Continuous glucose monitoring

Continuous glucose monitoring uses a coin-sized sensor worn on the skin, with a filament sitting in the tissue beneath it, to report glucose concentration in interstitial fluid every one to five minutes. In type 1 diabetes and in insulin-treated type 2 diabetes the technology is established, reimbursed, and supported by randomized trials. Since 2024 the same class of sensor has been sold over the counter in the United States to adults who do not use insulin, including people without diabetes, a market in which no trial has shown that wearing one improves health.

§How it works

An applicator drives a flexible filament a few millimetres into subcutaneous tissue and leaves an adhesive patch on the skin. The filament is an enzyme electrode: an immobilized enzyme, usually glucose oxidase, reacts with glucose diffusing in from the surrounding fluid, and the electronics read the resulting current, which rises and falls with concentration. A transmitter sends a value to a phone every one to five minutes. Current consumer sensors are factory-calibrated, need a warm-up period after insertion before they display anything, and are discarded after roughly ten to fifteen days.

The crucial detail is what the sensor is standing in. Glucose reaches interstitial fluid by diffusion out of capillaries, so the interstitial value trails the blood value by something on the order of five to fifteen minutes, and manufacturers' smoothing algorithms add their own delay. When glucose is stable the two track closely. When it is moving fast, during exercise, in the half hour after a meal, or while treating a low, the displayed number describes the recent past. This is why the trend arrow on the screen carries more decision-relevant information than the digit beside it, and why insulin users are taught to treat a falling reading differently from a flat one at the same value.

The clinically useful outputs are aggregate rather than instantaneous: how much of the day is spent in a target band, how variable the readings are, how much time is spent low overnight. The same signal also closes a control loop. Paired with an insulin pump and a dosing algorithm, a monitor becomes the sensor half of an automated insulin delivery system, the sense-and-respond architecture also used by closed-loop Deep brain stimulation and proposed for nanoscale therapeutic devices.

Terminology

Time in range is the share of the day spent between 70 and 180 mg/dL (3.9–10.0 mmol/L). The glucose management indicator converts mean sensor glucose into an HbA1c-like percentage; it is an estimate, not a laboratory value, and the two can disagree. The ambulatory glucose profile is the standard one-page report that stacks two weeks of traces into a single median-and-percentile day.

§Development history

1. 1999
   First CGM approved
   The FDA approves MiniMed's system, which records glucose blind and is downloaded at a clinic days later. It is a diagnostic tool for physicians, not a display for patients.
2. 2001
   GlucoWatch reaches the market
   Cygnus wins approval for a wrist device that draws fluid through the skin by reverse iontophoresis. Skin irritation and unreliable readings sink it, and it is off the market within a few years.
3. 2005–2006
   Real-time systems arrive
   Medtronic's Guardian RT and DexCom's STS put a live number, a trend arrow, and alarms in the user's hand for the first time.
4. 2008
   Randomized benefit demonstrated
   A JDRF-funded trial in type 1 diabetes reports improved HbA1c with CGM, with the effect concentrated in adults aged 25 and over.
5. 2014
   FreeStyle Libre launches in Europe
   Abbott's factory-calibrated, scan-to-read sensor cuts cost and eliminates fingerstick calibration, and drives the first mass adoption outside intensive insulin users.
6. 2016
   Dosing from the sensor allowed
   The FDA permits the Dexcom G5 to be used for insulin dosing without a confirmatory fingerstick, making the sensor the primary measurement rather than an adjunct. Calibration fingersticks are still required.
7. 2017
   Trials confirm the effect on injections
   The DIAMOND and GOLD trials show HbA1c improvement in type 1 diabetes managed with multiple daily injections rather than pumps. FreeStyle Libre is approved in the US.
8. 2018
   Interoperable sensors defined
   The Dexcom G6 is authorized as an integrated CGM, a device category built so that sensors, pumps, and algorithms from different manufacturers can be combined.
9. 2024
   Over-the-counter sensors
   The FDA clears Dexcom's Stelo and then Abbott's Lingo for sale without a prescription to adults who do not use insulin.

Each step turned less on the electrochemistry than on trust in it. The 1999 system was blinded because nobody was prepared to let a patient act on the number; permitting insulin dosing from a sensor reading in 2016 was the decision that made the automated pumps, and eventually the retail shelf, possible.

§Evidence in diabetes

For people who take insulin, a reading changes an action. The 2008 JDRF trial found improved glycemic control in adults but not in the children and young adults enrolled alongside them, a split usually attributed to how much the younger groups actually wore the sensor.¹ The DIAMOND trial extended the result to adults on injections rather than pumps, the larger population, with a modest but consistent HbA1c reduction.² In type 2 diabetes treated with basal insulin in ordinary primary care, the MOBILE trial found a similar direction of effect.³ Less time spent hypoglycemic is a second and arguably more important benefit, since severe lows are the acute danger of insulin therapy and often go unnoticed overnight.

An international consensus group in 2019 converted the raw traces into targets that trials and clinics could share, chiefly the recommendation that most adults with diabetes spend more than 70% of the day between 70 and 180 mg/dL.⁴ Time in range is not a validated surrogate for complications in the way HbA1c is, but it correlates with it and captures variability that a three-month average hides.

