Relationship of continuous glucose monitoring-related metrics with HbA1c and residual β-cell function in Japanese patients with type 1 diabetes

The targets for continuous glucose monitoring (CGM)-derived metrics were recently set; however, studies on CGM data over a long period with stable glycemic control are limited. We analyzed 194,279 CGM values obtained from 19 adult Japanese patients with type 1 diabetes. CGM data obtained during stable glycemic control over four months were analyzed. CGM-related metrics of different durations “within 120, 90, 60, 30, and 7 days” were calculated from baseline. Time in range (TIR; glucose 70–180 mg/dL), time above range (TAR; glucose ≥ 181 mg/dL), and average glucose levels, but not time below range (TBR; glucose ≤ 69 mg/dL), strongly correlated with glycated hemoglobin (HbA1c) values (P < 0.0001). TBR correlated with glucose coefficient of variation (CV) (P < 0.01). Fasting serum C-peptide levels negatively correlated with glucose CV (P < 0.01). HbA1c of approximately 7% corresponded to TIR of 74% and TAR of 20%. The shorter the CGM period, the weaker was the relationship between HbA1c and CGM-related metrics. TIR, TAR, and average glucose levels accurately reflected HbA1c values in Japanese patients with type 1 diabetes with stable glycemic control. Glucose CV and TBR complemented the limitation of HbA1c to detect glucose variability and hypoglycemia. Stable glycemic control with minimal hypoglycemia depended on residual β-cell function.


Results
The results of univariate regression analysis between HbA1c (x-axis) and TIR (y-axis) measured within 120 days from baseline are shown in Fig. 1A. A strong inverse relationship between HbA1c and TIR was observed (P < 0.0001), suggesting that HbA1c data were sufficient to explain TIR. HbA1c of 7% corresponded to a TIR of approximately 74% (Table 1), and the coordinate (x = 7%, y = 70%) was included within the 95% confidence band of the best-fit line (Fig. 1A). An increase in TIR of 18% will cause a decrease in HbA1c of 1.0% (Supplementary Fig. 1A). In Fig. 1B, a strong positive relationship was observed between HbA1c and TAR (P < 0.0001), suggesting that HbA1c data were sufficient to explain TAR. HbA1c of 7% corresponded to a TAR of approximately 20% (Table 1), and the coordinate (x = 7%, y = 25%) was not included within the 95% confidence band of the best-fit line (Fig. 1B). A decrease of 19% in TAR will cause a decrease in HbA1c of 1.0% ( Supplementary  Fig. 1B). In Fig. 1C, unlike TIR and TAR, no significant relationship was observed between HbA1c and TBR (P > 0.05) ( Table 1).
In Fig. 2A, we observe a strong positive relationship between HbA1c and average glucose (P < 0.0001), suggesting that HbA1c had sufficient data to explain average glucose. HbA1c of 7% corresponded to an average glucose of approximately 138 mg/dL (7.7 mM) ( Table 1). A decrease of 39 mg/dL (2.6 mM) in average glucose will cause a decrease in HbA1c of 1.0% ( Supplementary Fig. 1C), indicating that HbA1c of 6% of the study population corresponded to an average glucose of approximately 100 mg/dL (5.6 mM). In the remaining CGMrelated metrics, glucose SD positively correlated with HbA1c (P < 0.0001) (Fig. 2B), whereas glucose CV and HbA1c were not related (P > 0.05) (Fig. 2C). The mean of glucose CV was 0.35 (range 0.27-0.47). Glucose CV did not correlate with TIR and TAR (P > 0.05 in both) (Fig. 3A,B) as in the case of HbA1c (Fig. 2C). However, glucose CV significantly correlated with TBR, indicating that the increase in glucose CV was associated with increase in hypoglycemia (Fig. 3C). Since 2 of the 19 patients (No. 3 and 10 in Tables 2 and 3) had fewer readings (< 70%) for "within 120 days, " we performed an analysis excluding them, but the results were similar to the analysis with all the patients.   [15][16][17] , the relationship between serum CPR and CGM-related metrics was investigated. Wide variations in TIR and TAR were observed in decreased fasting serum CPR (Fig. 4A,B). Markedly low TIR (Fig. 4A), high TAR (Fig. 4B) and high TBR (Fig. 4C) were observed in patients with fasting serum CPR < 0.1 ng/mL.    Fig. 5A, a significant inverse relationship was observed between fasting serum CPR and glucose CV (P < 0.005). An increase in fasting serum CPR was associated with a decrease in glucose CV (P < 0.01, Jonckheere-Terpstra trend test) (Fig. 5B).
