Page 166 - Glucose Monitoring Devices

Conclusions 167

Conclusions
CGM continues to evolve, and improvements in clinical and numerical accuracy and
performance can be anticipated. The criteria for market approval need to be
reviewed and updated to assure that they are appropriate for the sophisticated
intended use of these devices. Quantitative and qualitative measures for analyzing
the data streams generated by CGM will also need to be refined and standardized
to optimize their contribution to efficient and effective patient care. Although
some statistical techniques provide appropriate measures of the point accuracy of
CGM, they do not address the rate, or trend, components of the accuracy of CGM
systems. Continuous monitoring results in a data stream, a process in time, which
means that each data point is dependent on the prior data point and therefore not
an independent value as typically assumed by standard statistics [27,57]. Thus the
criteria for using standard statistical techniques to analyze these data are, generally,
not met. The order of the data points is important: the sequence of BG readings of
74, 80, 85, 86 mg/dL is not clinically the same as the sequence of BG readings of 86,
85, 80, and 74 mg/dL, even though the mean and standard deviation of the two data
series are identical. Thus the assessment of trend (or rate) accuracy is important for
the evaluation of CGM devices, both in terms of clinical and numerical metrics.
Although CG-EGA is admittedly complex, the information that it provides is
invaluable in deciding whether a particular CGM system would be effective in
different clinical situations. For instance, patients with a history of reduced aware-
ness of hypoglycemia might benefit from using a CGM system that is highly accu-
rate in the hypoglycemic range. Comparing the clinical accuracy of CGM systems in
the three clinically relevant ranges could help with the selection process. Systems
that feature alarms to signal impending hypo- or hyperglycemia would need to be
particularly accurate in measuring the rate and direction of change. CG-EGA has
been and is being used to evaluate the clinical accuracy of CGM systems as part
of the information required by regulatory agencies for marketing approval. To
date, the original EGA remains the only point error grid modified to support the eval-
uation of the clinical accuracy of CGM systems.

Acknowledgments
Author disclosures
BPK reports patents related to diabetes technology managed by the UVA Licensing and Ven-
tures Group, speaking engagements for Sanofi and Dexcom, consulting for Sanofi and Tan-
dem, and material support (to the University of Virginia) from Tandem Diabetes Care,
Roche Diagnostics, and Dexcom.
This work was supported in part by the UVA Strategic Investment in Precision Individualized
Medicine for Diabetes (PrIMeD Project).