Measures of Glycemic Variability in Type 1 Diabetes and the Effect of Real-Time Continuous Glucose Monitoring

Abstract

Objective:

To report the impact of continuous glucose monitoring (CGM) on glycemic variability (GV) indices, factors predictive of change, and to correlate variability with conventional markers of glycemia.

Methods:

Data from the JDRF study of CGM in participants with type 1 diabetes were used. Participants were randomized to CGM or self-monitored blood glucose (SMBG). GV indices at baseline, at 26 weeks in both groups, and at 52 weeks in the control group were analyzed. The associations of demographic and clinical factors with change in GV indices from baseline to 26 weeks were evaluated.

Results:

Baseline data were available for 448 subjects. GV indices were all outside normative ranges (P < 0.001). Intercorrelation between GV indices was common and, apart from coefficient of variation (CV), low blood glucose index (LBGI), and percentage of glycemic risk assessment diabetes equation score attributable to hypoglycemia (%GRADE_hypoglycemia), all indices correlate positively with HbA1c. There was strong correlation between time spent in hypoglycemia, and CV, LBGI, and %GRADE_hypoglycemia, but not with HbA1c. A significant reduction in all GV indices, except lability index and mean absolute glucose change per unit time (MAG), was demonstrated in the intervention group at 26 weeks compared with the control group. Baseline factors predicting a change in GV with CGM include baseline HbA1c, baseline GV, frequency of daily SMBG, and insulin pump use.

Conclusions:

CGM reduces most GV indices compared with SMBG in people with type 1 diabetes. The strong correlation between time spent in hypoglycemia and CV, LBGI, and %GRADE_hypoglycemia highlights the value of these metrics in assessing hypoglycemia as an adjunct to HbA1c in the overall assessment of glycemia.

Introduction

Clinical studies have demonstrated the relationship between HbA1c and the risk of micro- and macrovascular complications in type 1 diabetes,^1,2 and HbA1c is routinely used as a treatment target, to monitor glucose control, and for assessment of clinical response to therapeutic interventions. However, despite robust longitudinal data and their role in clinical care, HbA1c may not provide complete information on frequency, duration, or magnitude of short-term fluctuations in glycemia, and its validity is compromised in people with hemoglobinopathies, renal disease, and iron deficiency anemia.³ Furthermore, HbA1c is a poor predictor of the risk of severe hypoglycemia,⁴ and the threefold increase in severe hypoglycemia with intensive insulin treatment in the Diabetes Control and Complications Trial was in excess of that which could be explained by differences in HbA1c between the two groups.⁵

Despite evidence supporting a role for glycemic variability (GV),^6

–9 its contribution to diabetes-related complications remains controversial and its clinical value as an additional measure to HbA1c remains unclear.^10,11 A recent systematic review has suggested that GV is associated with an increased risk of microvascular complications in type 2 diabetes only and that the relationship between GV and vascular complications in type 1 diabetes is less clear.¹²

This analysis evaluates several indices of GV in a large cohort with type 1 diabetes with reference to ranges previously defined in people without diabetes and explores the primary hypothesis that the use of real-time continuous glucose monitoring (CGM) reduces GV. We have also explored the extent to which different GV measures positively correlate with HbA1c and time spent in hypoglycemia, and whether baseline factors predict GV. Our aim is to identify GV measures that could be a useful adjunct to HbA1c for a better assessment of the overall glycemic status of people with type 1 diabetes.

Methods

Data source

Data from the JDRF CGM study were used.^13
–15 The publicly accessible data were obtained from the Jaeb Center for Health Research (http://diabetes.jaeb.org/Dataset.aspx) and were processed using MATLAB (MathWorks).

The JDRF study protocol and clinical characteristics of enrolled subjects have been previously described in detail.^13
–15 In summary, the study is a 6-month randomized, parallel group, efficacy, and safety study designed to evaluate the impact of CGM on glycemic control in children and adults with type 1 diabetes. Participants underwent blinded CGM for 1 week before randomization to either standard self-monitored blood glucose (SMBG) (control) or use of unblinded CGM as a supplement to SMBG (intervention). All trial participants were provided with written instructions on how to use the data provided by CGM and capillary blood glucose meters to make real-time adjustments to insulin, and on the use of computer software to retrospectively review the glucose data to alter future insulin doses. Patients using CGM received additional instructions for modifying insulin doses and treatment of hypoglycemia on the basis of the glucose trend. Changes were made to diabetes management, as needed, for all participants during scheduled contacts. The randomized trial was followed by a 6-month extension study in which unblinded CGM was continued in the CGM group, but with unblinding to CGM data in the control group. CGM profiles were obtained at 26 weeks postrandomization from both control and intervention groups.