In addition to the analysis of CGM data "within 120 days" from baseline described above, we further studied multiple time period measurements from baseline: within 90, 60, 30, and 7 days ( Table 1). The results obtained were essentially the same as in the case of "within 120 days. " TIRs measured within 90 and 60 days from baseline showed strong relationship with HbA1c (coefficient of determination (R 2 ) = 0.8771 and R 2 = 0.8851, respectively)  www.nature.com/scientificreports/ as in the case of within 120 days (R 2 = 0.8881); however, TIRs measured within 30 and 7 days showed a slightly weaker correlation with HbA1c (R 2 = 0.8489 and R 2 = 0.7110, respectively). TARs and average glucose showed same tendency as TIRs, namely, a strong relationship with HbA1c for longer periods and slightly weaker for shorter periods. As for TBRs, no significant relationship with HbA1c for any period was observed. Glucose CVs measured within 90, 60, and 30 days from baseline showed strong relationship with fasting serum CPR (R 2 = 0.4092, R 2 = 0.3333, and R 2 = 0.2250, respectively) as in the case of "within 120 days" (R 2 = 0.4203); however, glucose CVs measured within 7 days showed no correlation with fasting serum CPR (R 2 = 0.1405).

Discussion
The present study was performed to clarify the advantages and disadvantages of CGM-related metrics relative to well-established laboratory-measured parameters, HbA1c and serum CPR, in type 1 diabetes. TIR, TAR, and average glucose levels, but not TBR, strongly correlated with HbA1c. TBR correlated with glucose CV. Fasting serum CPR was negatively correlated with glucose CV. HbA1c of approximately 7% corresponded to TIR of 74% and TAR of 20%. The shorter the CGM period, the weaker the relationship between HbA1c and CGM-related metrics. To date, numerous studies have been published on the relationship between CGM-related metrics and HbA1c 11,13 . However, most studies have been based on data extracted from published trials or articles for other purposes and does not exist as studies specified in measurement terms, to investigate for CGM data for a sufficiently long period of time 11,13,18 . To exactly investigate the relationship between HbA1c and CGMrelated metrics, the measurement periods were fixed as 120, 90, 60, 30, and 7 days from the baseline. To avoid discrepancy due to CGM-related metrics reflecting short-term glycemic status and HbA1c reflecting long-term glycemic status, only those patients with stable glycemic control and minimal changes in HbA1c four months prior to the baseline were considered for analysis, thus resulting in less than 15% variation in HbA1c. Despite a relatively small number of patients, a very strong relationship between CGM-related metrics, TIR, TAR, and particularly average glucose levels, and HbA1c was detected, indicating that CGM-related metrics TIR and TAR could adequately reflect laboratory-measured HbA1c provided that HbA1c levels have been stable within the past several months. ATTD has indicated three key CGM-related metrics: TIR, TAR, and TBR 2 . The ATTD report has described that CGM-based glycemic targets must be personalized to meet the needs of each person with diabetes. In addition, the report had proposed a glycemic range (a target range of 70-180 mg/dL) and targets of TIR, TAR, and TBR for adults with type 1 or type 2 diabetes (TIR > 70%, TAR < 25%, TBR < 4%) and older/high-risk individuals (TIR > 50%, TAR < 50%, TBR < 1%). Two studies have reported that TIR of 70% and 50% corresponded to an HbA1c of approximately 7% and 8% 11 , and 6.7% and 8.3% 13 , respectively. However, the present study on adult Japanese patients with type 1 diabetes showed that TIRs of 70% and 50% corresponded to HbA1c of approximately 7.3% and 8.4%, respectively (best-fit line: y = − 0.05412x + 11.06; when x = TIR, y = HbA1c) ( Supplementary  Fig. 1A). To achieve HbA1c of 7% and 8%, TIR should be more than 74% and 57%, respectively. TAR matched more with HbA1c compared to TIR or TBR, in our population. TAR of 25% and 50% corresponded with HbA1c of approximately 7.3% and 8.6%, respectively (best-fit line: y = 0.05211x + 5.969; when x = TAR, y = HbA1c) (Supplementary Fig. 1B). To achieve HbA1c of 7% and 8%, TAR should be less than 20% and 39%, respectively.
