Understanding MARD: The Primary Metric for CGM Accuracy Comparison
Mean Absolute Relative Difference (MARD) represents the industry standard for quantifying continuous glucose monitoring accuracy, expressing the average percentage deviation between CGM readings and simultaneous reference laboratory glucose measurements (typically obtained via venous blood sample from a laboratory analyzer or rapid-acting point-of-care glucose meter). MARD calculation compares paired CGM and reference measurements, determines the absolute difference, divides by the reference value, and averages across all paired measurements. For example, if a CGM reads 120 mg/dL when the reference is 140 mg/dL, the absolute difference is 20 mg/dL, and the relative difference is 20 divided by 140, or approximately 14.3%. MARD values historically considered acceptable for clinical use ranged from 10-15%, though contemporary standards have progressively tightened as technology improved. Dexcom G7 achieves a MARD of approximately 8.2% based on regulatory submission data, consistently outperforming FreeStyle Libre 2's MARD of approximately 9.3%. While this represents a meaningful difference, both systems fall substantially below the 10% threshold, indicating both qualify as clinically accurate. MARD limitations deserve emphasis: MARD provides an aggregate accuracy metric and obscures important nuances like accuracy variation across different glucose ranges, trends, and clinical conditions. A system might achieve 8% MARD while demonstrating poor accuracy specifically in hypoglycemic ranges or during rapid glucose changes-scenarios precisely when accuracy matters most for safe insulin dosing. Consequently, MARD should be interpreted alongside additional accuracy metrics including Clarke Error Grid Analysis, breakdown of accuracy by glucose range, and accuracy during dynamic glucose conditions.
Clarke Error Grid Analysis: Understanding Clinically Meaningful Accuracy
Clarke Error Grid Analysis provides clinically relevant accuracy assessment by plotting CGM readings against reference measurements and categorizing each reading into zones based on clinical consequences of errors. The grid is divided into five zones: Zone A includes readings within 20% of reference or within 15 mg/dL (whichever is larger) and represents clinically acceptable readings where patients would make identical treatment decisions regardless of whether they used the CGM or reference value. Zone B includes readings outside Zone A but would not lead to major treatment errors-a patient might make a suboptimal treatment decision but not one causing harm. Zone C encompasses readings that would cause unnecessary treatment changes, potentially producing hypoglycemia or overtreatment. Zone D includes readings where the CGM misses diabetic conditions (failing to detect hypoglycemia or severe hyperglycemia). Zone E represents contradictory errors where the CGM indicates opposite glucose direction compared to reference values. Clinically acceptable accuracy standards typically specify that at least 95% of readings fall within Zone A or B combined, with minimal readings in Zones C, D, and E. Published Clarke Error Grid analyses of Dexcom G7 typically demonstrate approximately 98-99% Zone A+B compliance across diverse patient populations, while FreeStyle Libre 2 studies show approximately 97-98% Zone A+B compliance. Both systems demonstrate excellent clinical accuracy by this metric, though Dexcom's superior performance in Zone A (tight agreement) versus Zone B (acceptable but less precise) becomes evident upon detailed analysis. Crucially, Clarke Error Grid analysis highlights that CGM accuracy varies substantially by glucose range: both Dexcom and Libre 2 demonstrate lower accuracy in hypoglycemic ranges (less than 70 mg/dL) compared to normoglycemic ranges, a critically important consideration for patients relying on their CGM for hypoglycemia detection.
Accuracy Across Glucose Ranges: Hypoglycemia, Normal, and Hyperglycemia Zones
CGM accuracy varies substantially across different glucose ranges, with profound implications for clinical safety. In hypoglycemic ranges (below 70 mg/dL), both Dexcom G7 and Libre 2 demonstrate reduced accuracy compared to their overall MARD values, though the patterns differ. Dexcom G7 maintains approximately 12-15% MARD in hypoglycemic ranges-higher than its overall 8.2% but substantially better than many competing systems. Critically, Dexcom demonstrates superior specificity for detecting true hypoglycemia (correctly identifying low readings without false alarms) compared to Libre 2. FreeStyle Libre 2 achieves approximately 13-16% MARD in hypoglycemic ranges, with particular challenges accurately predicting impending lows through rate-of-change analysis. This accuracy difference becomes clinically significant: if a true glucose is 60 mg/dL, Dexcom may read 65-75 mg/dL (relatively close), whereas Libre 2 might read 55-75 mg/dL (wider uncertainty), creating situations where Libre 2 users cannot reliably distinguish mild from severe hypoglycemia without fingerstick confirmation. In normoglycemic ranges (70-180 mg/dL), both systems achieve substantially superior accuracy: Dexcom maintains approximately 7-8% MARD while Libre 2 achieves approximately 8-10% MARD. This represents the "sweet spot" where both systems perform reliably and make treatment decisions with confidence. In hyperglycemic ranges (above 180 mg/dL), both systems again demonstrate acceptable accuracy around 8-9% MARD, though neither system excels at detecting extreme highs (above 300 mg/dL). The practical implication: both systems perform optimally in normal glucose ranges but show reduced reliability specifically when hypoglycemia occurs-precisely when decision accuracy is most safety-critical. This reality drives the clinical practice recommendation that when CGM readings enter hypoglycemic ranges, confirmation with fingerstick testing remains advisable, particularly if the hypoglycemic reading is surprising or inconsistent with symptoms.
