Page 60 - Glucose Monitoring Devices

58 CHAPTER 4 Consequences of SMBG systems inaccuracy

Finally, EGA meter reading classifications have limitations that can lead to inaccu-
rate estimations [58]. In other words, while it is possible to observe patient’s use of a
specific meter and even observe short-term clinical benefits [3,49,57], doing so for
a long enough time to observe long-term complications would be costly. In fact, it
would be near impossible to simultaneously assess all available meters. Such an
experiment would simply be too costly and impractical. Furthermore, to understand
corner cases in accuracy, a study may need to put a patient at risk.
An alternative to clinical studies is to resort to computer simulations or in silico
studies to assess meter performance and its consequences. Computer simulation is
first used to understand the relationship between meter errors and glycemic control,
which can be translated into short-term clinical outcomes. Additional models are
then applied to translate these clinical outcomes, such as changes in HbA1c, into
long-term complications of diabetes. Finally, fingerstick and insulin costs, together
with treatment costs associated with severe hypoglycemia treatment in hospitals,
along with the costs of treating other diabetes-related complications is estimated
using country and healthcare system-specific figures.

Modeling and simulation
In this section, we present a model and simulation-based framework to characterize
the consequences of BGM systems inaccuracy. In the discussion, we will detail the
steps that were required to develop the simulation framework and how the available
tools allow to satisfy all the requirements discussed in Requirements section.

Metabolic models and simulators (metabolic variability)
A first step requires to connect an erroneous meter reading to the treatment decision
and immediate effect on glycemic control. For this purpose, models of the glucose-
insulin regulation system are needed, such as the so-called minimal model that
greatly improved our understanding and measurement of glucose-insulin meta-
bolism [59,60]. This model and variants were then extended with the availability
of models of meal absorption [61], glucagon secretion [62], and insulin/glucagon
transport [63e66].
Metabolic modeling efforts laid the groundwork for metabolic simulators such as
the University of Virginia (UVA)/Padova [67e69] and other compartmental simula-
tors [70]. Although originally designed with the specific goal to support the
development of artificial pancreas systems [71e73], metabolic simulators have
been extensively applied to other forms of insulin dosing therapy modalities, and
accuracy studies in particular.
Behavioral modeling and simulation (behavioral variability and therapy
modes)
Parallel efforts were made to develop behavioral models of self-treatment of hypo-
glycemia [39,74,75], eating, and bolusing [18,76]. These models were applied to the
UVA/Padova simulator in several accuracy studies [18,76], but are still of limited