Is there a recipe for Comparability of MARD values from Clinical Performance Evaluation of CGM Sensors?
Sprache des Vortragstitels:
Diabetes Technology Meeting 2015
Sprache des Tagungstitel:
The most widely used measure for the performance of continuous glucose monitoring (CGM) systems is the mean absolute relative difference (MARD). However, different MARD values have been published for the same CGM system, which was explained by the fact that MARD does not reflect only the performance of the CGM system, but is heavily influenced by the study design. As study protocols will never be identical, the aim of this publication is to define conditions under which comparability is given.
We first examine by a Monte Carlo simulation how reliable a MARD is (quantified by its confidence interval) according to the number of measurements, the choice of the reference (or comparison) quantity and the properties of paired points. The results of this Monte Carlo study (performed using CGM data from a clinical trial with 12 patients) are then used to define simple rules to fix the key study parameters according to the desired confidence.
A critical pre-condition is a sufficient number of data points in all clinically relevant blood glucose ranges. An irregular distribution of points (e.g. fewer values in euglycemia and more in hyperglycemia) can be taken into account subsequently and, if so, will not affect MARD. The choice of the comparison method is critical, as it defines a reliability threshold which cannot be improved independently from the number and distribution of points.
As CGMs are continuously improving, and their difference will become the longer the smaller, a proper performance assessment of the continuous glucose readings is necessary. This is achievable using the MARD, but care must be taken to ensure comparability of results already at the study design phase.