TBR showed only marginal correlation with HbA1c. This may be due to a limited TBR compared to TIR and TAR. However, a tendency of reducing HbA1c with increasing TBR suggests the limitation of HbA1c and the advantage of the differentiation to increase TIR and simultaneously decrease or at least have unchanged TBR. In addition, as shown in Fig. 3C, a significant relationship between glucose CV and TBR was observed; therefore, higher the glucose CV, higher was TBR (risk of hypoglycemia). These results demonstrated the need to maintain glucose CV below a certain level (e.g., < 0.36 19 ) to avoid hypoglycemia in patients at an increased risk. However, it is often difficult to regulate glucose CV in type 1 diabetes, particularly, in the subset of patients with unstable (brittle) diabetes. Even in patients with stable HbA1c in our study, the mean glucose CV was 0.35. One Fasting serum CPR levels are divided into three groups: < 0.1 ng/mL (low), > 0.1 ng/mL and < 0.6 ng/mL (moderate), and > 0.6 ng/mL (high). Glucose CV was measured within 120 days from baseline. Glucose CV among different levels of CPR groups was tested using the Jonckheere-Terpstra trend test. A trend toward lower glucose CV with higher serum CPR (p < 0.01) is noted. CPR C-peptide, CV coefficient of variation, R 2 coefficient of determination. www.nature.com/scientificreports/ major factor contributing to glucose CV is residual beta-cell function [15][16][17] . Wide variations in TIR and TAR in patients with low CPR (Fig. 4A,B) together with the inverse relationship between glucose CV and serum CPR (Fig. 5A,B) indicate the importance of residual β-cell function on glycemic control stabilization. Our data regarding CGM-derived glucose CV are consistent with those of previous reports on the negative correlation between serum CPR and glucose variability assessed in pre-CGM era [15][16][17] . It emphasizes the importance of residual CPR, which is reported to be associated with the progression of diabetes complications 20 , on glycemic stability in type 1 diabetes. Most Japanese patients with type 1 diabetes were reported to completely lose endogenous insulin during the disease duration 21 ; inversely most patients with long-duration type 1 diabetes in Western countries preserved endogenous insulin levels 22,23 , and treatment to prevent CPR decline has been reported to inhibit the progression of diabetes complications in patients with residual CPR 24 . Therefore, CGM-derived TBR and glucose CV together with CPR are essential metrics supplementing HbA1c in patients with type 1 diabetes, particularly in the Japanese population. Average glucose, rather than other CGM-related metrics, exactly matched with HbA1c, particularly, in the CGM data collected from the last 60 days or more. CGM data collected "within 7 days" showed a significant but weaker relationship with HbA1c and average glucose than those from longer periods (Table 1). These results confirmed that HbA1c is an index for long-term monitoring of blood glucose and reflects mean glucose levels during the last 2 months, as reported previously 1,2 . This study was designed to use CGM data only in patients with stable glycemic control with minimal changes in HbA1c during the 4 months before the baseline. Even then, a duration dependency was observed regarding the correlation between CGM-related metrics and HbA1c, shorter duration of CGM-derived data, weaker was the correlation with HbA1c. The correlation would be even weaker in case CGM-derived data are extracted from clinical trials or articles in which glycemic control changes due to interventions or treatments resulting in CGM-derived data reflecting short-term glycemic control changes much faster than HbA1c. In fact, only a moderate correlation was reported between CGM-related metrics and HbA1c with a correlation coefficient of 0.67-0.73 for TIR in 545 patients with type 1 diabetes from 4 randomized trials 11 . From this point of view, previous studies on the relationship between HbA1c and CGM-related metrics and estimation of HbA1c from CGM-related metrics using short-term CGM data should be carefully interpreted.
Petersson et al. reported the relationship between HbA1c and CGM-related metrics, focusing on time in target range (TIT) defined as 70-140 mg/dL 25 ; this is the first report to identify a nonlinear relationship between TIT and HbA1c in pediatric type 1 diabetes. However, in the same paper, there is a description of a linear analysis of TIT 30, 60, and 90 days and HbA1c. All of them showed significant correlations, but the correlation coefficients (R 2 : 0.59-0.63) were lower than R 2 for TIR in the present study. These differences may be due to the difference in race (Japanese vs. Swedish), subjects (adults vs. children and adolescents), selection criteria (stable HbA1c population vs. general population), and CGM-metrics (TIR vs. TIT). The relationship between HbA1c and CGM-metrics relative to CGM duration, which is our main objective in the present study, is unknown because the relationship in shorter (7 days) and longer (120 days) CGM duration is not available in their study.
The main strength of this study is that it was specifically designed to test the correlation between CGMrelated metrics and HbA1c; it evaluated the accuracy of CGM-related metrics for different CGM measurement periods. Moreover, this is the first report on adult Japanese patients with type 1 diabetes, a population prone to insulin depletion.