Rapid Glucose Changes and Glucose Trend Accuracy
Accuracy during rapid glucose fluctuations represents another critical but often underappreciated dimension of CGM performance. Both Dexcom and Libre 2 rely on enzymatic sensors that measure glucose diffusion into subcutaneous tissue, not glucose in blood directly; consequently, subcutaneous glucose lags behind blood glucose by approximately 5-15 minutes, creating inherent "phase lag." This physiological delay becomes particularly problematic during rapid glucose changes-aggressive insulin doses causing brisk glucose decline, or large carbohydrate consumption producing swift glucose rise. When glucose is declining rapidly (greater than 2-3 mg/dL per minute), both systems tend to overestimate glucose because the subcutaneous reading lags behind the declining blood glucose. A patient experiencing rapid hypoglycemia (such as during prolonged intense exercise) might see a CGM reading of 85 mg/dL while true blood glucose is actually 65 mg/dL-a clinically important discrepancy. Similarly, during rapid rise (such as immediately after fast-acting carbohydrate intake), the CGM underestimates the peak glucose rise. Dexcom's algorithmic processing attempts to predict glucose trajectory and partially correct for phase lag using sophisticated mathematical models incorporating prior glucose trends and rate-of-change data. This predictive capability provides some advantage during rapid changes, though predictive errors can occur if glucose trajectories change unexpectedly (such as when exercise suddenly decreases glucose decline rate). FreeStyle Libre 2's trend prediction relies on simpler algorithms given its passive scanning architecture, rendering it somewhat less adaptable to unexpected trajectory changes. Practically, during rapid glucose changes, neither system should be absolutely trusted without fingerstick confirmation, particularly if treatment decisions hinge on exact glucose values. This limitation particularly impacts users of advanced insulin pumps with predictive algorithms: the CGM data feeding the pump's predictive model will have inherent lag, which the pump algorithm attempts to correct for, but imperfect correction can occasionally produce suboptimal insulin dosing during periods of rapid glucose change.
Calibration Requirements, Accuracy Drift, and Sensor Longevity Effects
Calibration practices substantially influence CGM accuracy and differ between systems. Dexcom G7 requires two calibrations per day: users enter fingerstick glucose measurements into the Dexcom app approximately every 12 hours, which mathematically adjusts the sensor's raw electrochemical signal to align with reference measurements. This calibration process enables Dexcom to correct for inter-individual variations in sensor behavior (some sensors drift more than others, some require steeper signal-to-glucose conversion) and sensor aging (accuracy typically diminishes slightly across the ten-day sensor life). FreeStyle Libre 2 employs factory calibration only-no user-performed calibrations are conducted, though the system underwent extensive factory calibration during manufacturing. Libre's approach trades calibration burden (more convenient for users) for potential reduced accuracy, as individual sensor variation and aging cannot be corrected in real-time. Published data suggests this calibration difference contributes measurably to Dexcom's superior accuracy: Dexcom's twice-daily calibration enables systematic accuracy improvement across the sensor life, whereas Libre 2's accuracy gradually declines across the fourteen-day wear period. By day 10-14 of Libre 2 wear, accuracy may deteriorate by approximately 0.5-1.0% MARD, prompting some clinicians to recommend limiting Libre 2 wear to twelve days rather than utilizing the full fourteen-day rated duration. Importantly, calibration accuracy critically depends on fingerstick glucose measurement quality: if calibration readings from your fingerstick meter are inaccurate, your Dexcom calibration will propagate those inaccuracies. Best-practice calibration technique requires: using meters with high accuracy ratings (ideally ISO 15189 certified), ensuring adequate blood sample size, avoiding hematocrit extremes that can bias fingerstick readings, and calibrating at stable glucose (not during rapid changes when fingerstick-to-CGM timing discrepancies create artificial misalignment).
