Zou HY, Jiang H, Guo XD, Wen ZY, Shi Y, Zhou AD, Wang MM, Xu F, Cai MY. Dynamic glycemic exposure and trajectory profiling for incident diabetic kidney disease: A longitudinal cohort study. World J Diabetes 2026; 17(6): 120343 [DOI: 10.4239/wjd.120343]
Corresponding Author of This Article
Meng-Yin Cai, PhD, Chief Physician, Department of Endocrinology and Metabolism, The Third Affiliated Hospital of Sun Yat-sen University, No. 600 Tianhe Road, Tianhe District, Guangzhou 510630, Guangdong Province, China. caimengyin@mail.sysu.edu.cn
Research Domain of This Article
Endocrinology & Metabolism
Article-Type of This Article
research-article
Open-Access Policy of This Article
This article is an open-access article which was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Baishideng Publishing Group Inc, 7041 Koll Center Parkway, Suite 160, Pleasanton, CA 94566, USA
Share the Article
Zou HY, Jiang H, Guo XD, Wen ZY, Shi Y, Zhou AD, Wang MM, Xu F, Cai MY. Dynamic glycemic exposure and trajectory profiling for incident diabetic kidney disease: A longitudinal cohort study. World J Diabetes 2026; 17(6): 120343 [DOI: 10.4239/wjd.120343]
Hui-Yu Zou, Xiao-Di Guo, Zhe-Yao Wen, Man-Man Wang, Fen Xu, Meng-Yin Cai, Department of Endocrinology and Metabolism, The Third Affiliated Hospital of Sun Yat-sen University, Guangzhou 510630, Guangdong Province, China
Hui-Yu Zou, Xiao-Di Guo, Zhe-Yao Wen, Man-Man Wang, Fen Xu, Meng-Yin Cai, Guangdong Provincial Key Laboratory of Diabetology and Guangzhou Municipal Key Laboratory of Mechanistic and Translational Obesity Research, The Third Affiliated Hospital of Sun Yat-sen University, Guangzhou 510630, Guangdong Province, China
Hao Jiang, Department of Gastroenterology and Hepatology, Laboratory for Clinical Medicine, Beijing You'an Hospital, Capital Medical University, Beijing 100069, China
Yi Shi, Department of Endocrinology and Metabolism, The Fourth Hospital of Changsha, Changsha 410006, Hunan Province, China
An-Dong Zhou, Department of Pharmacology, Zhongshan School of Medicine, Sun Yat-sen University, Guangzhou 510080, Guangdong Province, China
Co-corresponding authors: Fen Xu and Meng-Yin Cai.
Author contributions: Zou HY and Jiang H contributed to the conception and design of the study, data collection, data analysis and interpretation, and manuscript drafting; Guo XD and Wen ZY assisted with data collection; Shi Y, Zhou AD and Wang MM contributed to the study’s conception and design; Xu F and Cai MY participated in manuscript revision and editing; Zou HY and Jiang H contributed equally to this manuscript and are co-first authors; Cai MY and Xu F contributed equally to this manuscript and are co-corresponding authors; all authors reviewed and approved the final version of the manuscript.
AI contribution statement: ChatGPT was used only to assist with English language polishing, grammar correction, and translation of author-prepared text. No AI tool was used to generate scientific content, perform data or statistical analyses, interpret results, create figures or tables, generate references, or participate in the study design. All authors have carefully reviewed and approved the final manuscript.
Supported by the National Natural Science Foundation of China, No. 82270942; the Natural Science Foundation of Guangdong Province, No. 2026A1515012884; and the Noncommunicable Chronic Diseases-National Science and Technology Major Project, No. 2024ZD0523200.
Institutional review board statement: The study was performed according to the guidelines of the Helsinki Declaration and was approved by the Ethics Committee of the Third Affiliated Hospital of Sun Yat-sen University (Guangzhou, Guangdong Province, China), No. II2024-352-01.
Informed consent statement: Written informed consent was obtained from all participants prior to their involvement in the study.
Conflict-of-interest statement: The authors declare that they have no conflict of interest.
STROBE statement: The authors have read the STROBE Statement—a checklist of items, and the manuscript was prepared and revised according to the STROBE Statement-a checklist of items.
Data sharing statement: The dataset used during the current study is available from the corresponding author upon reasonable request.
Corresponding author: Meng-Yin Cai, PhD, Chief Physician, Department of Endocrinology and Metabolism, The Third Affiliated Hospital of Sun Yat-sen University, No. 600 Tianhe Road, Tianhe District, Guangzhou 510630, Guangdong Province, China. caimengyin@mail.sysu.edu.cn
Received: February 24, 2026 Revised: March 13, 2026 Accepted: April 17, 2026 Published online: June 15, 2026 Processing time: 107 Days and 19.9 Hours
Abstract
BACKGROUND
While glycated hemoglobin (HbA1c) remains the clinical cornerstone for monitoring type 2 diabetes (T2D), the prognostic value of the cumulative glycemic burden and its dynamic trajectories for predicting incident diabetic kidney disease (DKD) remains insufficiently characterized.
