for a fingerstick BHB reading, which then appears in the LibreView export as a Ketone mmol/L row. That is a discrete blood measurement, not continuous sensing.

Are continuous ketone monitors available in the USA? The Libre Duo 10 Day has FDA authorization as the first wearable continuously measuring both ketone and glucose. Commercial availability and prescription status differ from authorization, so check current Abbott distribution.

At what glucose level are you in ketosis? There is no glucose threshold that defines ketosis. Ketosis is defined by blood BHB, conventionally above about 0.5 mmol/L. People with low glucose can have negligible ketones and people eating carbohydrate can show mild ketones after a long fast or hard exercise. Measure the ketone, not the glucose.

Can my glucose meter check ketones? Only if it is a dual-strip meter such as the Precision Xtra or Keto-Mojo GK+, and only with ketone strips. A glucose strip cannot report BHB, since it carries the wrong enzyme 1.

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. Sergei Tarasov, Yulia Plekhanova, Mahendra Rai, et al. Nano- and Microelectrochemical Biosensors for Determining Blood Glucose. 2021. https://doi.org/10.1007/978-3-030-55490-3_15 ↩ ↩2

  2. Udo Hoss, Shridhara Alva, Heather Pryor, et al. Continuous Dual Glucose–Ketone Sensing Technology. Diabetes Technology & Therapeutics, 2025. https://doi.org/10.1177/15209156251390820 ↩

  3. wanmeng wang, Chongyu Liu, Peiwen Lin, et al. Toward continuous and non-invasive monitoring: a scoping review of in vitro blood glucose devices from electrochemistry to optics and micro-system integration. Journal of Materials Chemistry B, 2026. https://doi.org/10.1039/d5tb02338f ↩

  4. Kevin T. Nguyen, Nicole Y. Xu, Jennifer Y. Zhang, et al. Continuous Ketone Monitoring Consensus Report 2021. Journal of Diabetes Science and Technology, 2021. https://doi.org/10.1177/19322968211042656 ↩

  5. Satish K. Garg, Christopher G. Parkin. Role of Continuous Dual Glucose–Ketone Monitoring in Diabetes Care. Diabetes Technology & Therapeutics, 2025. https://doi.org/10.1177/15209156251369529 ↩ ↩2