Body Site Variation and Compression Lows: Understanding False Readings
CGM accuracy varies by body site, with abdomen consistently demonstrating superior accuracy compared to arm, leg, or other sites for both Dexcom and Libre 2. Abdominal placement typically yields approximately 1-2% better accuracy (lower MARD) compared to arm placement, attributed to better blood flow consistency and more stable interstitial glucose equilibration in abdominal subcutaneous tissue. Arm placement remains acceptable for most users but may increase false readings, particularly pressure-induced "compression lows." Compression lows represent a specific artifact where external pressure on the sensor (such as sleeping on an arm with a sensor, leaning against a table, or tight clothing) artificially depresses the glucose reading without true glucose change. Dexcom sensors appear relatively resistant to compression artifacts (approximately 2-5% of users report occasional compression lows), while Libre 2 sensors exhibit higher susceptibility (approximately 10-15% of users report intermittent compression-related false readings), likely related to differences in sensor geometry and pressure transduction. Compression artifacts typically present as unexplained sudden glucose drops (such as reading 120 mg/dL then abruptly dropping to 60 mg/dL without symptoms), followed by spontaneous recovery once pressure is relieved. Clinically, compression lows create management challenges: patients may treat apparent hypoglycemia unnecessarily, consuming carbohydrates and subsequently rebounding to hyperglycemia. Strategies to minimize compression artifacts include: preferring abdominal placement when feasible, avoiding tight clothing or pressure on arm sensors, not sleeping directly on sensors, and learning to recognize the characteristic artifact pattern (sudden drop without symptom correlation, rapid recovery). For patients experiencing frequent compression lows with Libre 2, switching to Dexcom's more pressure-resistant design may improve quality of life and reduce unnecessary hypoglycemia treatment.
Moisture, Sweat, and Environmental Effects on Sensor Accuracy
Environmental factors including moisture, sweat, temperature, and humidity influence CGM accuracy, though the magnitude of effects remains modest for modern systems. Excessive moisture, such as from prolonged swimming or saturation with sweat, can degrade sensor adhesion and occasionally affect electrochemical signal quality, potentially decreasing accuracy by 1-3%. Both Dexcom and Libre 2 sensors are designed with water-resistant membranes, but prolonged submersion (greater than one hour) or high-pressure water exposure (such as from shower jets directed at the sensor) may occasionally compromise the membrane's barrier function. Practical implications suggest: keeping sensors dry during normal activities, protecting sensors with waterproof patches during swimming or bathing, and being mindful of sensor placement if you regularly engage in water sports or work in wet environments. High ambient temperature can theoretically affect enzyme kinetics in the sensor's glucose oxidase layer, potentially degrading accuracy; however, the magnitude of effect in real-world use remains minimal, with published studies showing less than 0.5% MARD change across typical ambient temperatures (15-35 degrees Celsius). More practically, extreme heat (such as leaving a sensor in direct sunlight in a hot car) may accelerate enzymatic degradation, prompting recommendations to store spare sensors below 25 degrees Celsius and avoid exposure to extreme heat. Conversely, extreme cold (below freezing) is not typically problematic during normal wear but may affect stored sensors' calibration or enzymatic activity if prolonged. The practical bottom line: for typical users in typical climates, environmental factors have minimal impact on CGM accuracy; only users in extreme environmental conditions (very hot climates, water sports professionals, arctic exploration) need to consider these factors substantially when selecting or maintaining their CGM system.
iCGM Designation and Its Accuracy Implications for Insulin Dosing
The iCGM (integrated Continuous Glucose Monitoring) designation, established by the FDA in 2016 and updated in 2024, certifies that a CGM system has met rigorously defined accuracy standards and can be used for insulin dosing decisions without necessarily requiring fingerstick verification. Both Dexcom G7 and FreeStyle Libre 2 carry the iCGM designation, signifying that regulatory review determined their accuracy meets clinical standards sufficient for autonomous insulin decision-making. Achieving iCGM status requires manufacturers to demonstrate their systems maintain acceptable accuracy across diverse conditions: multiple body sites, various user populations (diverse body compositions, ages, ethnic backgrounds), different climate conditions, and various baseline HbA1c levels. Manufacturers must provide Clarke Error Grid Analysis showing that the vast majority of readings fall within clinically acceptable zones and specify accuracy in critical glucose ranges (particularly hypoglycemia less than 70 mg/dL and hyperglycemia greater than 180 mg/dL). iCGM designation does not certify that devices are equally accurate (Dexcom's superior MARD still represents an objective advantage) but rather that both exceed minimum accuracy thresholds. Critically, iCGM status enables "no-fingerstick" insulin dosing: FDA recognizes these devices as sufficiently accurate that patients with iCGM-designated CGMs may legally and safely dose insulin exclusively based on CGM readings without confirmatory fingerstick testing. This represents a dramatic shift from prior requirements where fingerstick confirmation was mandatory before treating hypoglycemia or making insulin dose adjustments. For practical application: iCGM designation provides regulatory assurance that your CGM is sufficiently accurate for autonomous insulin management, though individual clinical judgment remains essential-if a CGM reading seems inconsistent with your physical symptoms or clinical context, prudent practice still suggests fingerstick verification.
