Page 50 - Glucose Monitoring Devices

Clinical evaluation of SMBG systems 47

significance of postmarketing SMBG system-generated clinical errors. To address
this need, a group of representatives from FDA, NIH, academia, US Army, and
industry convened to develop a new error grid, the surveillance error grid (SEG),
which would assess the degree of clinical risk associated with clinically inaccurate
SMBG systems in postmarket decision-making [7]. Diabetes clinicians were
recruited to provide their assessments of the clinical risk of errors associated with
four different patient scenarios [1]: T1D patient using an insulin pump [2]; type 2
diabetes (T2D) patient using insulin [3]; T2D patient not using insulin [4]; T1D
patient using multiple daily injections and continuous glucose monitoring system;
and supply a minimum and maximum BG range for each of the following types
of clinical action: (A) emergency treatment for low BG; (B) take oral glucose;
(C) no action needed; (D) take insulin; (E) emergency treatment of high BG.
Scenario [3], that considered a noninsulin-treated patient, had a different description
for clinical action (D), which was exercise and eat less. Clinicians were then asked to
compare SMBG measurements from each of the five possible BG ranges (AeE) to
actual BG concentrations from the same five ranges and select on a scale of 1e9 the
magnitude of clinical risk for each possible combination. If the measured BG and the
actual BG were in the same range, that square was prefilled as a no-risk outcome. Of
note, each respondent used their own glucose ranges for AeE, which they had
previously defined. A grid was then created for each respondent, such that each point
on the grid represented a data pair consisting of reference glucose on the x-axis and
measured glucose on the y-axis. Each data point was then integrated and averaged
for the entire set of respondents such that each data point could be assigned a unique
mean score according to the mean perception of clinical risk for that data pair. This
calculation generated a clinical risk for each combination of measured and reference
BG and a gradual spectrum of risk within each risk zone that was now defined by a
range of risk scores. The color-coded final surveillance error grid displays a risk
estimate for each SMBGereference pair (Fig. 3.4). This analysis differs from
both the EGA and the CEG where data points within the same risk zone are all
assigned the same risk and data pairs that are categorized similarly may differ signif-
icantly. Comparisons of the EGA/CEG and the SEG are shown in Fig. 3.5. Using a
set of 10,000 simulated reference versus SMBG data pairs, the SEG was shown to
correlate significantly with either the EGA or CEG, while the correlation between
the EGA and the CEG was stronger. It is unclear whether or not or how the contin-
uous risk information provided by the SEG will be used by clinicians and diabetes
educators in patient management. However, the SEG has been used to help identify
clinical risk in postmarketing studies of SMBG accuracy [8].
In summary, the clinical accuracy of SMBG systems is critical to the utility of the
information generated in the management of patient with diabetes. Understanding
the potential risk associated with differences between reference and SMBG results
can be of immediate importance to clinical decision-making. Error grids have
become a useful way to both visually quantitate and present clinical accuracy.