Measures of GV

We computed 13 measures of GV using EasyGV (v8.8.2.R2) software. Evaluated GV measures included standard deviation (SD), coefficient of variation (CV), mean amplitude of glycemic excursions (MAGE), continuous overall net glycemic action (CONGA), mean of daily differences (MODD), lability index (LI), and mean absolute glucose change per unit time (MAG), glycemic risk assessment diabetes equation (GRADE), M-value, average daily risk range (ADRR), J-Index, low blood glucose index (LBGI), and high blood glucose index (HBGI). GRADE score is also reported as %GRADE_hypoglycemia, %GRADE_euglycemia, and %GRADE_{hyperglycemia} representing percentages of GRADE scores attributable to glucose values <3.9 mmol/L, and between 3.9–7.8 mmol/L and >7.8 mmol/L, respectively. EasyGV is a Microsoft Excel workbook that has a number of options to define the sampling interval, CONGA length, LI interval, reference value of M-value, and whether SMBG or CGM is used for MAGE calculation.¹⁶ A description of the GV measures, formulae used for their calculations, and a critical review of their limitations is described elsewhere.^6,17

Analysis design

GV indices were assessed in comparison to previously published GV reference ranges in people without diabetes.⁶ To evaluate the effect of CGM on GV, between-group differences in GV indices were evaluated at 26 weeks and glucose profile data collected at 52 weeks from subjects in the control group who crossed over to CGM (unblinded control) were compared with data collected at 26 weeks in the same group (blinded control). The relationship between baseline GV indices and HbA1c was explored. Finally, baseline predictors of change in GV indices from baseline to 26 weeks in the CGM group were evaluated.

Statistical analysis

Data were examined for normality; non-normally distributed variables were logarithmically transformed with the use of geometric mean, SD, and 95% confidence interval (CI) for descriptive statistics of baseline data (n = 448).

An unpaired t-test was used to compare the effect of CGM on the change of GV indices from baseline to 26 weeks in the control and intervention arms, while a paired t-test was used for analysis of within-group differences in each of those arms. To study the effect on GV of unmasking CGM in the control group, a paired t-test was used to compare GV indices in the 207 participants in the control group with unblinded CGM at 52 weeks with their blinded GV indices at 26 weeks. Spearman correlation was used to examine the relationship between HbA1c and GV indices at baseline.

The associations of baseline demographic and clinical factors with change in GV indices from baseline to 26 weeks were evaluated in the CGM group (n = 231) using regression analysis. The analysis was constructed using the following predictor variables: age, gender, race, education level of caregiver, insulin modality (pump or multiple daily injections), frequency of daily self-reported blood glucose monitoring, occurrence of one or more episodes of severe hypoglycemia in last 6 months, diabetes duration, baseline HbA1c, and baseline GV. The analysis was performed with change in GV measures and continuous variables expressed as z scores. Categorical variables were included as dummy variables. Baseline factors with P ≤ 0.2 in the univariate analysis were carried forward to multivariable analysis.

Data are presented as means (SD), unless otherwise stated. Statistical tests were two tailed and for descriptive and exploratory analyses, a significance level of P < 0.05 was adopted. Where significance levels were consistently <0.001, t values were reported to aid evaluation of relative differences in magnitude and scatter of the differences observed. For exploring our primary hypothesis that the use of real-time CGM reduces GV, 13 different measures of glycemic variability were tested. Therefore, to test our primary hypothesis, we have adopted a significance level of P < 0.004 (i.e., 0.05/13). Statistical tests were performed using SPSS 21.0 for Mac (SPSS Inc., Chicago, IL).

Results

Participants

CGM profile data were available for 448 subjects at baseline (54.9% women, 94.4% white race), following exclusion of three subjects due to missing HbA1c data at baseline. Mean age was 25.1 years (SD 15.8) and mean diabetes duration 13.6 years (SD 11.7). Two hundred thirty-one subjects were randomized to the CGM group and 217 subjects to the control group. At 26 weeks, CGM profile data were available for all the 231 subjects in the CGM group and 214 subjects in the control group (blinded CGM). At 52 weeks, data were available for 207 subjects in the control group (unblinded CGM). Baseline characteristics were similar between the two groups (Table 1). There was no statistically significant difference in HbA1c and the evaluated GV measures between the two groups at baseline.

Table 1.

Baseline Characteristics of Participants in the Control and Intervention Groups

	Control	CGM
Number	217	231
Female sex, n (%)	124 (57.1)	122 (52.8)
Age, n. (%)
8–14 years	68 (31.35)	74 (32)
15–24 years	70 (32.25)	71 (30.7)
≥25 years	79 (36.4)	86 (37.2)
White race, n (%)	205 (94.4)	218 (94.3)
Duration of diabetes, years	9.2 (5.3, 17.5)	9.6 (5.6, 18.2)
Insulin administration, n (%)
Pump	175 (80.6)	190 (82.2)
MDI	42 (19.4)	41 (17.8)
HbA1c, mmol/mol (%)
<53 (7.0), n (%)	61 (28.1)	67 (29)
53–64 (7.0–8.0), n (%)	109 (50.2)	109 (47.2)
65–74 (8.1–8.9), n (%)	39 (18)	44 (19)
≥75 (9.0), n (%)	8 (3.7)	11 (4.8)
≥one episode of severe hypoglycemia in last 6 months, n (%)	17 (7.8)	21 (9.1)
Self-reported daily SMBG, n/day	6.26 (2.73)	6.11 (2.42)
College graduate (patient or primary caregiver), n (%)	194 (89.4)	197 (85.3)

Data are presented as number (%), mean (SD), or median (IQR).

