Longitudinal associations of cumulative and repeated-measure patterns of insulin resistance surrogate indices with biological age acceleration and incident cardiovascular disease across cardiovascular-kidney-metabolic syndrome stages 0–3: evidence from CHARLS 2011–2020

Abstract

Background

Cardiovascular-kidney-metabolic (CKM) syndrome stages 0–3 represent a critical window for preventing progression to overt cardiovascular disease (CVD). Insulin resistance (IR) is central to CKM pathophysiology, yet the comparative utility of longitudinal IR patterns (cumulative burden and longitudinal pattern groups) across multiple surrogates, and the mediating role of biological ageing, remain unexamined in this preclinical population.

Methods

We included 3948 participants with CKM stages 0–3 from the CHARLS 2011–2020. Twelve IR surrogates (TyG and derivatives, METS-IR, CTI, eGDR, TG/HDL-C) were assessed via cumulative exposure and K-means-derived pattern groups. Associations with incident CVD were evaluated using Fine-Gray competing risk models, spline regression, receiver operating characteristic analyses, and quantile-based models. Mediation analyses quantified the contribution of biological age acceleration.

Results

756 (19.1%) of 3948 participants with CKM stages 0–3 developed incident CVD. Higher cumulative levels and the least favorable pattern groups of TyG-based indices, METS-IR, CTI, and TG/HDL-C were consistently associated with increased CVD risk, whereas elevated cumulative eGDR was protective. Cumulative eGDR demonstrated superior predictive performance for CVD risk (AUC: 0.613; all DeLong P < 0.05 vs. other indices) and was consistently identified as the top contributor by both WQS and Qgcomp analyses. Both KDM- and Light-BioAgeAccel partially mediated several IR-CVD associations (up to 45.8% and 25.9%, respectively).

Conclusions

Sustained IR burden and unfavorable longitudinal IR patterns are linked to higher CVD risk in CKM stages 0–3, partly through accelerated biological aging. Integrating longitudinal IR profiling with aging metrics may sharpen early risk stratification and support scalable prevention targeting upstream metabolic drivers.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12933-026-03206-z.

Research insights

What is currently known about this topic?

1. Single-time-point IR surrogates predict CVD, but evidence on their longitudinal patterns in preclinical CKM stages 0–3 is lacking.

2. Biological age acceleration reflects systemic aging burden and predicts cardiometabolic outcomes, but its role in linking longitudinal IR burden to CVD remains unclear.

What is the key research question?

Do cumulative exposure and longitudinal patterns of multiple IR surrogate indices predict incident CVD in individuals with CKM stages 0–3, and is this association mediated by biological age acceleration?

What is new?

1. First systematic comparison of 12 IR surrogates integrating cumulative exposure and longitudinal patterns in CKM 0–3.

2. Biological age acceleration (KDM/Light BioAgeAccel) mediates up to 45.8% of the IR-CVD association, identifying ageing pathways as a mechanistic link.

3. Cumulative eGDR shows superior predictive performance and is the top contributor by WQS/Qgcomp.

4. IR indices confer stronger CVD risk in biologically younger individuals, revealing a sensitive window for early intervention.

How might this study influence clinical practice?

Longitudinal monitoring of IR dynamics using routine clinical metrics, combined with biological aging assessment, may improve early cardiovascular risk stratification and enable preventive interventions targeting metabolic drivers in preclinical CKM populations.

Introduction

Cardiometabolic, renal, and vascular disorders increasingly manifest as interconnected trajectories of systemic dysfunction rather than discrete clinical entities, forming a tightly coupled network of systemic dysfunction [1, 2]. The American Heart Association (AHA) recently introduced the cardiovascular-kidney-metabolic (CKM) framework to capture this multidimensional risk continuum [3, 4]. Epidemiological evidence indicates that CKM syndrome is highly prevalent in adult populations, with nearly 90% of adults falling somewhere along the CKM risk spectrum (stage 1 or higher), highlighting a substantial population-level health burden [5–10]. Importantly, the clinical burden attributable to cardiovascular disease (CVD) does not accrue uniformly across this spectrum but escalates disproportionately as CKM severity increases. In this context, the AHA has highlighted the prevention of progression from predominantly subclinical stages (CKM 0–3) to overt disease as a critical public health priority. This imperative underscores the pressing need to identify sensitive and reliable early markers of CVD risk among individuals in the preclinical phases of CKM syndrome to effectively attenuate downstream morbidity and mortality [4].

Insulin resistance (IR) represents a central metabolic disturbance underlying the CKM continuum and a major determinant of vascular vulnerability [11–13]. While HOMA-IR remains a reference standard in research settings, its reliance on fasting insulin limits scalability in population-based studies [14–16]. Consequently, several pragmatic IR surrogate indices derived from routine clinical parameters, including the TyG index, its anthropometric derivatives, and newer metrics such as METS-IR, eGDR, CTI, and TG/HDL-C, have been developed and independently linked to vascular injury and incident CVD [17–19]. However, existing evidence has largely focused on isolated indices assessed at a single time point. Systematic comparisons across a broad spectrum of IR surrogates, particularly incorporating cumulative exposure and repeated-measure longitudinal patterns, remain scarce. Given overlapping yet distinct metabolic components and index proliferation, head‑to‑head benchmarking is required to clarify comparative utility and reduce fragmented inference across studies. A recent cohort study by Wang et al. reported that the adverse longitudinal patterns of TyG and its obesity derivatives were closely associated with increased CVD risk among individuals on the CKM spectrum [20]. Moreover, such evidence is notably absent among individuals with CKM syndrome stages 0–3. Whether distinct repeated-measure IR patterns differentially inform future CVD risk remains unresolved.

Metabolic dysfunction is closely linked to systemic aging trajectories, positioning insulin resistance as a key contributor to accelerated biological aging [21, 22]. Biological age metrics integrate cumulative metabolic, inflammatory, and vascular stress and often provide prognostic value beyond chronological age [23–25]. Established approaches, including the Klemera–Doubal method and the more parsimonious Light BioAge index, use routinely collected biomarkers to estimate biological aging and have shown strong associations with cardiovascular morbidity and mortality across populations [26, 27]. Mechanistically, insulin resistance promotes endothelial dysfunction, chronic inflammation, oxidative stress, and mitochondrial impairment, thereby accelerating vascular and systemic aging [28–30]. Notably, these aging processes exhibit substantial interindividual heterogeneity, offering a potential explanation for the increasingly observed premature onset of CVD among younger adults. Despite these insights, it remains unclear whether the cumulative burden and longitudinal trajectories of different IR surrogate indices exhibit heterogeneous relationships with biological age acceleration, or to what extent accelerated aging mediates the longitudinal association between IR burden and incident CVD. Critically, these relationships have not been systematically examined across the preclinical CKM spectrum (stages 0–3), where disentangling metabolic–aging interactions may yield novel opportunities for early prevention.

To address these gaps, we used repeated-measure longitudinal data from the China Health and Retirement Longitudinal Study (CHARLS, 2011–2020) to investigate, among individuals with cardiovascular-kidney-metabolic (CKM) syndrome stages 0–3: (i) the associations of longitudinal patterns of multiple IR surrogate indices, defined by cumulative exposure and data-driven pattern classification based on two repeated measurements, with the risk of incident cardiovascular disease; (ii) the associations of these IR pattern groups with Wave 3 biological age acceleration, assessed using both Klemera–Doubal method-based and Light biological age acceleration metrics; and (iii) the mediating role of biological age acceleration in the relationships between these IR pattern groups and cardiovascular disease incidence. We hypothesized that sustained or worsening repeated-measure IR patterns would be associated with accelerated biological aging and elevated cardiovascular risk, and that biological age acceleration would partially mediate these associations.

Methods

Study population

This investigation was based on data derived from the China Health and Retirement Longitudinal Study (CHARLS), an ongoing nationally representative prospective cohort comprising Chinese adults aged 45 years or older who reside in community settings. CHARLS employs a multistage, stratified probability sampling strategy with probability proportional to size, recruiting participants from both urban and rural areas across 28 provinces and 150 counties or urban districts throughout China [31]. During each survey wave, standardized in-person interviews assisted by computer technology were administered by trained personnel, together with uniform physical examinations. The study collected comprehensive data on demographic characteristics, lifestyle factors, clinical conditions, anthropometric measurements, and a wide range of laboratory biomarkers, allowing for the derivation of CKM stages, IR surrogate indices, and biomarker-based indicators of biological aging. Detailed descriptions of the sampling design, measurement procedures, and biospecimen collection and laboratory assays have been reported in prior publications [31, 32]. In the current study, all analytical variables were defined according to established criteria, with detailed operational definitions provided in Material S1. The baseline survey was conducted in 2011 to 2012, followed by subsequent assessment waves in 2013, 2015, 2018, and 2020. Ethical approval for CHARLS was granted by the Institutional Review Board of Peking University (IRB00001052-11015), and written informed consent was obtained from all participants in accordance with the principles of the Declaration of Helsinki. Additional information regarding CHARLS is publicly accessible through the project website (http://charls.pku.edu.cn/en).

For the present study, we initially considered 17,708 participants who were enrolled in the 2011 to 2012 baseline Wave 1 of the CHARLS. Participants were sequentially excluded if they: (1) were younger than 45 years at baseline or had missing age information (n = 424); (2) had incomplete data required for the derivation of biological age acceleration measures (n = 7960); (3) lacked information necessary to calculate IR surrogate indices (n = 321); (4) were missing key baseline sociodemographic or lifestyle characteristics (n = 61); (5) reported a history of CVD at or before baseline (n = 1215); or (6) had insufficient data to classify CKM stages 0 to 3 (n = 1349). After these exclusions, 6,378 participants remained and were eligible for follow-up through Wave 3. Further exclusions were applied if participants: (7) had missing data for biological age acceleration related biomarkers at Wave 3 (n = 2176); (8) lacked sufficient information to construct IR surrogate indices at Wave 3 (n = 140); or (9) developed CVD between 2011 and 2015 (n = 114). The final analytic cohort therefore consisted of 3,948 participants. The detailed participant selection process is illustrated in Fig. 1.

Fig. 1.

Flowchart of participant selection and study timeline in the CHARLS

Definition of CKM syndrome stages 0–3

Consistent with the 2023 AHA Presidential Advisory Statement on cardiovascular kidney metabolic health [4], we operationalized CKM stages 0 through 3 using the following criteria. Stage 0 was defined as the absence of CKM-related risk factors, characterized by normal adiposity, glycaemic status, blood pressure, lipid profile, and kidney function, with no evidence of cardiovascular disease. Stage 1 indicated early CKM risk, defined by abdominal adiposity and/or dysglycaemia within the prediabetes range. Stage 2 captured established cardiometabolic disease and/or kidney impairment, including type 2 diabetes, hypertension, elevated triglycerides, and/or CKD. Stage 3 reflected subclinical cardiovascular involvement occurring in the setting of CKM abnormalities, defined as markedly elevated 10-year cardiovascular risk or very advanced kidney dysfunction. Specifically, stage 3 was assigned if the Framingham 10-year risk score was at least 21.5% for women or at least 21.6% for men, or if estimated glomerular filtration rate (eGFR) was below 30 mL/min/1.73 m². Kidney function was quantified using eGFR, and CKD categorization followed KDIGO recommendations [4, 33]. Framingham risk scores were calculated using established predictors (Table S1). Comprehensive operational rules for CKM staging and the diagnostic definitions for cardiometabolic conditions are presented in Tables S2 and S3.

Definitions of exposure and outcome variables

IR was characterized using a comprehensive set of surrogate markers that have been extensively validated in previous studies and can be constructed from routinely obtained clinical data [34–37]. The exposure of interest comprised twelve IR related indices, including TyG, TyG-BMI, TyG-WC, TyG-WHtR, TyG-BRI, TyG-ABSI, TyG-WWI, TyG CVAI, METS IR, CTI, eGDR, and TG/HDL-C. These indices were selected on the basis that they have been widely adopted in prior epidemiological and clinical investigations, are readily applicable in population-based cohorts owing to their reliance on standard anthropometric and fasting biochemical measurements, and collectively reflect partially overlapping yet non-identical aspects of IR and early cardiometabolic dysfunction relevant to CKM stages, including glycaemic burden, overall and central adiposity, visceral adiposity, inflammation, and estimated insulin sensitivity. They were therefore evaluated in parallel for comparative benchmarking of pragmatic IR surrogates, rather than as mutually adjusted exposures or as interchangeable representations of a single construct. At the Waves 1 and 3 of CHARLS, fasting venous blood samples were collected by trained staff from the Chinese Center for Disease Control and Prevention and centrally processed at an accredited laboratory at Youanmen Clinical Trial Center, Capital Medical University, Beijing. Biochemical assays were performed using standardized enzymatic methods with established internal quality control procedures to ensure analytical reliability. Anthropometric assessments, including body height, body weight, and waist circumference, were conducted by certified health technicians according to standardized protocols. Each IR metric was generated from anthropometric and biochemical data, applying calculation rules documented in prior research. Full methodological details and formulas are provided in Table S4.

Long-term cumulative exposure to IR related indices was quantified using a time-weighted integration approach based on repeated measurements. For each IR surrogate index, the cumulative value was calculated according to the following formula: cumulative index = (index₂₀₁₂ + index₂₀₁₅)/2 × time interval (2015–2012). Here, index₂₀₁₂ and index₂₀₁₅ represent measurements obtained in 2012 and 2015, respectively, with a 3-year interval between assessments. This formulation incorporates both the average level and duration of exposure, thereby approximating sustained metabolic burden over time rather than relying on a single observation. The same computational framework was consistently applied to all twelve IR surrogate indices examined in the present study. Such cumulative metrics have been widely used in longitudinal epidemiological research to characterize long-term cardiometabolic exposure and have demonstrated improved relevance for risk prediction compared with cross-sectional measures [38, 39].

The primary endpoint was incident CVD occurring during follow-up, defined as the first occurrence of physician-diagnosed heart disease or stroke reported after Wave 3 [40–42]. At each follow-up interview, participants were systematically queried about whether a medical professional had ever diagnosed them with a serious cardiac condition, including myocardial infarction, angina pectoris, coronary heart disease, congestive heart failure, or other clinically significant heart disorders, as well as stroke [43]. A cardiovascular event was considered incident when such a diagnosis was reported for the first time following study entry. Participants were prospectively observed from the 2015 survey and followed across subsequent survey waves through 2020. Follow-up time was accumulated from Wave 3 until the earliest occurrence of a cardiovascular event or censoring at the end of the observation period, whichever occurred first. To ensure data consistency and outcome ascertainment quality, the CHARLS coordinating center implemented standardized protocols for questionnaire administration, data processing, and quality control of self-reported health outcomes.

Definition of AgeAccel

Biological ageing was characterized using two complementary biomarker-based ageing clocks, namely the Klemera-Doubal method biological age (KDM-BA) and the Light biological age (Light-BA), both of which have been validated in large population studies, including CHARLS [26, 27]. These measures were selected for their reliance on routinely available CHARLS biomarkers with low missingness, enabling stable estimation, whereas less consistently available biomarkers for indices such as PhenoAge would reduce sample size and increase selection bias. The Klemera-Doubal approach synthesizes information from multiple physiological biomarkers into a single age-equivalent estimate and has demonstrated robust performance in capturing interindividual variation in biological ageing [44–47]. Consistent with prior analyses conducted within the CHARLS framework [45], eight biomarkers measured at Wave 3 were incorporated, including total cholesterol (TC), triglycerides (TG), glycated haemoglobin, blood urea nitrogen (BUN), creatinine, high-sensitivity C reactive protein (hs-CRP), platelet count (PLT), and systolic blood pressure (SBP). These markers reflect cardiometabolic status, inflammatory activity, and renal function. Biological age estimates based on the Klemera-Doubal algorithm were calculated following established procedures, with TG and hs-CRP logarithmically transformed prior to model fitting to improve distributional properties. In parallel, Light-BA was derived as a streamlined ageing metric designed for application in large-scale epidemiological settings and previously validated in multiple cohorts, including CHARLS [26]. This model integrates chronological age with serum creatinine, blood glucose, and hs-CRP, thereby capturing key aspects of metabolic regulation, kidney function, and systemic inflammation using routinely collected clinical measures. All equations, variable coding schemes, and parameter specifications used to compute both ageing clocks in the current analysis are detailed in Material S2.

To quantify ageing pace beyond chronological age, we constructed biological age acceleration (BioAgeAccel) indices for each clock. Specifically, KDM BioAgeAccel and Light BioAgeAccel were calculated as the residual difference between biological age and chronological age for each participant at Wave 3. These metrics capture deviations in ageing trajectories among individuals of the same chronological age. Positive values indicate advanced biological ageing relative to chronological age, whereas negative values reflect a biologically younger profile. BioAgeAccel measures were subsequently treated as intermediate variables to evaluate their potential mediating role in the associations between longitudinal insulin resistance patterns and incident CVD.

Statistical analysis

Continuous variables approximating normality are reported as mean with standard deviation (SD), whereas skewed variables are expressed as median with interquartile range (IQR). Categorical variables are presented as frequencies and proportions. Group differences were evaluated using the chi-square test or Fisher exact test for categorical variables, and Student t test or Wilcoxon Mann–Whitney tests for continuous variables, as appropriate. Spearman's rank correlation test was used to quantify pairwise associations among IR surrogate indices.

To characterize longitudinal patterns of IR, K-means clustering was applied to repeated measurements obtained in 2012 and 2015 for each insulin resistance surrogate index. The operational principles of the K-means algorithm have been previously described in detail [48]. This unsupervised algorithm groups individuals by minimizing within-cluster Euclidean distances, thereby capturing heterogeneity in repeated-measure patterns defined by the combination of exposure level and short-term directional change between 2012 and 2015. To enhance interpretability and consistency with prior repeated-measure studies of metabolic risk, the minimum number of clusters was prespecified as four. The final cluster structure for each index was determined using the elbow method, which assesses incremental reductions in within-cluster variance with increasing cluster numbers. This approach balanced model parsimony with adequate representation of longitudinal pattern heterogeneity while minimizing overfitting. Given that only two repeated measurements were available, the resulting groups were interpreted as data-driven longitudinal pattern groups rather than formal long-term trajectories. To further evaluate clustering quality and stability, internal clustering validity was quantified using the silhouette coefficient, Calinski-Harabasz index, and Davies-Bouldin index, and clustering stability was assessed using the mean and minimum Jaccard similarity based on 1,000 bootstrap resamples. All participants were assigned to the nearest cluster centroid, and the resulting pattern groups were further evaluated for clinical and biological plausibility.

All cumulative IR indices were examined both as standardized continuous variables and categorized into quartiles. Fine-Gray competing risk models were fitted to estimate subdistribution hazard ratios (sHRs) and 95% confidence intervals (CIs) for the associations of cumulative IR indices and longitudinal IR surrogate pattern groups with incident CVD, with non-CVD mortality treated as a competing event. Each index was evaluated in a separate model to avoid multicollinearity and to facilitate head-to-head benchmarking rather than inference on mutually adjusted independent effects. For continuous analyses, sHRs are per 1‑SD increase; for quartile analyses, Q2–Q4 versus Q1 as the reference. Four models were specified: Crude model included no covariates. Model 1 was adjusted for age and sex. Model 2 additionally controlled for residential status, smoking status, alcohol consumption, marital status, and educational level. Model 3 further incorporated major cardiometabolic comorbidities, including diabetes, hypertension, hypertriglyceridaemia, and CKD. We assessed collinearity among covariates by calculating variance inflation factors (VIFs) for every term entered into the multivariable models (Table S5). All VIFs were below 5, indicating no evidence of meaningful multicollinearity among the included covariates. Restricted cubic spline (RCS) analyses with three knots placed at the 10th, 50th, and 90th percentiles were applied to evaluate potential non-linear associations between cumulative levels of IR indices and incident CVD. For indices with statistically significant nonlinearity, a two-piecewise linear regression model was applied to estimate approximate inflection points.

To compare the predictive capacity of IR surrogate indices in assessing CVD risk in individuals with CKM stages 0–3, we conducted receiver operating characteristic (ROC) analysis and applied the DeLong test to evaluate and compare the areas under the curve (AUC) for various IR surrogate indices. Subsequently, the net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to assess incremental predictive value, and decision curve analysis (DCA) was performed to evaluate clinical net benefit. In addition, quantile-based mixture models were implemented to quantify the joint contribution of multiple indices. Specifically, quantile g-computation (Qgcomp) was performed using the qgcomp.noboot procedure, and weighted quantile sum (WQS) regression with 1,000 bootstrap iterations was conducted. Both approaches generated index-specific weights, enabling assessment of the relative importance of cumulative IR surrogates in predicting cardiovascular risk [49, 50].

To investigate potential mechanistic pathways, exploratory mediation analyses, including CKM stage-stratified analyses in stages 0–1 and 2–3, were performed to determine whether BioAgeAccel statistically accounted for part of the association between cumulative IR surrogate indices and incident CVD, with the mediated proportion quantitatively estimated. Temporal ordering was partially preserved in the analytic framework, with cumulative IR exposure and repeated-measure pattern groups defined from the 2012 and 2015 assessments, BioAgeAccel assessed at Wave 3, and incident CVD ascertained only thereafter. Furthermore, subgroup analyses and effect modification assessments were conducted according to BioAgeAccel status, sex, smoking and drinking behaviours, major cardiometabolic comorbidities, and CKM stages. Effect heterogeneity across strata was examined by incorporating interaction terms into regression models, and statistical significance was evaluated using likelihood ratio tests. To account for multiple comparisons, P values from subgroup and interaction analyses were further adjusted using the Benjamini-Hochberg false discovery rate procedure.

Sensitivity analyses were conducted to evaluate the robustness of the primary findings: (1) Missing-data robustness: missing baseline covariates were handled using multiple imputation by chained equations (MICE), and baseline characteristics of participants excluded because of missing covariates were compared with those of the final analytic cohort; (2) Weighted reanalysis: to further reduce residual imbalance in baseline risk profiles, inverse probability of treatment weighting (IPTW) was implemented in the imputed datasets, and the primary Fine–Gray models were refitted; (3) Time-varying medication adjustment: separate Cox models additionally incorporated follow-up use of antihypertensive, lipid-lowering, and glucose-lowering medications as time-varying covariates; (4) Unmeasured confounding: E-values were calculated for the primary point estimates and lower confidence limits [51, 52]; (5) Cumulative-exposure validity: Pearson correlation analyses were performed to compare cumulative exposure with baseline, mean, and change values, and the discriminatory performance of models based on 2012 indices, 2015 indices, and cumulative indices was further compared with that of the corresponding principal component analysis (PCA)-based models; (6) Stroke-specific outcome: stroke was analyzed separately as an outcome; (7) Baseline- and change-based exposure: analyses were repeated using baseline IR surrogate indices alone and absolute changes between 2012 and 2015; and (8) Lag analysis: analyses were repeated after excluding early CVD events occurring before 2016 to minimize potential reverse causation.

All analyses were conducted using R software version 4.5.2. RCS regression, qgcomp analysis, WQS analysis, K-means clustering, and mediation analyses were implemented using the rms (version 6.8.0), qgcomp (version 2.15.2), gWQS (version 0.2.2), stats (version 4.5.2), and mediation (version 4.5.1) packages, respectively. Multiple imputation was performed using the mice (version 3.18.0) package. Fine–Gray competing risk analyses and IPTW were implemented using the tidycmprsk (version 1.1.2) and WeightIt (version 1.5.0) packages, respectively.

All statistical tests were two-sided, and statistical significance was defined as a P value below 0.05.

Results

Baseline characteristics of study participants

As shown in Tables 1 and S6, over a median follow-up of 8 years, 756 (19.1%) developed incident CVD among 3,948 participants. The average age of the participants was 58.80 ± 8.8 years, and males accounted for 47.0%. Compared to those who remained free of CVD, participants who developed CVD tended to exhibit an older biological ageing profile and showed less favorable cardiometabolic and inflammatory status, characterized by higher levels of FBG, LDL-C, TC, hs-CRP, HbA1c, SBP, DBP, Framingham score, BMI, and waist circumference. They also had greater long-term cumulative exposure across multiple IR surrogates, including TyG, TyG-BMI, TyG-WC, TyG-WHtR, TyG-BRI, TyG-ABSI, TyG-WWI, TyG-CVAI, METS-IR, and CTI. In contrast, incident CVD cases exhibited lower eGFR and reduced cumulative eGDR. Moreover, the CVD group had a higher prevalence of cardiometabolic comorbidities and risk states, including overweight or obesity, abdominal obesity, diabetes, hypertension, metabolic syndrome, subclinical CVD, and more advanced CKM stages. All between-group comparisons reached statistical significance (all P < 0.05). Additional comparisons stratified by CKM stages and biological age acceleration metrics are reported in Tables S7–S9. Of note, each cumulative IR surrogate index was significantly associated with both KDM BioAgeAccel and Light BioAgeAccel.

Table 1.

Baseline characteristics of participants

Variables	Total (N = 3948)	Non-CVD (N = 3192)	CVD (N = 756)	P value	SMD
Age (years)	58.8 (8.8)	58.5 (8.8)	60.2 (8.6)	< 0.001	0.201
KDM BioAgeAccel2012	0.0 (20.2)	− 1.2 (19.3)	4.2 (23.3)	< 0.001	0.256
KDM BioAgeAccel2015	0.0 (21.4)	− 1.3 (21.0)	4.6 (22.7)	< 0.001	0.273
Light BioAgeAccel2012	0.0 (6.6)	− 0.2 (6.4)	1.0 (7.0)	< 0.001	0.177
Light BioAgeAccel2015	0.0 (6.8)	− 0.2 (6.8)	0.8 (6.7)	0.001	0.139
Gender, n (%)
Male	1856 (47.0)	1538 (48.2)	318 (42.1)	0.003	0.123
Female	2092 (53.0)	1654 (51.8)	438 (57.9)
Residence, n (%)
Rural	3426 (86.8)	2765 (86.6)	661 (87.4)	0.595	0.024
Urban	522 (13.2)	427 (13.4)	95 (12.6)
Educational level, n (%)
No formal education	1122 (28.4)	886 (27.8)	236 (31.2)	0.009	0.140
Primary school	1678 (42.5)	1341 (42.0)	337 (44.6)
Middle or high school	1107 (28.0)	932 (29.2)	175 (23.1)
College or above	41 (1.0)	33 (1.0)	8 (1.1)
Marital status, n (%)
Married	2800 (70.9)	2227 (69.8)	573 (75.8)	0.001	0.136
Other	1148 (29.1)	965 (30.2)	183 (24.2)
Smoking status, n (%)
Current	1233 (31.2)	1016 (31.8)	217 (28.7)	0.248	0.068
Ever	315 (8.0)	252 (7.9)	63 (8.3)
Never	2400 (60.8)	1924 (60.3)	476 (63.0)
Drinking status, n (%)
Current/ever	1314 (33.3)	1075 (33.7)	239 (31.6)	0.298	0.044
Never	2634 (66.7)	2117 (66.3)	517 (68.4)
IR surrogate indices
TyG2012	8.8 (0.7)	8.7 (0.7)	8.9 (0.7)	< 0.001	0.164
TyG2015	8.8 (0.6)	8.8 (0.6)	8.8 (0.7)	0.001	0.138
TyG-BMI2012	208.7 (41.9)	207.0 (41.4)	216.1 (43.2)	< 0.001	0.216
TyG-BMI2015	210.2 (41.1)	208.8 (40.9)	216.2 (41.5)	< 0.001	0.180
TyG-WC2012	754.8 (120.7)	748.6 (118.7)	780.8 (126.0)	< 0.001	0.263
TyG-WC2015	760.7 (123.4)	755.4 (121.6)	783.3 (128.8)	< 0.001	0.223
TyG-WHtR2012	4.8 (0.8)	4.7 (0.8)	5.0 (0.8)	< 0.001	0.269
TyG-WHtR2015	4.8 (0.8)	4.8 (0.8)	5.0 (0.8)	< 0.001	0.237
TyG-BRI2012	38.0 (13.4)	37.3 (13.0)	41.1 (14.4)	< 0.001	0.271
TyG-BRI2015	39.2 (14.0)	38.5 (13.5)	42.1 (15.3)	< 0.001	0.248
TyG-ABSI2012	0.7 (0.1)	0.7 (0.1)	0.7 (0.1)	< 0.001	0.207
TyG-ABSI2015	0.7 (0.1)	0.7 (0.1)	0.7 (0.1)	< 0.001	0.192
TyG-WWI2012	98.2 (11.9)	97.7 (11.6)	100.5 (12.5)	< 0.001	0.237
TyG-WWI2015	98.9 (11.8)	98.4 (11.7)	101.1 (12.4)	< 0.001	0.218
TyG-CVAI2012	886.9 (390.6)	866.9 (385.5)	971.2 (401.1)	< 0.001	0.265
TyG-CVAI2015	949.2 (393.7)	930.5 (389.7)	1028.0 (401.0)	< 0.001	0.247
METS-IR2012	36.5 (8.6)	36.2 (8.5)	37.8 (9.0)	< 0.001	0.177
METS-IR2015	36.0 (7.4)	35.8 (7.4)	36.9 (7.5)	< 0.001	0.154
CTI2012	4.8 (0.6)	4.8 (0.6)	4.9 (0.6)	< 0.001	0.215
CTI2015	4.9 (0.6)	4.9 (0.6)	5.0 (0.6)	< 0.001	0.176
eGDR2012	8.3 (2.1)	8.4 (2.1)	7.6 (2.0)	< 0.001	0.388
eGDR2015	8.1 (2.2)	8.3 (2.2)	7.5 (2.2)	< 0.001	0.355
TG/HDL-C2012	3.7 (5.8)	3.7 (5.8)	3.9 (5.7)	0.328	0.040
TG/HDL-C2015	3.3 (2.7)	3.2 (2.6)	3.4 (2.8)	0.044	0.080
Cumulative TyG	26.3 (1.8)	26.2 (1.7)	26.6 (1.9)	< 0.001	0.170
Cumulative TyG-BMI	628.4 (117.7)	623.6 (116.4)	648.5 (121.2)	< 0.001	0.209
Cumulative TyG-WC	2273.3 (344.5)	2256.0 (338.5)	2346.1 (359.9)	< 0.001	0.258
Cumulative TyG-WHtR	14.4 (2.2)	14.3 (2.2)	14.9 (2.4)	< 0.001	0.268
Cumulative TyG-BRI	115.8 (39.0)	113.7 (37.9)	124.7 (42.5)	< 0.001	0.273
Cumulative TyG-ABSI	2.2 (0.2)	2.2 (0.2)	2.2 (0.2)	< 0.001	0.228
Cumulative TyG-WWI	295.7 (32.1)	294.2 (31.5)	302.4 (33.8)	< 0.001	0.252
Cumulative TyG-CVAI	2754.1 (1118.1)	2696.2 (1104.1)	2998.9 (1144.6)	< 0.001	0.269
Cumulative METS-IR	108.8 (22.7)	108.0 (22.5)	112.0 (23.6)	< 0.001	0.176
Cumulative CTI	14.6 (1.6)	14.5 (1.5)	14.8 (1.6)	< 0.001	0.224
Cumulative eGDR	24.7 (5.8)	25.1 (5.8)	22.7 (5.6)	< 0.001	0.417
Cumulative TG/HDL-C	10.5 (11.1)	10.4 (11.0)	11.0 (11.6)	0.136	0.059
Overweight/obesity, n (%)
No	1843 (46.7)	1528 (47.9)	315 (41.7)	0.002	0.125
Yes	2105 (53.3)	1664 (52.1)	441 (58.3)
Abdominal obesity, n (%)
No	1827 (46.3)	1557 (48.8)	270 (35.7)	< 0.001	0.267
Yes	2121 (53.7)	1635 (51.2)	486 (64.3)
Prediabetes, n (%)
No	2257 (57.2)	1839 (57.6)	418 (55.3)	0.263	0.047
Yes	1691 (42.8)	1353 (42.4)	338 (44.7)
Hypertension, n (%)
No	1363 (34.5)	1186 (37.2)	177 (23.4)	< 0.001	0.302
Yes	2585 (65.5)	2006 (62.8)	579 (76.6)
Chronic kidney disease, n (%)
No	3604 (91.9)	2918 (92.0)	686 (91.3)	0.610	0.023
Yes	319 (8.1)	254 (8.0)	65 (8.7)
Hypertriglyceridemia, n (%)
No	2370 (60.0)	1939 (60.7)	431 (57.0)	0.065	0.076
Yes	1578 (40.0)	1253 (39.3)	325 (43.0)
Subclinical CVD, n (%)
No	3853 (97.6)	3128 (98.0)	725 (95.9)	0.001	0.122
Yes	95 (2.4)	64 (2.0)	31 (4.1)
CKM stage, n (%)
Stage 0–1	571 (14.5)	515 (16.1)	56 (7.4)	< 0.001	0.294
Stage 2–3	3377 (85.5)	2677 (83.9)	700 (92.6)

All values are presented as number and proportion for categorical variables, and mean (SD) for continuous variables

CVD, Cardiovascular disease; IR, Insulin resistance; KDM, Klemera–Doubal method; BioAgeAccel, Biological age acceleration; TyG, Triglyceride-glucose index; WC, Waist circumference; WHtR, Waist-to-height ratio; BMI, Body mass index; BRI, Body roundness index; ABSI, A body shape index; WWI, Weight-adjusted waist index; CVAI, Chinese visceral adiposity index; CKM, Cardiovascular-kidney-metabolic; METS-IR, Metabolic score for insulin resistance; CTI, C-reactive protein-triglyceride glucose index; eGDR, Estimated glucose disposal rate; TG/HDL-C, Triglyceride/high-density lipoprotein cholesterol ratio; SD, Standard deviation; SMD, Standardized mean difference

The P values marked in bold indicate statistical significance

Strong correlations were observed between IR surrogate indices measured at Wave 1 and at Wave 3. Moreover, the cumulative IR surrogate indices exhibited pronounced interrelationships with one another, as illustrated in Fig. 2A. Both biological ageing metrics, including KDM BioAge and Light BioAge, demonstrated a robust positive association with chronological age (all P < 0.001). We calculated biological age difference (BioAgeDiff) as the difference between biological age estimates and chronological age to estimate the magnitude of how biological age deviates from chronological age. BioAgeDiff values were distributed symmetrically and uniformly around zero across the full chronological age range, demonstrating the absence of any systematic directional bias in biological age deviation relative to chronological age (Fig. 2B). In addition, both biological age measures were significantly higher among individuals with metabolic disorders (e.g., obesity, hypertension, dyslipidaemia) than among those without these conditions (Fig. 2C).

Fig. 2.

Relationships among cumulative insulin resistance surrogate indices and distributions of biological age measures. A Pearson correlation heatmap illustrating pairwise correlations among cumulative IR surrogate indices and correlations between measurements at Wave 1 and Wave 3 for each index. B Scatterplots of biological age and BioAgeDiff against chronological age for KDM-BA and Light-BA. C Comparisons of KDM-BA and Light-BA between participants with and without metabolic disorders

K-means classification of longitudinal patterns of IR surrogate indices

Using K-means clustering combined with the elbow criterion, distinct longitudinal pattern groups of insulin resistance surrogate indices were identified based on measurements obtained in 2012 and 2015 (Fig. 3A). For TyG-BMI, TyG-WC, TyG-WHtR, TyG-BRI, TyG-CVAI, and METS-IR, four ordered pattern groups were observed, corresponding to low-stable (Cluster 1), lower-stable (Cluster 2), higher-stable (Cluster 3), and high-stable levels (Cluster 4) across the two visits. In contrast, TyG, TyG-ABSI, CTI, and TG/HDL-C exhibited more dynamic patterns, comprising a low-stable pattern (Cluster 1), a decreasing pattern (Cluster 2), an increasing pattern (Cluster 3), and a high-stable pattern (Cluster 4). TyG-WWI and eGDR demonstrated greater heterogeneity and were classified into six distinct pattern groups. Density plots depicting the distributions of TyG, TyG-derived indices, METS-IR, CTI, and TG/HDL-C at both time points illustrate clear separation across pattern groups, with marked differences in central tendency and dispersion, supporting the robustness of the clustering results (Fig. 3B). Additional clustering metrics supported the overall quality and stability of the derived pattern groups (Table S10), with silhouette coefficients ranging from 0.414 to 0.574, Davies-Bouldin indices from 0.731 to 0.987, mean Jaccard similarity from 0.767 to 0.982, and minimum Jaccard similarity from 0.690 to 0.975.

Fig. 3.

Longitudinal patterns of insulin resistance surrogate indices identified by K-means clustering. A Pattern groups for each IR surrogate index based on measurements at Wave 1 (2012) and Wave 3 (2015). For TyG-BMI, TyG-WC, TyG-WHtR, TyG-BRI, TyG-CVAI, and METS-IR, four stable pattern groups were identified: low-stable (Cluster 1), lower-stable (Cluster 2), higher-stable (Cluster 3), and high-stable levels (Cluster 4). For TyG, TyG-ABSI, CTI, and TG/HDL-C, four dynamic groups were observed: low-stable pattern (Cluster 1), a decreasing pattern (Cluster 2), an increasing pattern (Cluster 3), and a high-stable pattern (Cluster 4). TyG-WWI and eGDR exhibited greater heterogeneity and were classified into six distinct groups. B Density plots showing the distribution of TyG, TyG-derived indices, METS-IR, CTI, and TG/HDL-C at Wave 1 and Wave 3 across the identified clusters. Clear separation in central tendency and dispersion across clusters supports the robustness of the clustering results

Associations among cumulative IR surrogate indices, pattern groups, and new-onset CVD risk

Among individuals with CKM syndrome stages 0–3, Fine–Gray competing risk models were used to quantify the associations between cumulative IR surrogate indices and incident CVD, treating death as a competing event and analyzing the indices both as continuous variables and as quartiles (Table 2). Across all models, higher cumulative levels of TyG-based indices, METS-IR, CTI and TG/HDL-C were consistently related to greater risk of incident CVD, whereas cumulative eGDR was inversely associated with CVD risk, with progressively lower risk at higher eGDR values. These patterns remained statistically robust after stepwise multivariable adjustment (all P for trend < 0.05). We also evaluated the associations between longitudinal pattern groups of IR surrogate indices and CVD (Table 3). We observed that the least favorable pattern groups of TyG, TyG-based indices, METS-IR, CTI and TG/HDL-C exhibited the strongest associations with CVD, whereas the most favorable pattern groups of eGDR showed significant inverse associations with CVD (all P < 0.05). ROC analysis in Fig. S1 and Table S11 showed that the cumulative eGDR exhibited relatively superior predictive performance (AUC: 0.613) for CVD risk compared to other indices (AUC range: 0.525–0.577, all DeLong test P < 0.05) among individuals with CKM stages 0–3. Pairwise NRI and IDI analyses consistently indicated that cumulative eGDR provided greater incremental predictive value than the other individual indices, including both single-wave and cumulative measures (Tables S12–S13; all P < 0.05). DCA further showed that both cumulative eGDR and single-wave eGDR yielded greater net benefit than the other indices over a broad range of threshold probabilities (Fig. S2). Both WQS and Qgcomp analyses yielded consistent results, identifying eGDR as the top contributor to CVD risk prediction (Fig. S3).

Table 2.

Association between cumulative IR surrogate Indices and CVD incidence in individuals with CKM syndrome stages 0–3

Cumulative IR surrogate indices (quartile)	Total (n)	Event (%)	Crude model		Model 1		Model 2		Model 3
			sHR (95% CI)	P	sHR (95% CI)	P	sHR (95% CI)	P	sHR (95% CI)	P
Cumulative TyG
Standardized continuous	3948	756 (19.1)	1.16 (1.08, 1.25)	< 0.001	1.17 (1.09, 1.26)	< 0.001	1.17 (1.09, 1.26)	< 0.001	1.10 (1.00, 1.22)	0.050
Q1	987	153 (20.2)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	197 (26.1)	1.30 (1.05, 1.60)	0.016	1.26 (1.02, 1.56)	0.033	1.27 (1.02, 1.57)	0.030	1.21 (0.97, 1.50)	0.090
Q3	987	182 (24.1)	1.24 (1.00, 1.54)	0.052	1.22 (0.98, 1.52)	0.069	1.23 (0.99, 1.53)	0.068	1.15 (0.89, 1.47)	0.288
Q4	987	224 (29.6)	1.53 (1.25, 1.88)	< 0.001	1.56 (1.27, 1.92)	< 0.001	1.57 (1.27, 1.93)	< 0.001	1.34 (1.01, 1.79)	0.044
Ptrend			< 0.001		< 0.001		< 0.001		0.070
Cumulative TyG-BMI
Standardized continuous	3948	756 (19.1)	1.19 (1.10, 1.28)	< 0.001	1.24 (1.15, 1.33)	< 0.001	1.25 (1.16, 1.34)	< 0.001	1.15 (1.05, 1.25)	0.002
Q1	987	143 (18.9)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	191 (25.3)	1.31 (1.05, 1.62)	0.016	1.36 (1.10, 1.70)	0.005	1.38 (1.10, 1.71)	0.004	1.27 (1.01, 1.59)	0.038
Q3	987	181 (23.9)	1.22 (0.98, 1.52)	0.076	1.32 (1.06, 1.65)	0.014	1.34 (1.07, 1.68)	0.011	1.18 (0.93, 1.50)	0.182
Q4	987	241 (31.9)	1.70 (1.38, 2.09)	< 0.001	1.91 (1.54, 2.36)	< 0.001	1.95 (1.57, 2.42)	< 0.001	1.58 (1.23, 2.02)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		0.001
Cumulative TyG-WC
Standardized continuous	3948	756 (19.1)	1.26 (1.17, 1.36)	< 0.001	1.27 (1.18, 1.37)	< 0.001	1.28 (1.19, 1.38)	< 0.001	1.20 (1.09, 1.31)	< 0.001
Q1	987	138 (18.3)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	191 (25.3)	1.41 (1.13, 1.76)	0.002	1.39 (1.12, 1.74)	0.003	1.41 (1.13, 1.75)	0.002	1.32 (1.05, 1.65)	0.016
Q3	987	182 (24.1)	1.31 (1.05, 1.63)	0.018	1.31 (1.05, 1.63)	0.019	1.33 (1.06, 1.66)	0.014	1.20 (0.94, 1.52)	0.139
Q4	987	245 (32.4)	1.89 (1.54, 2.33)	< 0.001	1.94 (1.57, 2.39)	< 0.001	1.97 (1.59, 2.44)	< 0.001	1.66 (1.29, 2.13)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		< 0.001
Cumulative TyG-WHtR
Standardized continuous	3948	756 (19.1)	1.27 (1.17, 1.37)	< 0.001	1.25 (1.16, 1.36)	< 0.001	1.26 (1.16, 1.37)	< 0.001	1.16 (1.06, 1.28)	0.002
Q1	987	142 (18.8)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	172 (22.8)	1.20 (0.96, 1.50)	0.106	1.18 (0.94, 1.48)	0.145	1.18 (0.95, 1.48)	0.141	1.08 (0.86, 1.36)	0.525
Q3	987	190 (25.1)	1.32 (1.06, 1.64)	0.013	1.30 (1.04, 1.63)	0.020	1.32 (1.06, 1.66)	0.015	1.17 (0.92, 1.49)	0.188
Q4	987	252 (33.3)	1.86 (1.51, 2.28)	< 0.001	1.80 (1.45, 2.24)	< 0.001	1.83 (1.47, 2.27)	< 0.001	1.49 (1.15, 1.92)	0.002
Ptrend			< 0.001		< 0.001		< 0.001		0.002
Cumulative TyG-BRI
Standardized continuous	3948	756 (19.1)	1.26 (1.17, 1.37)	< 0.001	1.24 (1.14, 1.35)	< 0.001	1.25 (1.15, 1.36)	< 0.001	1.15 (1.05, 1.26)	0.004
Q1	987	142 (18.8)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	177 (23.4)	1.24 (1.00, 1.55)	0.055	1.22 (0.97, 1.52)	0.085	1.22 (0.98, 1.53)	0.078	1.10 (0.87, 1.39)	0.427
Q3	987	191 (25.3)	1.31 (1.05, 1.63)	0.016	1.29 (1.03, 1.61)	0.024	1.31 (1.04, 1.63)	0.020	1.13 (0.90, 1.43)	0.293
Q4	987	246 (32.5)	1.81 (1.48, 2.23)	< 0.001	1.73 (1.39, 2.17)	< 0.001	1.76 (1.40, 2.21)	< 0.001	1.40 (1.08, 1.80)	0.009
Ptrend			< 0.001		< 0.001		< 0.001		0.008
Cumulative TyG-ABSI
Standardized continuous	3948	756 (19.1)	1.25 (1.16, 1.34)	< 0.001	1.19 (1.11, 1.28)	< 0.001	1.19 (1.11, 1.28)	< 0.001	1.12 (1.02, 1.22)	0.014
Q1	987	150 (19.8)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	171 (22.6)	1.17 (0.94, 1.46)	0.157	1.11 (0.89, 1.38)	0.373	1.12 (0.89, 1.39)	0.328	1.06 (0.85, 1.33)	0.617
Q3	987	193 (25.5)	1.35 (1.09, 1.67)	0.006	1.27 (1.02, 1.58)	0.029	1.28 (1.03, 1.59)	0.026	1.18 (0.93, 1.49)	0.169
Q4	987	242 (32.0)	1.79 (1.46, 2.19)	< 0.001	1.57 (1.27, 1.95)	< 0.001	1.58 (1.28, 1.95)	< 0.001	1.37 (1.06, 1.77)	0.015
Ptrend			< 0.001		< 0.001		< 0.001		0.010
Cumulative TyG-WWI
Standardized continuous	3948	756 (19.1)	1.27 (1.18, 1.36)	< 0.001	1.21 (1.12, 1.31)	< 0.001	1.21 (1.12, 1.31)	< 0.001	1.13 (1.02, 1.24)	0.017
Q1	987	139 (18.4)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	179 (23.7)	1.31 (1.05, 1.64)	0.017	1.26 (1.01, 1.58)	0.040	1.26 (1.01, 1.58)	0.040	1.15 (0.92, 1.45)	0.229
Q3	987	205 (27.1)	1.50 (1.21, 1.86)	< 0.001	1.41 (1.13, 1.76)	0.002	1.42 (1.14, 1.77)	0.002	1.24 (0.98, 1.57)	0.076
Q4	987	233 (30.8)	1.81 (1.46, 2.23)	< 0.001	1.61 (1.28, 2.01)	< 0.001	1.61 (1.29, 2.02)	< 0.001	1.30 (1.01, 1.69)	0.047
Ptrend			< 0.001		< 0.001		< 0.001		0.037
Cumulative TyG-CVAI
Standardized continuous	3948	756 (19.1)	1.29 (1.18, 1.41)	< 0.001	1.24 (1.13, 1.35)	< 0.001	1.25 (1.14, 1.37)	< 0.001	1.15 (1.04, 1.28)	0.007
Q1	987	130 (17.2)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	185 (24.5)	1.50 (1.20, 1.88)	< 0.001	1.44 (1.15, 1.80)	0.002	1.45 (1.15, 1.82)	0.001	1.36 (1.07, 1.71)	0.010
Q3	987	193 (25.5)	1.54 (1.24, 1.93)	< 0.001	1.44 (1.15, 1.81)	0.002	1.47 (1.17, 1.85)	0.001	1.34 (1.05, 1.70)	0.018
Q4	987	248 (32.8)	2.08 (1.68, 2.57)	< 0.001	1.88 (1.51, 2.34)	< 0.001	1.92 (1.54, 2.40)	< 0.001	1.61 (1.24, 2.08)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		< 0.001
Cumulative METS-IR
Standardized continuous	3948	756 (19.1)	1.16 (1.08, 1.25)	< 0.001	1.20 (1.12, 1.30)	< 0.001	1.21 (1.13, 1.31)	< 0.001	1.12 (1.03, 1.22)	0.008
Q1	987	155 (20.5)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	181 (23.9)	1.15 (0.92, 1.42)	0.216	1.21 (0.97, 1.50)	0.088	1.22 (0.98, 1.51)	0.072	1.14 (0.91, 1.42)	0.253
Q3	987	188 (24.9)	1.19 (0.97, 1.48)	0.102	1.28 (1.04, 1.59)	0.022	1.31 (1.05, 1.63)	0.015	1.18 (0.94, 1.49)	0.153
Q4	987	232 (30.7)	1.51 (1.23, 1.85)	< 0.001	1.67 (1.35, 2.05)	< 0.001	1.70 (1.38, 2.10)	< 0.001	1.40 (1.10, 1.78)	0.006
Ptrend			< 0.001		< 0.001		< 0.001		0.007
Cumulative CTI
Standardized continuous	3948	756 (19.1)	1.24 (1.16, 1.34)	< 0.001	1.23 (1.15, 1.33)	< 0.001	1.23 (1.15, 1.33)	< 0.001	1.18 (1.08, 1.29)	< 0.001
Q1	987	145 (19.2)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	178 (23.5)	1.26 (1.01, 1.57)	0.039	1.23 (0.99, 1.54)	0.061	1.23 (0.99, 1.54)	0.063	1.17 (0.94, 1.47)	0.162
Q3	987	203 (26.9)	1.43 (1.16, 1.78)	< 0.001	1.38 (1.12, 1.71)	0.003	1.38 (1.11, 1.71)	0.003	1.25 (0.99, 1.57)	0.056
Q4	987	230 (30.4)	1.74 (1.42, 2.15)	< 0.001	1.71 (1.39, 2.11)	< 0.001	1.71 (1.39, 2.11)	< 0.001	1.51 (1.19, 1.93)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		< 0.001
Cumulative eGDR
Standardized continuous	3948	756 (19.1)	0.70 (0.66, 0.75)	< 0.001	0.72 (0.67, 0.76)	< 0.001	0.71 (0.67, 0.76)	< 0.001	0.77 (0.70, 0.84)	< 0.001
Q1	987	257 (34.0)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	212 (28.0)	0.80 (0.67, 0.96)	0.018	0.79 (0.66, 0.95)	0.013	0.79 (0.66, 0.95)	0.011	0.83 (0.69, 1.00)	0.056
Q3	987	178 (23.5)	0.64 (0.53, 0.78)	< 0.001	0.66 (0.54, 0.80)	< 0.001	0.65 (0.54, 0.79)	< 0.001	0.72 (0.58, 0.89)	0.003
Q4	987	109 (14.4)	0.38 (0.30, 0.47)	< 0.001	0.40 (0.32, 0.50)	< 0.001	0.40 (0.32, 0.50)	< 0.001	0.50 (0.36, 0.69)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		< 0.001
Cumulative TG/HDL-C
Standardized continuous	3948	756 (19.1)	1.07 (1.00, 1.15)	0.038	1.10 (1.02, 1.18)	0.009	1.10 (1.02, 1.18)	0.008	1.05 (0.96, 1.15)	0.312
Q1	987	163 (21.6)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	987	196 (25.9)	1.20 (0.98, 1.48)	0.079	1.20 (0.97, 1.47)	0.093	1.21 (0.98, 1.49)	0.079	1.22 (0.98, 1.50)	0.073
Q3	987	191 (25.3)	1.22 (0.99, 1.50)	0.067	1.24 (1.01, 1.53)	0.044	1.25 (1.01, 1.55)	0.037	1.20 (0.94, 1.53)	0.147
Q4	987	206 (27.2)	1.28 (1.04, 1.57)	0.019	1.35 (1.10, 1.66)	0.005	1.35 (1.10, 1.67)	0.004	1.25 (0.94, 1.66)	0.119
Ptrend			0.024		0.005		0.005		0.128

Q1: Quartile 1, Q2: Quartile 2, Q3: Quartile 3, Q4: Quartile 4. Standardized continuous: denotes analyses of z-standardized cumulative IR surrogate indices; the corresponding sHRs represent the relative change in incident CVD associated with a 1 SD increase in each index

Crude model: unadjusted for covariates; model 1: age, gender; model 2: age, gender, residence, smoking, alcohol consumption, marital status and educational attainment; model 3: age, gender, residence, smoking, alcohol consumption, marital status, educational attainment, diabetes, hypertension, hypertriglyceridaemia, and CKD

The P values marked in bold indicate statistical significance

Table 3.

Association between Change in the IR surrogate Indices and CVD incidence in individuals with CKM syndrome stages 0–3

Change in the IR surrogate indices	Total (%)	Event (%)	Crude model		Model 1		Model 2		Model 3
			sHR (95% CI)	P	sHR (95% CI)	P	sHR (95% CI)	P	sHR (95% CI)	P
Change in the TyG
Cluster 1	1361 (34.5)	228 (30.2)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	935 (23.7)	181 (23.9)	1.19 (0.98, 1.44)	0.086	1.16 (0.96, 1.42)	0.129	1.17 (0.96, 1.42)	0.123	1.07 (0.83, 1.37)	0.605
Cluster 3	1017 (25.8)	198 (26.2)	1.21 (1.00, 1.47)	0.047	1.21 (1.00, 1.47)	0.050	1.22 (1.00, 1.47)	0.048	1.13 (0.92, 1.39)	0.257
Cluster 4	635 (16.1)	149 (19.7)	1.47 (1.20, 1.81)	< 0.001	1.52 (1.23, 1.87)	< 0.001	1.52 (1.23, 1.87)	< 0.001	1.24 (0.91, 1.68)	0.166
Ptrend			< 0.001		< 0.001		< 0.001		0.039
Change in the TyG-BMI
Cluster 1	775 (19.6)	116 (15.3)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1241 (31.4)	225 (29.8)	1.17 (0.93, 1.46)	0.178	1.23 (0.98, 1.54)	0.078	1.24 (0.99, 1.56)	0.065	1.14 (0.90, 1.43)	0.277
Cluster 3	1241 (31.4)	248 (32.8)	1.30 (1.04, 1.63)	0.019	1.42 (1.13, 1.78)	0.002	1.45 (1.15, 1.82)	0.001	1.24 (0.97, 1.58)	0.090
Cluster 4	691 (17.5)	167 (22.1)	1.61 (1.27, 2.04)	< 0.001	1.81 (1.42, 2.31)	< 0.001	1.86 (1.45, 2.38)	< 0.001	1.43 (1.08, 1.90)	0.011
Ptrend			< 0.001		< 0.001		< 0.001		0.010
Change in the TyG-WC
Cluster 1	667 (16.9)	88 (11.6)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1203 (30.5)	222 (29.4)	1.42 (1.11, 1.82)	0.005	1.43 (1.12, 1.84)	0.004	1.44 (1.12, 1.84)	0.004	1.36 (1.06, 1.76)	0.016
Cluster 3	1319 (33.4)	240 (31.7)	1.37 (1.07, 1.76)	0.011	1.39 (1.08, 1.78)	0.009	1.41 (1.10, 1.80)	0.007	1.27 (0.98, 1.65)	0.070
Cluster 4	759 (19.2)	206 (27.2)	2.20 (1.71, 2.83)	< 0.001	2.28 (1.77, 2.93)	< 0.001	2.33 (1.80, 3.00)	< 0.001	1.95 (1.45, 2.62)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		< 0.001
Change in the TyG-WHtR
Cluster 1	767 (19.4)	97 (12.8)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1206 (30.5)	219 (29.0)	1.47 (1.15, 1.87)	0.002	1.44 (1.13, 1.84)	0.003	1.45 (1.13, 1.84)	0.003	1.32 (1.03, 1.69)	0.028
Cluster 3	1207 (30.6)	234 (31.0)	1.54 (1.22, 1.96)	< 0.001	1.52 (1.19, 1.94)	< 0.001	1.54 (1.21, 1.97)	< 0.001	1.36 (1.05, 1.77)	0.020
Cluster 4	768 (19.5)	206 (27.2)	2.29 (1.79, 2.91)	< 0.001	2.22 (1.72, 2.86)	< 0.001	2.25 (1.75, 2.90)	< 0.001	1.84 (1.37, 2.46)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		< 0.001
Change in the TyG-BRI
Cluster 1	562 (14.2)	79 (10.4)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1093 (27.7)	178 (23.5)	1.17 (0.90, 1.53)	0.244	1.18 (0.90, 1.55)	0.220	1.18 (0.90, 1.54)	0.220	1.10 (0.84, 1.44)	0.495
Cluster 3	1327 (33.6)	259 (34.3)	1.37 (1.06, 1.76)	0.015	1.36 (1.06, 1.76)	0.018	1.38 (1.06, 1.78)	0.015	1.18 (0.90, 1.54)	0.234
Cluster 4	966 (24.5)	240 (31.7)	1.87 (1.45, 2.42)	< 0.001	1.81 (1.38, 2.37)	< 0.001	1.83 (1.40, 2.40)	< 0.001	1.43 (1.07, 1.92)	0.016
Ptrend			< 0.001		< 0.001		< 0.001		0.008
Change in the TyG-ABSI
Cluster 1	960 (24.3)	146 (19.3)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	930 (23.6)	160 (21.2)	1.16 (0.93, 1.45)	0.198	1.07 (0.85, 1.34)	0.556	1.08 (0.86, 1.35)	0.514	1.00 (0.79, 1.26)	0.972
Cluster 3	1207 (30.6)	249 (32.9)	1.42 (1.16, 1.74)	< 0.001	1.32 (1.07, 1.62)	0.009	1.33 (1.08, 1.63)	0.008	1.19 (0.96, 1.48)	0.119
Cluster 4	851 (21.6)	201 (26.6)	1.71 (1.38, 2.12)	< 0.001	1.49 (1.19, 1.86)	< 0.001	1.49 (1.20, 1.86)	< 0.001	1.21 (0.93, 1.58)	0.157
Ptrend			< 0.001		< 0.001		< 0.001		0.043
Change in the TyG-WWI
Cluster 1	616 (15.6)	83 (11.0)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	581 (14.7)	108 (14.3)	1.42 (1.07, 1.90)	0.016	1.33 (1.00, 1.78)	0.049	1.34 (1.00, 1.78)	0.048	1.26 (0.94, 1.69)	0.127
Cluster 3	868 (22.0)	144 (19.0)	1.27 (0.97, 1.66)	0.083	1.22 (0.92, 1.60)	0.162	1.21 (0.92, 1.60)	0.170	1.19 (0.83, 1.45)	0.527
Cluster 4	800 (20.3)	171 (22.6)	1.64 (1.26, 2.14)	< 0.001	1.49 (1.14, 1.96)	0.004	1.50 (1.14, 1.96)	0.003	1.31 (0.95, 1.76)	0.099
Cluster 5	639 (16.2)	133 (17.6)	1.64 (1.25, 2.16)	< 0.001	1.51 (1.13, 2.01)	0.005	1.51 (1.14, 2.01)	0.005	1.39 (0.97, 1.76)	0.077
Cluster 6	444 (11.2)	117 (15.5)	2.19 (1.66, 2.91)	< 0.001	1.87 (1.39, 2.51)	< 0.001	1.88 (1.40, 2.53)	< 0.001	1.47 (1.04, 2.08)	0.030
Ptrend			< 0.001		< 0.001		< 0.001		0.040
Change in the TyG-CVAI
Cluster 1	425 (10.8)	60 (7.9)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1041 (26.4)	168 (22.2)	1.14 (0.85, 1.54)	0.377	1.08 (0.80, 1.45)	0.612	1.09 (0.81, 1.47)	0.561	1.06 (0.78, 1.43)	0.723
Cluster 3	1415 (35.8)	261 (34.5)	1.32 (1.00, 1.75)	0.051	1.19 (0.90, 1.59)	0.228	1.21 (0.91, 1.62)	0.187	1.11 (0.82, 1.50)	0.505
Cluster 4	1067 (27.0)	267 (35.3)	1.88 (1.42, 2.49)	< 0.001	1.64 (1.23, 2.19)	< 0.001	1.68 (1.25, 2.25)	< 0.001	1.38 (1.00, 1.91)	0.050
Ptrend			< 0.001		< 0.001		< 0.001		0.018
Change in the METS-IR
Cluster 1	760 (19.3)	115 (15.2)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1281 (32.4)	234 (31.0)	1.17 (0.93, 1.46)	0.174	1.24 (0.99, 1.55)	0.060	1.26 (1.00, 1.57)	0.046	1.16 (0.93, 1.46)	0.193
Cluster 3	1251 (31.7)	241 (31.9)	1.24 (0.99, 1.55)	0.059	1.34 (1.07, 1.68)	0.010	1.37 (1.09, 1.72)	0.006	1.20 (0.94, 1.53)	0.137
Cluster 4	656 (16.6)	166 (22.0)	1.69 (1.33, 2.14)	< 0.001	1.93 (1.51, 2.46)	< 0.001	1.98 (1.55, 2.53)	< 0.001	1.60 (1.21, 2.11)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		0.002
Change in the CTI
Cluster 1	898 (22.7)	129 (17.1)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1073 (27.2)	221 (29.2)	1.49 (1.20, 1.85)	< 0.001	1.43 (1.15, 1.78)	0.001	1.43 (1.15, 1.78)	0.001	1.31 (1.04, 1.66)	0.023
Cluster 3	1043 (26.4)	187 (24.7)	1.28 (1.02, 1.60)	0.033	1.28 (1.02, 1.60)	0.034	1.27 (1.01, 1.59)	0.038	1.22 (0.97, 1.53)	0.089
Cluster 4	934 (23.7)	219 (29.0)	1.81 (1.46, 2.25)	< 0.001	1.77 (1.42, 2.20)	< 0.001	1.77 (1.42, 2.20)	< 0.001	1.56 (1.21, 2.01)	< 0.001
Ptrend			< 0.001		< 0.001		< 0.001		0.003
Change in the eGDR
Cluster 1	138 (3.5)	62 (8.2)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	829 (21.0)	192 (25.4)	0.46 (0.35, 0.61)	< 0.001	0.47 (0.35, 0.63)	< 0.001	0.47 (0.35, 0.62)	< 0.001	0.56 (0.41, 0.76)	< 0.001
Cluster 3	365 (9.2)	63 (8.3)	0.32 (0.23, 0.45)	< 0.001	0.35 (0.24, 0.49)	< 0.001	0.34 (0.24, 0.49)	< 0.001	0.63 (0.32, 1.22)	0.170
Cluster 4	959 (24.3)	206 (27.2)	0.42 (0.32, 0.56)	< 0.001	0.42 (0.32, 0.56)	< 0.001	0.42 (0.31, 0.55)	< 0.001	0.51 (0.37, 0.70)	< 0.001
Cluster 5	685 (17.4)	124 (16.4)	0.34 (0.25, 0.46)	< 0.001	0.35 (0.26, 0.48)	< 0.001	0.35 (0.25, 0.47)	< 0.001	0.42 (0.30, 0.58)	< 0.001
Cluster 6	972 (24.6)	109 (14.4)	0.20 (0.15, 0.27)	< 0.001	0.22 (0.16, 0.30)	< 0.001	0.22 (0.16, 0.30)	< 0.001	0.40 (0.21, 0.77)	0.006
Ptrend			< 0.001		< 0.001		< 0.001		< 0.001
Change in the TG/HDL-C
Cluster 1	1142 (28.9)	193 (25.5)	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Cluster 2	1134 (28.7)	226 (29.9)	1.19 (0.98, 1.44)	0.075	1.20 (0.99, 1.46)	0.064	1.20 (0.99, 1.46)	0.058	1.17 (0.94, 1.45)	0.153
Cluster 3	1061 (26.9)	209 (27.6)	1.19 (0.98, 1.45)	0.076	1.23 (1.01, 1.50)	0.038	1.24 (1.01, 1.51)	0.037	1.19 (0.96, 1.47)	0.120
Cluster 4	611 (15.5)	128 (16.9)	1.25 (1.00, 1.57)	0.047	1.33 (1.06, 1.66)	0.013	1.33 (1.06, 1.67)	0.012	1.18 (0.88, 1.58)	0.263
Ptrend			0.039		0.010		0.010		0.223

Crude model: unadjusted for covariates; model 1: age, gender; model 2: age, gender, residence, smoking, alcohol consumption, marital status and educational attainment; model 3: age, gender, residence, smoking, alcohol consumption, marital status, educational attainment, diabetes, hypertension, hypertriglyceridaemia, and CKD

The P values marked in bold indicate statistical significance

The RCS model analysis further evaluated the association between cumulative IR indices and incident CVD, with all models adjusted for potential confounders (Fig. 4). All cumulative IR surrogate indices showed strong overall associations with new-onset CVD (P _overall < 0.05). For cumulative TyG-BRI and cumulative TyG-CVAI, non-linear associations with incident CVD were observed (P _nonlinearity = 0.043 and 0.031, respectively), with approximate inflection points at 111.138 and 2675.647; beyond these thresholds, CVD risk increased more steeply. The remaining indices exhibited approximately monotonic risk gradients across their observed ranges.

Fig. 4.

The RCS analysis between the cumulative IR surrogate indices and CVD incidence in a population with CKM syndrome stages 0–3. The model was adjusted for age, gender, residence, smoking, alcohol consumption, marital status, educational attainment, diabetes, hypertension, hypertriglyceridaemia, and CKD

Mediation analyses and subgroup analyses

Mediation models suggested that both KDM and Light BioAgeAccel partially explained the associations of cumulative IR indices with incident CVD (Fig. 5; Table S14). For cumulative TyG-BMI, TyG-WC, TyG-WHtR, TyG-BRI, TyG-ABSI, TyG-WWI, TyG-CVAI, METS‑IR, and CTI, indirect effects mediated through KDM BioAgeAccel were significant and explained 29.5–45.8% of the total association (all P < 0.05). Meanwhile, cumulative TyG, TyG-BMI, TyG-BRI, TyG-ABSI, TyG-CVAI, and METS‑IR showed significant mediation via Light BioAgeAccel, with proportions ranging from 11.2% to 25.9% (all P < 0.05). Moreover, stage-stratified mediation showed no significant indirect effects in CKM stages 0–1, whereas mediation remained evident in the more advanced stages (Tables S15–S16).

Fig. 5.

Mediating role of BioAgeAccel in the associations between cumulative IR surrogate indices and incident CVD. A Mediation by KDM BioAgeAccel. B Mediation by Light BioAgeAccel. ADE, average direct effect; ACME, average causal mediation effect; PE, proportion mediated (PE = ACME/[ADE + ACME]). All models were adjusted for age, gender, residence, smoking, alcohol consumption, marital status, educational attainment, diabetes, hypertension, hypertriglyceridaemia, and CKD. *P < 0.05; **P < 0.01; ***P < 0.001

To evaluate potential heterogeneity, we conducted stratified analyses across KDM BioAgeAccel, Light BioAgeAccel, gender, smoking and drinking status, overweight/obesity, hypertension, diabetes, hypertriglyceridemia, subclinical CVD, and CKM stages (Tables S17–S27). Both biological age measures interacted significantly with several IR indices (P_FDR for interaction < 0.05). Specifically, higher cumulative TyG-BMI, TyG-WHtR, and TyG-CVAI conferred stronger CVD risk in the KDM BioAge‑younger subgroup than in the KDM BioAge‑older subgroup. Similarly, elevated cumulative TyG-WHtR was more strongly associated with CVD in the Light BioAge‑younger subgroup. No marked heterogeneity was observed across other stratification variables.

Sensitivity analyses

Sensitivity analyses were performed to evaluate the robustness of the findings.

1. Missing-data robustness: Participants excluded because of missing baseline covariates were broadly comparable to those included in the final analytic cohort (Table S28). After multiple imputation by chained equations for participants with missing baseline covariates, the expanded analytic sample (n = 6630) showed a baseline distribution of participants with and without incident CVD that was similar to that observed in the primary analysis (Table S29).

2. Weighted reanalysis: In the imputed datasets, associations of cumulative IR surrogate indices and their pattern groups with incident CVD remained materially unchanged after inverse probability of treatment weighting and refitting of the primary Fine–Gray models (Tables S30–S31), indicating that the main findings were unlikely to be driven by residual imbalance in baseline risk profiles.

3. Time-varying medication adjustment: Additional Cox models incorporating follow-up use of antihypertensive, lipid-lowering, and glucose-lowering medications as time-varying covariates yielded results consistent with the primary analysis across the crude and adjusted models (Tables S32–S33).

4. Unmeasured confounding: E-value analyses supported the robustness of the observed cumulative and pattern-group associations to potential unmeasured confounding (Tables S34–S35). In the fully adjusted model, the largest E-values were observed for cumulative eGDR [Q4: 3.40 (2.26)] and the most adverse eGDR pattern group [Cluster 6: 4.39 (2.09)], followed by TyG-WC, TyG-CVAI, and TyG-BMI for their corresponding associations.

5. Cumulative-exposure validity: Cumulative IR surrogate indices were significantly correlated with their corresponding 2012, 2015, and mean values (Fig. S4). Cumulative IR surrogates showed larger SMDs and Cohen’s d between CVD and non‑CVD groups than their 2012 or 2015 counterparts (Table S36). Moreover, cumulative-index models showed the highest discriminatory performance for incident CVD among the three full models (AUC: 0.629 [0.607–0.651]), and the same pattern was observed for the PCA-based models (cumPCA: 0.620 [0.598–0.642]), supporting the rationale for using cumulative exposure rather than relying solely on single-time-point measures (Fig. S5).

6. Stroke-specific outcome: During follow-up, 262 participants developed incident stroke. Baseline characteristics according to stroke status are presented in Table S37. Fine–Gray models showed that higher cumulative levels and adverse pattern groups of IR surrogate indices were also significantly associated with greater stroke risk (Tables S38–S39). In addition, both KDM BioAgeAccel and Light BioAgeAccel significantly mediated the associations between several cumulative IR surrogate indices and incident stroke (Table S40).

7. Baseline- and change-based exposure: When the analyses were repeated using baseline IR surrogate indices alone and absolute changes between 2012 and 2015, the associations with incident CVD remained directionally consistent, and the corresponding AUCs were also significant; however, the discriminatory performance was consistently lower than that of the corresponding cumulative indices (Table S41, Fig. S1).

8. Lag analysis: After excluding participants who developed incident CVD before 2016 (n = 247), the main associations of cumulative IR surrogate indices and their pattern groups with incident CVD remained essentially unchanged (Tables S42–S43), arguing against substantial reverse causation.

Discussion

In this nationwide longitudinal cohort of adults with CKM syndrome stages 0–3, we demonstrate that sustained IR burden, captured by both cumulative exposure and data-driven longitudinal pattern groups across multiple surrogate indices, is robustly associated with accelerated biological ageing and increased risk of incident CVD. Higher long-term levels and adverse longitudinal pattern groups of TyG-based indices, METS-IR, CTI, and TG/HDL-C were consistently linked to excess CVD risk, whereas favorable eGDR patterns were protective. Importantly, biological age acceleration quantified by both KDM and Light ageing clocks showed partially mediating patterns in these associations, supporting ageing-related pathways as one plausible biological explanation linking metabolic dysfunction to cardiovascular events. The stronger associations observed in biologically younger individuals further underscore the value of early dynamic risk profiling. Our findings support longitudinal monitoring of IR dynamics as a clinically actionable strategy for early cardiovascular risk stratification and prevention across the preclinical CKM spectrum.

Most prior evidence has focused on cross-sectional associations between baseline IR surrogate levels and subsequent CVD risk [18, 40, 53–56]. Only a small body of recent work has begun to explore whether longitudinal features of selected indices add prognostic information beyond single-time-point assessments [20, 57–59]. For example, higher cumulative CTI and persistently elevated CTI trajectories have been linked to greater stroke risk [57]; similarly, adverse cumulative profiles of AIP and its modified derivatives [58, 59], cumTyG and its obesity-related extensions [20] have been associated with increased CVD events among individuals on the CKM spectrum. Wang et al. further reported an inverse association between cumulative eGDR and stroke risk across CKM stages 0 to 4 [60]. However, comprehensive evaluations that simultaneously integrate a broad panel of IR surrogates, leverage both cumulative burden and data-driven trajectory phenotypes, and directly compare their predictive performance remain limited, particularly in CKM stages 0 to 3. Addressing this gap, our results indicate that sustained exposure and unfavourable longitudinal patterns across multiple IR indices are consistently related to higher incident CVD risk, with discernible differences in discrimination across indices. The identification of non-linear associations and threshold effects further implies that keeping cumulative TyG-BRI or TyG-CVAI within favorable ranges may be an important modifiable target for reducing CVD risk. Mechanistically, IR may promote cardiovascular injury through converging pathways. Elevated free fatty acids and chronic hyperglycaemia can impair mitochondrial function, heighten oxidative stress, and reduce nitric oxide bioavailability, compromising endothelial integrity and vasomotor control [61–63]. Endothelial dysfunction then facilitates lipid accumulation and inflammatory cell recruitment, accelerating atherogenesis [64, 65]. In parallel, adipose-derived cytokines such as IL-6 amplify systemic inflammation and further weaken insulin signalling, reinforcing a metabolic inflammatory loop that increases plaque vulnerability and thrombogenic potential [66, 67]. IR may also shift haemostasis toward hypercoagulability and impaired fibrinolysis, thereby favouring intravascular thrombosis [68]. In CKM stages 2 to 3 complicated by hypertension, diabetes, or CKD, these mechanisms likely interact with haemodynamic load and renal stress, accelerating myocardial remodelling and kidney dysfunction and predisposing to myocardial infarction and heart failure.

