Glycemic Variability Bridges Time in Range and Time in Tight Range: A Unified Equation for Both Type 1 and Type 2 Diabetes Based on Large‐Scale Continuous Glucose Monitoring Data

ABSTRACT

Aims

This study aimed to establish a regression model for the relationship between time in range (TIR) and time in tight range (TITR) in individuals with type 1 diabetes (T1D) and type 2 diabetes (T2D) based on real‐world continuous glucose monitoring (CGM) data.

Materials and Methods

A cross‐sectional analysis was conducted on over 200 000 CGM users with diabetes. Participants self‐reported basic demographic and clinical details via in‐app fields. Exponential regression models were constructed to examine the TIR‐TITR association for individuals with T1D and T2D, respectively. After controlling for coefficient of variation (CV), the model was extended to provide more precise glycemic targets for clinical use. Model performance was evaluated using the coefficient of determination (R ²), root mean square error (RMSE), and Akaike information criterion (AIC).

Results

The TIR‐TITR relationship exhibited a nonlinear relationship. Exponential models (TITR_T1D = 8.54436 × exp[0.02414 × TIR]; TITR_T2D = 5.52189 × exp[0.02839 × TIR]) provided the best fit compared to linear and quadratic models. A TIR of 70% corresponded to TITR values of 40.3%–46.3%, whereas achieving TITR of 50% required TIR of 73.2%–77.6%. For TIR below 60%, each 5% TIR increment boosted TITR by less than 5% points; above 60%, gains exceeded 5% points. Additionally, the inclusion of CV in the model was associated with reduced differences between the fitted T1D and T2D curves and improved the model's performance (TITR = 2.18448 × exp[0.03749 × TIR] +0.94018 × CV−14.99420).

Conclusions

This study established the exponential model for TIR‐TITR relationship in individuals with T1D and T2D, using a real‐world CGM dataset. The model may provide new insights into the setting of individualized treatment goals.

1. Introduction

With the rapid development and widespread application of continuous glucose monitoring (CGM) technology, CGM‐based glycemic metrics have become important tools in diabetes management. Among these, time in range (TIR, 3.9–10.0 mmol/L [70–180 mg/dL]) has been widely recognized as a core indicator for assessing glucose control, and its association with diabetic‐related adverse outcomes has been well established [1, 2, 3, 4, 5]. In recent years, researchers have proposed the concept of time in tight range (TITR, 3.9–7.8 mmol/L [70–140 mg/dL]), which reflects normoglycemia better and is suitable for specific populations requiring near‐normal glycemic control or intensive glucose management, such as individuals with prediabetes, gestational diabetes, or type 2 diabetes (T2D) treated with novel antidiabetic agents [6, 7]. Furthermore, multiple studies have demonstrated that TITR is independently associated with the risk of diabetic complications and adverse outcomes [8, 9, 10, 11], and performs significantly better than TIR in predicting Haemoglobin A1c (HbA1c) levels < 6.5% and < 7.0% [12]. These findings highlight the clinical value of TITR and underscore the importance of further exploring the dynamic relationship between TIR and TITR.

Previous studies have shown a nonlinear relationship between TIR and TITR, which is significantly influenced by the glucose coefficient of variation (CV) and the time below range (TBR, < 3.9 mmol/L [< 70 mg/dL]) [7, 13, 14, 15]. Beck et al. [13] reported that, because individuals with type 1 diabetes (T1D) exhibited higher CV and TBR, their TITR values were greater than those of individuals with T2D at the same TIR level. Nevertheless, prior studies have included limited numbers of individuals with T2D, underscoring the need for large‐scale real‐world datasets encompassing substantial proportions of individuals with T2D to investigate this relationship. Moreover, while previous studies have recognized CV as an influencing factor, they have not yet integrated it into a unified mathematical model to bridge the TIR–TITR gap between T1D and T2D. Additionally, previous analyses have predominantly relied on non‐parametric approaches, and clinically applicable parametric models capable of describing this relationship quantitatively have yet to be established.

Therefore, using data from the Sibionics GS1 CGM database, a large real‐world CGM database, we included over 200 000 adult CGM users with T1D and T2D to: (1) establish a fitted equation describing the TIR‐TITR relationship in participants with diabetes in our study; (2) characterize the dynamic patterns of TIR‐TITR changes; and (3) analyse the TIR–TITR correspondence across CV subgroups and extend the model by incorporating CV as a covariate, thereby generating evidence to support more precise glycemic target recommendations.

