MARD stands for Mean Absolute Relative Difference. It's the standard way the continuous glucose monitoring industry summarizes how closely a device's readings agree with a trusted reference — a lab-grade reference monitor, or a fingerstick meter, depending on the study.
How it works
For every paired reading — one from the device being tested, one from the reference at the same moment — researchers calculate the percentage difference between them, take the absolute value (so a reading that runs high and one that runs low count the same way), and average that across every pair in the study. The result is a single percentage: the MARD. A lower MARD means the device tracked the reference more closely on average. It is a statement about average agreement across a whole study, not a wrong-answer rate for any one reading.
Why it can be misread
That last distinction matters, because MARD is a term of art, and a bare percentage tends to read as a failure rate to anyone unfamiliar with it. A MARD figure on its own also tells you nothing about the study behind it — how many people, how many paired readings, what the readings were compared against, and how consistent the sample was. All of that context changes how much weight a MARD figure deserves.
MARD is not the only number that matters
A single average can hide how a device behaves at the edges. Two studies with the same MARD could differ a great deal in how often a reading lands close to the reference versus how often it lands far off. That is why a proper validation record usually reports more than the headline average — commonly, the share of readings that fall within a defined percentage of the reference, and separately, the share that fall well outside it. All three tell a fuller story together than any one of them does alone.
Where to find LifeLeaf's own numbers
Because a MARD figure needs that context to mean anything, LifeLeaf publishes its own MARD results — study by study, with the sample size, the reference used, the supporting agreement and outlier figures, and where each result was presented — on its clinical evidence page rather than as a stripped-out number on a general page like this one. That is the one place on the site built to carry the statistic responsibly.
Not a medical device · does not diagnose or treat