CGM, continuous glucose monitoring; IQR, interquartile range; MDI, multiple daily injections; SD, standard deviation; SMBG, self-monitored blood glucose.

Measures of GV and glycemic control at baseline

Reference means and 95% CIs for several measures of GV have been previously described by analyzing CGM profiles of 70 subjects without diabetes.⁶ As shown in Table 2, the mean values of GV indices at baseline in the 448 subjects with type 1 diabetes were all appreciably higher than the reference means, with complete nonoverlap of 95% CIs throughout (P < 0.001).

Table 2.

Descriptive Statistics (Mean, 95% CI) for GV Indices in 70 Subjects Without Diabetes Representing Normal Reference Range ⁶ and in the 448 Subjects with Type 1 Diabetes at Baseline

GV measure	Reference (mean, 95% CI)	Type 1 DM (mean, 95% CI)
Mean	5.10 (4.98, 5.22)	9.27 (9.10, 9.43)
SD	1.50 (1.34, 1.66)	3.73 (3.64, 3.82)
CV	Calculated as 29.4%	40.43% (39.66, 41.19)
CONGA₁	4.60 (4.48, 4.72)	8.29 (8.13, 8.44)
LI	0.40 (−0.12, 0.92)	6.17 (5.84, 6.51)
J index	14.30 (13.20, 15.40)	52.72 (50.83, 54.67)
LBGI	3.10 (2.65, 3.55)	4.10 (3.79, 4.43)
HBGI	0.20 (−0.69, 1.09)	12.03 (11.53, 12.53)
GRADE	0.40 (−0.09, 0.89)	8.37 (7.98, 8.76)
MODD	0.80 (0.47, 1.13)	4.06 (3.95, 4.16)
MAGE-CGM	1.40 (1.24, 1.56)	6.76 (6.54, 6.98)
ADRR	0.40 (−0.56, 1.36)	33.35 (32.21, 34.48)
M-value	4.70 (3.81, 5.59)	14.70 (13.83, 15.62)
MAG	1.30 (1.21, 1.39)	2.70 (2.61, 2.79)

Difference between measures of GV and glycemic control in the two groups is statistically significant (P < 0.001). Measures are expressed in mmol/L.

CI, confidence interval; GV, glycemic variability.

Intercorrelation between measures of GV and glycemic control at baseline and correlation with HbA1c

Statistically significant intercorrelation between measures of GV and HbA1c was common. The GV index, ADRR, showed the strongest intercorrelation with other measures with r > 0.7 in 9 out of 13 intercorrelations. By contrast, CV, LBGI, and %GRADE_hypoglycemia showed the weakest intercorrelation. Similarly, all these measures, apart from CV, LBGI, and %GRADE_hypoglycemia, correlated significantly, but moderately (r 0.35–0.66), with HbA1c as shown in Table 3. The relationship between percent of time spent in hypoglycemia (glucose level <3.9, <3.3, and <2.8 mmol/L), CV, LBGI, %GRADE_hypoglycemia, and HbA1c was analyzed using Spearman correlation. CV, LBGI, and %GRADE_hypoglycemia correlated strongly with time spent in hypoglycemia (<3.9, 3.3, and 2.8 mmol/L) (r > 0.88, P < 0.001). No relationship between HbA1c and time spent in hypoglycemia was seen (Table 4).

Table 3.

Correlation Matrix Examining the Relationship Between Various Measures of Glycemic Variability and Quality of Glycemic Control in 448 Subjects with Type 1 Diabetes at Baseline(Data on %GRADE_hypoglycemi _a,%GRADE_euglycemia, and %GRADE_hyperglycemi _a Were Only Available on 340 Subjects at Baseline)

