Useful Measures to Assess Glucose Dynamics from Continuous Glucose Monitoring Data
Authored by Kohnert KD
Abstract
Blood glucose control is a fundamental element in
preventing micro- and macrovascular complications of diabetes. Many
patients with diabetes can improve glycemic control by use of continuous
glucose monitoring. Analysis of continuously monitored data further
allows glucose dysregulation to reveal early in the development of
diabetes or the metabolic syndrome and response to therapeutic
intervention. But traditional measures of glycemia currently used in
clinical practice are insufficient to characterize the various aspects
of glucose profile complexity that may be important in different states
of dysglycemia and the design of optimal antidiabetes therapy. This
article discusses several dynamical complexity measures, their
association with metabolic characteristics, and possible implications
for glucose control.
Abbreviations:
CGM: Continuous Glucose Monitoring; DFA: Detrended Fluctuation
Analysis; PCP: Poincare Plots; SFE: Shape of the Fitting Ellipse; AFE:
New Metrics Area; ApEn: Approximate Entropy; SampEn: Sample Entropy;
MSE: Multiscale Entropy
Introduction
Although useful in clinical practice, metrics proposed for continuous glucose monitoring (CGM) in diabetes control [1]
such as mean glucose, SD of glucose, and time or percentage of time
spent in hyper- and hypoglycemia, are not appropriate to reveal the
internal dynamics of CGM time series. These metrics are derived from
linear models of glucose analysis and mostly fail to characterize
glucose dynamics [2].
As a matter of fact, the variation of blood glucose levels is not
linear, and glucose profiles contain nonlinear, non-stationary
components [3].
Several nonlinear analytical methods have been recently applied to
quantify the complexity of blood glucose signals, including detrended
fluctuation analysis (DFA) [4-11], Poincare plots (PCP) and various entropy measures [12-15].
These studies show that the complexity of blood
glucose variations is lower in patients with diabetes as compared with
nondiabetic subjects. Beyond traditional estimates of glycemia and
glycemic variability, complexity measures target the glucoregulatory
system and have the potential to assess how treatment modalities can
modify the dynamics of glucose. The clinical importance of the current
dynamical measures, however, it is not yet clear. And it is neither
known whether the complexity measures are affected by the diabetes
therapy nor have correlations been examined between various dynamical
indices.
Quantification of Glucose Dynamics and Association with Glycemic Control Measures
Table1 summarizes the measures of glucose dynamics recently used in studies on hyperglycemia.
Detrended fluctuation analysis
The Detrended fluctuation analysis (DFA) is useful to
assess long-range correlations in time series. The use of this method
yields the DFA scaling exponent α that reflects the degree of
complexity. Alpha values <1.5 indicate long-range negatively and
α>1.5 positively correlated fluctuation. Churruca et al. [4] and Ogata et al. [5]
reported a loss of complexity in glucose profiles of patients with the
metabolic syndrome and with diabetes, e.g., and the scaling exponents α
were higher than in healthy persons. An exploratory study conducted by
Khovanova et al. [6] and Thomas et al. [7]
provided a dynamical definition of glycemic stability and supported the
potential of DFA for time series analysis in diabetic patients.
Yamamoto et al. [8]
showed that the loss of glucose profile complexity stretched from the
short-range (α1) to the long-range scaling exponent (α2) with the
worsening glycemia. The results of these studies led to the conclusion
that changes in glucose dynamics may already occur before full-blown
hyperglycemia develops. Indeed, Varela's group [9,10]
has recently demonstrated that DFA α was capable of indicating the risk
of developing T2D. But it remains unknown whether dynamical alterations
are primarily dependent on exogenous factors, e.g., antihyperglycemics,
or endogenous factors such as the residual β-cell function and insulin
resistance. A retrospective analysis by Kohnert et al. [11]
in patients with T2D found that the loss of dynamical complexity
relates to the decline in β-cell reserve and increasing glycemic
variability.
Poincaré plot analysis
The standard oincaré plot (PCP) used to visualize the
non-linear pattern of glucose dynamics is a scattergram constructed by
locating data points from the CGM time series [12,13].
Quantification of the plots is done using SD1 and SD2 statistics, where
the minor axis (SD1) comprises the data points perpendicular to the
line of identity and SD2 those dispersed along the major axis of the
fitting ellipse. Crenier [13]
has validated these standard measures for the geometry of PCP in T1D
and introduced the new metrics area (AFE) and shape of the fitting
ellipse (SFE). As he reported, all the metrics of the PCP geometry were
higher in diabetic subjects than in the healthy control group. Worthy of
note, these parameters decreased upon continuous subcutaneous insulin
infusion therapy, indicating that they were modifiable by exogenous
factors. More recently, Garcia Maset et al. [14]
have reported that the loss of complexity in glucose time series of
pediatric patients with T1D, measured as DFA α and PCP parameters, was
correlated with increased glycemic variability.