AIM
To determine whether longitudinal glycemic metrics outperform static baseline HbA1c in DKD risk stratification.
METHODS
In this retrospective cohort study, 1057 patients with T2D were enrolled. Incident DKD was defined based on the urinary albumin-to-creatinine ratio (UACR) or estimated glomerular filtration rate. To mitigate immortal time bias, we calculated the time-weighted average HbA1c (TWA-HbA1c) and incorporated it as a time-varying covariate into time-dependent Cox proportional hazards models. Furthermore, group-based trajectory modeling (GBTM) was utilized to uncover distinct longitudinal HbA1c phenotypic patterns, and multivariable Cox models were used to assess their independent associations with DKD risk.
RESULTS
Over a median follow-up of 7.0 years, 244 (23.1%) patients developed DKD. In fully adjusted Cox models [accounting for age, sex, body mass index, hypertension, major adverse cardiovascular and cerebrovascular events, hyperlipidemia, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker treatment, diabetes duration, blood urea nitrogen, serum creatinine, uric acid, low-density lipoprotein cholesterol, lipoprotein(a), UACR and measurement intensity], each 1% increment in TWA-HbA1c yielded a hazard ratio (HR) of 1.37 [95% confidence interval (CI): 1.29-1.45], compared to 1.20 (95%CI: 1.14-1.27) in baseline HbA1c. TWA-HbA1c showed superior discrimination compared to baseline HbA1c (overall C-index: 0.724 vs 0.651, P = 0.007). GBTM identified four trajectories: “Low-stable” (73.6%), “quickly declining-stable” (12.0%), “quickly increasing-stable” (5.8%), and “high-slow decline” (8.7%). Compared to “low-stable”, risk was highest for “high-slow decline” (HR = 4.72, 95%CI: 3.29-6.77) and “quickly increasing-stable” (HR = 3.62, 95%CI: 2.41-5.43).
CONCLUSION
TWA-HbA1c provides superior prognostic utility for DKD over baseline HbA1c. Continuous glycemic surveillance to identify adverse trajectories early may facilitate long-term renal preservation.
Core Tip: This study demonstrates that dynamic glycemic tracking using time-weighted average glycated hemoglobin (HbA1c) significantly outperforms static baseline measurements in predicting incident diabetic kidney disease (DKD) in type 2 diabetes. We identified a critical non-linear risk inflection at 7.0% HbA1c, beyond which renal risk escalates precipitously. Furthermore, longitudinal trajectory profiling revealed that rapid early correction of severe hyperglycemia is associated with an attenuated risk of long-term renal deterioration. These findings highlight the prognostic superiority of quantifying cumulative glucotoxicity and emphasize that early, intensive glycemic intervention may be crucial to overcome therapeutic inertia and potentially mitigate the risk of DKD.
Citation: Zou HY, Jiang H, Guo XD, Wen ZY, Shi Y, Zhou AD, Wang MM, Xu F, Cai MY. Dynamic glycemic exposure and trajectory profiling for incident diabetic kidney disease: A longitudinal cohort study. World J Diabetes 2026; 17(6): 120343
Despite continuous optimization of diabetes management strategies, diabetic kidney disease (DKD) remains a major global burden as one of the most common and serious microvascular complications of type 2 diabetes (T2D)[1]. Epidemiological data indicate that DKD has become the leading cause of end-stage renal disease, significantly increasing the risk of cardiovascular events and all-cause mortality, and thereby posing a severe challenge to healthcare systems worldwide[2]. Notably, chronic hyperglycemia is a core driver of DKD development and progression[3]. The resulting metabolic disturbances and hemodynamic changes collectively damage the glomerular filtration barrier and promote renal interstitial fibrosis[4]. Consequently, in the long-term management of patients with T2D, accurate assessment and dynamic monitoring of glycemic control are crucial for predicting DKD risk and evaluating clinical outcomes.
Glycated hemoglobin (HbA1c), recognized as the gold-standard biomarker for reflecting the average blood glucose levels over the preceding 2-3 months, is widely used for the risk stratification of diabetic complications[5]. HbA1c can be utilized as a simplified surrogate indicator of cumulative glycemic load. Due to its straightforward calculation and high reproducibility, it has been recommended by several clinical guidelines for assessing long-term glycemic control quality[6]. However, the sensitivity and specificity of these conventional HbA1c metrics for longitudinal risk assessment, as well as their dynamic association with DKD risk, remain subjects of debate. Their predictive accuracy may be particularly compromised in patients with significant glycemic variability or a long duration of diabetes[7]. To more comprehensively assess the long-term glycemic load, the concept of “cumulative HbA1c” has been proposed, defined as the area under the HbA1c-time curve[8,9]. This metric integrates both the intensity and duration of glycemic exposure during follow-up and is considered to provide a more precise quantification of the total glycemic burden[10]. However, because follow-up durations inherently vary in real-world cohorts, unadjusted cumulative HbA1c values can be confounded by the observation time. Therefore, we used the time-weighted average HbA1c (TWA-HbA1c) as the primary metric to evaluate long-term glycemic exposure[11]. Furthermore, recent advances in longitudinal data analysis methods, such as group-based trajectory modeling (GBTM), have provided new perspectives for understanding the relationship between long-term dynamic changes in HbA1c and the risk of diabetes complications[12]. GBTM identifies latent patient subgroups with similar HbA1c trajectory patterns, thereby helping to reveal heterogeneity in disease progression and potentially uncovering high-risk populations that may be overlooked by static glycemic metrics. However, most current studies still focus on baseline or single-point HbA1c measurements.