In this study, we observed a positive and independent association between cumulative IR surrogate indices and BioAgeAccel. Li et al. have reported a significant relationship between METS-IR and biological aging markers [28]. IR is linked to various age-related diseases, including coronary heart disease [69], Alzheimer's disease [70], and CKD [71], with evidence suggesting that IR accelerates the aging of tissues like adipose and neuronal cells [72]. Additionally, the Forkhead box O (FOXO) transcription factors, which help mitigate oxidative stress and preserve cellular function, are impaired in an insulin-resistant environment, accelerating aging processes [73]. Notably, our study is the first to show that accelerated biological aging may account for part of the association between IR accumulation and increased CVD risk. Importantly, KDM-BA and Light-BA place somewhat different emphasis on multisystem physiological dysregulation versus parsimonious metabolic inflammatory ageing, yet yielded directionally concordant mediation patterns, suggesting that the observed IR to CVD linkage through ageing is not confined to a single operational definition of biological ageing. The underlying mechanisms likely involve the interplay of obesity and IR through oxidative stress, mitochondrial dysfunction, and chronic inflammation, which accelerate biological aging; this process, in turn, exacerbates metabolic disturbances, such as mitochondrial DNA loss and telomere attrition, and contributes to the transition from subclinical or asymptomatic CKM to overt CVD [74–76]. Our stage-stratified mediation analyses suggest that the pathway linking IR burden, BioAgeAccel, and CVD is more evident in later CKM stages than in stages 0–1, supporting the rationale that early interventions targeting the core metabolic drivers of this process, such as lifestyle modifications, blood glucose control, and insulin resistance management, could potentially delay the progression from CKM to clinical CVD by reshaping biological aging and counteracting the synergistic deterioration of metabolism and aging, thus effectively preventing CVD development in the preclinical stage.

Our subgroup analyses indicate that several cumulative IR surrogate indices exhibited stronger associations with incident CVD among individuals with biologically younger profiles. In relatively younger biological strata, baseline structural vascular damage and competing risk burden may be lower, rendering the cardiovascular system more sensitive to metabolic perturbations. In contrast, among biologically older individuals, cumulative exposure to multiple risk factors, subclinical organ injury, and survival selection may attenuate the relative contribution of IR, leading to smaller effect estimates. Mechanistically, early metabolic dysregulation in biologically younger individuals may initiate endothelial dysfunction, inflammatory activation, and vascular remodeling before irreversible damage accumulates [77], thereby exerting a proportionally greater influence on event risk. These findings underscore the importance of dynamic IR monitoring even in ostensibly low-risk or biologically younger populations and support earlier preventive intervention across the CKM continuum.

Among the evaluated IR surrogates, cumulative eGDR exhibited the best predictive performance, although its absolute discriminative ability remained modest. This observation aligns with prior work by Jiang et al., who reported that eGDR outperformed several alternative indices, including CVAI, TyG, TyG-BMI, METS-IR, and AIP, in predicting stroke risk among individuals with dysglycaemia. The relatively stronger performance of eGDR may reflect its composite structure integrating glycaemic status, central adiposity, and blood pressure, thereby capturing key pathophysiological domains of CKM syndrome. HbA1c reflects chronic glucotoxic exposure driving oxidative stress and endothelial dysfunction [11]; waist circumference indexes visceral adiposity linked to lipotoxicity, inflammation, and insulin resistance [78]; and blood pressure reflects hemodynamic consequences of impaired insulin signalling and reduced nitric oxide bioavailability [79]. These components approximate the convergence of metabolic, adipose-inflammatory, and vascular stress underlying CKM progression, offering greater pathophysiological relevance than indices centered on a single dimension. At the same time, because several surrogates are mathematically related and highly correlated, this apparent advantage should be interpreted as relative comparative performance within this dataset rather than as evidence that eGDR captures a wholly independent biological signal. Accordingly, cumulative eGDR may be better viewed as a relatively stronger comparator or pragmatic component of broader risk stratification. These findings suggest that cumulative eGDR may serve as a pragmatic, scalable reference benchmark for early cardiovascular risk stratification within the preclinical CKM continuum, and that newly proposed markers should demonstrate incremental predictive value beyond cumulative eGDR rather than against simpler indices alone.

Several limitations should be acknowledged. First, incident cardiovascular events were ascertained through participant-reported physician diagnoses, which may introduce recall bias and potential misclassification. Nonetheless, prior validation studies have demonstrated relatively low false positive rates for self-reported CVD, supporting the overall reliability of this approach [80]. Second, high covariate missingness may have introduced selection bias, although MICE and inverse probability weighting yielded consistent results. Third, although we adjusted for a broad range of sociodemographic and clinical covariates, residual confounding arising from unmeasured factors and inherent measurement variability cannot be completely excluded. However, sensitivity analyses based on E-values suggested that such residual confounding was unlikely to materially alter the main findings. Fourth, causal inference from the mediation analyses is limited by the observational design and uncertain temporal ordering. Fifth, the biological age metrics applied here do not encompass the entire spectrum of aging-related biomarkers [46]. However, the multi-marker algorithms incorporated indicators reflecting immune activity, inflammation, metabolic status, cardiovascular function, and hepatic and renal physiology. Sixth, as all analyses were based on a single Chinese cohort (CHARLS), the generalisability of our findings to other ethnicities, age groups, and clinical settings may be limited, and external validation in independent population-based and clinical cohorts is warranted. Finally, only two blood assessments were available for constructing cumulative exposure and pattern groups, which may limit the full characterization of temporal metabolic fluctuations. Accordingly, the K-means-derived groups should be interpreted as short-term longitudinal pattern classifications that jointly reflect level and directional change, rather than formal long-term trajectories. Future studies with more frequent repeated measurements are needed to refine dynamic risk estimation and strengthen causal inference.

Conclusion

In conclusion, this study demonstrates that sustained IR burden, captured by cumulative exposure and adverse longitudinal pattern groups across multiple surrogate indices, is independently associated with higher incident CVD risk in individuals with CKM stages 0–3. Notably, BioAgeAccel partially mediated these associations, implicating ageing pathways as a mechanistic bridge linking metabolic dysfunction to cardiovascular events. Integrating longitudinal IR profiling with ageing metrics may sharpen early risk stratification and enable timely, scalable prevention strategies that target upstream metabolic drivers, thereby delaying biological ageing and curbing progression from subclinical CKM to overt CVD.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ABSI

A body shape index

AHA

American Heart Association

AUC

Areas under the curve

BA

Biological age

BioAgeAccel

Biological ageing acceleration

BMI

Body mass index

BRI

Body roundness index

BUN

Blood urea nitrogen

CHARLS

China Health and Retirement Longitudinal Study

CI

Confidence interval

CKD

Chronic kidney disease

CKM

Cardiovascular-kidney-metabolic

CRP

C-reactive protein

CTI

C-reactive protein-triglyceride glucose index

CVD

Cardiovascular disease

CVAI

Chinese visceral adiposity index

DBP

Diastolic blood pressure

DCA

Decision curve analysis

eGDR

Estimated glucose disposal rate

eGFR

Estimated glomerular filtration rate

FPG

Fasting plasma glucose

HbA1c

Glycated haemoglobin

HGB

Hemoglobin

HOMA-IR

Homeostasis model assessment of insulin resistance

hs-CRP

High-sensitivity C-reactive protein

IDI

Integrated discrimination improvement

IPTW

Inverse probability of treatment weighting

IR

Insulin resistance

IQR

Interquartile range

KDM-BA

Klemera–Doubal method biological age

Light-BA

Light biological age

MCV

Mean corpuscular volume

METS-IR

Metabolic score for insulin resistance

MICE

Multiple imputation by chained equations

NRI

The net reclassification improvement

OR

Odds ratio

PCA

Principal component analysis

PLT

Platelet count

Qgcomp

Quantile g-computation

RCS

Restricted cubic spline

ROC

Receiver operating characteristic

SBP

Systolic blood pressure

SD

Standard deviation

sHRs

Subdistribution hazard ratios

TC

Total cholesterol

TG

Triglycerides

TG/HDL-C

Triglyceride to high-density lipoprotein cholesterol ratio

TyG

Triglyceride-glucose index

TyG-ABSI

Triglyceride-glucose-a body shape index

TyG-BMI

Triglyceride-glucose-body mass index

TyG-BRI

Triglyceride-glucose-body roundness index

TyG-CVAI

Triglyceride-glucose-Chinese visceral adiposity index

TyG-WC

Triglyceride-glucose-waist circumference

TyG-WHtR

Triglyceride-glucose-waist-to-height ratio

TyG-WWI

Triglyceride-glucose-weight-adjusted waist index

WBC

White blood cell count

WC

Waist circumference

WHtR

Waist-to-height ratio

WQS

Weighted quantile sum

WWI

Weight-adjusted waist index

Author contributions

HSS, LQ and ZZM conceived and designed the study; HSS and ZZM extracted and curated the data; HSS and LQ performed the statistical analyses and drafted the manuscript; ZMQ, LY, GTT, HYK, GTA and ZZM critically reviewed the data and revised the manuscript for important intellectual content. All authors read and approved the final version of the manuscript. All authors have reviewed and approved the final version of the manuscript.

Funding

This study was supported by Shenzhen High-level Hospital Construction Fund, Youth Program of National Natural Science Foundation of China (62201015), and Clinical Medicine Plus X-Young Scholars Project, Peking University, the Fundamental Research Funds for the Central Universities.

Data availability

Data for this study were obtained from the China Health and Retirement Longitudinal Study (CHARLS) and are accessible via the project website (http://charls.pku.edu.cn).

Declarations

Ethics approval and consent to participate

The study protocol was reviewed and approved by the Biomedical Ethics Review Committee of Peking University (IRB00001052–11015). All participants provided written informed consent prior to enrolment.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.