	Mean	SD	CV	GRADE	GRADE%Hypo	GRADE%Eugly	GRADE%Hyper	CONGA	LI	JINDEX	LBGI	HBGI	MODD	MAGE	ADDR	M-value	MAG
HbA1c	0.679^**	0.514^**	0.06	0.655^**	−0.266^**	−0.535^**	0.424^**	0.664^**	0.402^**	0.676^**	−0.07	0.659^**	0.483^**	0.446^**	0.613^**	0.548^**	0.345^**
Mean	1.00	0.648^**	−0.04	0.922^**	−0.566^**	−0.769^**	0.781^**	0.976^**	0.515^**	0.956^**	−0.192^**	0.890^**	0.620^**	0.547^**	0.843^**	0.725^**	0.431^**
SD		1.00	0.701^**	0.674^**	0.286^**	−0.603^**	−0.06	0.677^**	0.768^**	0.833^**	0.378^**	0.888^**	0.936^**	0.710^**	0.844^**	0.892^**	0.658^**
CV			1.00	0.06	0.746^**	−0.114^*	−0.646^**	0.01	0.536^**	0.224^**	0.695^**	0.356^**	0.649^**	0.448^**	0.336^**	0.520^**	0.472^**
GRADE				1.00	−0.290^**	−0.926^**	0.546^**	0.897^**	0.577^**	0.911^**	0.04	0.892^**	0.654^**	0.608^**	0.826^**	0.820^**	0.492^**
GRADE%Hypo					1.00	0.09	−0.909^**	−0.515^**	0.283^**	−0.297^**	0.940^**	−0.133^*	0.317^**	0.207^**	−0.145^**	0.141^**	0.317^**
GRADE%Eugly						1.00	−0.385^**	−0.765^**	−0.420^**	−0.831^**	−0.136^*	−0.879^**	−0.554^**	−0.495^**	−0.724^**	−0.861^**	−0.318^**
GRADE%Hyper							1.00	0.728^**	−0.121^*	0.549^**	−0.776^**	0.410^**	−0.09	−0.02	0.358^**	0.129^*	−0.177^**
CONGA								1.00	0.423^**	0.954^**	−0.173^**	0.900^**	0.636^**	0.511^**	0.803^**	0.741^**	0.325^**
LI									1.00	0.646^**	0.358^**	0.680^**	0.741^**	0.728^**	0.806^**	0.710^**	0.926^**
JINDEX										1.00	0.00	0.975^**	0.788^**	0.654^**	0.921^**	0.860^**	0.546^**
LBGI											1.00	0.120^*	0.389^**	0.310^**	0.09	0.362^**	0.360^**
HBGI												1.00	0.841^**	0.691^**	0.913^**	0.917^**	0.570^**
MODD													1.00	0.683^**	0.798^**	0.836^**	0.640^**
MAGE														1.00	0.721^**	0.701^**	0.623^**
ADDR															1.00	0.844^**	0.727^**
M-value																1.00	0.615^**
MAG																	1.00

Correlation is significant at the 0.05 level (two tailed).

Correlation is significant at the 0.01 level (two tailed).

Table 4.

Correlation Matrix Examining the Relationship Between Time Spent in Various Glucose Ranges, HbA1c, SD, CV, LBGI, and %GRADE_hypoglycemia in 448 Subjects with Type 1 Diabetes at Baseline (Data on %GRADE_hypoglycemi _a were Only Available on 340 Subjects at Baseline)

	HbA1c	CV	LBGI	GRADE %Hypo	≤2.8 mmol/L	≤3.3 mmol/L	≤3.9 mmol/L
HbA1c	1.00	0.06	−0.07	−0.266^**	−0.101^*	−0.213^**	−0.285^**
CV		1.00	0.695^**	0.746^**	0.639^**	0.681^**	0.693^**
LBGI			1.00	0.940^**	0.894^**	0.923^**	0.889^**
GRADE%Hypo				1.00	0.923^**	0.980^**	0.977^**
≤2.8 mmol/L					1.00	0.888^**	0.822^**
≤3.3 mmol/L						1.00	0.965^**
≤3.9 mmol/L							1.00

Correlation is significant at the 0.01 level (two tailed).

Correlation is significant at the 0.05 level (two tailed).

Effect of CGM on GV measures in the intervention group

Analysis of between-group differences demonstrated a significant difference between the intervention and control group in all measures of GV, with the exception of CV, LI, and MAG (Table 5). At 26 weeks, there was a significant reduction in all measures of GV, with the exception of LI and MAG, from baseline in the intervention group. There was a significant reduction in HbA1c in the intervention group from 58 mmol/mol (7.4%) to 55 mmol/mol (7.2%) (P < 0.001). Correspondingly, the achieved relative reductions in M-value, LBGI, and GRADE were 25.7%, 24.9%, and 16.5%, respectively (P < 0.001) (Table 5). In contrast, there was no statistically significant reduction in any of the measures of GV and glycemic control in the control group at 26 weeks compared to baseline.

Table 5.

Showing Mean (SD) Values of Measures of GV at 26 Weeks Compared to Baseline in the Intervention Group, Control Group, and Between-Group Differences in Change of These Measures at 26 Weeks from Baseline

