Abstract
Estimated glucose disposal rate (eGDR), a surrogate marker of insulin resistance, has been associated with all-cause mortality, but most studies have treated eGDR as a static exposure based on a single baseline measurement. Using data from the China Health and Retirement Longitudinal Study (CHARLS), we examined the associations of short-term eGDR change patterns and cumulative eGDR with all-cause mortality in middle-aged and older Chinese adults. Short-term eGDR change-pattern classes between 2012 and 2015 were identified using K-means clustering based on two repeated measurements. Cox proportional hazards models were used to evaluate the associations of eGDR change-pattern classes and cumulative eGDR with all-cause mortality, and restricted cubic spline regression was applied to assess the potential nonlinear association of cumulative eGDR with mortality risk. Over 52,853.24 person-years of follow-up, 428 deaths occurred among 5,966 participants. Compared with Class 1, the adjusted hazard ratios (95% confidence intervals) for all-cause mortality were 1.25 (0.87–1.79) for Class 2, 1.68 (1.25–2.24) for Class 3, and 1.77 (1.40–2.24) for Class 4. Cumulative eGDR was inversely associated with all-cause mortality. These findings suggest that dynamic and cumulative assessment of eGDR may provide prognostic information beyond a single baseline measurement.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-56908-5.
Keywords: Estimated glucose disposal rate, Insulin resistance, All-cause mortality, Cumulative exposure, Cohort study
Subject terms: Cardiology, Diseases, Endocrinology, Health care, Medical research, Risk factors
Introduction
Population aging and the increasing burden of cardiometabolic disorders have made the identification of early, modifiable risk factors for mortality a major public health priority1,2. Insulin resistance (IR), a pathophysiological state characterized by reduced responsiveness and sensitivity to insulin and impaired glucose utilization, has been consistently associated with adverse clinical outcomes and premature death3–5. Therefore, reliable and accessible markers of insulin resistance are of substantial clinical and epidemiological importance.
Although the hyperinsulinemic–euglycemic clamp remains the gold standard for assessing insulin sensitivity, its complexity, cost, and limited feasibility preclude its use in large-scale population studies6,7. Consequently, surrogate indices derived from routinely collected clinical parameters, such as the triglyceride–glucose (TyG) index, HOMA-IR, and estimated glucose disposal rate (eGDR), have gained increasing attention8–11. Among these, eGDR offers several advantages for population-based research. Unlike HOMA-IR, which requires fasting insulin measurements that are often unavailable in large cohort studies, eGDR is derived from three routinely collected parameters: waist circumference, hypertension status, and HbA1c12,13. This composition captures not only glycemic control but also central obesity and blood pressure, both of which are key determinants of cardiometabolic risk14,15. Compared with the TyG index, which primarily reflects glucose and lipid metabolism, eGDR may provide a broader representation of insulin resistance by incorporating anthropometric and vascular components. These features make eGDR particularly attractive for large-scale epidemiological studies and for risk stratification in aging populations13,16.
Accumulating evidence suggests that lower eGDR levels are associated with cardiovascular events and all-cause mortality across various populations12,17,18. However, most existing studies have relied on a single baseline measurement, implicitly assuming that insulin resistance remains stable over time. Such an approach may fail to capture the dynamic and progressive nature of metabolic dysfunction. In other cardiometabolic domains, including blood pressure, glycemic control, and lipid profiles, cumulative exposure and longitudinal patterns have often shown stronger prognostic value than baseline measurements alone. Whether similar longitudinal information improves risk stratification for eGDR remains unclear.
In addition, the eGDR formula was originally developed in Western cohorts of individuals with type 1 diabetes19. Its prognostic relevance in Asian populations, particularly among middle-aged and older adults from the general population, has not been fully characterized. Given established ethnic differences in body composition, fat distribution, and cardiometabolic risk profiles20,21, it is important to evaluate whether dynamic changes and cumulative exposure of eGDR are informative in a large Chinese population.
To address these gaps, we used data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative prospective cohort of middle-aged and older Chinese adults. Specifically, we aimed to: (1) identify short-term eGDR change-pattern classes between 2012 and 2015 using a data-driven clustering approach; (2) examine the association between these change-pattern classes and subsequent all-cause mortality; and (3) evaluate the relationship between cumulative eGDR exposure and mortality risk. By moving beyond a static, single-time-point assessment to incorporate both short-term change patterns and cumulative exposure, this study extends the current literature on eGDR and mortality. We hypothesized that less favorable eGDR change patterns and lower cumulative eGDR would be associated with a higher risk of all-cause mortality.
Methods
Study population
The CHARLS is a prospective, nationally representative cohort study conducted in China. A detailed description of its study design has been published previously22. Briefly, the study employed a multistage stratified probability-proportional-to-size sampling method to recruit 17,708 participants from 10,257 households. The baseline survey was conducted between June 2011 and March 2012 across 150 counties or districts and 450 villages within 28 provinces. Since then, participants have been followed up every two years, with five waves of data collected to date (2011, 2013, 2015, 2018, and 2020). Blood samples were obtained during the baseline wave and the third wave (2015). Ethical approval for the CHARLS study was granted by the Ethics Review Committees of Peking University, and informed consent was obtained from all participants. Further information about the CHARLS is available on its official website.