2. Materials and Methods

2.1. Data Source and Processing

This study was a cross‐sectional analysis based on real‐world CGM data. The study included 193 956 adults with T2D and 10 524 with T1D, all of whom used a real‐time CGM device (GS1, Sibionics, China) between January 1, 2022, and December 31, 2023. Basic demographic and clinical information, including age, sex, diabetes type, duration of diabetes, and treatment regimen, were self‐reported in‐app during the initial device and application registration process without laboratory confirmation. The application incorporated a data‐privacy agreement through which users provided informed consent for the collection and analysis of de‐identified data. All participants and CGM data were collected with informed consent and anonymized prior to analysis in compliance with applicable data‐protection regulations, including the General Data Protection Regulation (GDPR). Identifiable information was removed or transformed to prevent re‐identification, and only de‐identified data were used for analyses. The study was approved by the Ethics Committee of Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine.

During data cleaning, we excluded records with missing information on sex or age, inconsistent data (e.g., disease duration exceeding age, usage of insulin without a diabetes diagnosis, or self‐reported T1D without subcutaneous insulin injections or insulin pump therapy), individuals younger than 18 years, individuals with gestational diabetes, non‐diabetic individuals, and other specific or unknown diabetes types. The GS1 system utilizes Bluetooth transmission and on‐sensor data storage, enabling continuous glucose recording and complete data synchronization throughout the activation period. Users were included only if their CGM wear duration was at least 10 days (a minimum of 70% of 14 days) [16]. In total, 193 956 adults with T2D (356 879 CGM reports; an average of 1.84 sensors per user) and 10 524 insulin‐ or pump‐treated individuals with T1D (37 872 CGM reports; an average of 3.6 sensors per user) were included. A flowchart summarizing the CGM data cleaning and participant inclusion process is provided in Figure S1.

For each CGM sensor, we calculated various glycemic metrics, including mean glucose (MG), TIR, TITR, time above range (TAR, > 10.0 mmol/L [> 180 mg/dL] and > 13.9 mmol/L [> 250 mg/dL]), time below range (TBR, < 3.9 mmol/L [< 70 mg/dL] and < 3.0 mmol/L [< 54 mg/dL]), CV, and glucose management indicator (GMI). These metrics were recalculated directly from raw sensor readings rather than derived from web‐based platform summaries. GMI was calculated using the standard formula: GMI (%) = 3.31 + 0.02392 × MG (mg/dL) [17]. For users who wore multiple CGM sensors between 2022 and 2023, we aggregated all sensor‐derived parameters for each individual and calculated the mean values to obtain user‐level averaged CGM metrics.

Two additional independent datasets (from the REPLACE‐BG [18], and from the ShanghaiT2DM [19]) were used to externally validate the CV‐adjusted model. Participants in REPLACE‐BG and ShanghaiT2DM used the CGM device (Dexcom G4 Platinum Sensor, Dexcom Inc., USA) and the flash glucose monitoring device (FreeStyle Libre H, Abbott Diabetes Care, UK), respectively. Individuals with device wear time < 70% of the intended monitoring period were excluded. Finally, we enrolled 225 individuals with T1D from REPLACE‐BG and 79 individuals with T2D from ShanghaiT2DM.

2.2. Statistical Analysis

Exponential, linear, and quadratic regression models were constructed for individuals with T1D and T2D to quantify the relationship between TIR and TITR. Model performance was evaluated and compared using the coefficient of determination (R ²), root mean square error (RMSE), and Akaike information criterion (AIC). Calibration plots were used for external validation. The exponential regression model was determined to be the best model.

The adjustment of CV followed the approach described by Beck et al. [13]. Specifically, exponential models between TITR and TIR were fitted within each of the five CV strata, followed by a weighted average of the results to generate the adjusted prediction curve. The final curves were plotted within the overlapping TIR intervals of all groups.

According to the recommended CV target for patients with diabetes in China (< 33%) [20, 21], participants were categorized into five CV subgroups: < 21% (Group 1, n = 62 158); 21% to < 25% (Group 2, n = 61 217); 25% to < 29% (Group 3, n = 46 486); 29% to < 33% (Group 4, n = 24 197); and ≥ 33% (Group 5, n = 10 422). We examined the relationship and heterogeneity of TIR and TITR across the CV subgroups. For the overall estimate, the predictions from each subgroup were combined using a sample‐size‐weighted average based on the number of participants in each group. Subsequently, CV was introduced as a covariate into the baseline model to construct an extended regression equation. To estimate the contribution of CV to the model, we conducted a variable importance analysis based on R ². R ² in the nonlinear model was computed as a pseudo‐R ² (1−sum of squared errors/total sum of squares).

The Kolmogorov–Smirnov (K–S) test was used to assess the normality of continuous variables. Data with skewed distributions were expressed as median with interquartile range (IQR), whereas normally distributed variables were presented as mean ± standard deviation (SD). Categorical variables were summarized as frequencies and percentages. Statistical significance was defined as p < 0.05. All statistical analyses were performed using SPSS version 29.0.1 (IBM Corp., Armonk, NY, USA), R version 4.2.2 (http://www.r‐project.org/), and Python version 3.13 (https://www.python.org/).

3. Results

3.1. Characteristics of CGM Users

A total of 394 751 CGM reports from 204 480 users were included in the analysis. Each user was matched with a set of averaged CGM parameters, of whom 10 524 were self‐identified as having T1D and 193 956 as having T2D. In individuals with T2D, 2.2% received insulin pump ± oral hypoglycemic agent (OHA) (n = 4203), 11.1% received insulin ± OHA (n = 21 501), 27.4% were treated with OHA (n = 53 125), 58.7% underwent lifestyle interventions (n = 113 845), and 0.6% (n = 1282) had missing data regarding treatment type. In individuals with T1D, 28.9% received insulin pump (n = 3039), and 71.1% received insulin (n = 7485). Overall, the participants in this study exhibited good glycemic control, with adults with T2D showing better control than those with T1D. Specifically, 80.6% of adults with T2D had a TIR greater than 70%, and 95.8% had a CV below 33%, whereas 70.5% of adults with T1D achieved a TIR > 70% and 81.9% had a CV < 33%. Figure 1 illustrates the demographic characteristics and glycemic control profiles of self‐reported adults with T1D, adults with T2D, and the overall adult diabetic population.

FIGURE 1.

Characteristics and glycemic control of adult CGM users with diabetes. Data are the mean ± standard deviation, median (25th–75th percentile) or n (%). T1D, type 1 diabetes. T2D, type 2 diabetes. MG, mean glucose. TIR_3.9–10.0, time in range, 3.9–10.0 mmol/L (70–180 mg/dL). TITR_3.9–7.8, time in tight range, 3.9–7.8 mmol/L (70–140 mg/dL). TAR_10.0, time above range, > 10.0 mmol/L (> 180 mg/dL). TAR_13.9, time above range, > 13.9 mmol/L (> 250 mg/dL). TBR_3.9, time below range, < 3.9 mmol/L (< 70 mg/dL). TBR_3.0, time below range, < 3.0 mmol/L (< 54 mg/dL). CV, coefficient of variation. GMI, glucose management indicator.

3.2. Relationship Between TIR and TITR and the Exponential Regression Model

The scatter plot in Figure 2 illustrates the relationship between TIR and TITR, and the Spearman correlation coefficient in the whole population was 0.957 (95% confidence interval [CI], 0.957–0.958; p < 0.001). To minimize potential confounding by baseline characteristics, we conducted 1:1 propensity score matching (PSM) adjusting for age, sex, and diabetes duration. This matched cohort (9998 individuals with T1D and 9998 with T2D) yielded a similar Spearman correlation (r = 0.961; 95% CI, 0.960–0.962; p < 0.001). Figure S2 shows the scatter plot of TIR–TITR after PSM. Notably, at the same TIR level, individuals with T1D exhibited higher TITR values than those with T2D.

FIGURE 2.

Scatter plot of TIR versus TITR in individuals with T1D and T2D. Scatter plot of TIR (3.9–10.0 mmol/L [70–180 mg/dL]) versus TITR (3.9–7.8 mmol/L [70–140 mg/dL]) in individuals with T1D (red) and T2D (blue), with exponential fitting curves shown as dashed lines for T1D and solid lines for T2D. T1D, type 1 diabetes; T2D, type 2 diabetes; TIR, time in range; TITR, time in tight range.

The scatter plot also revealed that TITR increased at an accelerating rate as TIR rose. Exponential, linear, and quadratic regression models were applied separately to fit the TIR‐TITR relationships in adults with T1D and T2D. Comparison among the three regression models showed that both the exponential and quadratic models achieved substantially higher R ² values and markedly lower AIC scores than the linear model, indicating superior goodness of fit (Table S1). Although the exponential and quadratic models demonstrated comparable goodness of fit, the exponential model was ultimately selected for data fitting because of its simplicity and interpretability:

$$\begin{array}{c}{\text{TITR}}_{\mathrm{T}1\mathrm{D}}=8.54436\times \exp \left(0.02414\times \mathrm{TIR}\right);\hfill \\ {\text{TITR}}_{\mathrm{T}2\mathrm{D}}=5.52189\times \exp \left(0.02839\times \mathrm{TIR}\right)\hfill \end{array}$$

Table 1 presents the quantitative correspondence between TITR and different TIR levels. When TIR reached 70%, the corresponding TITR values were 46.3% for T1D and 40.3% for T2D. Table S2 presents the quantitative correspondence between different TITR levels and TIR. When TITR reached 50%, the corresponding TIR values were 73.2% for T1D and 77.6% for T2D. When TITR reached 70%, the corresponding TIR values were 87.1% for T1D and 89.5% for T2D.