This study aims to systematically compare the predictive performance of static baseline HbA1c, dynamically updated TWA-HbA1c, and GBTM-derived longitudinal trajectories for incident DKD. We hypothesize that continuous, time-updated glycemic metrics will provide superior prognostic utility, ultimately yielding a more precise risk stratification tool to mitigate the escalating clinical burden of DKD in patients with T2D.
MATERIALS AND METHODS
Data collection and definitions
This retrospective cohort study included patients aged 20-80 years with T2D who visited The Third Affiliated Hospital of Sun Yat-sen University between January 2012 and December 2018. From an initial cohort of 9622 patients, we excluded those meeting the following criteria: (1) Baseline DKD (n = 1751); (2) Fewer than three HbA1c measurements during the follow-up period (n = 6602); (3) Known non-DKD, history of acute kidney injury requiring dialysis, or kidney transplantation (n = 48); (4) Pregnancy or lactation (n = 12); (5) Presence of malignant tumors (n = 26); and (6) Missing data for key baseline variables (n = 126). Ultimately, 1057 eligible patients were enrolled in the study (Supplementary Figure 1). For all enrolled patients, follow-up began on the date of the first available HbA1c record. The primary endpoint was the occurrence of DKD. The follow-up cutoff date was October 30, 2024. For patients who did not reach the endpoint by this date (including those lost to follow-up), their data were censored at the date of their last visit or when they were confirmed as lost to follow-up. The study was performed according to the guidelines of the Helsinki Declaration and was approved by the Ethics Committee of the Third Affiliated Hospital of Sun Yat-sen University (Guangzhou, Guangdong Province, China), No. II2024-352-01. All participants provided written informed consent prior to enrollment.
Baseline demographic and clinical characteristics were collected from electronic medical records, including age, sex, body mass index, duration of diabetes, medical history (including hypertension, hyperlipidemia, and major adverse cardiovascular and cerebrovascular events, defined as coronary heart disease and stroke) and medication history. Laboratory parameters at baseline included fasting plasma glucose, blood urea nitrogen, serum creatinine, uric acid, total cholesterol, triglycerides, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), lipoprotein(a) [Lp(a)], and urinary albumin-to-creatinine ratio (UACR). DKD was diagnosed based on an UACR > 30 mg/g or an estimated glomerular filtration rate < 60 mL/minute/1.73 m2, confirmed by two consecutive measurements at least 3 months apart[13].
HbA1c metrics and TWA-HbA1c
Baseline HbA1c was defined as the initial measurement recorded at enrollment. To address irregularly spaced measurements and potential healthcare utilization bias, we calculated the TWA-HbA1c for each participant[11]. Unlike a static baseline or a simple unweighted arithmetic mean, the TWA-HbA1c was computed using the trapezoidal rule to integrate the area under the longitudinal HbA1c curve from baseline up to each specific time point, divided by the respective follow-up duration up to that point. Incorporated into the time-dependent Cox regression models as a time-varying covariate[11], this metric dynamically updates the cumulative glycemic burden for each risk interval. This approach strictly avoids future data leakage and naturally penalizes transient, high-frequency measurements driven by acute clinical events, providing a highly robust estimation of longitudinal glycemic exposure. For clinical interpretation, both baseline HbA1c and TWA-HbA1c levels were categorized into three predefined clinical classes: < 7% (class 1), 7%-8% (class 2), and > 8% (class 3).
GBTM
The GBTM approach was implemented using the “lcmm” package in R to identify distinct longitudinal HbA1c trajectories. GBTM, a specific type of finite mixture model, was designed to identify latent subgroups of individuals who follow similar developmental trajectories over time[14]. To determine the optimal number and shape of trajectories, we estimated models with 2-7 potential trajectory groups. Model selection was based on comparisons of the Bayesian information criterion (BIC) and Akaike information criterion (AIC), with the model exhibiting the lowest BIC and AIC values preferred. The adequacy of the final model was assessed using two key criteria. First, for the assignment of individuals to specific trajectory groups to be considered accurate, the average posterior probability of assignment for each group had to be 70% or greater. Second, each trajectory group in the selected model was required to comprise at least 5% of the total sample to ensure reliable estimation and clinical interpretability. Furthermore, to specifically evaluate the predictive value of HbA1c trajectories for incident DKD, all HbA1c measurements taken after the diagnosis of DKD were excluded from the GBTM analysis.