This study had limitations. First, only a limited number of patients were eligible for investigation because we considered a study design for data collection to increase the accuracy of the results; herein, we selected for each patient the period with stable glycemic control with small changes in HbA1c. In addition, we excluded patients with clinical conditions unrelated to glycemia but affecting HbA1c levels, such as unstable blood hemoglobin values, gestational period, surgery, and diagnosis of malignant tumors. Due to the exclusion of subjects with these conditions, the results of the present study may not apply to patients with type 1 diabetes in general. Second, the relationship between HbA1c and average glucose was studied only with respect to CGM-derived glucose value, but not with laboratory-measured plasma glucose, as our study was performed under daily life conditions. Considering the recent developments in CGM technology with improvement in sensor accuracy, the data obtained with CGM for real-life glucose in normal daily use provide information difficult to be obtained with the laboratory-measured glucose, under restricted settings such as hospitals. Third, as our patients use FreeStyle Libre, which has dual function, CGM and FGM, there may be biases based on patients' habits. For example, when unexpected high or low glucose values were detected using FGM, patients willingly compensated their blood glucose, leading to better than the expected TIR levels. Fourth, the analysis of the relationship between glucose CV and fasting serum CPR was performed using only univariate regression analysis because of the limited number of the subjects in this study. For this reason, the observed relationship can be indirect and secondary to other unknown factors. Although our data on CGM-derived glucose CV are consistent with those of previous reports of glucose variability estimated using different methods in pre-CGM era [15][16][17] , further studies with larger number of participants are required to clarify the observed relationship with the consideration for other confounding factors. Fifth, although the relationship between glucose CV and TBR was detected in this study, the relationship may be overrated or underrated. This is because residual CPR may potentially affect both glucose CV and TBR. Additionally, Japanese patients with type 1 diabetes are known to be prone to CPR depletion.
In conclusion, this study provided evidence regarding the reliability of CGM-related metrics in adult Japanese patients with type 1 diabetes in a usual daily life setting. TIR and TAR strongly correlated with HbA1c and complemented or even replaced HbA1c for short-term monitoring of blood glucose. While TIR, TAR, and HbA1c reflected mean blood glucose, TBR and glucose CV complemented the mean glucose-related metrics by reflecting hypoglycemia. The negative correlation between glucose CV and fasting CPR suggests the importance of residual β-cell function for stable glycemic control without hypoglycemia, in type 1 diabetes. Study design. Patients' data were collected in their daily life settings. The patients at their visit to our hospital, brought with them their FreeStyle Libre device (Abbott Japan Diabetes Care Inc., Tokyo, Japan), which is a flash glucose monitoring (FGM) system that has dual function, CGM and FGM. The FreeStyle Libre data were extracted and stored in a computer solely for this purpose. CGM measures every 15 min the interstitial fluid glucose and converts it to venous blood glucose; however, in FGM, glucose data are not constantly shown and are available only on demand. In this study, only CGM data were used for analysis.
Patients who met all of the following five criteria were included in the study: (1) stable HbA1c, which was defined as minimum difference in HbA1c levels for the past 4 months in each patient, and stable blood hemoglobin levels in the past 4 months, (2) no gestational period, (3) no operations within the last 6 months and no planned operations, (4) no diagnosis for malignant tumors, (5) no clinical conditions unrelated to glycemia but affecting HbA1c levels, such as anemia, renal failure and liver cirrhosis. HbA1c at baseline and at 1 to 4 months before baseline is shown in Table 3. In our institution, about 200 patients with type 1 diabetes are provided treatment through regular outpatient clinic visits, and 26 patients used Freestyle Libre during the period of this study. Of these, 7 patients were excluded because of the following reasons: CGM data of less than 4 months were available, blood data were not available, and/or inclusion criteria were not met. Eventually, 19 patients were included in the analysis.
The CGM-related metrics were calculated, namely TIR (glucose 70-180 mg/dL), TAR (glucose ≥ 181 mg/dL), TBR (glucose ≤ 69 mg/dL), average glucose, glucose standard deviation (SD), and glucose CV for each patient using the CGM data within 120 days from baseline. Overall, 194,279 CGM values (average 10,225 per patient) were analyzed. In addition, we calculated CGM-related metrics based on CGM data closer to the baseline, "within 90 days, " "within 60 days, " "within 30 days, " and "within 7 days" from baseline. The number of data analyzed and the percentage mounting time of CGM in the corresponding period are provided in Table 3. On an average, the percentage mounting time of CGM was over 84%.
In addition to the relationship between CGM-related metrics and HbA1c, the relationship between CGMrelated metrics and serum CPR was analyzed to study the effect of residual β-cell function on glycemic control, particularly fluctuation (glucose CV) and hypoglycemia (TBR).