In this study, we used data from the baseline and third survey waves to evaluate short-term changes in eGDR. Subsequent follow-up surveys were used to ascertain all-cause mortality through the final follow-up in 2020. The participant selection process is shown in Fig. 1. Among 17,708 participants who attended the baseline survey, 9,642 were excluded because they lacked blood examination data in both the 2011 and 2015 surveys. Of the remaining 8,066 participants, 1,497 were further excluded because of missing data required for eGDR calculation, 185 were excluded because they were younger than 45 years, and 418 were excluded because of missing baseline covariates. Finally, 5,966 participants were included in the primary analysis of eGDR change-pattern classes. To assess the potential impact of these exclusions, baseline characteristics were compared between included and excluded participants (Supplementary Table S1).
Fig. 1.
Flowchart of participant selection for the primary analysis of eGDR change-pattern classes.
Assessment of eGDR changes
The exposures of this study were the short-term change in eGDR between 2012 and 2015 and cumulative eGDR over the same interval. eGDR (mg/kg/min) was calculated using the following formula: eGDR = 21.158 − (0.09 × WC) − (3.407 × HT) − (0.551 × HbA1c), where WC denotes waist circumference (cm), HT denotes hypertension (yes = 1, no = 0), and HbA1c is expressed as a percentage. Cumulative eGDR was calculated as the average of the 2012 and 2015 eGDR values multiplied by the time interval (3 years), providing an approximation of short-term cumulative exposure burden. Because only two repeated measurements were available, this metric was intended to summarize overall exposure across the interval rather than to fully capture more complex temporal variation or long-term trajectories. Waist circumference was measured at the natural waist position. Hypertension was defined as self-reported physician-diagnosed hypertension, use of antihypertensive medication, or blood pressure ≥ 140/90 mmHg23. Venous blood samples were collected to measure HbA1c, which was expressed as a percentage in the formula.
Covariates
Covariates included demographic and lifestyle factors, including age, sex, marital status, education level, smoking status, and drinking status. We also considered self-reported physician-diagnosed diabetes, dyslipidemia, and cardiovascular diseases, as well as use of antidiabetic and lipid-lowering medications. In addition, descriptive characteristics included body mass index (BMI), waist circumference (WC), systolic blood pressure (SBP), diastolic blood pressure (DBP), fasting plasma glucose (FPG), HbA1c, total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-c), low-density lipoprotein cholesterol (LDL-c), blood urea nitrogen (BUN), uric acid (UA), and C-reactive protein (CRP).
Statistical analysis
To identify distinct short-term change-pattern groups of eGDR between 2012 and 2015, we applied K-means clustering analysis. We selected this approach pragmatically because only two repeated measurements were available. Under this setting, more complex longitudinal trajectory-modeling approaches, such as latent class growth modeling (LCGM) or group-based trajectory modeling (GBTM), were not ideal. K-means provided a transparent, data-driven classification based on Euclidean distance without imposing parametric assumptions24. This approach has also been used in previous studies to summarize change patterns based on two repeated measurements25. Importantly, because only two observations were available for each participant, the identified groups should be interpreted as summaries of baseline level and short-term directional change rather than true long-term trajectories.
To further evaluate the robustness of the clustering solution, we examined candidate K-means solutions across k = 2–6 using the total within-cluster sum of squares and average silhouette width. We also compared cluster centroids for the k = 3, 4, and 5 solutions to assess whether the overall change-pattern structure was preserved across alternative cluster numbers. The four-cluster solution was selected on the basis of statistical performance, subgroup size, and clinical interpretability.
Based on the selected K-means solution, participants were classified into four eGDR change-pattern classes according to their baseline level and short-term directional change between 2012 and 2015. Class labels were assigned based on the corresponding centroid patterns.
For descriptive analyses, continuous variables were summarized as mean (SD) or median (interquartile range), as appropriate, and categorical variables were presented as n (%). Differences across eGDR change-pattern classes were compared using one-way analysis of variance or the Kruskal–Wallis test for continuous variables, according to data distribution, and the chi-square test for categorical variables.
To examine the associations of eGDR change-pattern classes and cumulative eGDR with all-cause mortality, we fitted Cox proportional hazards regression models. Because the publicly available CHARLS dataset does not provide exact dates of death, survival time for deceased participants was estimated using the midpoint imputation approach26,27. Specifically, survival time was defined as the midpoint between the date of the last survey wave at which the participant was confirmed alive and the date of the survey wave at which death was first recorded. For participants who remained alive, survival time was calculated from baseline to the date of the last follow-up survey in 2020. Participants lost to follow-up were censored at the date of their last known contact. As sensitivity analyses, we additionally evaluated alternative event-time assumptions using the earliest plausible date, the midpoint, and the latest plausible date of death. Hazard ratios (HRs) and 95% confidence intervals (CIs) were reported.
We constructed four models with progressive adjustment. Model 1 adjusted for age and sex. Model 2 additionally adjusted for education, marital status, smoking status, and drinking status. Model 3 was prespecified as the primary inferential model and further adjusted for history of diabetes, dyslipidemia, and cardiovascular diseases, as well as use of antidiabetic and lipid-lowering agents. Model 4 was treated strictly as a sensitivity analysis and additionally adjusted for metabolic variables, including total cholesterol, triglycerides, HDL-c, LDL-c, blood urea nitrogen, uric acid, and C-reactive protein. These additional metabolic variables may lie on the causal pathway between eGDR and mortality and therefore could introduce over-adjustment. A simplified conceptual framework for covariate selection is provided in Supplementary Figure S3. To assess potential multicollinearity, variance inflation factors (VIFs) were calculated for predictors included in Models 3 and 4. To flexibly model the potential nonlinear association between cumulative eGDR and mortality risk, we used restricted cubic spline regression with three knots placed at the 10th, 50th, and 90th percentiles.
