Page 68 - Glucose Monitoring Devices

66 CHAPTER 4 Consequences of SMBG systems inaccuracy

We assessed the influence of error and bias on the clinical outcomes derived from
the simulation results; these included HbA1c, severe hypoglycemia, insulin, and
fingerstick use. Then, we proceeded to estimate the impact of HbA1c changes on
the incidence of DKA and long-term complications. Finally, we estimated the
impact of the observed inaccuracy on the overall financial costs. This included the
costs related to long-term complications, DKA and severe hypoglycemic events,
as well as the costs of insulin and fingerstick use.

Results: clinical outcomes

The results of the study are summarized in Fig. 4.10. Each dot in the figure corre-
sponds to the average outcomes of the entire in silico population using one particular
BGM system and represents four values: BGM system noise and bias (respectively
x-axis and dot color), and associated HbA1c and incidence of severe hypoglycemia
(respectively y-axis and dot size).
For example, consider the scenario where patients use an ideal sensor, that is, one
reporting the exact plasma glucose value. The x-coordinate value is 0 (no noise), and
the dot has a green color (no bias). The use of perfect information results in an
estimated HbA1c of almost 8.8% with an incidence in severe hypoglycemia of
around 1 event every 6 months (as per the size of the dot). Now consider the top-
right small dark blue dot. This corresponds to an erratic BGM system (one with
more than 80% of its measurements with errors greater than 5% of the true value).
This BGM system also has a large negative bias ( 20 mg/dL), hence the dark blue
color. Using this BGM system will drive the in silico population to a reduction of
hypoglycemia with respect to the ideal case, but at the cost of an increase of
HbA1c of 0.5%, up to a 9.2% level. An alternate example is to consider the big yel-
low dot at the bottom right of Fig. 4.10. This BGM system is noisy with an x-axis of
70 meaning that 70% of its readings have an ARD of more than 5%. It is deep yellow
because of its large positive bias. As a result of this bias (the BGM system overre-
ports glucose levels), patients using the meter will experience a lower HbA1 (y-axis
around 8.4% down from 8.8% HbA1c) at the cost of increased hypoglycemia (size of
the bubble z 2 hypoglycemic events every 6 months).
Overall, the results show a clear inverse relationship between bias (color) and
HbA1c. Note that the same effect is apparent in the incidence of severe hypoglyce-
mia, as BGM systems with large positive bias will more frequently drive patients to
over bolus and severe hypoglycemia. Our error metric alone seems to only very
mildly affect severe hypoglycemia. In this next section, we present regression
models that formalize and quantify these relationships.

Accuracy and clinical outcomes: a regression model
In an effort to further clarify the accuracy-to-outcome relationship, a set of linear
regression models was developed. The models explain HbA1c, severe hypoglyce-
mia, total daily insulin, and fingerstick count as a function of error and bias [76].