Page 59 - Glucose Monitoring Devices

Accuracy and its consequences 57

(when a CGM is used in adjunctive mode), or only for calibration (CGM in nonad-
junctive mode). The effects of accuracy will be different in each therapy mode and
use case, and this information must be clearly recorded and reported as part of the
study.

Time span
A system study must somehow assess the effects of accuracy as it pertains accrued
costs and more importantly the development of diabetes-related complications.
The rest of the chapter focuses on describing a systems approach that has been
applied to this problem and an example of its application to pump users. First, we
motivate our approach by discussing the result of direct observational studies and
their limitations. We then discuss modeling and simulation as an alternative
approach and how it addresses the stated requirements.

Accuracy and its consequences
The first studies relating accuracy and its clinical effects coined the term clinical
accuracy [23,25]. The rationale here is to directly transform measurement errors
into quantities that assign these errors a clinical significance. This is the approach
followed in Refs. [24,56] where (measurement, reference) pairs are classified into
zones in the plane. Each zone is associated with certain clinical significance. For
example, meter measurements that are extremely large with respect to the reference
value are labeled erroneous (D or E) as opposed to accurate or benign. The approach
lends itself well to a graphical display and is an excellent communication tool. There
are two potential shortcomings of the approach. On one hand, the boundaries of
these regions are discrete and the same for all the population. Considering the meta-
bolic variability of patients, it is unlikely that such boundaries are equally accurate
for all. On the other hand, the clinical consequences of inaccuracy represent a static
view of a dynamic system. It is conceivable that repeated errors, even mild ones, can
have long-term and cascading effects.
An alternate approach is to directly observe the clinical outcomes in a popula-
tion. For example, in Ref. [49], meter accuracy was assessed by comparing
SMBG meter measurements to a reference such as Yellow Springs Instrument
(YSI) results taken during patient visits. Measurement errors were classified into
accurate, benign, or erroneous according to the error grid analysis (EGA) [23].
The study reports significant correlations between erroneous reports and HbA1c
levels and the incidence of severe hypoglycemia. Other studies [57] show that the
improper use of meters requiring calibration to match a specific strip code may
also lead to insulin dosing errors that can have life-threatening consequences, for
example, can lead to an increased number of hypoglycemia levels under 50 mg/
dL. Among the limitations of the approach are the reduced number of meter brands
that can be observed, and accuracy variability across lots [27]. In addition, results are
aggregated making it difficult to understand the performance of individual meters.