TABLE 1.

Predicted TITR (3.9–7.8 mmol/L [70–140 mg/dL]) based on TIR (3.9–10 mmol/L [70–180 mg/dL]).

TIR (%)	TITR (%)
	T1D	T2D	All
20	13.8	9.7	10
30	17.6	12.9	13.2
40	22.4	17.2	17.5
50	28.6	22.8	23.1
60	36.4	30.3	30.6
70	46.3	40.3	40.6
80	59	53.5	53.8
90	75.1	71.1	71.3
100	95.6	94.4	94.4

Abbreviations: T1D, type 1 diabetes; T2D, type 2 diabetes; TIR, time in range; TITR, time in tight range.

Finally, we performed subgroup analyses stratified by treatment modality, age, and diabetes duration to evaluate the consistency of the TIR–TITR relationship. As shown in Figure S3, the exponential regression curves remained consistent across subgroups for both T2D (Figure S3a–c) and T1D (Figure S3d–f). The substantial overlap of these curves indicated that the exponential relationship is robust and independent of these clinical characteristics. Characteristics and CGM‐derived glycemic metrics stratified by treatment modalities, age groups, and diabetes duration are detailed in Table S3.

3.3. Patterns of Change Between TIR and TITR

As shown in Table 2, across different stages of glycemic control, both the increase in TITR and the incremental change in its increase (i.e., Δ, which denotes the difference in TITR gain (in percentage points) between consecutive 5%‐point increments in TIR) progressively rose with higher baseline TIR, exhibiting a phase‐dependent acceleration pattern. Overall, the data indicate that the slope of the TIR–TITR relationship may become larger at higher TIR.

TABLE 2.

Average increase in TITR (3.9–7.8 mmol/L [70–140 mg/dL]) corresponding to a fixed glycemic improvement (+5%‐point) across different baseline TIR (3.9–10.0 mmol/L [70–180 mg/dL]) levels.

Baseline TIR (%)	Increase in TIR (%)	Average Increase in TITR (%)
		T1D	T2D	All
0–10	5	1.2	1	1
10–20	5	1.6 (Δ = 0.4)	1.3 (Δ = 0.3)	1.3 (Δ = 0.3)
20–30	5	2 (Δ = 0.4)	1.7 (Δ = 0.4)	1.7 (Δ = 0.4)
30–40	5	2.6 (Δ = 0.6)	2.3 (Δ = 0.6)	2.3 (Δ = 0.6)
40–50	5	3.3 (Δ = 0.7)	3 (Δ = 0.7)	3 (Δ = 0.7)
50–60	5	4.1 (Δ = 0.8)	4 (Δ = 1)	4 (Δ = 1)
60–70	5	5.3 (Δ = 1.2)	5.3 (Δ = 1.3)	5.3 (Δ = 1.3)
70–80	5	6.7 (Δ = 1.4)	7.1 (Δ = 1.8)	7.1 (Δ = 1.8)
80–90	5	8.6 (Δ = 1.9)	9.4 (Δ = 2.3)	9.4 (Δ = 2.3)
90–100	5	10.9 (Δ = 2.3)	12.5 (Δ = 3.1)	12.4 (Δ = 3)

Note: Δ denotes the difference in TITR gain (in percentage points) between consecutive 5%‐point increments in TIR.

Abbreviations: T1D, type 1 diabetes; T2D, type 2 diabetes; TIR, time in range; TITR, time in tight range.

3.4. Exponential Regression Models of TIR–TITR–CV Relationships and the Contribution of CV

After adjustment for CV, the difference between the exponential fitting curves for T1D and T2D was reduced, and the two curves tended to overlap (Figure S4). Both groups could then be described by the same fitted equation.