	CGM							Control				Difference ^a
	n	Baseline	26 weeks	Change	% Change	t	P vs. baseline	n	Baseline	26 weeks	P vs. baseline	Mean	P
HbA1c (mmol/mol)	229	58 (10)	55 (10)	−3	−3.28	−5.4	<0.001	212	58 (9.2)	58 (8.2)	0.75	−3	<0.001
HbA1c (%)	229	7.4 (0.9)	7.2 (0.9)	−0.24	−3.28	−5.4	<0.001	212	7.46 (0.84)	7.44 (0.75)	0.75	−0.23	<0.001
Mean glucose	231	9.24 (1.85)	8.89 (1.34)	−0.34	−3.69	−3.9	<0.001	214	9.29 (1.7)	9.25 (1.5)	0.6	−0.29	0.018
SD	231	3.65 (0.97)	3.42 (0.79)	−0.24	−6.46	−5.3	<0.001	214	3.82 (0.98)	3.83 (0.85)	0.969	−0.24	<0.001
CV	231	39.66 (8.03)	38.23 (5.76)	−1.43	−3.61	−3.9	<0.001	214	41.34 (8.4)	41.55 (7.4)	0.683	−1.65	0.011
CONGA₁	231	8.26 (1.76)	7.83 (1.29)	−0.43	−5.18	−4.9	<0.001	214	8.32 (1.62)	8.25 (1.44)	0.454	−0.36	0.003
LI	231	6.94 (4.27)	6.77 (3.6)	−0.18	−2.53	−0.9	0.35	214	7.56 (4.1)	7.66 (4.3)	0.701	−0.27	0.383
J-index	231	56.01 (22.16)	50.43 (16.78)	−5.58	−9.96	−5.1	<0.001	214	57.61 (21.2)	56.89 (19.3)	0.5	−4.85	0.002
LBGI	231	5.38 (4.36)	4.06 (2.23)	−1.32	−24.50	−6.0	<0.001	214	5.63 (4.37)	5.43 (4.14)	0.521	−1.12	0.003
HBGI	231	11.79 (5.5)	10.03 (4.21)	−1.76	−14.91	−6.7	<0.001	214	12.29 (5.3)	11.91 (4.7)	0.153	−1.37	<0.001
GRADE	231	8.34 (4.31)	6.97 (3.1)	−1.38	−16.49	−6.4	<0.001	214	8.40 (4.1)	8.08 (3.65)	0.167	−1.06	0.001
MODD	231	3.98 (1.12)	3.66 (0.89)	−0.33	−8.16	−6.0	<0.001	214	4.14 (1.16)	4.52 (0.73)	<0.001	−0.7	<0.001
MAGE	222	6.75 (2.1)	6.29 (1.35)	−0.46	−6.81	−4.0	<0.001	205	7.17 (2.1)	7.37 (1.9)	0.140	−0.66	<0.001
ADRR	231	32.38 (12.03)	30.35 (9.99)	−2.03	−6.27	−3.6	<0.001	214	34.43 (12.5)	41.09 (9.6)	<0.001	−8.68	<0.001
M-value	231	17.51 (11.12)	12.99 (7.61)	−4.52	−25.81	−8.0	<0.001	214	18.33 (10.7)	17.43 (10.23)	0.151	−3.62	<0.001
MAG	231	2.84 (1.22)	2.89 (1.19)	0.05	1.76	0.8	0.40	214	2.96 (1.1)	3.08 (1.3)	0.108	−0.66	0.474

The table also shows the absolute and relative change in the mean values in the intervention group. Measures are expressed in mmol/L.

Between-group difference for change in measures of GV and quality of glycemic control at end of 26 weeks from baseline.

Effect of unmasking of CGM in the control group

The effect of unmasking CGM in the control group at 26 weeks was evaluated by comparing GV measures at 52 weeks to those at 26 weeks (immediately before unmasking CGM). Despite the nonsignificant change in HbA1c from 26 to 52 weeks, there was a significant reduction in all measures of GV, with the exception of LI and MAG, from baseline (Supplementary Table S1; Supplementary Data are available at www.liebertonline.com/dia).

Factors predictive of response in the CGM group

Univariate analysis showed that baseline GV was a significant predictor of change in all measures of GV at 26 weeks in the intervention group, with higher baseline GV associated with greater reduction in GV at 26 weeks. Similarly, baseline HbA1c was also a significant predictor of change in GV in the majority of the evaluated GV measures (all except LBGI, CV, LI, and MAG), with higher baseline HbA1c associated with greater reduction in GV measures at 26 weeks. However, multivariable analysis showed that higher HbA1c at baseline was associated with less of a reduction of GV measures at 26 weeks (Supplementary Table S2). In the multivariable analysis, treatment with insulin pump predicted a reduction in LBGI and M-value, while frequent use of SMBG predicted a reduction in LI. Other variables, including education level of caregiver, did not predict change in any of the evaluated GV measures.

Discussion

The increased availability of CGM provides a wealth of data and enables assessment of GV. Several measures of GV have been described.^18

–23 These can be broadly subdivided into measures based on glucose distribution (e.g., SD, CV, MAGE, CONGA, MODD, LI, and MAG), and measures based on risk and quality of glycemic control that are also sensitive to GV (e.g., GRADE, M-value, ADRR, J-Index, LBGI, and HBGI).²⁴ There are several challenges related to GV measurements and interpretation,²⁴ including correlation with mean glycemia, making it difficult to evaluate if the change in GV measures following an intervention is related to change in mean glycemia, GV, or both. These challenges are further complicated by the plethora of GV measures proposed in the literature with the lack of a “gold standard” measure.^18

–23

The data reported here are from the largest available CGM data set, with evaluation of measures of GV, the effect of real-time CGM on these measures, and predictors of changes to variability. The analysis demonstrates the magnitude of variability in people with type 1 diabetes and the impact that real-time CGM has on GV measures. Measures of dispersion (SD and CV), glucose risk indices (LGBI, HGBI, ADRR, GRADE), measures of unknown significance (CONGA, MAGE, J-index, M-value), and day to day change (MODD) all fell significantly in the intervention group over 26 weeks, and (with the exception of CV) fell significantly more than in the control group. Moreover, over a further 26 weeks, in the control group, unblinding resulted in significant falls in these measures of GV. However, two measures of variability over time (LI and MAG) were entirely unchanged. This may suggest that the use of real-time glucose trends and alarms addresses glucose dispersion and risk while the underlying temporal glucose variability remains unchanged. It could also reflect the impact of interventions based on CGM use such as more frequent insulin correction boluses or carbohydrate ingestion, resulting in more high-frequency (rapid) glucose fluctuations that LI and MAG are more sensitive to.