Subgroup analyses were performed according to age (< 60 years vs. ≥60 years), sex, marital status, education level, smoking status, drinking status, and history of diabetes, dyslipidemia, and cardiovascular diseases. Interaction terms were included to assess potential effect modification. The main analyses were conducted using a complete-case analytic sample. As an additional sensitivity analysis, inverse probability weighting (IPW) was applied to assess the potential impact of selection into the complete-case sample.
The proportional hazards assumption was assessed using Schoenfeld residuals. All statistical analyses were performed using R software (version 4.4.2). All tests were two-sided, and P < 0.05 was considered statistically significant.
Results
Baseline characteristics of study participants
Baseline characteristics according to eGDR change-pattern classes are shown in Table 1. The mean age of the participants was 59.0 ± 8.7 years, and 46.1% were men. The mean eGDR was 9.3 ± 2.2 in 2012 and 8.6 ± 2.3 in 2015, with a cumulative eGDR of 26.9 ± 6.2. Compared with Class 1, participants in the other classes generally had higher adiposity and blood pressure levels and a greater prevalence of diabetes, dyslipidemia, and cardiovascular diseases. Additional laboratory characteristics are provided in Supplementary Table S2. Baseline characteristics of included and excluded participants are compared in Supplementary Table S1. Excluded participants had higher BMI and a higher prevalence of diabetes and dyslipidemia, whereas cardiovascular disease was more common among included participants. These findings suggest potential selection differences between included and excluded participants.
Table 1.
Baseline characteristics of participants according to eGDR change-pattern classes.
| Total (N = 5966) | Class 1 (N = 2885) | Class 2 (N = 463) | Class 3 (N = 781) | Class 4 (N = 1837) | P value | |
|---|---|---|---|---|---|---|
| Age, years | 59.0 ± 8.7 | 57.2 ± 8.3 | 60.9 ± 9.6 | 60.0 ± 8.8 | 61.1 ± 8.5 | < 0.001 |
| Sex, n(%) | 0.018 | |||||
| Male | 2749(46.1%) | 1330(46.1%) | 212(45.8%) | 397(50.8%) | 810(44.1%) | |
| Female | 3217(53.9%) | 1555(53.9%) | 251(54.2%) | 384(49.2%) | 1027(55.9%) | |
| Marital status, n(%) | < 0.001 | |||||
| Married or partnered | 5100(85.5%) | 2537(87.9%) | 372(80.3%) | 674(86.3%) | 1517(82.6%) | |
| Other marital status | 866(14.5%) | 348(12.1%) | 91(19.7%) | 107(13.7%) | 320(17.4%) | |
| Education level, n(%) | < 0.001 | |||||
| Primary school and lower | 4226(70.8%) | 1959(67.9%) | 343(74.1%) | 5639(72,0.1%) | 1361(74.1%) | |
| Junior high school and higher | 1740(29.2%) | 926(32.1%) | 120(25.9%) | 218(27.9%) | 476(25.9%) | |
| Smoking status, n(%) | 0.321 | |||||
| Never smokers | 2330(39.1%) | 2885(61.1%) | 463(62.0%) | 781(58.0%) | 1837(61.7%) | |
| Ever smokers | 3636(60.9%) | 1122(38.9%) | 176(38.0%) | 328(42.0%) | 704(38.3%) | |
| Drinking status, n(%) | 0.052 | |||||
| Never drinkers | 3674(61.6%) | 1073(37.2%) | 179(38.7%) | 333(42.6%) | 707(38.5%) | |
| Ever drinkers | 2292(38.4%) | 1812(62.8%) | 284(61.3%) | 448(57.4%) | 1130(61.5%) | |
| BMI, mean (SD), kg/m2 | 23.6 ± 3.8 | 22.6 ± 3.4 | 23.4 ± 3.7 | 23.6 ± 3.6 | 25.2 ± 3.9 | < 0.001 |
| WC, mean (SD), cm | 85.5 ± 10.0 | 82.2 ± 8.7 | 85.8 ± 9.7 | 85.5 ± 10.1 | 90.5 ± 9.8 | < 0.001 |
| SBP, mean (SD), mmHg | 129.8 ± 20.6 | 116.9 ± 11.7 | 143.1 ± 15.3 | 126.3 ± 11.0 | 148.2 ± 20.0 | < 0.001 |
| DBP, mean (SD), mmHg | 75.5 ± 11.6 | 69.7 ± 8.8 | 82.5 ± 9.9 | 74.1 ± 8.5 | 83.6 ± 11.4 | < 0.001 |
| Diabetes, n(%) | 995(16.7%) | 291(10.1%) | 92(19.9%) | 143(18.3%) | 469(25.5%) | < 0.001 |
| Dyslipidemia, n(%) | 578(9.7%) | 152(5.4%) | 48(10.7%) | 56(7.3%) | 322(17.9%) | < 0.001 |
| Cardiovascular diseases, n (%) | 788(13.2%) | 257(8.9%) | 59(12.7%) | 94(12.0%) | 378(20.6%) | < 0.001 |
| Antidiabetic agents, n (%) | 227(3.8%) | 37(1.3%) | 21(4.5%) | 24(3.1%) | 145(7.9%) | < 0.001 |
| Antihyperlipidemic agents, n (%) | 347(5.8%) | 79(2.7%) | 31(6.7%) | 23(2.9%) | 214(11.6%) | < 0.001 |
| eGDR in 2012, mean(SD) | 9.3 ± 2.2 | 10.9 ± 0.8 | 7.3 ± 1.1 | 10.5 ± 1.0 | 6.8 ± 1.0 | < 0.001 |
| eGDR in 2015, mean(SD) | 8.6 ± 2.3 | 10.4 ± 0.9 | 10.2 ± 1.1 | 6.6 ± 1.1 | 6.2 ± 1.2 | < 0.001 |
| Cumulative eGDR, mean(SD) | 26.9 ± 6.2 | 32.0 ± 2.5 | 26.2 ± 3.3 | 25.7 ± 3.1 | 19.5 ± 3.2 | < 0.001 |
Abbreviations: BMI, body mass index; WC, waist circumference; SBP, systolic blood pressure; DBP, diastolic blood pressure; eGDR, estimated glucose disposal rate.