Where the indication ends

The strongest evidence is in people whose reading changes an insulin dose. In type 2 diabetes managed with diet, Metformin and the TAME trial, or a GLP-1 receptor agonist, trials are fewer, effects on HbA1c smaller, and reviewers disagree about whether continuous wear is worth the cost against periodic testing. That boundary, rather than the wellness market, is where most of the live clinical argument sits.

§Use without diabetes

In 2024 the FDA cleared the first monitors sold without a prescription, Dexcom's Stelo followed by Abbott's Lingo, both indicated for adults who do not use insulin.⁵ The consumer proposition is that watching one's own curve teaches which foods, workouts, and bad nights produce which responses, and that flattening it is a route to longer healthy life. Companies including Levels and Signos sell subscriptions built on Dexcom and Abbott hardware, adding food logging, meal scores, and coaching rather than sensors of their own. The sensor joins the other consumer physiological sensors as a fixture of Biohacking and grinders and of the self-tracking culture that surrounds it, alongside direct-to-consumer blood panels and supplement regimens.

The scientific basis is thinner than the marketing. The most-cited support is a 2015 Weizmann Institute study which showed that postprandial glucose responses to identical meals differ markedly between people, and that an algorithm trained on personal data could design meals that blunted those responses.⁶ The endpoint there was the glucose curve itself. The larger PREDICT study later reported the same between-person spread and, importantly, substantial variation within the same person eating the same food on different days.⁷ Reference data from people without diabetes wearing blinded sensors show that normal glucose tolerance means spending nearly all of the day in a narrow band, with brief post-meal excursions that a consumer app may flag as alarming.⁸ A monitor that pathologizes normal physiology has a specificity problem, not a sensitivity one.

What a spike does not mean

No threshold for a post-meal glucose excursion has been validated as a risk factor in a person with normal glucose tolerance. The same meal, eaten by the same person on two days, can produce visibly different curves through sleep, prior activity, stress, sensor site, and simple noise. A number that moves is not the same as a number that matters, and the inference from a curve to a health outcome runs through evidence that does not exist yet.

§Accuracy and its limits

Sensor accuracy is reported as mean absolute relative difference (MARD): the average percentage gap between sensor readings and a reference measurement taken at the same time. Manufacturer-reported MARD for current sensors sits in the high single digits, close to the reproducibility of fingerstick meters. The figure is easy to over-read: it depends on the reference method, on how many comparisons fell during rapid change, and on the distribution of glucose values in the study population, so numbers from different manufacturers' studies are not directly comparable. Accuracy is generally worse in the low range, which is where errors matter most.

Physical artifacts are common. Lying on a sensor compresses the tissue around it and can produce a false low overnight, the best-known failure mode among users. Some sensors are affected by common substances: high-dose vitamin C has been reported to raise readings on certain devices, and acetaminophen interfered with older Dexcom generations. Adhesive failure, insertion-site variability, and the first day of a sensor's life all contribute noise that a smooth on-screen line conceals.

§Risks and second-order effects

The physical risks are minor: adhesive dermatitis, occasional infection, rarely a retained filament fragment. The interesting risks are behavioural. Clinicians have raised concerns that continuous feedback encourages food restriction and anxiety in people predisposed to disordered eating, and alarm fatigue is a documented problem even among those with a clinical reason to wear a sensor. Attention spent on curves is also attention not spent on structured exercise, which has randomized evidence on function in older adults, or on sleep, whose mortality epidemiology is far larger than anything assembled for glucose curves in people without diabetes.

A glucose trace also reveals meal timing, sleep, illness, and activity, and it sits with manufacturers and app companies under commercial privacy policies rather than the rules that govern clinical records. Cost pushes in the other direction: sensors are a recurring out-of-pocket expense, so a technology with proven benefit in diabetes is distributed partly by ability to pay, an instance of the pattern examined under Access and inequality.

§Outlook

Hardware is moving toward longer wear and more analytes. Senseonics' implanted fluorescence-based sensor runs for months to a year rather than days, and manufacturers have announced sensors reporting ketones alongside glucose, which matters for users of automated insulin delivery. Non-invasive optical measurement remains the field's perennial promise; the FDA warned in 2024 against smartwatches and rings claiming to measure blood glucose without piercing the skin, and no such device has been authorized.

The more consequential question is interpretive. Sensors generate dense longitudinal metabolic data on large numbers of people outside clinical research, the kind of input that individualized physiological models require. Whether that yields anything beyond diabetes depends on studies that have not been run: trials in people without diabetes, powered on health outcomes rather than glucose curves, long enough to show whether early detection of impaired glucose tolerance changes anything. Until those exist, wellness monitoring occupies the same position as consumer methylation age tests and other biomarker panels sold on the promise of measuring biological aging: a real measurement of a real quantity, with an unestablished link between moving the number and living better. That gap is the standing challenge to the geroscience programme as a whole, and glucose is simply its most widely worn instance.

§See also

• Wearable health sensors
• Consumer blood testing
• Quantified self
• GLP-1 receptor agonists
• Metformin and the TAME trial
• Biohacking and grinders
• Aging biomarkers
• Human digital twins

§References