The multiple glycometric data provided by CGM systems have allowed the development of new indices to measure the glycemic control of patients with T1D. Although HbA1c is the parameter with the greatest evidence for predicting chronic complications to date, it is insufficient to optimally assess the degree of glycemic control of a given individual, and there are already studies that relate the TIR to the risk of long-term complications [12, 13]. Nevertheless, the TIR alone shows certain limitations; it does not take into account whether the time out of range constitutes time in hypo- or hyperglycemia, it is not sensitive to time in hypoglycemia, and it does not give greater weight to the most extreme deviations from the TIR [2,3,4]. The large amount of data that the CGM provides demands a significant amount of time and effort from the professionals involved in the management of these patients [6, 14], requiring new metrics that synthesize all this available information.

The appearance of the GRI attempts to solve some of these drawbacks since it is a single parameter whose value ranges from 0 (best degree of glycemic control) to 100 (worst control), is actionable, calculated from a simple formula that gives greater weight to extreme glycemic values, is easily interpretable, and whose changes can be evaluated over time. Moreover, it arises with the support of many international experts in T1D [11]. The GRI uses only four parameters from the AGP report (TBR < 54, TBR 54–70, TAR 180–250, and TAR > 250 mg/dL) and can be easily applied in a wide variety of study designs and settings that use CGM to assess outcomes. Recently, a 14-day period of CGM data has been established as the most appropriate for the calculation of the GRI [15].

To date, published results extensively analyzing the correlation of GRI with other glycometric parameters correspond to data collected from clinical trials in adults with Dexcom G4 and G6 CGM systems (Dexcom Inc, San Diego, CA, USA). Therefore, to our knowledge, this is the first study that investigates the consistency of these relationships in an actual clinical practice setting using isCGM and a pediatric population. It also stratifies the GRI correlation results according to GV, whose influence has already been demonstrated in the relationship between TIR and HbA1c by different authors [9, 10].

Our results demonstrate the correlation between the different glycemic metrics most commonly used in clinical practice and the GRI and its components (Table 2). The statistically significant relationship between practically all the variables analyzed highlights the important interrelationship between the different parameters and emphasizes once again the difficulty of interpreting them independently. Logically, glycemic parameters could be grouped into measurements related to hyperglycemia (HbA1c, mean glucose, SD, GMI, TIR, TAR, and CHyper) and hypoglycemia (CV, TBR, and CHypo). The relationship between HbA1c and TIR was congruent with that found by Vigersky et al. (R = −0.84; R2 = 0.71; p < 0.001) [7] and Diaz-Soto et al. (R = −0.746; R2 = 0.557; p < 0.001) [10]. The GRI correlated significantly (and in most cases, strongly) with all the parameters analyzed, related to both hyper- and hypoglycemia, unlike its predecessors (glycosylated hemoglobin A1c and TIR), which did not correlate significantly or correlated weakly with hypoglycemia parameters. This ability of the GRI to better reflect changes in the area of hypoglycemia, which is derived from its formula (focused on the most extreme values of glycemia rather than on values of centrality), is one of the fundamental differences with respect to HbA1c and TIR and has been described in recent publications [16, 17]. Our correlations are similar to those described by Klonoff et al. [11] in the original GRI article, in which a smaller number of variables were analyzed, agreeing on the low correlation of TIR with parameters related to hypoglycemia such as TBR < 45 mg/dl (R = −0.11), TBR 70–45 mg/dl (R = 0.1), and CV (R = −0.27). Recently, a new study has evaluated the GRI correlation against CHypo and CHyper in T1D patients on an automated insulin delivery system. The results found a significant GRI correlation with TAR but not TBR [18]. These results are not surprising due to the low risk of hypoglycemia in the population evaluated even before the use of the automatic insulin system (TBR around 3.9%). In fact, our findings are in line with these results. Those subgroups of patients at higher risk of hypoglycemia, especially the pediatric group, showed significantly stronger correlations of GRI with CHypo/TBR. Moreover, following this approach, we should consider the lack of usefulness of the GRI in those subgroups of patients with practically nonexistent TBR or TAR.