The impact of CGM on GV measures is consistent with previous studies.^25
–27 However, these studies have been for a shorter duration^25,26 or in limited numbers of participants.²⁷ Similarly, the correlation between GV measures reported in this analysis confirms previous findings. A correlation analysis of GV measures in 48 subjects with type 1 diabetes showed similar relationships to our study between MAGE and other measures of GV, with a significant correlation of 0.74 between MAGE and SD and a correlation of 0.58 between SD and CONGA₁.²⁸ Rodbard also reported a strong correlation between SD and MAGE (r = 0.89), SD and CONGA₁ (r = 0.71), SD and MODD (r = 0.81), and between MAGE and MODD (r = 0.74).²⁹ Compared to our analysis, other studies also showed similar relationships between HbA1c and HBGI (r = 0.63)³⁰ and between HbA1c and MAGE (r = 0.49).³¹ In another retrospective analysis of 72-h CGM data in 815 outpatients (48 with type 1 diabetes and 767 with type 2 diabetes), correlation between MAG and SD was stronger (r = 0.88 compared to 0.66 in our analysis), as was the relationship between MAG and other GV measures.³²

Common intercorrelation between evaluated GV measures is demonstrated alongside a moderate correlation between most of the evaluated GV measures and HbA1c (Table 3). This suggests that these measures convey similar information, reflecting mean glycemia, as well as information on glucose variability.

LBGI and %GRADE_hypoglycemia, which are sensitive to hypoglycemia alone, correlated poorly with other GV measures and not at all with HbA1c, supporting the hypothesis that HbA1c has a limited role in reflecting or predicting the risk of hypoglycemia²⁰ and suggesting that these two metrics may offer additional information when assessing CGM data. The relationship between time spent in hypoglycemia, LBGI, and %GRADE_hypoglycemia may suggest that direct assessment of time spent in hypoglycemia, routinely reported from CGM data, is sufficient. However, these two metrics provide a continuous scale for hypoglycemia with a different risk score assigned to glucose values rather than the categorical information provided by time spent below a glucose threshold, which may not consider severity, with all values assigned an equal weight.³³ Similarly, CV correlated with time spent in hypoglycemia (Table 4) and not with HbA1c or mean glucose (Table 3). CV has been previously proposed as the best parameter to characterize GV since it is corrected for the mean and avoids the dependency of SD and other measures of GV on mean glucose or HbA1c.³⁴ The weaker impact of CGM on CV demonstrated in the intervention group supports the role of CV in providing distinct information from the other GV measures, and from HbA1c.

There was a complex of associations between HbA1c and GV at baseline and subsequent change in GV in the CGM group. In univariate analyses, higher baseline HbA1c was associated with greater GV (except for CV and LGBI), and higher baseline HbA1c and GV were associated with greater reductions in GV over 26 weeks. However, multivariable analysis showed that the association between higher baseline HbA1c and greater reduction in GV was secondary to the association between baseline GV and reduction in GV. In fact, when variation in baseline GV was taken into account, the association between baseline HbA1c and subsequent reduction in GV changed direction: higher baseline HbA1c was associated with an increase in the majority of GV measures over 26 weeks, indicating that among those with high baseline HbA1c, the subsequent reduction in GV was not as great as would have been expected from their baseline GV.

Limitations of this analysis include the reliability of some of the evaluated GV measures. ADRR requires data collected typically over 1 month with a minimum of 14 days and with typical frequency of 3–5 glucose measurements per day, whereas CGM data used to compute ADRR in control group at 26 weeks in this analysis were collected for only 1 week.

Another limitation is the possible different outcomes when calculating GV measures using different automated GV calculators. Several methods and automated calculators have been described for calculating MAGE.^35

–38 EasyGV uses a modified method (MAGE-CGM) for MAGE calculation. The MAGE-CGM formula selects a peak or trough based on direction of change (rising or falling) of the preceding and succeeding data points. It also contains a 15-min lag window for the direction of change based on the lag between interstitial fluid glucose measurement and plasma glucose concentrations. It also contains an algorithm that eliminates short-term fluctuations related to sensor inaccuracies.⁶ The correlation between MAGE calculated using EasyGV and MAGE calculated based on the methods of Fritzsche or Baghurst was reported to be of 0.87.³⁹ Comparison between EasyGV (v8.8.2.R2) and a validated automated GV calculator showed a correlation of 0.76 in calculating MODD but poor correlation in calculating CONGA₁.⁴⁰ EasyGV has now been updated to address an issue with the MODD calculation but the reason for discrepancies in CONGA₁ between calculators remains unclear and is the subject of ongoing investigation.