Statistical analysis
Continuous variables were first assessed for normality using the Shapiro-Wilk test (Supplementary Table 1). Normally distributed data were expressed as mean ± SD and compared using the unpaired Student’s t-test or one-way analysis of variance. Non-normally distributed variables were reported as medians with interquartile ranges (IQR) and compared using the Mann-Whitney U test or Kruskal-Wallis test, depending on the number of groups. Categorical variables were expressed as n (%) and compared using the χ2 or Fisher’s exact test. Survival analysis was estimated using standard Kaplan-Meier curves for baseline HbA1c and trajectory groups, with differences assessed by the log-rank test. For TWA-HbA1c, which incorporates time-updating information, Simon-Makuch cumulative incidence curves were generated to prevent immortal time bias. Unadjusted and multivariable-adjusted Cox proportional hazards regression models were utilized to evaluate the associations between different HbA1c metrics (or trajectory groups) and the risk of incident DKD. TWA-HbA1c was strictly treated as a time-dependent covariate in the Cox models. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated. Sensitivity analyses were performed by excluding DKD events that occurred within the initial 3 years of follow-up to minimize potential reverse causality. To explore potential non-linear dose–response relationships between continuous HbA1c metrics and DKD risk, restricted cubic spline (RCS) curves were fitted within the fully adjusted Cox models. The overall predictive performance of baseline HbA1c vs TWA-HbA1c was compared using the overall Harrell’s C-index. Time-dependent receiver operating characteristic (ROC) curves were constructed to estimate the areas under the curve at 3 years, 5 years, and 7 years of follow-up, and compared using DeLong’s test. To further quantify the incremental predictive value, we calculated the continuous net reclassification improvement and integrated discrimination improvement. The agreement between predicted probabilities and observed DKD events at 7.0 years was visually assessed using calibration curves. The clinical utility of the models was evaluated via decision curve analysis across a range of threshold probabilities. Finally, a multivariable multinomial logistic regression model was performed to identify baseline clinical predictors for HbA1c trajectory membership, calculating odds ratios and 95%CIs. All statistical analyses were performed using R software (version 4.3.2, R Foundation for Statistical Computing, Vienna, Austria). A two-sided P < 0.05 was considered statistically significant.
RESULTS
Clinical characteristics of participants
A total of 1057 participants with T2D were included; during a median follow-up of 7.0 (IQR: 4.9-9.2) years, 244 (23.1%) patients developed DKD (Table 1). Compared with the non-DKD group, the patients who progressed to DKD were significantly older (median: 58.6 years vs 53.8 years, P < 0.001), had a longer duration of diabetes (12.0 years vs 3.5 years, P < 0.001), and showed higher prevalence of major adverse cardiovascular and cerebrovascular events (18.9% vs 12.9%, P = 0.026) and angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker (ACEI/ARB) utilization (17.7% vs 12.4%, P = 0.047). Metabolically, the DKD cohort exhibited significantly inferior glycemic control, as evidenced by elevated baseline HbA1c (9.7% vs 7.6%, P < 0.001) and fasting plasma glucose levels (9.6 mmol/L vs 8.0 mmol/L, P < 0.001). Furthermore, evaluation of baseline renal function indices revealed a lower estimated glomerular filtration rate (96.2 mL/minute/1.73 m2 vs 99.3 mL/minute/1.73 m2, P = 0.009) coupled with a higher UACR (15.8 mg/g vs 10.0 mg/g, P < 0.001) in the DKD group. Conversely, no significant differences were observed between the two groups regarding sex, body mass index, the prevalences of hypertension and hyperlipidemia, or other metabolic and lipid measures, including blood urea nitrogen, creatinine, uric acid, total cholesterol, triglycerides, LDL-C, HDL-C, and Lp(a) (all P > 0.05).
Table 1 Baseline and clinical characteristics of participants according to diabetic kidney disease status, n (%)/median (interquartile range).
Association of baseline and TWA-HbA1c with incident DKD
We stratified the participants into three groups according to their baseline HbA1c levels: < 7.0%, 7.0%-8.0%, and > 8.0%. The baseline characteristics and cumulative DKD incidence rates among groups are presented in Supplementary Tables 2 and 3. The cumulative incidence curves demonstrated overall differences among the three categories for both metrics (log-rank P < 0.001; Figure 1), which was further confirmed by pairwise comparisons (Supplementary Table 4).
Figure 1 Cumulative incidence curves for incident diabetic kidney disease according to baseline and time-weighted average glycated hemoglobin.
A: Standard Kaplan-Meier cumulative incidence curves stratified by baseline glycated hemoglobin (HbA1c); B: Simon-Makuch cumulative incidence curves stratified by time-weighted average HbA1c, analyzed as a time-dependent covariate. Participants were categorized by predefined clinical thresholds: < 7% (class 1), 7%-8% (class 2), and > 8% (class 3). Differences among groups were evaluated by the log-rank test (overall P < 0.001). HbA1c: Glycated hemoglobin.
