Your glucose monitor is not measuring your blood, and that explains almost every reading you disagreed with

Someone wearing a continuous glucose monitor checks it against a fingerprick and finds the numbers do not match. The usual conclusion is that one of them is wrong. Usually neither is. They are measuring glucose in two different fluids, in two different compartments of the body, and the difference between those readings is not error — it is information, provided you know what the sensor is actually doing.

The sensor is under your skin, not in your vein

What is stuck to the arm is a transmitter. The part that measures is a filament thinner than a hair, sitting a few millimetres into the subcutaneous tissue, bathed in interstitial fluid — the fluid that fills the spaces between cells.

Glucose reaches that fluid from the capillaries by diffusion. When blood glucose rises, interstitial glucose rises shortly afterwards; when it falls, interstitial glucose follows down. The two track each other closely at steady state and diverge whenever glucose is changing quickly, which is exactly when people tend to check.

That gives a delay of roughly five to fifteen minutes, depending on the system, part of it physiological diffusion and part of it the smoothing the device applies to a noisy signal. The practical consequence is specific and useful: after a meal, when glucose is climbing fast, the sensor will read lower than a fingerprick. During a rapid fall — after insulin or exercise — it will read higher for a while, then catch up. A disagreement during a rapid change is the system behaving correctly. A disagreement when glucose has been flat for an hour is worth investigating.

Turning sugar into an electric current

The filament is an enzyme electrode, and the chemistry is elegant.

Immobilised on it is an enzyme that reacts specifically with glucose — glucose oxidase in the classical design, glucose dehydrogenase in many modern ones. The enzyme strips electrons from each glucose molecule it encounters. In the older chemistry, that produces hydrogen peroxide, which is then oxidised at the electrode, releasing electrons and generating a small current. In newer designs a mediator — often an osmium or ferrocene complex bound into a polymer — shuttles electrons from the enzyme directly to the electrode, which lets the sensor run at a lower voltage and avoids depending on how much oxygen is around.

That last point matters more than it sounds. A peroxide-based sensor needs oxygen as a co-substrate, and oxygen in tissue is far scarcer than glucose, so the reaction can become limited by oxygen supply rather than by the thing you are trying to measure. Wiring the enzyme to a mediator sidesteps that entirely.

Either way, the current is proportional to the glucose concentration, and the device converts current to a number.

The enzyme decides what the sensor responds to. The membrane over it decides whether the reading means anything.

The unglamorous component doing most of the work is that outer diffusion-limiting membrane. It deliberately restricts how much glucose reaches the enzyme, which sounds counterproductive and is essential: it keeps the enzyme from being saturated, keeps the response linear across the physiological range, ensures the reaction is limited by glucose rather than by oxygen, and blocks larger interfering molecules. It is also the biocompatible face the body reacts to. Most of the engineering difficulty in these devices is in a coating a few micrometres thick, not in the biology.

Modern sensors are factory calibrated — the manufacturing process is controlled tightly enough that a code shipped with the sensor sets the conversion, with no fingerprick calibration needed. Accuracy is quoted as MARD, the mean absolute relative difference from a laboratory reference, and current systems sit under about ten per cent. That is good enough to dose insulin from and not good enough to treat a single reading as a laboratory result.

Why it dies after two weeks

Sensor life is capped at around ten to fifteen days, and the limit is biological rather than electronic.

Anything inserted into tissue triggers a foreign-body response. Proteins adsorb onto the surface within seconds, immune cells arrive, and over days the body begins walling the object off with fibrous tissue. That capsule increasingly obstructs the diffusion of glucose to the sensing surface, so the signal drifts downward and the calibration slowly stops holding. Add gradual enzyme degradation at body temperature, and there is a point where the manufacturer can no longer guarantee the number. Extending sensor life is largely a materials problem — coatings that provoke less encapsulation — which is why this device sits as much in materials science as in medicine.