Real-World Accuracy Studies and Published Evidence Synthesis
Beyond regulatory submission data, numerous peer-reviewed studies have evaluated Dexcom G7 and Libre 2 accuracy in real-world patient populations, providing additional evidence beyond manufacturing claims. Dexcom G7 studies consistently report MARD between 7.5-8.5% across diverse populations, with Clarke Error Grid Zone A+B compliance exceeding 97%, and particularly strong performance in detecting hypoglycemia and hyperglycemia. Studies published in Diabetes Technology & Therapeutics, Diabetes Care, and other major journals consistently rank Dexcom G7 among the most accurate CGM systems, with superiority most pronounced in hypoglycemic ranges. FreeStyle Libre 2 studies report MARD between 9.0-10.0% depending on population characteristics and use conditions, with Clarke Error Grid Zone A+B compliance of 96-98%. Notably, Libre 2 accuracy appears more variable across populations than Dexcom, with lower accuracy reported in adolescents compared to adults, in type 1 compared to type 2 diabetes, and in patients with lower baseline HbA1c (suggesting less robust sensor performance when glucose is well-controlled). Meta-analyses synthesizing multiple studies confirm Dexcom G7's consistent 1-2% MARD advantage over Libre 2, though both systems perform substantially better than historical CGM technology. Importantly, published real-world studies generally show slightly lower accuracy than regulatory submission data (real-world MARD often runs 0.5-1.0% higher than claimed specifications), likely reflecting more diverse calibration practices, storage conditions, and user populations than occurred in controlled clinical trials. This reality reinforces that published manufacturer specifications should be viewed as best-case scenarios; actual accuracy will vary individually based on calibration technique, body site selection, and individual sensor sensitivity variation.
When to Trust Your CGM and When Fingerstick Verification Remains Essential
Despite excellent overall accuracy, clinical judgment requires knowing when your CGM reading can be trusted absolutely versus when confirmatory fingerstick testing is prudent. CGM readings can be trusted with confidence when: glucose is in the normoglycemic range (70-180 mg/dL), glucose trend has been stable (not changing rapidly), symptoms are consistent with the CGM reading (you feel fine and CGM is normal; you feel shaky and CGM is low), and the reading makes contextual sense given recent insulin, food, or activity changes. Conversely, fingerstick verification becomes advisable when: CGM indicates hypoglycemia, particularly if you lack hypoglycemia symptoms (hypoglycemia unawareness situations) or if the low reading is unexpected; CGM indicates rapid change inconsistent with your clinical context; you're making critical treatment decisions such as bolus doses for meals or corrective insulin for hyperglycemia; or the CGM reading conflicts with your physical symptoms. Specific scenarios mandating fingerstick verification include: after changing sensor sites (confirming calibration and sensor function), during exercise when glucose is changing rapidly, when diabetic ketoacidosis is suspected (CGM can underestimate severity), and whenever you're uncertain and accuracy significantly impacts your safety. Healthcare providers increasingly use the phrase "treat your symptoms, not your CGM reading"-meaning if you feel hypoglycemic, treat hypoglycemia regardless of what CGM shows; if you feel fine, don't treat based on CGM alone. This nuanced approach recognizes that CGMs are extraordinarily useful decision-support tools but imperfect systems that occasionally produce false readings. For most routine glucose management (calculating meal boluses, adjusting basal rates, overnight glucose trending), trusting your CGM represents appropriate practice; however, maintaining a backup fingerstick meter and knowing how to verify readings when uncertainty arises represents essential safety practice.