Data are presented as mean (SD) for continuous variables and n (%) for categorical variables.
P values were derived from one-way analysis of variance for continuous variables and the chi-square test for categorical variables. Additional laboratory characteristics are presented in Supplementary Table S2.
Evaluation of the clustering solution
Across candidate K-means solutions (k = 2–6), the two-cluster model yielded the highest average silhouette width (0.558), whereas the four-cluster solution also showed acceptable cluster separation (average silhouette width = 0.512) together with a substantial reduction in total within-cluster sum of squares compared with the three-cluster solution (10,267 vs. 14,616). The reduction in total within-cluster sum of squares became less pronounced beyond four clusters, supporting an elbow around k = 4 (Supplementary Figure S1 and Supplementary Table S3). In addition, the four-cluster solution retained adequate subgroup sizes (minimum cluster size = 754; maximum cluster size = 2,545). Comparison of cluster centroids across the k = 3, 4, and 5 solutions showed that the broad change-pattern structure was preserved, while the four-cluster solution provided a clinically interpretable separation of persistently high, increasing, decreasing, and persistently low eGDR patterns (Supplementary Figure S2).
Based on the four-cluster solution, participants were classified into four distinct eGDR change-pattern classes (Fig. 2). Class 1 consisted of individuals with consistently high eGDR levels, showing a slight decrease from 10.91 in 2012 to 10.44 in 2015. Class 2 included individuals with a notable increase from 7.26 to 10.20. Class 3 comprised individuals with a marked decline from 10.51 to 6.62. Class 4 consisted of individuals with persistently low eGDR levels that decreased further from 6.77 to 6.24.
Fig. 2.
Identification of eGDR change-pattern classes using K-means clustering. (A) Individual-level eGDR values across the two repeated measurement waves (2012 and 2015), colored by cluster assignment. (B) Cluster centroids showing the mean eGDR values for each of the four change-pattern classes.
Association of eGDR change-pattern classes with all-cause mortality
Over a total follow-up of 52,853.24 person-years, 428 deaths were recorded among 5,966 participants. Table 2 presents the associations between eGDR change-pattern classes and all-cause mortality across four models. Model 4 additionally adjusted for metabolic variables (TC, TG, HDL-c, LDL-c, BUN, UA, CRP) as a sensitivity analysis; these variables may lie on the causal pathway between eGDR and mortality, and thus the estimates from Model 4 should be interpreted with caution. Compared with Class 1, the adjusted HRs (95% CIs) from the primary model (Model 3) were 1.25 (0.87–1.79) for Class 2, 1.68 (1.25–2.24) for Class 3, and 1.77 (1.40–2.24) for Class 4. Model 4 yielded similar estimates.
Table 2.
Association between eGDR change-pattern classes and all-cause mortality.
| eGDR change-pattern class | Events/person-years of follow-up | Model 1 | Model 2 | Model 3 | Model 4 | ||||
|---|---|---|---|---|---|---|---|---|---|
| HR(95% CI) | P value | HR(95% CI) | P value | HR(95% CI) | P value | HR(95% CI) | P value | ||
| Class 1 | 129/25721.01 | Reference | Reference | Reference | Reference | ||||
| Class 2 | 39/4090.17 | 1.33 (0.93–1.91) | 0.120 | 1.26 (0.88–1.81) | 0.206 | 1.25 (0.87–1.79) | 0.236 | 1.21 (0.84–1.75) | 0.307 |
| Class 3 | 72/6878.18 | 1.64 (1.23–2.19) | < 0.001 | 1.65 (1.24–2.20) | < 0.001 | 1.68 (1.25–2.24) | < 0.001 | 1.66 (1.23–2.23) | 0.001 |
| Class 4 | 188/16163.88 | 1.80 (1.43–2.25) | < 0.001 | 1.77 (1.41–2.22) | < 0.001 | 1.77 (1.40–2.24) | < 0.001 | 1.67 (1.31–2.13) | < 0.001 |
Association of cumulative eGDR with all-cause mortality
As shown in Table 3, in the primary model (Model 3), cumulative eGDR was significantly associated with mortality risk (HR = 0.94, 95% CI: 0.89–0.98, P = 0.005). The quartile analysis showed a significant trend (P for trend = 0.001). In the sensitivity analysis (Model 4), which additionally adjusted for metabolic variables, the HRs for Quartile 2, Quartile 3, and Quartile 4 (compared with Quartile 1) were 0.81 (0.62–1.04), 0.77 (0.57 − 1.02), and 0.55 (0.41–0.76), respectively. The restricted cubic spline regression model suggested an overall linear inverse association between cumulative eGDR and all-cause mortality risk (P for nonlinear = 0.3655). The apparent decline in risk at higher cumulative eGDR values should be interpreted descriptively rather than as evidence of a precise threshold (Fig. 3).