In the fully adjusted Cox proportional hazards model (model 3), each 1% increment in baseline HbA1c was associated with an HR of 1.20 (95%CI: 1.14-1.27, P < 0.001) for incident DKD. When TWA-HbA1c was analyzed as a time-updating covariate, each 1% increment corresponded to an HR of 1.37 (95%CI: 1.29-1.45, P < 0.001) (Table 2). Compared to the < 7% reference group, the > 8% category yielded HRs of 3.10 (95%CI: 2.09-4.59) for baseline HbA1c and 4.86 (95%CI: 3.34-7.10) for TWA-HbA1c. The sensitivity analysis results excluding events within the initial three years of follow-up were similar to the primary analysis (Supplementary Table 5).
Table 2 The association of baseline and time-weighted average glycated hemoglobin with incident diabetic kidney disease.
Furthermore, RCS analysis revealed non-linear dose–response relationships between HbA1c, evaluated as both a baseline measurement and a TWA, and the risk of incident DKD (both P for nonlinearity < 0.05; Figure 2). The risk of developing DKD increased continuously with ascending levels of HbA1c. The RCS curves identified an inflection point at approximately 7%, above which the estimated HRs increased sharply.
Figure 2 Restricted cubic spline curves for the association of glycated hemoglobin with the risk of incident diabetic kidney disease.
A: Baseline glycated hemoglobin (HbA1c); B: Time-weighted average HbA1c. The solid blue lines indicate the estimated hazard ratios, and the shaded regions represent the 95% confidence intervals. The models were fully adjusted for age, sex, body mass index, hypertension, major adverse cardiovascular and cerebrovascular events, hyperlipidemia, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker treatment, diabetes duration, blood urea nitrogen, creatinine, uric acid, low-density lipoprotein cholesterol, lipoprotein(a), urinary albumin-to-creatinine ratio, and measurement intensity. The vertical dashed line indicates the inflection point. The horizontal dashed lines represent the reference hazard ratio of 1.0. HbA1c: Glycated hemoglobin; CI: Confidence interval.
Predictive values of HbA1c and related parameters for DKD
Table 3 and Figure 3 detail the predictive performance of baseline HbA1c vs TWA-HbA1c for incident DKD. Overall, TWA-HbA1c demonstrated a superior discriminative ability compared to baseline HbA1c throughout the follow-up period, as evidenced by a higher overall C-index (0.724 vs 0.651, P = 0.007). In time-dependent ROC analyses, TWA-HbA1c consistently yielded significantly larger areas under the curve at 3 years, 5 years, and 7 years (0.692, 0.730, and 0.726, respectively) compared to baseline HbA1c (0.617, 0.662, and 0.653) (all DeLong test P < 0.001; Figure 3A and C). Focusing on the 7-year landmark (approximating the cohort’s median follow-up), integrating TWA-HbA1c substantially enhanced risk reclassification, yielding a net reclassification improvement of 0.301 (95%CI: 0.046-0.557, P < 0.001) and an integrated discrimination improvement of 0.058 (95%CI: 0.017-0.099, P < 0.001) (Table 3 and Figure 3B). Calibration curves at 7 years illustrated the agreement between the predicted probabilities and observed DKD events for both evaluations of HbA1c (Figure 3D and E). Finally, decision curve analysis demonstrated that utilizing TWA-HbA1c provided a higher net clinical benefit than baseline HbA1c across the evaluated threshold probabilities for predicting the 7-year DKD risk (Figure 3F).
Figure 3 Predictive performance and clinical utility of baseline glycated hemoglobin vs time-weighted average glycated hemoglobin for incident diabetic kidney disease.
A: Comparison of time-dependent areas under the receiver operating characteristic curves (AUCs) at 3 years, 5 years, and 7 years of follow-up (P values derived from DeLong’s test); B: Dynamic variations of integrated discrimination improvement and net reclassification improvement metrics over time; C: Time-dependent receiver operating characteristic curves comparing the discriminative ability of the two metrics for predicting diabetic kidney disease at 7 years; D: Calibration curves assessing the agreement between predicted and observed 7-year survival probabilities for baseline glycated hemoglobin (HbA1c); E: Calibration curves assessing the agreement between predicted and observed 7-year survival probabilities for time-weighted average HbA1c; F: Decision curve analysis evaluating the net clinical benefit of the predictive models at 7 years across a range of threshold probabilities. AUC: Area under the receiver operating characteristic curve; HbA1c: Glycated hemoglobin; IDI: Integrated discrimination improvement; NRI: Net reclassification improvement.
Table 3 Predictive performance of baseline glycated hemoglobin vs time-weighted average glycated hemoglobin for incident diabetic kidney disease, median (interquartile range).