Closed-Loop AID Systems: How CGM Accuracy Directly Impacts Automated Insulin Delivery Safety
Closed-loop artificial pancreas systems (such as Tandem's Control-IQ, Medtronic's 780G, or Insulet's Omnipod 5) depend critically on accurate, frequent CGM data-if the CGM reading is inaccurate, the insulin pump's automated algorithm receives corrupted input and may deliver inappropriate insulin. This relationship creates a direct coupling between CGM accuracy and AID safety: poor CGM accuracy doesn't just affect your treatment decisions; it directly impairs your pump's ability to deliver correct insulin quantities. Dexcom's superior accuracy (8.2% MARD) and faster data transmission frequency provide advantages for AID systems: the pump receives frequent, accurate glucose data and can make precise real-time adjustments with confidence. FreeStyle Libre 2's lower accuracy (9.3% MARD) and the scanning requirement mean that AID integration with Libre 2 remains limited and less optimal compared to Dexcom integration. This reality has driven clinical guidelines recommending Dexcom preferentially for AID users, as the superior accuracy margin, though seemingly modest in percentage terms, translates into meaningful differences in automated insulin delivery precision and safety. Closed-loop systems employ sophisticated algorithms attempting to account for CGM lag and accuracy variation, but imperfect compensation means underlying CGM accuracy substantially influences final insulin delivery appropriateness. For patients committed to closed-loop automation, CGM accuracy should represent a primary consideration in system selection, making Dexcom's objective accuracy advantage clinically meaningful. Conversely, patients using insulin pumps for non-closed-loop hybrid approaches (the pump includes basal-bolus automation but the patient makes high-level treatment decisions) experience somewhat reduced impact from marginal accuracy differences, as human judgment provides an additional layer of error-checking.
Practical Strategies for Optimizing CGM Accuracy in Your Daily Use
While individual CGM system design fundamentally determines accuracy potential, numerous practical strategies substantially influence your experienced accuracy. For Dexcom G7 users: perform calibrations at stable glucose, ideally in normoglycemic ranges rather than during hypo- or hyperglycemia when CGM-to-fingerstick timing discrepancies introduce calibration error; use a high-quality fingerstick meter (check your meter's accuracy claims or ask your provider which meters they recommend); ensure adequate blood sample size for calibration; and consider calibrating on the abdomen if possible, avoiding calibration when sitting on a sensor or applying pressure. Place new sensors on the abdomen when feasible; if using arm sites, minimize pressure during sleep and choose sites with lower physical activity. For FreeStyle Libre 2 users: scan regularly (at least every eight hours) to retrieve and align stored glucose data; store sensors properly at room temperature before use; avoid compression by not sleeping on sensor sites; and consider limiting sensor wear to twelve days rather than fourteen days if you notice accuracy degradation in the latter wear period. For all users: verify your CGM calibration/sensor placement technique by comparing CGM readings to fingerstick measurements at different times of day; if discrepancies exceed 15%, investigate whether calibration technique needs improvement, meter accuracy is questionable, or sensor placement is suboptimal. Learn your individual sensor variation: some users report their particular sensors run systematically high or low relative to fingerstick; accounting for this individual pattern helps contextualize readings. Finally, participate in quality improvement by reporting suspected sensor failures or unusual accuracy patterns to your CGM manufacturer, as such feedback contributes to technological refinement and provides you with replacement sensors if true sensor defects occurred.
Frequently Asked Questions
What is MARD and why does it matter for CGM accuracy?
MARD (Mean Absolute Relative Difference) is the primary metric for CGM accuracy, measuring the average percentage deviation between CGM readings and simultaneous reference blood glucose values. A MARD of 8–10% means readings are within 8–10% of the true value on average. Lower MARD = higher accuracy.
How accurate is the Dexcom G7 vs FreeStyle Libre 2?
Dexcom G7 has a MARD of approximately 8.2% in pivotal trials; FreeStyle Libre 2 has a MARD of approximately 9.3%. Both meet FDA iCGM accuracy criteria (<10% MARD). Real-world performance is typically slightly lower than trial data for both devices.
Why does CGM accuracy drop during hypoglycaemia?
CGM sensors measure glucose in interstitial fluid rather than blood, and during hypoglycaemia (rapidly falling glucose) there is a physiological time lag of 5–15 minutes between blood and interstitial glucose values. This lag causes CGMs to read higher than actual blood glucose during acute hypoglycaemia, potentially underestimating the severity.
When should I verify my CGM reading with a fingerstick?
Always verify with a fingerstick before treating hypoglycaemia (if symptoms don't match the reading), before making major insulin dose decisions when glucose is rapidly changing, when the CGM sensor is newly inserted (first 2–3 hours), and whenever you experience CGM symptoms without a matching reading.