Table 3.
Association between cumulative eGDR all-cause mortality.
| Cumulative eGDR | Events/person-years | Model 1 | Model 2 | Model 3 | Model 4 | ||||
|---|---|---|---|---|---|---|---|---|---|
| HR(95% CI) | P-value | HR(95% CI) | P-value | HR(95% CI) | P-value | HR(95% CI) | P-value | ||
| Continuous | 428/52853.24 | 0.93(0.89–0.97) | < 0.001 | 0.93(0.89–0.97) | 0.001 | 0.94(0.89–0.98) | 0.005 | 0.94(0.90–0.99) | 0.017 |
| Q1 (< 7.22) | 140/13146.45 | Reference | Reference | Reference | Reference | ||||
| Q2 (7.23–9.89) | 128/13162.29 | 0.83(0.65–1.05) | 0.122 | 0.81(0.64–1.03) | 0.086 | 0.81(0.63–1.04) | 0.095 | 0.81(0.62–1.04) | 0.098 |
| Q3 (9.90–11.07) | 86/13310.96 | 0.71(0.54–0.93) | 0.014 | 0.71(0.54–0.93) | 0.013 | 0.75(0.56–0.99) | 0.035 | 0.77(0.57–1.02) | 0.068 |
| Q4 (≥ 11.08) | 74/13233.54 | 0.57(0.43–0.76) | < 0.001 | 0.55(0.42–0.74) | < 0.001 | 0.55(0.41–0.74) | < 0.001 | 0.55(0.41–0.76) | < 0.001 |
| P for trend | < 0.001 | < 0.001 | 0.001 | 0.005 | |||||
Values are hazard ratios (HRs) and 95% confidence intervals (CIs) from Cox proportional hazards models. Cumulative eGDR was analyzed per 1-unit increase and by quartiles.
Model 1 was adjusted for age and sex.
Model 2 was additionally adjusted for education, marital status, smoking status, and drinking status.
Model 3 was additionally adjusted for history of diabetes, dyslipidemia, and cardiovascular diseases, as well as use of antidiabetic and lipid-lowering agents, and was prespecified as the primary inferential model.
Model 4 was additionally adjusted for total cholesterol (TC), triglycerides (TG), HDL-c, LDL-c, blood urea nitrogen (BUN), uric acid (UA), and C-reactive protein (CRP), and was treated as a sensitivity analysis because these metabolic variables may lie on the causal pathway between eGDR and mortality.
Quartile 1 was used as the reference group in quartile analyses. P for trend was calculated by modeling quartiles as an ordinal variable.
Fig. 3.
Restricted cubic spline analysis of the association between cumulative eGDR and all-cause mortality. The solid line represents the adjusted hazard ratio, and the shaded area represents the 95% confidence interval. The dashed horizontal line indicates a hazard ratio of 1.0, and the dashed vertical line indicates the reference value. The model was adjusted for age, sex, education, marital status, smoking status, drinking status, history of diabetes, dyslipidemia, and cardiovascular diseases, as well as use of antidiabetic and lipid-lowering agents.
Sensitivity and supplementary analyses
Subgroup analyses were performed according to age, sex, marital status, education level, smoking status, drinking status, and history of diabetes, dyslipidemia, and cardiovascular diseases. Overall, the associations of eGDR change-pattern classes and cumulative eGDR with all-cause mortality were broadly consistent across subgroups, and no significant interaction was observed for any of the prespecified subgroup variables. Detailed subgroup-specific effect estimates and interaction P values are presented in Supplementary Table S4.
Additional sensitivity analyses supported the stability of the primary results. First, analyses using alternative event-time assumptions (earliest, midpoint, and latest plausible dates) yielded results consistent with those of the main analysis (Supplementary Table S5). Second, complete-case and inverse probability weighting analyses produced nearly identical estimates, suggesting that selection into the analytic sample was unlikely to have materially influenced the observed associations (Supplementary Table S6). Collinearity diagnostics showed no problematic multicollinearity in Model 3, whereas higher variance inflation factors were observed for several lipid-related variables in Model 4, supporting its interpretation as a sensitivity analysis rather than the primary inferential model (Supplementary Table S7). In addition, Schoenfeld residual tests and scaled Schoenfeld residual plots did not indicate a material violation of the proportional hazards assumption for the primary Cox models (Supplementary Table S8 and Supplementary Figure S4).
Discussion
In this nationally representative cohort of middle-aged and older Chinese adults, we found that both unfavorable short-term eGDR change patterns and lower cumulative eGDR were associated with a higher risk of all-cause mortality. Participants with persistently low or declining eGDR had the highest mortality risk, whereas higher cumulative eGDR was linearly associated with lower mortality. These findings suggest that, beyond a single baseline measurement, dynamic and cumulative assessment of eGDR may provide additional prognostic information.
Most previous studies have evaluated eGDR as a static exposure and have shown that lower baseline eGDR is associated with cardiovascular events and mortality, particularly in individuals with diabetes28–31. By incorporating repeated measurements, our study extends this literature and suggests that short-term deterioration in eGDR may identify individuals at increased risk even over a relatively short interval. In parallel, the inverse association between cumulative eGDR and mortality is consistent with the broader concept that cumulative metabolic burden may be more informative than single-time-point assessment32–34.
The change-pattern classes may also have different clinical interpretations. Class 2 included individuals whose eGDR increased from a relatively low baseline level; however, because this pattern was not significantly associated with mortality, improvement in eGDR should not be interpreted as evidence of fully reversible risk in the present study. This finding may reflect limited follow-up duration, residual effects of prior metabolic burden, or incomplete reversal of cumulative exposure. By contrast, Class 3 captured individuals with declining eGDR over time, and Class 4 represented those with persistently low eGDR, both of which may reflect sustained or worsening metabolic dysfunction. Nevertheless, these classes should be interpreted cautiously because they were derived from only two repeated measurements and do not represent true long-term trajectories.
Several biological processes may help explain the observed associations. Insulin resistance has been linked to endothelial dysfunction, oxidative stress, chronic low-grade inflammation, atherogenic dyslipidemia, ectopic fat deposition, and adverse cardiovascular remodeling14,35–41. Because eGDR incorporates waist circumference, hypertension status, and HbA1c, lower eGDR may reflect a broader adverse cardiometabolic milieu rather than glycemic dysregulation alone. These mechanisms provide a biologically plausible context for our findings, but they were not directly examined in the present study and should therefore be regarded as hypothesis-generating rather than established causal pathways.
A practical strength of eGDR is that it is derived from routinely collected parameters, which may support its use in large-scale clinical and epidemiological settings. Our findings suggest that repeated assessment of eGDR may help identify individuals with persistently unfavorable or worsening insulin resistance profiles and may provide complementary information for longitudinal risk stratification beyond baseline evaluation alone. However, its practical utility should be interpreted alongside simpler metabolic markers, and comparative studies are still needed to determine whether dynamic eGDR assessment offers incremental predictive value in routine practice.
Over-adjustment should also be considered when interpreting the extended model. In Model 4, we additionally adjusted for total cholesterol, triglycerides, HDL-c, LDL-c, blood urea nitrogen, uric acid, and C-reactive protein. Because some of these variables may lie on the causal pathway between eGDR and mortality, this model was interpreted strictly as a sensitivity analysis. The broadly similar estimates across models nevertheless support the overall robustness of the findings.
Several limitations should be acknowledged. First, eGDR remains a surrogate rather than a direct measure of insulin resistance, and the formula was originally developed in Western populations with type 1 diabetes. Second, both the change-pattern analysis and cumulative eGDR metric were based on only two repeated measurements, which may not fully capture more complex temporal variation. In addition, cumulative eGDR represents an approximation of short-term cumulative exposure burden rather than a full characterization of long-term metabolic dynamics. Third, residual confounding cannot be excluded because of the observational design. Fourth, questionnaire-based variables are subject to recall bias. Fifth, a substantial proportion of participants were excluded because of missing laboratory data. Because excluded participants had higher BMI and a higher prevalence of diabetes and dyslipidemia, selection bias cannot be excluded and the generalizability of the findings may be limited. However, complete-case and inverse probability weighting analyses yielded nearly identical estimates, suggesting that selection into the analytic sample was unlikely to have materially influenced the observed associations. Finally, survival times for deceased participants were estimated using midpoint imputation because exact dates of death were unavailable, although sensitivity analyses using alternative event-time assumptions yielded results consistent with the main analysis.
Conclusion
In conclusion, both unfavorable short-term eGDR change patterns and lower cumulative eGDR were associated with a higher risk of all-cause mortality in a nationally representative cohort of middle-aged and older Chinese adults. By moving beyond a single baseline measurement, this study shows that dynamic and cumulative assessment of eGDR may provide additional prognostic information for mortality risk stratification. These findings extend the current evidence on eGDR in a general Chinese population and support the potential value of repeated eGDR assessment in longitudinal cardiometabolic risk evaluation. Further studies are needed to confirm its incremental clinical utility and to clarify the mechanisms underlying these associations.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
The authors thank all the members of the CHARLS for their contributions and the participants who contributed their data.
Author contributions
T.F.L.: Conceptualization, Methodology, Formal analysis, Writing – original draft. X.J.H.: Conceptualization, Methodology, Formal analysis, Writing – review & editing. R.Z.: Data curation, Interpretation of data, Writing – review & editing. Z.S.Z.: Interpretation of data, Writing – review & editing. Z.F.Z.: Interpretation of data, Writing – review & editing. X.T.C.: Conceptualization, Supervision, Interpretation of data, Writing – review & editing.
Funding
This work was supported by the Program for Wuhan Preventive Medicine Research (WY22A08) and the Young Investigator Guidance Project of the Scientific Research Program of the Health Management Innovation Talent Cultivation Action of the Hubei Preventive Medicine Association (2025SWGKY535). The funders had no role in study design, data analysis, data interpretation, manuscript preparation, or the decision to submit the manuscript for publication.
Data availability
Online repositories contain the datasets used in this investigation. The names of the repositories and accession numbers can be found at https://charls.pku.edu.cn/en.
Declarations
Competing interests
The authors declare no competing interests.
Consent for publication
All authors have approved the manuscript and agree with its submission to the Scientific Reports.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Tengfei Long, Email: tengfl@yeah.net.
Xiaoting Chen, Email: WHSYY_ggwsk@126.com.
References
- 1.GBD 2023 Disease and Injury and Risk Factor Collaborators. Burden of 375 diseases and injuries, risk-attributable burden of 88 risk factors, and healthy life expectancy in 204 countries and territories, including 660 subnational locations, 1990–2023: a systematic analysis for the Global Burden of Disease Study 2023. Lancet406(10513), 1873–1922 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Chong, B. et al. The Global Syndemic of Modifiable Cardiovascular Risk Factors Projected From 2025 to 2050. J. Am. Coll. Cardiol.86(3), 165–177 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Lee, S. H., Park, S. Y. & Choi, C. S. Insulin resistance: From mechanisms to therapeutic strategies. Diabetes Metab J202246(1), 15–37 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Hill, M. A. et al. Insulin resistance, cardiovascular stiffening and cardiovascular disease. Metabolism2021119:154766 . [DOI] [PubMed]
- 5.Duan, M. et al. Metabolic score for insulin resistance (METS-IR) predicts all-cause and cardiovascular mortality in the general population: evidence from NHANES 2001–2018. Cardiovasc. Diabetol.23 (1), 243 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Tam, C. S. et al. Defining insulin resistance from hyperinsulinemic-euglycemic clamps. Diabetes Care201235(7), 1605–10 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.DeFronzo, R. A., Tobin, J. D. & Andres, R. Glucose clamp technique: A method for quantifying insulin secretion and resistance. Am J Physiol237(3), E214-23 (1979). [DOI] [PubMed] [Google Scholar]
- 8.Guerrero-Romero, F. et al. The product of triglycerides and glucose, a simple measure of insulin sensitivity. Comparison with the euglycemic-hyperinsulinemic clamp. J Clin Endocrinol Metab201095(7), 3347–51 (2010). [DOI] [PubMed] [Google Scholar]
- 9.Januszewski, A. S. et al. Independent euglycaemic hyperinsulinaemic clamp studies validate clinically applicable formulae to estimate insulin sensitivity in people with type 1 diabetes. Diabetes Metab Syndr202317(1), 102691, 102691. (2023). [DOI] [PubMed]
- 10.Tahapary, D. L. et al. Challenges in the diagnosis of insulin resistance: Focusing on the role of HOMA-IR and tryglyceride/glucose index. Diabetes Metab Syndr202216(8), 102581, 102581. (2022). [DOI] [PubMed]
- 11.Qiao, T. et al. Association between abdominal obesity indices and risk of cardiovascular events in Chinese populations with type 2 diabetes: a prospective cohort study. Cardiovasc. Diabetol.21 (1), 225 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Zhang, Z., Zhao, L., Lu, Y., Xiao, Y. & Zhou, X. Insulin resistance assessed by estimated glucose disposal rate and risk of incident cardiovascular diseases among individuals without diabetes: findings from a nationwide, population based, prospective cohort study. Cardiovasc. Diabetol.23 (1), 194 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Zhou, Y. et al. Estimated glucose disposal rate and cardiovascular outcomes in overweight/obese adults: a nationwide cross-sectional study with prospective cohort follow-up. J. Diabetes Metab. Disord. 25 (1), 26 (2026). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Choe, H. J., Almas, T., Neeland, I. J., Lim, S. & Després, J. P. Obesity phenotypes and atherogenic dyslipidemias. Eur J Clin Invest202656(1), e70151 (2026). [DOI] [PubMed] [Google Scholar]
- 15.de la Sierra, A. et al. Resistant hypertension and mortality: An observational cohort study. Hypertension202481(11), 2350–2356 (2024). [DOI] [PubMed] [Google Scholar]
- 16.Sun, R. et al. Association of Insulin Resistance With Cardiovascular Disease and All-Cause Mortality in Type 1 Diabetes: Systematic Review and Meta-analysis. Diabetes Care47(12), 2266–2274 (2024). [DOI] [PubMed] [Google Scholar]
- 17.Fu, C. et al. Association between estimated glucose disposal rate with the all-cause and cause-specific mortality among the population with cardiometabolic syndrome. Diabetol. Metab. Syndr.17 (1), 73 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Yao, J. et al. Association between estimated glucose disposal rate control level and stroke incidence in middle-aged and elderly adults. J Diabetes202416(8), e13595 (2024). [DOI] [PMC free article] [PubMed]
- 19.Williams, K. V., Erbey, J. R., Becker, D., Arslanian, S. & Orchard, T. J. Can clinical factors estimate insulin resistance in type 1 diabetes?. Diabetes200049(4), 626–32 (2000). [DOI] [PubMed] [Google Scholar]