Therefore, CV was used as a grouping variable and participants were stratified in 4% increments to allow a more detailed analysis of the TIR–TITR relationship across different CV intervals. Exponential models of the TIR–TITR relationship were fitted for all T1D and T2D individuals across different CV subgroups, establishing the corresponding relationships for each CV category (Table 3, Figure S5). The proportion of individuals with T1D increased markedly across CV groups, from 0.9% in the < 21% group to 21.4% in the ≥ 33% group (p < 0.001). The source of heterogeneity among subgroups was related to diabetes type.

TABLE 3.

Relationship between TIR (3.9–10 mmol/L [70–180 mg/dL]) and TITR (3.9–7.8 mmol/L [70–140 mg/dL]) across different CV levels (group 1, < 21%; group 2, 21% to < 25%; group 3, 25% to < 29%; group 4, 29% to < 33%; and group 5, ≥ 33%).

TIR (%)		TITR (%)
	CV (%)	< 21	21–25	25–29	29–33	≥ 33	All
20		4.3	8	11.2	13.7	16.2	8.7
30		6.3	10.8	14.6	17.5	20.3	11.6
40		9.3	14.8	19.1	22.3	25.5	15.5
50		13.8	20.1	24.9	28.5	32	20.9
60		20.3	27.5	32.6	36.4	40.1	28.2
70		30	37.4	42.6	46.5	50.2	38.1
80		44.3	51	55.7	59.4	62.9	51.7
90		65.5	69.6	72.8	75.8	78.9	70.3
100		96.8	94.9	95.1	96.9	98.9	95.9

Abbreviations: CV, coefficient of variation; TIR, time in range; TITR, time in tight range.

To ensure comparability, we also used 36% as the cutoff value to analyse the relationship between TIR and TITR across different CV subgroups (Table S4, Figure S6). Notably, for a given TIR, a higher CV yielded a greater TITR. Figure S7 illustrates that for a given TIR, a higher CV is associated with a greater fluctuation amplitude, resulting in a larger proportion of glucose values falling within the 3.9–7.8 mmol/L range.

Given the significant influence of CV on the TIR–TITR relationship, CV was further incorporated as a covariate in the regression model, yielding the following equation:

$$\text{TITR}=2.18448\times \exp \left(0.03749\times \mathrm{TIR}\right)+0.94018\times \mathrm{CV}-14.99420$$

The unified model showed satisfactory performance in the study population, with an RMSE of 5.49% and an R ² of 0.942. However, in participants with TIR < 20%, the median observed TITR value was 1.1 (0.2–3), while the median predicted TITR value was 4 (2.1–5.9), with an RMSE of 3.24%, showing a relatively large prediction error. This subgroup comprised 588 participants (0.3% of the total study sample).

To further evaluate the model's performance across varying levels of glycemic variability, we assessed the performance metrics within each CV stratum (Table S5). The model demonstrated robust performance across all CV subgroups.

To estimate the contribution of CV to the model, we used a variable importance analysis based on R ². In the overall population, TIR explained 96.9% of the model fit (R ² = 0.913), CV explained 7.1% (R ² = 0.067), and the shared component was 4%. Among individuals with T1D, TIR explained 97.5% of the model fit (R ² = 0.927), CV explained 15.5% (R ² = 0.147), and the shared component was 13%. In individuals with T2D, TIR explained 97.1% of the model fit (R ² = 0.915), CV explained 6.4% (R ² = 0.060), and the shared component was 3.6%.

3.5. External Validation

The demographic characteristics and glycemic control of the participants are detailed in Table S6. The fitted model: TITR = 2.18448 × exp(0.03749 × TIR) + 0.94018 × CV−14.99420 yielded an RMSE of 6.54% and 8.81%, and an R ² of 0.681 and 0.847 in the REPLACE‐BG and ShanghaiT2DM validation datasets, respectively. Calibration analysis yielded slopes of 1.09 and 1.04, respectively, demonstrating high trend consistency. However, the substantial deviations in the intercepts (−10.24 and −8.18) revealed a persistent systematic bias within the model. These discrepancies may arise from differences in sample size, study population, and possibly the CGM devices used. However, after adjusting the intercepts, the model demonstrated improved fitting performance across both datasets (RMSE = 2.32% and 6.63%, R ² = 0.96 and 0.913) (Figure S8).

4. Discussion

Based on the large‐scale real‐world real‐time CGM database, this study included 204 480 adult individuals with diabetes (10 524 with T1D and 193 956 with T2D). We systematically characterized the dynamic and nonlinear relationship between TIR and TITR. The study is the first to establish an exponential model describing the relationship between TIR and TITR in individuals with diabetes. After incorporating CV, the model performance was further improved, effectively reducing the TITR differences between T1D and T2D at a given TIR level.