Entropy measures
In addition to the previous algorithms the complexity
of glucose time series can be accessed through several other methods,
including approximate entropy (ApEn) and sample entropy (SampEn) [15,16].
Both require for computation the three parameters length of the data
segment (m), similarity criterion (r) and length of the data (N).
However, a consensus is still lacking about the proper selection of the
parameters. Using ApEn, Lytrivi & Crenier [17]
found an increase of glucose profile complexity in T1D upon switching
therapy from multiple daily insulin injections to continuous
subcutaneous insulin infusion. The ApEn increase was inversely related
with the DFA exponent α and reduced glycemic variability, indicating
that therapy closer to physiological insulin secretion can improve
glycemic complexity. SampEn is a modification of ApEn but has the
advantage of being less sensitive to the step length within the time
series [15].
Regarding the analysis of glucose complexity, studies on patients with
T1D indicated that SampEn was associated with insulin resistance and
with time in hypoglycemia (<3.9 mol/l) [18].
Since its introduction [19] multiscale entropy (MSE) represents the predominating method to characterize the complexity of physiological signals [19].
The MSE approach is based on sample entropy computation over a range of
timescales and has been proposed for assessment of glucose dynamics.
Costa et al. [20]
introduced the term "dynamical geometry" to set the framework for the
analysis of blood glucose time series using computation methods. Using
the MSE approach, recent studies have shown that the temporal structure
of glucose fluctuations is more complex in nondiabetic subjects than in
those with T2D. One study reported a significant correlation between the
MSE index with conventional measures of glycemia [20], i.e., glycated hemoglobin A1c (HbA1c) and mean blood glucose. Chen et al. [21]
identified no such correlations in their study, including a mixed
cohort of T1D and T2D patients. In a preliminary investigation, we
observed no significant association of the MSE index with mean glucose
and merely a weak correlation with HbA1c (Kohnert et al. unpublished).
Modifiable and Non-Modifiable Factors Associated with Glucose Dynamics
The internal structures of CGM time series imply a
complex dynamic process regulated by a set of interactions between
various hormones and metabolic components. And an important question is
whether individuals within a subgroup of diabetes have their dynamic
structure. As shown by Rahaghi et al. [22]
and confirmed in several studies, time-scale analysis does not reveal
major differences between diabetes types, even though they do show lower
glucose complexity compared with nondiabetic subjects. Insulin
resistance is expected to lower glucose dynamics, especially in T2D
patients not treated with insulin injections, and significant
correlation between insulin resistance and body mass index has been
reported [18]
However, insulin sensitivity is modifiable either by physical exercise
or administration of insulin sensitizers. Other modifiable factors
include meal intake and enteral feeding, which alter the individual
dynamic signature [22].
Presumably, any therapy, approaching the physiological glucose
regulation, is capable of shifting the glucose dynamics toward more
healthy conditions, as demonstrated in patients with T1D when switching
the therapy from daily insulin injections to insulin infusion [17].
Assessing the effects of non-modifiable factors such as age and
diabetes duration is complicated by the fact that close correlations
exist between aging, obesity, and insulin resistance.
Implications for Diabetes Control
Whereas the glycemic variability characterizes the
magnitude of the time series, the structural variability defines the
complexity of the time series. Both parameters represent two
complementary categories [23].
Despite the paucity of clinical data, the use of measures of blood
glucose dynamics appears to have a considerable potential in diabetes
control. Current study results consistently show that glucose complexity
decreases during progression from early glucoregulatory dysfunctions to
established diabetes. Thus, glucose series complexity measures may be
both beneficial in detecting the malfunctions in the glucoregulatory
system before the blood glucose reaches pathological levels [9]
and in characterizing the diabetes stability after therapeutic
intervention. The influence of antidiabetic agents is, however, as yet
unknown.
Clinical studies will need to be conducted to clarify
whether therapy modalities, beyond merely normalizing blood glucose
levels can reverse multiscale dynamics toward those of nondiabetic
subjects. Consequently, the glucoregulatory system could be the future
target for individualized diabetes therapy to restore healthy multiscale
dynamics. Furthermore, it remains an important goal to find out whether
glucose complexity measures are useful as markers for the
predisposition or the development of late diabetes complications.
Dynamic indices, measuring glucose dynamics on different time scales,
e.g., multiscale entropy, appear to be preferable for diabetes control.
There is no doubt that blood glucose dynamics is a fundamental concept
in the future management of diabetes.
To Know More About Current Research in Diabetes & Obesity
Journal Please click on:
https://juniperpublishers.com/crdoj/index.php
https://juniperpublishers.com/crdoj/index.php
Comments
Post a Comment