6782 The Myth of MARD: Limitations Of The Mean Absolute Relative Difference In The Analytical And Clinical Assessment Of Continuous Glucose Monitoring (CGM)
Year: 2024
Type: article
Abstract: Disclosure: R.A. Vigersky: Employee; Self; Medtronic Diabetes. Stock Owner; Self; Medtronic Diabetes. T.L. Cordero: Employee; Self; Medtronic Diabetes. Stock Owner; Self; Medtronic Diabetes. J. Shin: Employee; Self; Medtronic Diabetes. Stock Owner; Self; Medtronic Diabetes. Background: There are integrated CGM system criteria that describe the accuracy metrics for clearance as a Class II (special controls) device.[1] Mean absolute relative difference (MARD) is not a requisite criterion, yet MARD has been adopted by the diabetes community as the single value representing a sensor’s analytic accuracy despite the absence of clinical studies demonstrating that it differentiates the safety or clinical effectiveness of sensors in automated insulin delivery (AID) systems or standalone devices. While the calculation of MARD may be simple (i.e., the mean difference between sensor glucose [SG] and a reference blood glucose [BG] from a Yellow Springs Instrument or equivalent BG analyzer), that single value is materially affected by study design (e.g., subject demographics), type of diabetes, sensor location, and the percentage of values in each glycemic range (<70, 70-180 or >180 mg/dL). Methods: We compared the: A) insulin dose that would be delivered by an AID system or manual injection at an ambient SG of 180mg/dL with a MARD of 8%, 10% and 15% using typical settings of insulin sensitivity factor (1 unit/50mg/dL) and insulin-to-carb ratio (1 unit/10 grams) when a 50-gram carbohydrate meal is eaten; B) impact of having more points in the hypo- or hyperglycemic range during accuracy testing; and C) time-in-range (TIR) or A1C of AID systems using sensors with different MARDs. Results: A. When correcting to a SG of 120mg/dL, the difference in insulin dose using sensors with a MARD of 8% and 10% was -0.06 to +0.08 units, and with a MARD of 8% and 15% was -0.24 to +0.26 units, for lower and upper MARD boundaries. B. For a sensor with an overall MARD of 9.2%, adding 15% more SG-BG paired points in the hypo- or hyperglycemic range changes MARD from 9.2% to 10.3% and from 9.2% to 8.8%, respectively. C. Pivotal trials2-5 and real-world data6-8 of AID systems using sensors with MARDs of 9% (Dexcom’s G6) or 10% (Medtronic’s GS3) demonstrate that AID algorithms compensated for any of these minor differences with most system users achieving consensus recommended TIR or A1C targets. Conclusions: Using MARD to describe sensor accuracy has significant limitations, as a difference in study design (alone) can increase/decrease a MARD value. It has led many healthcare providers to inappropriately conclude that this criterion confers clinical superiority/inferiority of a CGM device. References: [1]U.S. Food and Drug Administration. 21CFR862.1355. Accessed December 4, 2023. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr=862.1355; 2Carlson A et al. Diabetes Technol Ther 2022;24:178-189; 3Breton M et al. NEJM 2020;383:836-45; 4Brown SA et al. NEJM 2019;381:1707-17; 5Brown SA et al. Diabetes Care 2021;44:1630-40; 6Castañeda J et al. Diab Obes Metab 2022;24:2212-21; 7Breton MD and Kovatchev BP. Diab Technol Ther 2021;23:601-8; 8Lal R et al. Diabetes 2023;72[Suppl.]:57-OR. Presentation: 6/1/2024
Source: Journal of the Endocrine Society
Language: English
Subfield: Endocrinology, Diabetes and Metabolism
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,418
OpenAlex ID: W4403148828