Comparative Accuracy and MARD Analysis

MARD is the number most often used to describe how accurate a continuous glucose monitor (CGM) is [1]. As of October 2026, the current sensors from Dexcom, Abbott, MiniMed and Senseonics report MARDs between about 8% and 10% in their own accuracy studies [2][3][4][5][6]. Those figures look close enough to rank, but they come from separate studies with different people, wear periods and blood references. When one research group put three sensors on the same 24 adults, all three scored worse than in earlier studies [7].

What MARD measures

MARD stands for mean absolute relative difference. A study pairs each sensor reading with a blood glucose value taken at the same moment, works out the gap as a percentage of the blood value, drops the plus or minus sign, and averages those percentages over every pair [1][8]. Lower is better. Because the sign is dropped, a MARD can't show whether a sensor tends to read high or low, so the FDA's summaries report that separately as bias [5]. A 2020 review of the metric concluded that MARD shouldn't be the only measure used to judge a CGM, especially for nonadjunctive use, meaning readings that stand in for a fingerstick check [1].

The FDA doesn't use MARD as a pass mark either. Its rule for integrated CGMs (iCGMs), sensors designed to send readings to connected devices such as automated insulin dosing systems, sets accuracy as shares of readings that must land within set ranges of the blood reference. Across the whole measuring range, for example, the lower 95% confidence bound for readings within 20% of the blood value has to exceed 87% [9]. Most of the studies below report those "within 20%" agreement rates next to MARD [2][3][4].

Published MARD for current sensors

Each row is the figure from that sensor's main accuracy study or FDA review. Read the notes column before comparing rows; it carries as much information as the number.

Calibration is part of the design, too. Eversense 365 asks for four fingerstick calibrations at the start, one a day through day 13, then one a week [6], while the Libre Plus sensor and Simplera were tested factory-calibrated, with no fingersticks from the user [4][5]. A review by Abbott scientists notes that mistakes made while calibrating with fingersticks can make a sensor less accurate, and lists removing them as one of the main advantages of factory calibration [12]. Our article on sensor calibration algorithms explains how both approaches work, and the one on implantable fluorescence sensors covers how Eversense measures glucose.

Earlier sensors, for context

The Dexcom G6 had an overall MARD of 9.0% in its 2018 study of 62 participants: 9.8% in 25 adults and 7.7% in 37 children and adolescents [8]. Participants calibrated once a day during the study, and the researchers reprocessed the raw data with a factory-calibration algorithm to produce those figures [8]. Dexcom will no longer manufacture the G6 after July 1, 2026, and is moving G6 users to its G7 sensors [13].

The first FreeStyle Libre reported an 11.4% MARD in 2015 [14]. That number was measured mainly against fingerstick readings from the meter built into the reader, with lab-analyzer checks at three clinic visits, so it doesn't line up directly with the lab-referenced figures in the table [14]. Abbott's current US sensors are the Libre 2 Plus and Libre 3 Plus listed above [10].

Why the figures don't rank cleanly

The reference method is the biggest variable. In the head-to-head study described below, the same sensors produced different MARDs against a YSI analyzer, a second lab analyzer and a fingerstick meter [7].

Who takes part matters. The G6 scored about two percentage points better in children than in adults [8], yet the first Libre study found that body mass index, age and diabetes type did not affect accuracy [14]. Placement matters as well: the same Dexcom G7 measured 8.2% on the upper arm and 9.1% on the abdomen [2].

The study protocol shapes the average. The G7 15 Day and Libre Plus studies pushed some participants' glucose up and down in clinic, so their figures include readings from across the sensor's measuring range [3][4]. Activity is another factor. A 2020 position statement from two diabetes societies, the EASD and ISPAD, put the average MARD across exercise studies in people with type 1 diabetes at about 13.6% [15].

What a head-to-head study found

The Institute for Diabetes Technology in Ulm, Germany, ran the comparison most readers want to see. Twenty-four adults with type 1 diabetes wore a FreeStyle Libre 3, a Dexcom G7 and a MiniMed Simplera at the same time for up to 15 days, with three 7-hour clinic sessions that included induced high and low glucose [7].

Libre 3 and G7 were more accurate in the normal and high ranges, and Simplera did better in the low range [7]. Here's what surprised us: all three sensors were less accurate than in earlier studies, and the authors called for shared guidelines on how CGM studies are designed [7]. One caveat for anyone comparing products: the Libre 3 tested here is the non-Plus sensor, which Abbott discontinued on September 30, 2025 [10].

Reading a MARD sensibly

A published MARD describes how one sensor did in one study, against one reference, in one group of people. We read the figures in the table as a rough tier rather than a ranking. The current sensors fall between about 8% and 10% in their own trials [2][3][4][5][6], and the gaps between them shift when the reference changes [7]. Wear time runs from 6 days to a year across the same table, and calibration needs and pump compatibility differ too. Weighing those trade-offs for a particular person is a decision to make with their care team.