Increased Glycemic Variability at the Onset and During Progression of Type 2 Diabetes—Commentary

Introduction

C ontinuous glucose monitoring (CGM) has been studied extensively in type 1 diabetes mellitus (T1DM) but has been less well studied in type 2 diabetes mellitus (T2DM), and only relatively recently has it been applied to evaluate patients with obesity, metabolic syndrome, and the natural history of prediabetes. This issue of Diabetes Technology and Therapeutics includes two articles ^1,2 that characterize glycemic variability (GV) in cross-sectional studies of patients with prediabetes ¹ and of patients with T2DM receiving various forms of therapy in comparison with T1DM. ² A third article examines the relationship of GV observed using CGM to serum levels of fructosamine and 1,5-anhydroglucitol (1,5-AG) in subjects with very well-controlled diabetes. ³ We shall discuss some of the principal findings, briefly consider pathophysiological mechanisms that might be involved, and discuss some of the methodological issues encountered in the measurement of GV using CGM.

Madhu et al. ¹ : Progressive Increase in GV During the Clinical Evolution of Prediabetes

Madhu et al. ¹ reported that there is a progressive increase in several measures of GV as one moves from subjects with normal glucose tolerance (NGT), to impaired fasting glucose (IFG), to impaired glucose tolerance (IGT), to newly diagnosed diabetes mellitus (NDDM). This progressive increase in GV was observed for SD, interquartile range (IQR), percentage coefficient of variation, minimum and maximum glucose levels, and range of glucose observed during CGM (cf. Table 2 in Madhu et al. ¹ ). These authors also examined the percentage of patients having excursions above prespecified thresholds (140 or 200 mg/dL), the number of excursions, the areas under the curve (AUCs) above those thresholds, and the duration of such excursions. There was a dramatic difference in these measures of GV in subjects with NDDM compared with IFG and IGT. The number of subjects in each group was small (n=20), making it difficult to achieve statistical significance (see Table 3 of Madhu et al. ¹ ). Other investigators have reported significant prevalence of dysglycemia in children with risk factors such as obesity ^4,5 or cystic fibrosis ⁶ but did not characterize GV using the array of criteria used by Madhu et al. ¹ The A1C-Derived Average Glucose (ADAG) study ⁷ found a significant frequency of glucose values above the threshold for IGT in adult subjects with low fasting plasma glucose, suggesting increased GV. It would be desirable to examine GV in a longitudinal study, perhaps using some elements of the experimental designs used by the Diabetes Prevention Program (DPP) ⁸ or the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness (GRADE) study. ^9,10

These findings ¹ might be explained in part by the progressive loss of the first phase of insulin secretion and progressive impairment of the second phase of insulin secretion. ¹¹ It may also reflect differences in dietary habits in parameters such as the number and frequency of meals, carbohydrate load, glycemic index of ingested carbohydrate, meal composition (protein, fat, fiber), and history of physical activity in these subjects.

Madhu et al. ¹ implicitly raise the following question: “Which measure of GV will be most reliable, sensitive, and specific for detection of early changes in subjects on the spectrum from NGT to IFG to IGT to NDDM?” Although only a couple of the parameters studied were statistically significant among NGT, IFG, and IGT, if we examine the “t-scores” for each of the parameters we obtain a suggestion that SD, AUC₁₄₀, AUC₂₀₀, and duration of excursions above 140 mg/dL are likely to be the most sensitive in terms of their ability to discriminate among NGT, IFG, and IGT. In contrast, maximum glucose, minimum glucose, range of glucose, SD, and AUC₂₀₀ appear to be the most sensitive criteria when comparing NDDM versus the other three categories (NGT, IFG, and IGT).

The IQR was consistently less sensitive than the SD as a criterion to discriminate among the four categories because the between-subject variability in the IQR was substantially (1.5–3.0-fold) larger than the corresponding variability in SD (cf. Table 2 of Madhu et al. ¹ ).

Kohnert et al. ² : Progressive Increase of GV in T2DM

Using a cross-sectional study, Kohnert et al. ² reported a progressive increase in GV as one moves between groups of patients with T2DM receiving different forms of therapy—from diet alone, to oral monotherapy, to combination oral therapy, to oral therapy combined with insulin, to insulin therapy alone. The nature of the therapy is likely to be a reflection of the duration of diabetes. ¹² One can infer that there is a progressive increase in GV with duration of disease, in large part because of the progressive loss of β-cell reserve. Both Madhu et al. ¹ and Kohnert et al. ² suggested a progression of GV as one moves from NGT to NDDM and through various stages of T2DM requiring more intensive therapy. Kohnert et al. ² also reported results for patients with T1DM, confirming and quantifying the well-known increase in GV in these patients compared with those with T2DM.

Kohnert et al. ² examined several measures of GV, and, like Madhu et al., ¹ also implicitly raised the question, “Which parameters are best for identification of the progression of GV in different groups of patients with T2DM?” Kohnert et al. ² examined the correlations among the various measures of GV and found extremely high correlations among SD, mean amplitude of glycemic excursions (MAGE), and several other measures of GV. They presented results graphically by treatment category using mean absolute change in glucose level per total time of observation (MAG) and MAGE (see Figure 3 of Kohnert et al. ² ). Based on the very high degree of correlation of MAGE and SD (r=+0.931), one would expect very similar patterns if SD had been displayed for each treatment category.