Two artefacts are worth recognising because they alarm people unnecessarily. Compression lows happen when you lie on the sensor: pressure squeezes fluid out of the tissue around the filament, local glucose delivery drops, and the device reports a fall that never happened in your blood. A low at 3 a.m. that recovers the moment you roll over was probably your own shoulder. And certain substances can interfere with the electrode chemistry — paracetamol was a classic cause of falsely high readings on older peroxide-based sensors, and high-dose vitamin C can affect some systems. Modern selective membranes have reduced this considerably, but it is still worth reading what your specific device lists.

What continuous data actually changed

The clinical value is not that the number is continuous. It is that a curve answers questions a single number cannot.

HbA1c, the standard measure of diabetes control, reflects average glucose over roughly three months. Two people with the same HbA1c can have entirely different lives — one steady, the other swinging between dangerous highs and lows that average out to the same figure. Worse, HbA1c depends on how long red blood cells survive, so it is distorted by anaemia, by haemoglobin variants and by iron deficiency. In India, where anaemia is widespread, that is not a footnote; it is a routine source of misleading control estimates.

Continuous data replaces the average with a distribution. Time in range — the proportion of the day spent within a target band, with widely used consensus targets of around 70 per cent between 70 and 180 mg/dL for many adults — captures variability directly, along with how much time is spent low, which is the number that matters most for immediate safety. It also makes cause and effect visible to the person living it: this specific meal, this walk, this bad night's sleep, this dose timing.

The honest caveat belongs at the end. CGMs are increasingly marketed to people without diabetes as a general wellness tool, and the evidence that this improves any health outcome in that group is thin. Glucose rises after eating in healthy people — that is physiology working, not a problem to be corrected, and a device with ten per cent MARD is not a precise instrument for detecting subtle metabolic differences. The strong evidence is for people with diabetes, particularly those on insulin, and it is genuinely strong there.

Why it matters for students and researchers

The continuous glucose monitor is the most successful biosensor ever deployed, and it is a compact lesson in why that is hard. The recognition chemistry has been understood since the 1960s. What took fifty years was everything around it: a membrane that controls diffusion and survives immersion in tissue, a surface the body tolerates for two weeks, manufacturing consistent enough to skip user calibration, algorithms that smooth noise without hiding a real fall, and a power budget that runs for a fortnight on a battery the size of a coin.

That pattern repeats across the field. Every proposed biosensor — for lactate, ketones, cortisol, drug levels, potassium — faces the same wall, and it is almost never the biology that stops it. It is stability, biocompatibility, calibration drift and the foreign-body response. For students in bioengineering and biotechnology, this is where the real problems are, and they are materials and systems problems wearing a biological label.

Frequently asked questions

Why does my CGM reading differ from a fingerprick test?

Because they measure different fluids. A CGM measures glucose in the interstitial fluid under the skin, which lags blood glucose by roughly five to fifteen minutes, so the two disagree most when glucose is rising or falling quickly and agree closely when it is stable.

How does a continuous glucose monitor actually measure glucose?

A thin filament under the skin carries an enzyme that reacts with glucose and releases electrons. Those electrons produce a small electric current, either via hydrogen peroxide or through a mediator molecule, and the current is proportional to the glucose concentration.

Why do CGM sensors only last about two weeks?

Because the body walls off anything inserted into tissue. Over days a fibrous capsule forms around the filament and restricts glucose from reaching it, causing signal drift, and the enzyme also degrades gradually at body temperature.

What is a compression low?

It is a false low reading caused by lying on the sensor. Pressure reduces fluid and glucose delivery to the tissue around the filament, so the device reports a drop that is not happening in the blood; the reading recovers once the pressure is removed.

Is a CGM useful for someone without diabetes?

The evidence is limited. Glucose naturally rises after meals in healthy people, and current sensors are not precise enough to detect subtle metabolic differences reliably. The strong clinical evidence is for people with diabetes, particularly those using insulin.