- 20.Kvist, A. A. S. et al. Differences in tissue-specific insulin resistance between South Asian and Nordic women with prediabetes after gestational diabetes. Diabetologia202568(12), 2696–2708 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Cuthbertson, D. J. et al. Differential Pathophysiological Drivers of Susceptibility to Type 2 Diabetes and Metabolic Dysfunction-Associated Steatotic Liver Disease: Ethnic Differences in Insulin Dynamics, Whole-Body Fat Metabolism, and Organ-Specific Lipid Deposition. Obes. Rev. 2026 Feb16:e70104 . [DOI] [PMC free article] [PubMed]
- 22.Zhao, Y., Hu, Y., Smith, J. P., Strauss, J. & Yang, G. Cohort profile: The China Health and Retirement Longitudinal Study (CHARLS). Int J Epidemiol201443(1), 61–8 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Williams, B. et al. 2018 ESC/ESH Guidelines for the management of arterial hypertension. Eur. Heart J.39 (33), 3021–3104 (2018). [DOI] [PubMed] [Google Scholar]
- 24.Setiawan, K. E., Kurniawan, A., Chowanda, A. & Suhartono, D. Clustering models for hospitals in Jakarta using fuzzy c-means and k-means. Procedia Comput Sci216, 356–363 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Wu, Y., Yang, Y., Zhang, J., Liu, S. & Zhuang, W. The change of triglyceride-glucose index may predict incidence of stroke in the general population over 45 years old. Cardiovasc. Diabetol.22 (1), 132 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Law, C. G. & Brookmeyer, R. Effects of mid-point imputation on the analysis of doubly censored data. Stat Med11(12), 1569–78 (1992). [DOI] [PubMed] [Google Scholar]
- 27.Nakagawa, Y. & Sozu, T. Improvement of midpoint imputation for estimation of median survival time for interval-censored time-to-event data. Ther Innov Regul Sci202458(4), 721–729 (2024). [DOI] [PubMed] [Google Scholar]
- 28.He, H. M. et al. The additive effect of the triglyceride-glucose index and estimated glucose disposal rate on long-term mortality among individuals with and without diabetes: a population-based study. Cardiovasc. Diabetol.23 (1), 307 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Peng, J. et al. Estimated glucose disposal rate for predicting cardiovascular events and mortality in patients with non-diabetic chronic kidney disease: a prospective cohort study. BMC Med.22 (1), 411 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Chen, Y. et al. Joint association of estimated glucose disposal rate and systemic inflammation response index with mortality in cardiovascular-kidney-metabolic syndrome stage 0–3: a nationwide prospective cohort study. Cardiovasc. Diabetol.24 (1), 147 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Zhu, B. et al. The predictive value of estimated glucose disposal rate for all-cause and cardiovascular mortality in the US non-diabetic population aged ≥ 60 years: A population-based cohort study. Diabetes Metab Syndr202519(1), 103182, 103182. (2025). [DOI] [PubMed]
- 32.Tian, X. et al. Association of cumulative blood glucose load with cardiovascular risk and all-cause mortality. Diabetes Metab Syndr202317(12), 102900, 102900. (2023). [DOI] [PubMed]
- 33.Ference, B. A., Braunwald, E. & Catapano, A. L. The LDL cumulative exposure hypothesis: Evidence and practical applications. Nat Rev Cardiol202421(10), 701–716 (2024). [DOI] [PubMed] [Google Scholar]
- 34.Cheng, J. et al. Comparison of the cumulative exposure to four measures of blood pressure for predicting cardiovascular disease risk in the Chinese Uyghurs. BMC Public. Health. 25 (1), 1214 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Pan, Y. & Si, J. Endothelial Metabolic Reprogramming Links Diabetes to Atherosclerosis. Diabetes Metab. Syndr. Obes.19, 565805 (2026). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Symons, J. D. Opportunity “nox”: A novel approach to preventing endothelial dysfunction in the context of insulin resistance. Diabetes201362(6), 1818–20 (1818). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Pellegrini, V. et al. Inflammatory Trajectory of Type 2 Diabetes: Novel Opportunities for Early and Late Treatment. Cells13 (19), 1662 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Rizvi, A. A. Inflammation markers as mediators of vasculo-endothelial dysfunction and atherosclerosis in the metabolic syndrome. Chin. Med. J. (Engl). 120 (21), 1918–1924 (2007). [PubMed] [Google Scholar]
- 39.Wang, X. et al. The close relationship between inflammation and insulin resistance: A comparative perspective from a new angle. Acta Diabetol202663(3), 423–434 (2026). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Sethi, P. et al. Lipid Metabolism-Signaling Crosstalk in Metabolic Disease and Aging: Mechanisms and Therapeutic Targets. Nutrients17 (23), 3699 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Ramirez, A. Y., Doman, E. R., Sanchez, K. & Chilton, R. J. Impact of insulin resistance and microvascular ischemia on myocardial energy metabolism and cardiovascular function: pathophysiology and therapeutic approaches. Cardiovasc. Endocrinol. Metab.14 (2), e00332 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Online repositories contain the datasets used in this investigation. The names of the repositories and accession numbers can be found at https://charls.pku.edu.cn/en.