Glycemic deviation index: a novel method of integrating glycemic numerical value and variability

Diabetic blood glucose disorders are mainly caused by abnormalities in the average glycemic level and variability, and the latter has been shown to independently affect diabetics-related complications [1‐4]. Clinical lab indexes and fingertip blood glucose monitoring are widely used to monitor blood glucose changes over a certain period. Among the clinical glycemic parameters, HbA1c concentration can be used to measure the average blood glucose level of individuals taken over 2 to 3 months, which is the “gold standard” to measure the control of diabetes [5‐7]. However, as a glycosylated product of hemoglobin, the concentration of HbA1c is affected by hemoglobin content, hemoglobin glycosylated rate, and erythrocyte clearance rate [8]. Additionally, it also does not reflect actual blood glucose changes and fluctuations, which limits its ability to measure glycemic variability and hypoglycemia [9].

In recent years, digital diabetic management has been regarded as a promising strategy, and commercial CGM devices that can record up to 10–14 days of persistent glycemic levels have been developed [10‐12]. Along with the development and prevalence of continuous glucose monitoring systems (CGMS), it is predictable that indexes calculated using CGM data will have better future application prospects. CGMS provides detailed blood glucose changes [13], while indexes calculated using CGM data are under development. Some indexes evaluate a single feature of glycemic variabilities, such as standard deviation of blood glucose (SDBG), mean amplitude of glycemic excursions (MAGE), or mean of daily differences (MODD) [14‐16]. In contrast, others, such as the area under the curve (AUC), low blood glucose index (LBGI) and high blood glucose index (HBGI), time in range (TIR), focus on representing a specific kind of glycemic exposure [8, 17, 18]. Although comprehensive glycemic indicators have also been proposed, none of them are as vital as the “gold standard” as to clinical decisions. There has also been some controversy regarding the integration of methods. For instance, M-value arbitrarily selects R as the blood glucose reference value. Simultaneously, the J-index presents the sum of elements heavily weighted towards hyperglycemia and performs poorly for hypoglycemia [4, 19].

Moreover, the choice of glycemic characters and the design of formula segmentation can have a crucial impact on these types of glycemic indicators. We found that indexes, such as the blood glucose risk index (BGRI) [20], index of glycemic control (IGC) [21], and glycemic risk assessment diabetes equation (GRADE) [22], do not independently consider glycemic variability. Additionally, BGRI and IGC do not calculate blood glucose exposure time. There are some glycemic metrics, such as continuous glucose monitoring index (COGI), Q-score, and comprehensive glucose pentagon (CGP), which cover glycemic characters more comprehensively than others [23‐25]. Due to the lack of comparable evidence and common consensus, it is challenging to decide on the best-integrated method to be applied for each type of patient [26]. On the other hand, glycemic parameters, such as personal glycemic state (PGS) [27] and GRADE, are used as the piecewise function, which complicates its integration with other factors.

To develop a better method of comprehensively integrating blood glucose abnormalities, an improved algorithm of glycemic parameters has been proposed in this study. We named it the glycemic deviation index (GDI), which mainly integrates the major glycemic trend and variability. Clinical practice was considered for the design process and derivation of formulas. Methods of blood glucose scale transformation [28] and an automatically adaptive weighted adjustment method were used for GDI derivation. The duration of hypoglycemic exposure was concerned for weighting. Further, we verified the effectiveness of GDI using the simulated 24 h glycemic profiles and the CGM data of T2D participants. This glycemic index may be a potential marker that can be used to identify overall blood glucose conditions in the future.

Figure 2 shows the 4 profiles of CGM over 24 h simulated under 4 typical glycemic conditions. Previous glycemic parameters (MBG, SDBG, eHbA1c, TBR, TIR, TAR) and novel parameters (MGI’, SDGI, GDI) were listed in Table 2. MBG reflects the objective average glycemic level, and SDBG reflects objective glycemic variability. eHbA1c was calculated using a method reported previously to estimate HbA1c and was listed as the clinical glycemic control reference of this study [32]. TBR, TIR, and TAR are intuitive CGM metrics that have been recommended by recent consensuses and have been thoroughly researched. The minimum and maximum glycemic level values were 3 mmol/L and 15 mmol/L, set for illustrative purposes only. Considering that postprandial hyperglycemia is prevalent in type 2 diabetes patients and increases the risk of diabetic complications [38, 39], Fig. 2a showed the fluctuating condition caused by postprandial hyperglycemia. Figure 2b demonstrated a matched high glycemic variability and average conditions with that of Fig. 2a, but with nocturnal hypoglycemia, common in diabetic patients under insulin treatment [40]. In Table 2, although both A and B showed high MBG and SDBG levels, the eHbA1c level of B was lower since the deviation of hypoglycemia was “neutralized” by hyperglycemia. However, the MGI’, SDGI, and GDI values of B were higher than that of A, using our method of overall deviation accumulation. Figure 2c simulated a condition in which the average blood glucose level was the same as in Fig. 2a, and the only difference was the glycemic variability. The results presented in Table 2 showed that C had lower SDBG, SDGI, and GDI values than A, even though the MBG, eHbA1c, TBR, TIR, and TAR values were similar. Figure 2d represented normal daily glycemic change control [41]. It showed a parallel normal glycemic variability as C. Compared with D, MBG and MGI’ were higher in C due to hyperglycemia, which also led to the rise of GDI. The GDI of D was 0.09 and was within the normal range of [0,1].

Given that GDI was designed to reflect the degree of glycemic deviation comprehensively, it should contain information on each glycemic characteristic separately at the same time. To verify its characteristics, one-way ANOVA and Brown–Forsythe test were used to exam the ability of the GDI to distinguish between separate glycemic control parameters. Using the quartiles of GDI (1.2, 2.2, 2.9), we divided the 30 participants into four groups based on their GDI score: normal (GDI ≤ 1), mild (1<GDI ≤ 2), moderate (2<GDI ≤ 3), and severe (3<GDI). Table 3 shows that the lower GDI values represent better glycemic control than the higher values. The values of the clinical glycemic parameters (HbA1c, GA, FBG), average glycemic parameters (MBG, AUC_H-L), and glycemic variability parameters (SDBG, MAGE, MODD) became significantly different as the GDI value increased. We also collected data on the average incidence of diabetic complications (diabetic retinopathy, diabetic nephropathy, diabetic neuropathies, and diabetic heart disease) for all 4 categories shown in Table 3. The results showed that the presence of diabetic complications increased as the GDI value increased.

This study aimed to develop a comprehensive index to assess the degree of deviation of blood glucose from normal levels. Therefore, a function containing the average glycemic level and glycemic variability was constructed and named as the GDI. It was composed of two main functions, MGI and SDGI, which were deduced separately. Ideally, the GDI should increase following the severity of hyperglycemia, hypoglycemia, and glycemic variability. GDI can be applied to quantitate the glycemic control of patients using CGM.

Four simulated glycemic profiles were tested to verify the efficacy of GDI, as shown in Fig. 2. The numerical results suggested that hypoglycemia, hyperglycemia, and abnormal glycemic variability could independently increase GDI values. In Table 2, MBG and SDBG, essential and widely used mathematical parameters, were computed to compare the transformed MGI’ and SDGI values [42, 43]. HbA1c is a clinical glycemic reference control, with eHbA1c as its counterpart in the simulated data. Hyperglycemia with high variability and occurrence of hypoglycemia showed a higher GDI value than the same condition without hypoglycemia, even though the MBG and eHbA1c values were at lower levels (A vs. B), indicating that hypoglycemia is an influential factor in increasing glycemic deviation in our formula. When glycemic variability or hyperglycemia was the only variable that changed in comparison with others (A vs. C, C vs. D), the GDI value remained consistent with that of the abnormal variable, indicating its ability in evaluating factors that deviate. Notably, although glycemic variability and GDI were higher (A vs. C), the TIR, TBR, and TAR remained at the same level. It indicated that the GDI could better reflect variability, while its clinical significance and the impact on diabetic complications needed to be researched further.

Additionally, we divided the thirty T2D patients into four groups based on their GDI severity to apply variance of analysis within clinical glycemic parameters, average glycemic parameters, and glycemic variability parameters. As clinical glycemic parameters, HbA1c, GA, and FBG represented glycemic control levels in separate periods [44‐46]. Statistically significant differences were found among the four GDI groups, even though clinical interventions performed on patients during hospitalization were not considered. Statistical differences were also found among the four GDI groups in terms of previously established indices of the average glycemic level and variabilities, such as MBG, AUC_H-L, SDBG, MAGE MODD. The ability to differentiate between these parameters indicated that GDI could adequately integrate corresponding glycemic information. Moreover, the increase in the average incidence of diabetic complications indicated potential correlations between GDI and diabetes complications.

Various methods have been used to evaluate overall glycemic control. However, none of them have been defined as the clinical gold standard or have been widely used in clinical practice. The TIR of 3.9–10 mmol/L, in addition to the HbA1c level, has recently been regarded as an essential metric in analyzing CGM data [9, 47]. Although TIR is not a comprehensive index, its clinical value indicates the importance of evaluating glycemic exposure time, which we have considered when designing GDI. However, specific comprehensive glycemic indicators [20, 21] have neglected this factor in their designs. In this study, the weighted coefficient “c” of MGI’ in the deduction of the GDI equation reflected glycemic exposure time (1–4). Moreover, the weighted coefficient also reflected the occurence of hypoglycemia, which proved to be effective in Fig. 2 and Table 2.

Additionally, comprehensive glycemic indicators, such as BGRI, IGC, and GRADE, tend not to add glycemic variability as an independent factor. The quantification of hyperglycemia and hypoglycemia indirectly reflected the variability, but there are some situations in which it cannot reflect. An example of this is the glycemic variability among patients with only hyperglycemia. In 1995, Wojcicki proposed a glycemic parameter to improve the M-value and named it as J-index [48]. The J-index was the first to use the idea of integrating MBG and SDBG to represent the mean level and the variability of glycemia. However, we found that without data conversion and appropriate weighting methods, hyperglycemia, hypoglycemia, and glycemic variability could interact with each other in the J-index. Thus, during the deduction of the GDI equation deducing, the standard quadratic risk function was superimposed onto the transformed blood glucose scale (1–2) [8], and value ranges were adjusted for both factors (1–3, 2–2). The adaptive weighting method was applied to ensure that the abnormal factor held the dominant position between MGI’ and SDGI (3–1).

This study has several limitations that should be noted. First, it was conducted at a single-center, which limited the number of patients who used iPro2 CGM. Second, because GDI was designed to evaluate the overall glycemic condition of individuals with CGM, it did not show transient glycemic changes. Third, there is no similar “gold standard” glycemic index, so diagnostic tests and further comparisons are required to identify GDI application limits. Finally, the relationship between the GDI and diabetic complications should be further addressed before it is applied in clinical practice.

The novel glycemic parameter, GDI, is a monitoring index that integrates multiple blood glucose measures. GDI mainly quantifies abnormalities in the average glycemic level and glycemic variability based on CGM data. Therefore, the GDI may be useful for the screening and evaluation of glycemic disorders. Studies using more extensive databases are needed to investigate the application of this novel index further.