The Myth of MARD (Mean Absolute Relative Difference): Limitations of MARD in the Clinical Assessment of Continuous Glucose Monitoring Data.

ABSTRACT

The mean absolute relative difference (MARD) is a numerical metric that has been adopted by the diabetes technology community as the main indicator that describes the accuracy of a glucose sensor at a single point in time. The appropriateness of this adoption is questionable because there is limited evidence that MARD has meaningful clinical relevance in the current era of sensor technology. The calculation may be simple, but evaluation of MARD can be very complex because it is substantially impacted by the design of the data collection in an accuracy study. Factors that can influence the overall MARD include participant demographics such as type of diabetes and age, site of sensor wear, and the percentage of collected values in each glycemic range during the study that is, in turn, a function of the study design. MARD is only one of several important statistical metrics such as bias and precision that are relevant to assessing accuracy of a sensor. Furthermore, these analytic metrics convey little information about the safety and effectiveness of sensor use with an automated insulin delivery system or a standalone device. There are no clinical studies in people with diabetes (PWD) proving that MARD can accurately differentiate between a safe and unsafe sensor or between a more and less clinically effective sensor. Moreover, there are alternatives to MARD that can do this in a clinically meaningful way, which include error grid analyses and clinical studies in PWD. This review attempts to demythologize the status of MARD for the diabetes community in an effort to shift the focus from MARD to using clinically relevant assessments.