Results using MAG ¹³ are highly dependent on the number of glucose measurements per 24-h period, which ranged from seven, to 24 (hourly measurements), to 288 (measurements every 5 min). There was a systematic bias in results for MAG as the sampling frequency fell below 24 per day. MAG₆₀ was highly correlated with SD, but the authors concluded that there was little or no advantage to MAG₅ relative to the use of the SD. (Rather than using MAG₅, MAG₆₀, and seven-point MAG, a more consistent nomenclature might be MAG₂₈₈, MAG₂₄, and MAG_7pt, each reflecting the number of glucose measurements per day used in the calculation.) They argued that one should use SD or MAGE in preference to MAG. Because the “distance traveled” criterion proposed by Marling et al. ¹⁴ is mathematically identical to the MAG when glucose measurements are equally spaced, their conclusions also imply that there would be little if any benefit for the use of “distance traveled” as an indicator of GV, relative to SD or MAGE.

Several interesting relationships among the various measures of GV are revealed in the correlation matrix provided by Kohnert et al. ² (see Table 1 of Kohnert et al. ² ). These correlations are similar to the ones presented previously by Fritzsche et al. ¹⁵ and by Rodbard et al. ¹⁶

The SD has the highest correlation with continuous overall net glycemic action (CONGA₆ and CONGA₃), IQR, and MAGE. The SD and IQR have a correlation of r=+0.950, but the IQR has lower correlations with all of the other parameters than does SD. The three measures of CONGA _n examined (CONGA₁, CONGA₃, and CONGA₆) are highly correlated among themselves. MAGE has the highest correlation with SD (r=0.931), followed by CONGA₆ and CONGA₃. MAG_7pt shows only a weak correlation with MAG₆₀ or MAG₅. CONGA₁ has a remarkably high correlation with MAG₆₀: both of these indices compare only those pairs of glucose values that are exactly 60 min apart.

Chon et al. ³ : Assessing Possible Relationships of GV to Fructosamine and 1,5-AG

Chon et al. ³ examined several indices of GV in a small group of patients with T2DM who were extremely well controlled (glycosylated hemoglobin level <7%). They were hoping to find a correlation of these indices with fructosamine, a biochemical marker of glycemic control, and with 1-5 AG, which is presumed to be a biochemical marker of GV. They found only very low to modest degrees of correlation. This was primarily due to a limited sample size and the narrow range of glycosylated hemoglobin in the selected patient population. Higher correlations might have been achieved if the patient population had been larger and expanded to include a much wider range of glycosylated hemoglobin.

The article by Chon et al. ³ is informative from a methodological standpoint in several respects. They used a modification of the original definition of MAGE, such that they manually (graphically) measured the average amplitude of both the upstrokes and downstrokes. Averaging of upstrokes and downstrokes, recently proposed by Baghurst, ¹⁷ provides better utilization of the CGM data than simply using only upstrokes or downstrokes, whichever occurs first, as proposed by Service et al. ¹⁸ in 1970 in their original definition of MAGE. Baghurst ¹⁷ developed a computer algorithm and program to automatically calculate MAGE ₊ for upstrokes, MAGE_ for the downstrokes, and MAGE_avge for both types of excursions with magnitude >1 SD. Chon et al. ³ calculated MAGE according to a relatively new and, to date, rarely used method, which they designated as MAGE_gos, where “gos” refers to “group of signs,” as developed by Zaccardi et al. ¹⁹ They found high correlations of MAGE_gos and MAGE_c, where MAGE_c represents the “classical” method as proposed by Service et al., ¹⁸ which was interpreted as a validation of the MAGE_gos method. There were also high correlations of both MAGE_c and MAGE_gos with the SD, with stronger correlation for MAGE_c than for MAGE_gos. This suggests that MAGE_c is likely to remain preferable to MAGE_gos. MAGE_gos first smooths the glucose versus time curve using a five-point running average. ¹⁹ This smoothing reduces the amount of random fluctuations [“noise”] and simplifies identification of local peaks and valleys and of excursions with magnitude >1 SD. Results depend on the somewhat arbitrary method and extent of curve smoothing. Chon et al. ³ also used a variation, MAGE_{abs gos}, ¹⁹ which is closely related mathematically both to MAG ¹³ and to “distance traveled.” ¹⁴

Discussion and Conclusion

The three studies ^{1 –3} reported in this issue of Diabetes Technology and Therapeutics examining GV in prediabetes and T2DM suggest that GV increases progressively as one moves from NGT to IFG, to IGT, to newly diagnosed T2DM ¹ and, furthermore, among patients with T2DM as one moves from diet to oral monotherapy, to combination oral therapy, to oral therapy combined with insulin, to therapies involving only insulin. ² GV in T2DM patients receiving insulin therapy is substantially smaller than in patients with T1DM. ² These findings are consistent with the interpretation that GV is largely a reflection of the extent of residual β-cell function. ¹¹ Methodologically, these three studies ^{1 –3} support the notion that SD of glucose using CGM remains the reference method for evaluating GV. The percentage coefficient of variation provides a way to express SD as a percentage of the mean glucose. Measures such as maximum, minimum, and range of glucose during a specified duration of CGM, AUC using 140 mg/dL as a threshold, and duration of excursions above the 140 mg/dL threshold are promising in terms of their ability to discriminate among NGT, IFG, IGT, and NDDM. ¹ Other measures of GV are less familiar among physicians treating patients with diabetes and prediabetes. Just as the use of CGM has led clinicians to appreciate the fact that hypoglycemia is vastly more prevalent than that revealed by intermittent or occasional blood glucose measurements with self-monitoring of blood glucose, so, too, CGM reveals that hyperglycemia is also considerably more prevalent than previously appreciated, especially during the early stages of prediabetes. ^{1 –7} This increased prevalence does not appear to be due to problems with accuracy and precision of the CGM. While benefiting from the clinical findings from each of these studies, ^{1 –3} we also seek to learn as much as possible about the methodologies for characterizing GV.