By analysing this CGM database, we found that overall, these participants exhibited relatively good glycemic control. In China, CGM systems primarily rely on out‐of‐pocket payments by patients, with most being over‐the‐counter (OTC) products purchased directly by patients rather than prescribed by physicians. This situation may result in the study population primarily comprising individuals who place a high priority on glucose management, thereby introducing potential selection bias. Furthermore, among participants with T2D, the proportion receiving non‐insulin therapy was high (86.1%), which may help explain the observed favourable overall glycemic control. Previous literature has reported on the glucose control of adult individuals with T1D and T2D using CGM in real‐world settings in Western countries [7, 22]. A study involving 789 adult individuals with T1D in the United Kingdom (UK) showed that the average TIR was approximately 55%–58.8%, and the CV was about 36.1%–37.6% [22]. Another study indicated an average TIR of about 83% and TITR around 60% in individuals with T2D receiving OHA, which closely resembled the results observed in this study for individuals with T2D [7]. The differences in glucose control may partially stem from variations in participant demographics and healthcare policies. Taking the UK as an example, since 2017, the National Health Service (NHS) has included CGM systems in its prescription reimbursement policy, resulting in a more clinically complex study population.

In terms of the relationship between TIR and TITR, compared with the study by Beck et al. [13], which included 912 individuals with T1D and 184 individuals with T2D from Western populations, we found that the participants in our study generally had lower TITR values at the same TIR levels (e.g., TITR 40.6% vs. 45% at TIR = 70%). This discrepancy was primarily due to the larger proportion of T2D participants and overall lower CVs (median CV 23.3%–29.6%) in our study, indirectly confirming the influence of CV on the TIR‐TITR relationship. At present, there is no unified international consensus on the recommended target for TITR. The international consensus statement published in The Lancet Diabetes & Endocrinology recommended setting both TIR and TITR targets above 70% [23]. Recent studies have suggested that a TITR greater than 50% is a more practical goal (approximately equivalent to an HbA1c of 6.5%) [24, 25]. A recent study in paediatric T1D also demonstrated that an optimal TIR cut‐off of 71.9% discriminated individuals achieving ≥ 50% TITR [26]. In this study, we found that when TIR reached 70%, the corresponding TITR values for T1D and T2D were approximately 46.3% and 40.3%, respectively. Conversely, to achieve a TITR of 70%, the required TIR levels for T1D and T2D would need to reach 87.1% and 89.5%, which would be clinically difficult to achieve. Therefore, our model indicated that achieving a TITR of approximately 50% may correspond to the widely accepted clinical target of TIR 70%. However, this TITR threshold has not been validated against clinical endpoints such as HbA1c or diabetes‐related complications in the current dataset.

TITR correlated strongly with TIR, and higher TIR corresponded to higher TITR. Passanisi et al. [26] reported that a 1% increase in TIR was associated with a 1.84‐fold higher likelihood of achieving TITR ≥ 50%. Furthermore, the exponential models established in this study (TITR_T1D = 8.54436 × exp[0.02414 × TIR]; TITR_T2D = 5.52189× exp[0.02839 × TIR]) and their fitted curves demonstrated that TITR increased more rapidly at higher TIR levels. This may be because glucose fluctuations decrease and its distribution becomes more concentrated as it approaches the normal range. The increase in TITR accelerated in a stepwise pattern with rising TIR levels, a feature similar to that reported by Beck et al. [13]. Compared with their study, our analysis more intuitively demonstrated the average TITR increment and its rate of change associated with each 5% points increase in TIR across different baseline TIR ranges. When baseline TIR was below 60%, a 5% points increase in TIR corresponded to less than a 5% points rise in TITR. However, when baseline TIR reached the 60%–70% range, both individuals with T1D and T2D exhibited TITR gains exceeding 5% points (equal to or greater than the defined TIR increment), indicating that higher TIR levels were associated with disproportionately larger gains in TITR at higher TIR levels. The magnitude of TITR increase also differed between diabetes types. At baseline, TIR levels below 60%, TITR increments were slightly higher in T1D than in T2D. However, when baseline TIR exceeded 70%, this trend reversed, with T2D showing significantly greater TITR gains. The rate of change in TITR increment (Δ) also exhibited a nonlinear acceleration. At baseline TIR levels below 30%, Δ values for both T1D and T2D were below 0.5% points. When baseline TIR increased to 50%–60%, Δ for T2D exceeded 1% point and remained consistently higher than that for T1D. In the TIR range of 80%–90%, Δ reached as high as 2.3% points in T2D, whereas T1D did not reach comparable levels until TIR exceeded 90%–100%. This difference arises because the fitted exponential curve for T2D is steeper than that for T1D at higher TIR levels. The steeper relationship is likely attributable to the lower glycemic variability in T2D. Consequently, at higher TIR levels (generally corresponding to lower MG), the more concentrated glucose distribution in T2D may enable further TIR improvements to produce larger increments in the 3.9–7.8 mmol/L range (TITR).

Beck et al. [13] reported that the CV and TBR were the primary factors influencing the relationship between TIR and TITR. After adjusting for CV or TBR separately, the predictive performance of TIR for TITR showed a smaller difference between T1D and T2D. Our study yielded similar findings after adjusting for CV. Because individuals with T2D in our study had very low TBR levels (only 0.6%), TBR was not included for subgroup comparisons in this study. Furthermore, this study represents an exploratory effort to estimate the differential contribution of CV to the model fit for TITR across different types of diabetes. In the T1D group, CV contributed 15.5% of the model fit, compared with only 6.4% in the T2D group. This difference may reflect underlying pathophysiologic and treatment differences between T1D and T2D. For example, the lack of endogenous insulin and the reliance on exogenous insulin in T1D may be associated with poorer glycemic regulation and greater glycemic variability compared with T2D. Importantly, our unified equation regardless of the type of diabetes showed comparable performance compared with the equations specifically fitted for T1D and T2D, respectively, further suggesting that CV may be an important factor associated with the TIR–TITR relationship [7, 13]. As shown in Table 3, at a fixed TIR, subgroups with higher CV showed higher TITR. The CV‐stratified analysis demonstrates that glycemic variability is an important modifier of the relationship between TIR and TITR. Individuals in our study with higher CV levels exhibited substantially higher TITR at a given TIR. These findings indicate that CV should be taken into consideration when interpreting the correspondence between TIR and TITR. Figure S7 provides an example of how CV impacts the TIR‐TITR relationship. Notably, despite an identical overall TIR (3.9–10.0 mmol/L), the high‐CV profile displays a broader distribution of glucose levels around the mean value, resulting in a larger fraction of readings within 3.9–7.8 mmol/L and a smaller fraction within the upper segment of the target range (7.8–10.0 mmol/L). However, in this case, the increase in TITR with CV is accompanied by an elevated risk of both hyper‐ and hypoglycemia due to more pronounced glucose fluctuations. This relationship between CV, TIR, and TITR is consistent with the study by Dunn et al. [7], which illustrated that, at an identical TIR, a higher CV may co‐occur with a lower MG, alongside an altered glucose distribution, resulting in a relatively higher proportion of values within 3.9–7.8 mmol/L. Consequently, an elevated TITR in the presence of increased CV may give a misleading impression of improved glycemic control, and the integration of other metrics is important for the accurate interpretation of glucose status.

External validation demonstrated an acceptable model fit in the ShanghaiT2DM dataset. However, the systematic bias emerged in both datasets. This discrepancy is likely attributable to limited sample sizes, differing population glycemic profiles, and variations in CGM devices. Notably, applying a simple intercept recalibration significantly improved both the RMSE and R ² values. This suggests that the dynamic relationship among TIR, TITR, and CV is fundamentally robust, and that initial predictive deviations arose primarily from systematic baseline shifts. Our findings support the feasibility of a unified conversion equation for TIR and TITR. However, recalibration will be required when extending this model to distinct populations or other CGM devices. Future large‐scale studies incorporating diverse CGM platforms are warranted to further validate this framework.

The major strength of this study lies in its large sample size, including 10 524 individuals with T1D and 193 956 with T2D, filling a critical gap in large‐scale data on T2D populations. Moreover, the large‐scale real‐world CGM dataset enabled the establishment of regression equations for the TIR–TITR relationship without relying on less interpretable nonparametric models. Several limitations should be noted. First, diabetes type was self‐reported, which may have introduced bias, as a proportion of self‐reported T1D cases might actually represent latent autoimmune diabetes in adults (LADA), and some self‐reported T2D users with lower CV might have had impaired glucose regulation or early‐stage T2D. Meanwhile, demographic and clinical characteristics, including sex, age, diabetes duration, and treatment modalities, were also self‐reported. Future studies should incorporate objective clinical biomarkers and detailed medical records to ensure more precise phenotypic classification. In addition, the accuracy of the exponential model was suboptimal at lower TIR levels, particularly when TIR < 20% where the relative prediction error was notably larger (median predicted TITR of 4.0% vs. observed 1.1%). The fitting discrepancies observed at the low TIR (< 20%) were primarily attributable to the limited sample size of individuals with T2D within this interval (0.3% of the total study sample). However, the compromised prediction performance in this range is unlikely to substantially diminish the clinical utility of the model, given that TITR is no longer a suitable metric for such patients with poor glucose control.

Although these observational findings do not directly impact current clinical practices, elucidating the dynamic relationships between TIR, TITR, and CV offers an evidence‐based foundation for future research. First, the exponential TIR–TITR equation provides a quantitative benchmark for clinical trial design. Due to the nonlinear relationship between TIR and TITR, studies focused on TIR must avoid assuming a simple linear relationship with TITR. This model enables the establishment of stratified targets based on baseline TIR values and population‐specific characteristics, thereby improving the precision of endpoint selection. Additionally, including CV in the model significantly reduces the disparities observed between T1D and T2D in the TIR–TITR relationship. The equation, adjusted for CV, may also facilitate a more consistent evaluation across various diabetes types. Thus, the proposed framework could help standardize efficacy criteria across mixed patient populations in the future.

In conclusion, by analysing the real‐world CGM data from more than 200 000 participants with T1D and T2D in this study, we constructed an exponential model depicting the relationship between TIR and TITR. Quantifying the dynamic association between TIR and TITR may provide new insights into the refined assessment of glycemic control and the setting of individualized glycemic targets.

Author Contributions

Y.Y. and Z.H. designed the study, collected and analysed the data, and drafted the manuscript. Y.M. and M.H. contributed to data acquisition and statistical analysis. J.L., L.C., J.C. and X.M. assisted with data interpretation and figure preparation. J.L., X.W., and J.Z. reviewed and edited the manuscript. All authors read and approved the final manuscript.

Funding

This article was supported by the Noncommunicable Chronic Diseases‐National Science and Technology Major Project (2024ZD0532000, 2024ZD0532001), the Shenzhen Medical Research Fund (C2406002), the Shanghai Leading Talent Program of Eastern Talent Plan (LJ2024121), the Sanming Project of Medicine in Shenzhen (SZSM202311019).

Conflicts of Interest

Z.H., Y.M. and M.H. were employed by the company Shenzhen Sibionics Co Ltd. The remaining authors declare no conflicts of interest.

Supporting information

Figure S1: Flowchart of the study.

Figure S2: Scatter plot of TIR versus TITR in individuals with T1D and T2D after 1:1 matching for age, sex, and diabetes duration.

Figure S3: Exponential fitting curves of the TIR–TITR relationship across clinical subgroups in individuals with T1D and T2D.

Figure S4: Exponential fitting curves of the TIR–TITR relationship for T1D and T2D after adjustment for CV.

Figure S5: Exponential fitting curves of the TIR–TITR relationship across different CV subgroups (Group 1, < 21%; group 2, 21% to < 25%; group 3, 25% to < 29%; group 4, 29% to < 33%; and group 5, ≥ 33%).

Figure S6: Exponential fitting curves of the TIR–TITR relationship across different CV subgroups (group 1, < 24%; group 2, 24% to < 28%; group 3, 28% to < 32%; group 4, 32% to < 36%; and group 5, ≥ 36%).

Figure S7: Representative CGM profiles illustrating the effect of glycemic variability on TITR at an identical TIR of 70%.

Figure S8: Calibration plots before and after intercept recalibration in REPLACE‐BG and ShanghaiT2DM dataset.

Table S1: Exponential, linear, and quadratic regression equations fitted to the TIR–TITR relationship in adults with T1D and T2D.

Table S2: Predicted TIR (3.9–10 mmol/L [70–180 mg/dL]) based on TITR (3.9–7.8 mmol/L [70–140 mg/dL]).

Table S3: Characteristics and glycemic control of CGM users with diabetes stratified by treatment modality, age, and diabetes duration.

Table S4: Relationship between TIR and TITR across different CV levels (group 1, < 24%; group 2, 24% to < 28%; group 3, 28% to < 32%; group 4, 32% to < 36%; and group 5, ≥ 36%).

Table S5: Performance of the unified equation (TITR = 2.18448 × exp.[0.03749 × TIR] +0.94018 × CV−14.99420) for predicting TITR across CV strata.

Table S6: Characteristics and glycemic control of participants in the REPLACE‐BG and ShanghaiT2DM dataset.

Data Availability Statement

Data are available from the authors upon reasonable request. The source of the data in the external validation is the REPLACE‐BG and ShanghaiT2DM, but the analyses, content and conclusions presented herein are solely the responsibility of the authors and have not been reviewed or approved by the trial group(s).