It is also important to note that the normative range of GV measures referred to in our article was based on the analysis of 55,000 data points from only 72 h of CGM obtained from 70 subjects with fasting plasma glucose of <120 mg/dL. Although this excludes subjects with diabetes, it is possible that some subjects had impaired glucose tolerance. This could explain why SD in this cohort, reported to be 1.5, is larger compared to SD reported in other studies in subjects with normal glucose tolerance (ranged 0.5–0.9). However, other GV measures were similar to those reported from analysis of other data sets.^{28,31,41
–43} The accuracy of the CGM systems used in the study, particularly in the hypoglycemic range, is a further limitation and may affect GV measurements. Since the CGM study was conducted, the accuracy of available systems has improved significantly and the large volume of data analyzed here mitigates much of this problem.

Conclusion

In summary, we report the largest study to date of the long-term impact of CGM on GV in people with type 1 diabetes. There was a significant reduction in the evaluated measures of GV with CGM use, with the exception of LI and MAG, suggesting that CGM reduces mean glycemia, glucose dispersion, and risk but not glucose fluctuation. The strong correlation between time spent in hypoglycemia, CV, LBGI, and %GRADE_hypoglycemia, but not with HbA1c, highlights the value of these metrics in assessing hypoglycemia and as a useful adjunct to HbA1c in overall assessment of glycemic status in people with type 1 diabetes. A large-scale longitudinal intervention study is required to evaluate the HbA1c-independent role of GV in the development of diabetes-related vascular complications and this will additionally require the definition of an agreed “gold-standard” GV measure.

Footnotes

Authors' Contributions

A.H.L. analyzed the data and wrote the manuscript; I.F.G. supervised data analysis and reviewed/edited the manuscript; D.G.J. contributed to the discussion and reviewed/edited the manuscript; N.S.O. processed original glucose profile data and reviewed/edited the manuscript.

Author Disclosure Statement

N.S.O. has received honoraria for consultancy from Abbott Diabetes Care and Roche.

References

UKPDS: Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). UK Prospective Diabetes Study (UKPDS) Group. Lancet, 1998; 352:837–853.

DCCT: The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. The Diabetes Control and Complications Trial Research Group. N Engl J Med, 1993; 329:977–986.

Coban

, Ozdogan

, Timuragaoglu

: Effect of iron deficiency anemia on the levels of hemoglobin A1c in nondiabetic patients. Acta Haematol, 2004; 112:126–128.

Miller

, Foster

, Beck

, et al.: Current state of type 1 diabetes treatment in the U.S.: updated data from the T1D Exchange Clinic Registry. Diabetes Care, 2015; 38:971–978.

Kilpatrick

, Rigby

, Goode

, Atkin

: Relating mean blood glucose and glucose variability to the risk of multiple episodes of hypoglycaemia in type 1 diabetes. Diabetologia, 2007; 50:2553–2561.

Hill

, Oliver

, Choudhary

, et al.: Normal reference range for mean tissue glucose and glycemic variability derived from continuous glucose monitoring for subjects without diabetes in different ethnic groups. Diabetes Technol Ther, 2011; 13:921–928.

Monnier

, Mas

, Ginet

, et al.: Activation of oxidative stress by acute glucose fluctuations compared with sustained chronic hyperglycemia in patients with type 2 diabetes. JAMA, 2006; 295:1681–1687.

Ceriello

, Esposito

, Piconi

, et al.: Oscillating glucose is more deleterious to endothelial function and oxidative stress than mean glucose in normal and type 2 diabetic patients. Diabetes, 2008; 57:1349–1354.

Krinsley

: Glycemic variability: a strong independent predictor of mortality in critically ill patients. Crit Care Med, 2008; 36:3008–3013.

10.

Siegelaar

, Barwari

, Kulik

, et al.: No relevant relationship between glucose variability and oxidative stress in well-regulated type 2 diabetes patients. J Diabetes Sci Technol, 2011; 5:86–92.

11.

Kilpatrick

, Rigby

, Atkin

: The effect of glucose variability on the risk of microvascular complications in type 1 diabetes. Diabetes Care, 2006; 29:1486–1490.

12.

Smith-Palmer

, Brandle

, Trevisan

, et al.: Assessment of the association between glycemic variability and diabetes-related complications in type 1 and type 2 diabetes. Diabetes Res Clin Pract, 2014; 105:273–284.

13.

Group

JCS

: JDRF randomized clinical trial to assess the efficacy of real-time continuous glucose monitoring in the management of type 1 diabetes: research design and methods. Diabetes Technol Ther, 2008; 10:310–321.

14.

Tamborlane

, Beck

, Bode

, et al.: Continuous glucose monitoring and intensive treatment of type 1 diabetes. N Engl J Med, 2008; 359:1464–1476.

15.

Beck

, Hirsch

, Laffel

, et al.: The effect of continuous glucose monitoring in well-controlled type 1 diabetes. Diabetes Care, 2009; 32:1378–1383.

16.

EasyGV software. www.phc.ox.ac.uk/research/diabetes/software/easygv/ (accessed August 22, 2016 ).

17.

Service

: Glucose variability. Diabetes, 2013; 62:1398–1404.

18.