The stratification of the correlation between the GRI and GMI according to the CV showed two lines with parallel slopes, with those patients with a CV ≥ 36% belonging to the line that runs along the lower part of Fig. 2. According to this, for the same HbA1c value, those patients with greater CV instability showed higher GRI values than those with lower GV. For example, for an estimated HbA1c of 7%, those individuals with a CV < 36% presented a mean GRI of 39.7, while those with a CV ≥ 36% showed a GRI of 48.7 (a GRI 18.5% higher). This parallel relationship between the two lines may be explained by the sensitivity of the CV in assessing an individual’s risk of hypoglycemia [18], the most penalized component in calculating the GRI. This parallel relationship is also different from the previously published relationships of the CV in relation to the TIR and GMI, where the lines crossed each other [9, 10], demonstrating the weighting provided by the effect of the GV all along the GRI calculation.

The relationship between the GRI and TIR in our work is practically analogous to the results published by Klonoff et al. in the original GRI article (R = −0.910; p < 0.001) [11]. Moreover, this correlation was maintained for patients with MDI and CSII. In the present work, we also found a similar correlation between these two parameters in the pediatric and adult groups, which had not been previously studied. When stratifying the data according to the CV, it was observed that both regression lines intersected at a value of GRI = 23 and TIR = 78%. For a TIR value greater than 78%, the higher the CV, the lower the GRI, while when the TIR is lower than 78%, the GRI is higher as the CV increases (Fig. 3). This relationship can be partially explained by the distribution of the CHypo and CHyper components that make up the GRI (Fig. 4) and explains much of the variability in the correlation between the GRI and TIR. When analyzing the relationship between the CHypo and CHyper components of the GRI according to GV, it was observed that those with lower GV presented a greater clustering around the hyperglycemia axis. In contrast, those with a high CV showed a greater component of hypoglycemia and variable of hyperglycemia, supporting the relationship between CV and CHypo [19]. A recent study on an automated insulin delivery system in adults supports our findings because its low global CV showed a GRI clustering around the hyperglycemia axis [18]

Finally, when analyzing the different glycometric parameters according to a CV greater or less than 36% (Table 3), significant differences were observed for all glycometric parameters except HbA1c, mean glucose and TAR 180–250. As for the TIR, those patients with greater variability showed a decrease of 11% with respect to those with CV < 36%; however, for the GRI, the difference was 27 percentiles more in the group with higher variability, which highlights the greater weight of variability (and, ultimately, of hypoglycemia) in this new index. The fact that significant differences were found for the GMI and not for HbA1c (given that the mean and SD values are similar) seems to be due to the study’s sample size. As shown in Table 3, those patients with a CV ≥ 36% presented greater GRI and lower TIR, with a marked component of hypoglycemia and a tendency to greater hyperglycemia than those with low GV.

Limitations of the present study include the relatively small sample size compared to large data studies; however, this is a real-life cohort with stable control and comprehensive knowledge of glycometric and clinical variables with a single CGM system. The non-incorporation of TIR in the GRI calculation can be seen as a potential limitation, being the only CGM parameter related to long-term complications at present [11]; however, the high correlation between TIR and GRI suggests that a similar correlation exists between GRI and the existence of long-term complications as recent studies support [20, 21]. More studies are needed to relate this parameter to future complications, its effect on the quality of life of patients with T1D, and its evaluation in other subpopulations (T2D, hospitalized patients). Finally, some studies have shown higher TBR in FreeStyle Libre isCGM users [22]. This could increase CHypo in our investigation. However, the use of the same isCGM model and version throughout the study in all patients ensures the representativeness of our results. As strengths, it is worth highlighting the results, in line with those previously published on the relationship of the GRI with other glycometrics [11], as well as the influence of the CV on the relationship between the different parameters of glycemic control in adult and pediatric patients in a non-selected population on different treatments [9, 10].

In conclusion, the GRI correlated significantly with all the glycometric parameters analyzed, related to both hypo- and hyperglycemia and especially closely with TIR. GV measured as the CV significantly affected the correlation of GRI with TIR and Glycosylated Hemoglobin A1c and should be considered when metabolic control is assessed.