Four patterns of HbA1c trajectories and their baseline characteristics
GBTM was utilized to identify distinct longitudinal HbA1c patterns. After evaluating models ranging from two to seven trajectories, a four-class trajectory model was selected as the optimal fit (Supplementary Table 6). As shown in Figure 4A, four distinct HbA1c trajectories were identified. Group 1, the “low-stable” trajectory (n = 777, 73.56%), presented with a lower baseline HbA1c (median 7.3%) and maintained stable glycemic levels during the follow-up period. Group 2, the “quickly declining-stable” trajectory (n = 127, 11.99%), presented with severe baseline hyperglycemia (median 12.6%) followed by rapid early correction within the initial years before stabilizing. Group 3, the “quickly increasing-stable” trajectory (n = 61, 5.76%), began with a moderate baseline HbA1c (median 8.4%) but showed a sharp, continuous increase over time. Group 4, the “high-slow decline” trajectory (n = 92, 8.69%), started with a high baseline HbA1c (median 12.1%) and demonstrated only a gradual, modest decrease. The baseline characteristics across the four trajectories are summarized in Supplementary Table 7.
Figure 4 Identification of glycated hemoglobin trajectories and their associations with incident diabetic kidney disease.
A: Class-specific mean predicted trajectories of glycated hemoglobin (HbA1c) over time, derived from group-based trajectory modeling; B: Kaplan-Meier cumulative incidence curves for diabetic kidney disease (DKD) stratified by the four identified HbA1c trajectory groups; C: Forest plot illustrating the hazard ratios and 95% confidence intervals for incident DKD across trajectory groups, with group 1 (low-stable) as the reference. Model 1 was unadjusted. Model 2 was adjusted for age and sex. Model 3 was adjusted for age, sex, body mass index, hypertension, major adverse cardiovascular and cerebrovascular events, hyperlipidemia, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker treatment, diabetes duration, blood urea nitrogen, creatinine, uric acid, low-density lipoprotein cholesterol, lipoprotein(a), urinary albumin-to-creatinine ratio, and measurement intensity; D: Forest plot showing the odds ratios and 95% confidence intervals from multivariable multinomial logistic regression analysis for baseline predictors of trajectory membership, with group 1 serving as the reference category. aP < 0.05 vs group 1. bP < 0.01 vs group 1. DKD: Diabetic kidney disease; HbA1c: Glycated hemoglobin; CI: Confidence interval; CREAT: Creatinine; UA: Uric acid; eGFR: Estimated glomerular filtration rate; UACR: Urinary albumin-to-creatinine ratio.
Association of the changing trajectory of HbA1c with incident DKD
During follow-up, the cumulative incidence of DKD varied significantly among the four trajectory groups (log-rank P < 0.001), with the 7-year incidence rates being highest in group 4 (49.1%) and group 3 (33.2%), compared to 17.0% in group 2 and 9.9% in group 1 (Figure 4B and Supplementary Table 8). Pairwise comparisons confirmed overall differences between the reference group (group 1) and all other trajectories (Supplementary Table 9). In the fully adjusted Cox proportional hazards model (model 3), compared to group 1, the group 4 and group 3 trajectories were associated with increased risks of incident DKD, yielding HRs of 4.72 (95%CI: 3.29-6.77, P < 0.001) and 3.62 (95%CI: 2.41-5.43, P < 0.001), respectively (Table 4 and Figure 4C). The association for group 2 was attenuated after full adjustment (HR = 1.57, 95%CI: 0.95-2.58, P = 0.076). Furthermore, multivariable multinomial logistic regression analysis identified baseline clinical predictors for trajectory membership (Figure 4D and Supplementary Table 10). A longer diabetes duration was associated with an increased likelihood of belonging to groups 2, 3, and 4 compared to group 1 (all P < 0.05). A higher UACR predicted membership in group 4 (P = 0.017), while higher uric acid levels were inversely associated with membership in groups 2 and 4.
Table 4 Association between glycated hemoglobin trajectories and the risk of incident diabetic kidney disease.
In this comprehensive longitudinal cohort study of patients with T2D followed for a median of 7.0 years, we systematically demonstrated the prognostic superiority of cumulative glycemic exposure and dynamic trajectories over traditional static assessments for incident DKD. Central to our findings was the robust performance of TWA-HbA1c, which significantly outperformed single baseline measurements in both discriminative accuracy and clinical net benefit. This predictive advantage was further refined by RCS analysis revealing a profound nonlinear dose-response relationship with a critical inflection threshold at approximately 7.0%, beyond which the risk of renal progression escalated exponentially. Expanding upon these cumulative metrics, the application of GBTM allowed us to decipher the temporal heterogeneity of glycemic control and successfully identify four distinct phenotypic patterns. Notably, patients following the quickly increasing-stable and high-slow decline trajectories, corresponding to groups 3 and 4, respectively, exhibited the highest hazard for disease progression. The clustering of specific clinical profiles, particularly a longer diabetes duration and an elevated baseline UACR, within these adverse trajectories further reinforces the clinical necessity of shifting from static to continuous glycemic surveillance to promptly identify vulnerable phenotypes.
The superiority of TWA-HbA1c over static baseline measurements is consistent with the current clinical consensus that cumulative glycemic burden and glycemic variability are the primary drivers of microvascular end-organ damage[15,16]. A single baseline HbA1c measurement captures a narrow window of several months and is inadequate to comprehensively reflect longitudinal glycemic fluctuations and cumulative glucotoxicity[17]. Consequently, metrics derived from multiple measurements offer superior prognostic value. The Rochester Diabetic Neuropathy Study identified longitudinal average HbA1c and diabetes duration as the primary risk covariates for microvascular complications, demonstrating that this cumulative assessment significantly outperforms static fasting plasma glucose levels in predicting diabetic neuropathy[17]. A secondary analysis of the Action to Control Cardiovascular Risk in Diabetes trial demonstrated that greater cumulative HbA1c exposure independently escalated the risk of major cardiovascular events and all-cause mortality, irrespective of the baseline glycemic status[18]. In alignment with these macrovascular findings, our study confirmed that sustained longitudinal glycemic burden, specifically quantified via TWA-HbA1c, is a principal determinant of microvascular deterioration and incident DKD. Mechanistically, this cumulative glucotoxicity drives renal injury through the relentless accumulation of advanced glycation end products, profound mitochondrial oxidative stress, and chronic low-grade inflammation. Over time, these sustained metabolic insults irreversibly disrupt the glomerular filtration barrier, ultimately culminating in glomerulosclerosis and tubulointerstitial fibrosis[19].
Our RCS analysis elucidated the non-linear dynamics of glycemic risk, revealing a critical inflection point at approximately 7.0% HbA1c. Below this threshold, the HR for incident DKD remained stable, escalating sharply thereafter. This provides robust longitudinal evidence supporting the American Diabetes Association and European Association for the Study of Diabetes guidelines, which advocate a generalized HbA1c target of < 7.0% to mitigate microvascular complications[20,21]. However, our continuous risk evaluation suggests that a more stringent target (e.g., 6.5%) may benefit carefully selected, low-hypoglycemia-risk populations, such as younger patients or those not requiring insulin[22]. Most crucially, the abrupt escalation in renal risk above 7.0% underscores the profound danger of therapeutic inertia. This reinforces the clinical mandate for early intensive intervention to prevent irreversible morphological damage, such as mesangial matrix expansion and glomerular basement membrane thickening, prior to a permanent functional decline[23-25].
While TWA-HbA1c effectively quantifies the integral glycemic burden, it may obscure the heterogeneity of temporal glycemic patterns. The application of GBTM allowed us to untangle these complex dynamics into clinically distinct phenotypic trajectories[26,27]. Similar to findings from previous longitudinal cohorts, our model demonstrated that while the majority of patients maintained a low-stable trajectory associated with a minimal risk of renal progression, those following the high-slow decline (group 4) and quickly increasing-stable (group 3) trajectories continued to exhibit a substantially increased hazard for incident DKD[26]. Intriguingly, despite presenting with severe baseline hyperglycemia (median HbA1c: 12.6%), the “quickly declining-stable” (group 2) trajectory exhibited an attenuated risk comparable to the “low-stable” group after multivariable adjustment. This phenomenon strongly advocates for aggressive early intervention, suggesting that rapid restoration of near-normoglycemia can effectively mitigate the progression of glucotoxicity, partially reversing early endothelial dysfunction[28]. Conversely, the “quickly increasing-stable” group may reflect a poor response to second-line T2D therapies or therapeutic inertia, where delayed treatment adjustments lead to worse long-term glycemic control and a higher complication risk[29]. Moreover, our multivariable multinomial logistic regression identified a longer diabetes duration and an elevated UACR as independent baseline predictors of adverse trajectory membership. An elevated UACR, even within the normal or microalbuminuric range, signifies established endothelial dysfunction and heightened glomerular permeability, which are precursors to overt DKD[30]. Notably, the paradoxically lower baseline uric acid in groups 2 and 4 likely reflects severe hyperglycemia-induced glycosuria and early glomerular hyperfiltration, both of which accelerate renal uric acid clearance, rather than a protective metabolic profile[31-34]. For these high-risk phenotypes, achieving glycemic targets alone is insufficient. The prompt initiation of therapies with independent cardiorenal benefits [e.g., sodium-glucose cotransporter 2 inhibitors (SGLT2is) or glucagon-like peptide-1 receptor agonists (GLP-1RAs)] is of paramount clinical importance[35,36].
Our study has several limitations. First, the retrospective design inherently precluded the ability to fully control for confounding effects related to longitudinal medication adjustments. Specifically, the dynamic initiation, titration, or discontinuation of reno-protective agents throughout the follow-up period (including ACEI/ARBs, SGLT2is, and GLP-1RAs) may have independently influenced both the glycemic trajectories and renal outcomes. Although we rigorously adjusted for baseline ACEI/ARB utilization, capturing time-varying prescription data was unfeasible in this retrospective framework. Future prospective cohorts incorporating medication changes as time-varying covariates are required to evaluate the isolated effects of these treatments and to ascertain causal relationships between glycemic exposure and incident DKD. Furthermore, the lack of renal biopsy data precludes the histological validation of tubulointerstitial fibrosis across distinct glycemic trajectories, thereby limiting a deeper mechanistic interpretation of the observed renal deterioration. Lastly, the observational nature of the dataset restricted our capacity to quantify the duration and severity of unrecognized hyperglycemia prior to the formal clinical diagnosis of T2D.
CONCLUSION
In conclusion, TWA-HbA1c provided superior prognostic accuracy for incident DKD in patients with T2D compared to static baseline evaluations. This longitudinal metric effectively captured the continuous intensity of cumulative glycemic exposure and delivered enhanced clinical utility for long-term risk stratification. Furthermore, the assessment of distinct glycemic trajectories elucidated the temporal heterogeneity of blood glucose control. This dynamic evaluation successfully identified high-risk clinical phenotypes that remain undetected by single cross-sectional measurements, ultimately underscoring the clinical necessity of continuous glycemic surveillance for the early prevention of microvascular complications.
ACKNOWLEDGEMENTS
We would like to thank all the participants in this study.
Adeshara K, Di Marco E, Bordino M, Gordin D, Bernardi L, Cooper ME, Groop PH; FinnDiane Study Group. Altered oxidant and antioxidant levels are associated with vascular stiffness and diabetic kidney disease in type 1 diabetes after exposure to acute and chronic hyperglycemia.Cardiovasc Diabetol. 2024;23:350.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 9][Reference Citation Analysis (0)]
Cho S, Ok Kim C, Cha BS, Kim E, Mo Nam C, Kim MG, Park MS. The effects of long-term cumulative HbA1c exposure on the development and onset time of dementia in the patients with type 2 diabetes mellitus: Hospital based retrospective study (2005-2021).Diabetes Res Clin Pract. 2023;201:110721.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 5][Reference Citation Analysis (0)]
Mao X, Cheung KS, Tan JT, Mak LY, Lee CH, Chiang CL, Cheng HM, Hui RW, Yuen MF, Leung WK, Seto WK. Optimal glycaemic control and the reduced risk of colorectal adenoma and cancer in patients with diabetes: a population-based cohort study.Gut. 2024;73:1313-1320.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 12][Reference Citation Analysis (0)]
ElSayed NA, Aleppo G, Aroda VR, Bannuru RR, Brown FM, Bruemmer D, Collins BS, Hilliard ME, Isaacs D, Johnson EL, Kahan S, Khunti K, Leon J, Lyons SK, Perry ML, Prahalad P, Pratley RE, Seley JJ, Stanton RC, Gabbay RA; on behalf of the American Diabetes Association. 6. Glycemic Targets: Standards of Care in Diabetes-2023.Diabetes Care. 2023;46:S97-S110.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 571][Cited by in RCA: 460][Article Influence: 153.3][Reference Citation Analysis (1)]
Davies MJ, Aroda VR, Collins BS, Gabbay RA, Green J, Maruthur NM, Rosas SE, Del Prato S, Mathieu C, Mingrone G, Rossing P, Tankova T, Tsapas A, Buse JB. Management of Hyperglycemia in Type 2 Diabetes, 2022. A Consensus Report by the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD).Diabetes Care. 2022;45:2753-2786.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in Crossref: 1463][Cited by in RCA: 1196][Article Influence: 299.0][Reference Citation Analysis (9)]
Russo GT, Nicolucci A, Lucisano G, Rossi MC, Ceriello A, Prattichizzo F, Manicardi V, Rocca A, Di Bartolo P, De Cosmo S, Di Cianni G, Candido R. When Does Metabolic Memory Start? Insights From the Association of Medical Diabetologists Annals Initiative on Stringent HbA1c Targets.Diabetes. 2025;74:75-81.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 1][Cited by in RCA: 9][Article Influence: 9.0][Reference Citation Analysis (0)]
Vaz-Luis I, Di Meglio A, Havas J, El-Mouhebb M, Lapidari P, Presti D, Soldato D, Pistilli B, Dumas A, Menvielle G, Charles C, Everhard S, Martin AL, Cottu PH, Lerebours F, Coutant C, Dauchy S, Delaloge S, Lin NU, Ganz PA, Partridge AH, André F, Michiels S. Long-Term Longitudinal Patterns of Patient-Reported Fatigue After Breast Cancer: A Group-Based Trajectory Analysis.J Clin Oncol. 2022;40:2148-2162.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in Crossref: 24][Cited by in RCA: 44][Article Influence: 11.0][Reference Citation Analysis (0)]
Creativity or innovation: Grade C, Grade C, Grade D
Scientific significance: Grade B, Grade C, Grade D
P-Reviewer: Gezh SAS, PhD, Researcher, Iraq; Pappachan JM, MD, Professor, United Kingdom; Pradhan S, Researcher, India; Tung TH, PhD, Associate Professor, Taiwan S-Editor: Fan M L-Editor: A P-Editor: Xu ZH