Hill

, Thompson

, Bruce

, et al.: Glycaemic risk assessment in children and young people with Type 1 diabetes mellitus. Diabet Med, 2009; 26:740–743.

19.

Service

, Molnar

, Rosevear

, et al.: Mean amplitude of glycemic excursions, a measure of diabetic instability. Diabetes, 1970; 19:644–655.

20.

Kovatchev

, Otto

, Cox

, et al.: Evaluation of a new measure of blood glucose variability in diabetes. Diabetes Care, 2006; 29:2433–2438.

21.

Ryan

, Shandro

, Green

, et al.: Assessment of the severity of hypoglycemia and glycemic lability in type 1 diabetic subjects undergoing islet transplantation. Diabetes, 2004; 53:955–962.

22.

Molnar

, Taylor

, Ho

: Day-to-day variation of continuously monitored glycaemia: a further measure of diabetic instability. Diabetologia, 1972; 8:342–348.

23.

Hermanides

, Vriesendorp

, Bosman

, et al.: Glucose variability is associated with intensive care unit mortality. Crit Care Med, 2010; 38:838–842.

24.

Rodbard

: The challenges of measuring glycemic variability. J Diabetes Sci Technol, 2012; 6:712–715.

25.

Danne

, de Valk

, Kracht

, et al.: Reducing glycaemic variability in type 1 diabetes self-management with a continuous glucose monitoring system based on wired enzyme technology. Diabetologia, 2009; 52:1496–1503.

26.

Rodbard

, Bailey

, Jovanovic

, et al.: Improved quality of glycemic control and reduced glycemic variability with use of continuous glucose monitoring. Diabetes Technol Ther, 2009; 11:717–723.

27.

Tumminia

, Crimi

, Sciacca

, et al.: Efficacy of real-time continuous glucose monitoring on glycaemic control and glucose variability in type 1 diabetic patients treated with either insulin pumps or multiple insulin injection therapy: a randomized controlled crossover trial. Diabetes Metab Res Rev, 2015; 31:61–68.

28.

Cameron

, Donath

, Baghurst

: Measuring glycaemic variation. Curr Diab Rev, 2010; 6:17–26.

29.

Rodbard

: New and improved methods to characterize glycemic variability using continuous glucose monitoring. Diabetes Technol Ther, 2009; 11:551–565.

30.

Kovatchev

, Cox

, Gonder-Frederick

, Clarke

: Symmetrization of the blood glucose measurement scale and its applications. Diabetes Care, 1997; 20:1655–1658.

31.

Borg

, Kuenen

, Carstensen

, et al.: Associations between features of glucose exposure and A1C: the A1C-Derived Average Glucose (ADAG) study. Diabetes, 2010; 59:1585–1590.

32.

Kohnert

, Heinke

, Fritzsche

, et al.: Evaluation of the mean absolute glucose change as a measure of glycemic variability using continuous glucose monitoring data. Diabetes Technol Ther, 2013; 15:448–454.

33.

Rodbard

: Glycemic variability: measurement and utility in clinical medicine and research—one viewpoint. Diabetes Technol Ther, 2011; 13:1077–1080.

34.

Rodbard

: Hypo- and hyperglycemia in relation to the mean, standard deviation, coefficient of variation, and nature of the glucose distribution. Diabetes Technol Ther, 2012; 14:868–876.

35.

Zaccardi

, Stefano

, Busetto

, et al.: Group of signs: a new method to evaluate glycemic variability. J Diabetes Sci Technol, 2008; 2:1061–1065.

36.

Baghurst

: Calculating the mean amplitude of glycemic excursion from continuous glucose monitoring data: an automated algorithm. Diabetes Technol Ther, 2011; 13:296–302.

37.

Fritzsche

, Kohnert

, Heinke

, et al.: The use of a computer program to calculate the mean amplitude of glycemic excursions. Diabetes Technol Ther, 2011; 13:319–325.

38.

Czerwoniuk

, Fendler

, Walenciak

, Mlynarski

: GlyCulator: a glycemic variability calculation tool for continuous glucose monitoring data. J Diabetes Sci Technol, 2011; 5:447–451.

39.

Sechterberger

, Luijf

, Devries

: Poor agreement of computerized calculators for mean amplitude of glycemic excursions. Diabetes Technol Ther, 2014; 16:72–75.

40.

El-Laboudi

, Oliver

: Agreement between glycaemic variability calculators: another limitation in measuring glycaemic variability. Diabetes Technol Ther, 2016; 18(S1):A-90.

41.

Wang

, Zhao

, Su

, et al.: Glycemic variability in normal glucose tolerance women with the previous gestational diabetes mellitus. Diabetol Metab Syndr, 2015; 7:82.

42.

Hanefeld

, Sulk

, Helbig

, et al.: Differences in glycemic variability between normoglycemic and prediabetic subjects. J Diabetes Sci Technol, 2014; 8:286–290.

43.

Zhou

, Li

, Ran

, et al.: Establishment of normal reference ranges for glycemic variability in Chinese subjects using continuous glucose monitoring. Med Sci Moni, 2011; 17:CR9–13.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB