Association between insulin resistance surrogate markers and cardiovascular disease risk in individuals with subclinical metabolic disorders: a prospective cohort study based on baseline levels, cumulative exposure, and longitudinal trajectories

Abstract

Background

Cardiovascular disease (CVD) has become an increasingly severe global public health issue, with its disease burden continuing to rise worldwide. Insulin resistance (IR) is a key metabolic process underlying CVD, but direct measurement is challenging in large populations.Various composite cardiometabolic indices are used to reflect IR-related metabolic alterations.Prediabetes, prehypertension, and predyslipidemia are early manifestations of subclinical metabolic dysfunction associated with CVD. However, their contributions to CVD risk and their relationship with dynamic changes in IR remain unclear.Therefore, this study aimed to comprehensively evaluate multiple insulin resistance–related surrogate markers and explore the associations of baseline IR, cumulative IR (cuIR), and IR trajectories with CVD risk in middle-aged and older adults.

Methods

Data were obtained from the China Health and Retirement Longitudinal Study (CHARLS, 2011–2020), with relevant information collected via standardized questionnaires during follow-up.Eleven IR-related surrogate markers (TyG, TyG-WC, TyG-WWI, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, CHG) were analyzed, with covariates pre-specified based on established cardiovascular risk factors. The primary outcome was the incidence of newly diagnosed CVD. In the CHARLS cohort, IR-related indicators were available at two time points (Wave 1: 2012 and Wave 3: 2015). Therefore, cumulative insulin resistance exposure (cuIR) was calculated as: cuIR = (IR2012 + IR2015) / 2 × time. Due to the limited number of repeated measurements, clustering analysis (K-means) was applied to classify cumulative exposure patterns of IR-related indicators.Multivariate Cox proportional hazards regression models estimated hazard ratios (HR) with 95% confidence intervals (95% CI), and restricted cubic splines (RCS) examined nonlinear associations. Predictive performance of IR-related surrogate indicators was assessed using C-statistic, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Subgroup analyses further tested robustness of findings. Additionally, external validation of baseline IR findings was conducted using the English Longitudinal Study of Ageing (ELSA, 2004–2016). In the ELSA cohort, longitudinal trajectories of IR-related surrogate markers were constructed using group-based trajectory modeling (GBTM) based on repeated measurements (Waves 2, 4, and 6).

Results

In the study examining the association between baseline IR-related surrogate markers and the risk of CVD occurrence, 7,299 participants were enrolled, with 1,870 new CVD events occurring during an 8-year follow-up period. Across three Cox proportional hazards models, all 11 surrogate markers were significantly associated with increased CVD risk. In Model 3, fully adjusted for confounders, TyG-ABSI demonstrated the strongest association with CVD risk (HR = 4.92, 95% CI 3.13–7.74). After stratification by quantiles, higher quantiles were associated with increased CVD risk (P for trend < 0.05). In the external validation cohort ELSA, all IR indicators except TyG showed positive correlations with CVD risk in unadjusted models; After multivariable adjustment TyG-WC, TyG-BMI, TyG-BRI, TyG-WHtR, METS-IR, and and CTI remained independently associated with CVD risk. Participants in the highest quantile showed significantly increased risk (P for trend < 0.05). In the study of cuIR and CVD risk, 773 out of 3,847 participants ultimately developed CVD. In the multivariable Cox proportional hazards model, all 11 cuIR-related surrogate markers significantly increased CVD risk. Compared to Q1, both Q2 and Q3 groups showed elevated risk, with a significant linear trend (P for trend < 0.05). Clustering pattern analysis based on cumulative exposure showed that individuals in the high-level groups had significantly higher CVD risk than those in the low-level groups. Trajectory analysis of the ELSA cohort revealed unique longitudinal patterns of insulin resistance-related indicators. In the fully adjusted model, compared with the low-risk trajectory group, individuals in the high-risk trajectory groups for TyG-WC, TyG-BMI, TyG-BRI, METS-IR, and CTI had a significantly increased risk of developing CVD. Among these, CTI showed the strongest association with CVD risk (HR = 1.90, 95% CI 1.25–2.89).In contrast, associations for TyG, TyG-WHtR, TyG-WWI, and TyG-ABSI were not significant after multivariable adjustment. No significant associations were observed for the trajectory groups of AIP and CHG.

Conclusion

This study demonstrates that baseline IR, cuIR, and longitudinal trajectory patterns are significantly associated with CVD risk in middle-aged and older populations. These measures provide incremental predictive value beyond traditional risk factors. Early identification and long-term monitoring of IR-related surrogate markers during the subclinical metabolic stage may enhance CVD risk stratification and support precision prevention and intervention strategies.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12933-026-03230-z.

Research insights

What is currently known about this topic?

• Existing evidence indicates that multiple surrogate markers of insulin resistance (IR) derived from conventional metabolic parameters are associated with elevated CVD risk. However, current studies primarily rely on single-time-point measurements, lacking multidimensional systematic comparisons of various IR-related indicators. Furthermore, the impact of cumulative exposure and longitudinal patterns of these markers on CVD risk in individuals with subclinical metabolic disorders remains unclear.

What is the key research question?

• How are baseline levels, cumulative exposure, and longitudinal trajectories of 11 IR surrogate markers (TyG, TyG-WC, TyG-WWI, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, CHG) associated with new-onset CVD in middle-aged and older adults with subclinical metabolic disorders? Which markers demonstrate optimal predictive performance in CVD risk assessment?

What is new?

• This large-scale multicohort study systematically compared the relationship between 11 IR surrogate markers and CVD risk across multiple dimensions—baseline levels, cumulative exposure, and longitudinal trajectory patterns. Key findings include: (1) All 11 baseline IR surrogate markers significantly associated with elevated cardiovascular risk, with TyG-ABSI showing the strongest association (HR = 4.92) (2) Individuals with higher cumulative IR exposure (reflecting long-term burden) and those in the persistently high IR trajectory group (capturing temporal changes) exhibited significantly increased CVD incidence risk. (3) Combining fat-related composite indices (especially TyG-WC and cuTyG-WC) demonstrated superior incremental predictive value beyond traditional risk factors.

How might this study influence clinical practice?

• Baseline levels, cumulative exposure, and longitudinal trajectory patterns (in ELSA) of IR surrogate markers may serve as practical biomarkers for identifying high-risk individuals with subclinical metabolic disorders. All assessed IR-related indicators are derived from routine, cost-effective laboratory tests and anthropometric measurements, providing readily applicable tools for cardiovascular risk stratification in primary care settings. Incorporating long-term monitoring of IR-related changes may improve early risk identification and support targeted preventive strategies, thereby contributing to the reduction of CVD burden.

Introduction

With the acceleration of population aging and urbanization, cardiovascular diseases (CVD) have become the leading cause of disability and death among the elderly population [1]. As a chronic non-communicable disease closely related to aging, the medical burden and economic costs of cardiovascular diseases are expected to increase significantly in the coming years, seriously affecting individual health and social economic conditions [2–3]. Epidemiological data show that the global mortality rate related to cardiovascular diseases increased by 62.6% from 1990 to 2022, reaching 19.8 million deaths in 2022 [4]. Predictions indicate that by 2025, the number of deaths related to cardiovascular diseases will rise to 20.5 million [5], highlighting the urgent need for early identification of high-risk individuals and the implementation of effective prevention strategies. It is worth noting that the occurrence and development of cardiovascular diseases is a long-term and gradual process, and before the appearance of obvious clinical symptoms and irreversible target organ damage, there is often a subclinical stage. At this time, the body has metabolic abnormalities but has not yet reached the clinical diagnostic criteria. If this stage can be identified and intervened in time, it can effectively delay or reverse the disease progression, reduce the risk of cardiovascular events, and alleviate the burden of the disease. Subclinical metabolic disorders are a concentrated manifestation of this state, including prediabetes, prehypertension, and prelipidemia, characterized by slight abnormalities in metabolic indicators. These states are relatively common and usually have no obvious symptoms, but already indicate that the body is in a high-risk state [6]. Prediabetes refers to a state where blood glucose levels exceed the normal range but have not yet reached the diagnostic criteria for diabetes (DM). It is closely associated with cardiovascular diseases (CVD) [7]. A prospective cohort study showed that during a 6.6-year observation period, the mortality rate and absolute risk of cardiovascular diseases (CVD) for individuals with prediabetes were 7.36 per 10,000 person-years and 8.75, respectively [8]. Mild hypertension affects approximately 25% to 50% of the adult population worldwide. Studies have shown that mild hypertension not only increases the risk of hypertension but also significantly increases the incidence of cardiovascular diseases, coronary heart disease, myocardial infarction, and stroke [9]. Prediabetes can lead to vascular endothelial dysfunction, promote lipid deposition in the vessel wall and form early atherosclerotic plaques. The gradual growth of plaques will cause the lumen to narrow and blood flow to be blocked, thereby triggering a series of clinical cardiovascular events, such as coronary heart disease and angina pectoris [10].

Insulin resistance (IR) represents a critical metabolic dysfunction characterized by diminished sensitivity and responsiveness to insulin in skeletal muscle, adipose tissue, or hepatocytes. This reduces glucose uptake and utilization, It disrupts glucose metabolism, increases free radical generation, activates the mitochondrial electron transport chain, and elevates reactive oxygen species (ROS) production. These mechanisms promote vascular smooth muscle cell proliferation and atherosclerosis while impairing myocardial glucose utilization, ultimately leading to metabolic disorders and cardiac injury [11–12]. Although the hyperinsulinemic euglycemic clamp (HEC) technique is considered the gold standard for assessing IR, its technical complexity, invasiveness, and high cost render it unsuitable for clinical practice and epidemiological studies [13]. In recent years, numerous studies have proposed simple, scalable, and cost-effective alternative markers for IR, including the triglyceride-glucose index (TyG) [14], triglyceride glucose-waist circumference (TyG-WC) [15], triglyceride glucose-weight adjusted waist index (TyG-WWI) [16], triglyceride glucose-body mass index (TyG-BMI) [17], triglyceride glucose–a body shape index (TyG-ABSI) [18], triglyceride glucose-body roundness index (TyG-BRI) [19], triglyceride glucose-waist-to-height ratio (TyG-WHtR) [20], metabolic score for insulin resistance (METS-IR) [21], atherogenic index of plasma (AIP) [22], C-reactive protein triglyceride glucose index (CTI) [23], and cholesterol-high-density lipoprotein-glucose (CHG) [24]. These indicators are derived from routine metabolic parameters (e.g., fasting glucose, lipids, and anthropometric measures) and are widely used as surrogate markers associated with insulin resistance. Previous studies have demonstrated their associations with CVD incidence and mortality. However, most existing studies are conducted in cross-sectional designs, focusing on single exposures or outcomes, and the associations between IR surrogate indicators and the specific incidence of CVD in subclinical metabolic disorder populations have not been widely studied [25–27]. Furthermore, the existing literature lacks systematic comparisons based on large-scale prospective cohorts to clarify the predictive efficacy of different IR surrogate markers for CVD events. This gap significantly limits the translation of such markers into clinical practice. Therefore, systematically evaluating and establishing optimal IR prediction surrogate indicators through large-scale prospective cohort studies will assist clinicians in identifying and intervening in early-stage IR using routine, low-cost biochemical testing. This holds significant public health implications for advancing early CVD prevention and reducing its overall disease burden.

This study integrated data from two population-based cohorts—the China Health and Retirement Longitudinal Study (CHARLS) and the English Longitudinal Study of Ageing (ELSA)—focusing on middle-aged and older adults aged 45 years and above. We examined the associations of baseline and cumulative IR-related surrogate markers with incident CVD and externally validated key findings in an independent cohort. In addition, long-term patterns of IR-related markers were evaluated using cumulative exposure patterns in CHARLS and longitudinal trajectory patterns in ELSA. The study aimed to: (1) examine the prospective association between IR surrogate markers and CVD risk among individuals with subclinical metabolic disorders; (2) evaluate the predictive performance of these markers for CVD outcomes; and (3) investigate whether long-term IR-related patterns were associated with subsequent CVD risk. These findings may provide a low-cost and readily accessible approach for early CVD risk identification and stratified prevention in individuals with subclinical metabolic disorders.

Materials and methods

Data source

This study used data from CHARLS and ELSA. CHARLS is a nationally representative longitudinal survey of Chinese adults aged 45 years and older, covering health status, socioeconomic status, family structure, and living conditions [28]. ELSA is a prospective cohort study of community-dwelling adults aged 50 years and older in England, with follow-up surveys conducted approximately every two years [29]. CHARLS was approved by the Biomedical Ethics Committee of Peking University, and ELSA was approved by the London Multi-Centre Research Ethics Committee. Written informed consent was obtained from all participants.

Study population

The analysis sample includes data from CHARLS Waves 1 through 5 (2011–2020) and ELSA Waves 2 through 8 (2004–2016). CHARLS Wave 1 and ELSA Wave 2 were used as the baseline, and cuIR was calculated using data from CHARLS Waves 1 and 3. Longitudinal trajectory analysis was conducted using data from ELSA Waves 2, 4, and 6. Exclusion criteria are as follows: (1) Individuals under 45 years of age; (2) Missing blood test results at the ELSA baseline and in CHARLS Waves 1 and 3 (triglycerides (TG), fasting blood glucose (FBG), total cholesterol (TC), glycated hemoglobin (HbA1c), C-reactive protein (CRP), high-density lipoprotein (HDL-C), low-density lipoprotein (LDL-C)); (3) Missing anthropometric measurements at baseline (body mass index (BMI), waist circumference (WC), height, systolic blood pressure (SBP), diastolic blood pressure (DBP)); (4) Missing demographic information; (5) Presence of cardiovascular disease (CVD) at baseline; (6) Missing follow-up data. The final analysis sample consisted of 7,299 participants from the CHARLS baseline cohort, 4,622 from the ELSA baseline cohort, 3,847 participants from the CHARLS cuIR clustering cohort in CHARLS Waves 1 and 3, and 2,740 participants from the ELSA trajectory cohort in ELSA Waves 2, 4, and 6. The process of selecting the study population is shown in Figs. 1 and 2.

Fig. 1.

Flowchart of the study participant selection based on the CHARLS dataset

Fig. 2.

Flowchart of the study participant selection based on the ELSA dataset

Assessment of subclinical metabolic disorders

Prediabetes was defined as FBG between 5.6 and 7.0 mmol/L or HbA1c between 5.7% and 6.5% according to ADA criteria [30]. Prehypertension was defined as SBP between 120 and 140 mmHg and/or DBP between 80 and 90 mmHg [31]. Predyslipidemia is defined according to the updated criteria of the National Cholesterol Education Program (NCEP) [32], defined as TC between 5.2 mmol/L and 6.19 mmol/L, or TG between 1.7 mmol/L and 2.29 mmol/L, or LDL-C between 3.38 mmol/L and 4.12 mmol/L.

Exposure assessment

This study evaluated 11 IR-related surrogate markers, including TyG, TyG-WC, TyG-WWI, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, and CHG. These indices were calculated using routine biochemical and anthropometric parameters, including fasting blood glucose, triglycerides, total cholesterol, HDL-C, CRP, BMI, waist circumference, weight, and height.

Although these markers share common underlying components, they capture distinct dimensions of metabolic dysfunction, including lipid–glucose metabolism, adiposity, fat distribution, and inflammation-related processes. Each marker was therefore analyzed separately to compare its predictive performance for CVD risk. To further assess the degree of overlap among these indicators, we performed a Spearman correlation analysis and visualized the results using a heatmap (Supplementary Fig. 1). The results indicate that there is moderate to high correlation among these indicators, while some heterogeneity remains. The calculation formulas for these indicators are as follows:

Assessment of covariates

To ensure the accuracy and reliability of the analysis results, we controlled for several potential confounding factors based on existing literature and clinical expertise, including sociodemographic factors, lifestyle, medical history, and laboratory test data. (1) Sociodemographic factors: age, gender (male, female), marital status (married/cohabiting, widowed/divorced/separated/unmarried), educational attainment (below high school, high school or equivalent, above high school), income; (2) Lifestyle: Smoking (never smoked, former smoker, current smoker), alcohol consumption (yes, no), life satisfaction (satisfied, dissatisfied), sleep problems (yes, no), physical activity (at least one session of moderate or vigorous activity per week, less than once per week), feelings of depression (yes, no); (3) Medical history: asthma, cancer, lung disease, arthritis; (4) Laboratory tests: hemoglobin (Hb).

Outcome definition

The baseline IR surrogate markers in this study utilized CVD events from waves 2–5 in the CHARLS database and waves 3–8 in the ELSA database. The cuIR outcome comprised CVD events identified during follow-up in waves 4 and 5 of the CHARLS survey For the trajectory-modeling cohort in ELSA, the outcome was assessed at Wave 6.

Consistent with previous studies [33], CVD events were assessed using a standardized set of questions: “Has a doctor ever told you that you have been diagnosed with a heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems?” or “Has a doctor ever told you that you have been diagnosed with a stroke?” Participants reporting a diagnosis of heart disease or stroke during follow-up were classified as having experienced a CVD event. When participants reported CVD during the follow-up period, the follow-up time was defined as the time interval between the onset of the disease and the baseline. For the participants who did not report CVD during the follow-up period, the follow-up duration was determined based on the interval between the baseline assessment and the final survey date [34].

Handling missing variable values

To reduce potential bias arising from incomplete covariate data while retaining participants with complete exposure and outcome information, Multiple Imputation with Chained Equations (MICE) was exclusively applied to selected covariates with missing data. The primary exposures in this study comprised 11 IR surrogate indicators. All core components required for calculating these IR surrogate markers, including fasting blood glucose, triglycerides, total cholesterol, high-density lipoprotein cholesterol, C-reactive protein, body mass index, waist circumference, weight and height, were not imputed. Participants with missing values in any component for surrogate IR markers calculation were excluded from the corresponding analyses. Assuming covariates were missing at random, continuous variables were imputed using predictive mean matching (pmm), binary variables using logistic regression (logreg), and ordinal categorical variables using polytomous regression (polyreg). Fifty iterations were performed to generate five imputed datasets, and Rubin’s rule was applied to obtain pooled estimates for subsequent statistical analyses [35].

Assessment and clustering of cuIR

This study applied k-means clustering with Euclidean distance to characterize patterns of insulin resistance (IR) proxy indicators from 2012 to 2015. The cumulative IR (cuIR) was calculated using a linear model: cuIR = (IR₂₀₁₂ + IR₂₀₁₅) / 2 × time interval (2015–2012), where IR₂₀₁₂ and IR₂₀₁₅ represent measurements at baseline and follow-up (3-year interval). This approach approximates long-term exposure by integrating average levels over time, consistent with previously reported cumulative metabolic indices [36–37].

Clustering was conducted in three steps. First, the optimal number of clusters (k) was determined using the elbow method, with k = 3 selected based on the inflection point of within-cluster sum of squares and its clinical interpretability, particularly in distinguishing a moderate-risk subgroup. Second, initial cluster centers were randomly assigned. Third, individuals were iteratively allocated to the nearest centroid, with cluster centers updated until convergence (minimization of within-cluster variance) [38].

Each participant was assigned to a cluster based on their Euclidean distance from the center of the final cluster. The resulting clusters reflected distinct patterns of insulin resistance surrogates between the two time points, capturing both cumulative exposure levels and short-term fluctuations. From a methodological perspective, the k-means algorithm was run independently of the outcome variable throughout the clustering process, ensuring the objectivity and unbiasedness of the analysis. Compared to traditional, manually predefined grouping strategies, this unsupervised machine learning method identifies natural, data-driven subgroups based on the intrinsic variability of IR biomarkers, thereby eliminating arbitrary classification boundaries. This approach effectively captures individual differences in IR exposure levels and their dynamic changes, which better aligns with the actual data distribution and reduces classification bias [39].

Modeling surrogate IR markers Trajectories

In this study, a group-based trajectory model (GBTM) was employed to identify different longitudinal patterns of surrogate markers for insulin resistance. The GBTM assumes population heterogeneity and classifies individuals into latent subgroups with similar developmental trajectories over time, assigning each individual to the group with the highest posterior probability [40].

Using this method, trajectory groups were identified based on surrogate markers of insulin resistance from waves 2, 4, and 6 of the ELSA study. To ensure model parsimony and interpretability, model selection was conducted in two stages [43]. First, the number of trajectory groups (2–5) and the order of the polynomial (linear, quadratic, cubic) were explored. Second, the optimal model was selected based on the following criteria: lower absolute values of the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC), odds ratio for correct classification > 5, average posterior probability (AvePP) > 0.70, and sample size within each group > 5%.

Although the four-group model yielded lower AIC and BIC values, it was ultimately rejected because it failed to strike a better balance between statistical fit and clinical interpretability; consequently, the three-group model was selected (Supplementary Table 1). This final model identified three trajectories (low, medium, and high), facilitating clinically meaningful risk stratification (Fig. 3).

Fig. 3.

Trajectories of the IR index from the ELSA database over an 8-year follow-up period (Waves 2, 4, and 6)

Cox regression models

Univariate and multivariate Cox regression models were applied to evaluate the associations between IR surrogate markers and incident CVD, with outcomes reported as hazard ratios (HRs) and corresponding 95% confidence intervals (CIs). The Schoenfeld residual test was used to examine the proportional hazards (PH) assumption prior to model establishment, and stratified analysis was performed for variables that did not satisfy the PH assumption. All IR-related metrics were incorporated into the models as both continuous and categorical variables. In the CHARLS cohort, baseline IR and cuIR levels were grouped into tertiles (Q1, Q2, and Q3), with Q1 set as the reference group. K-means clustering-derived IR clusters (low-level group as the reference) and GBTM-identified IR trajectories (low-level trajectory group as the reference) were also included in the analytical framework. Similarly, participants in the ELSA cohort were stratified into three tertiles according to baseline IR surrogate levels, with Q1 adopted as the reference group.

A three-step hierarchical modeling strategy was implemented to control for potential confounding effects. Model 1 was an unadjusted crude model. Model 2 adjusted for core baseline confounders, including age, sex, educational level, marital status, smoking status, alcohol consumption, and physical activity. On the basis of Model 2, Model 3 further adjusted for additional psychosocial and clinical covariates, namely life satisfaction, depressive symptoms, sleep problems, household income, hemoglobin levels, and disease histories of asthma, cancer, arthritis, and lung. To avoid multicollinearity bias, multiple metabolic indicators constituting the composite IR index were excluded from covariates, including BMI, WC, CRP, FBG, HbA1c, TG, TC, and HDL-C. This stepwise adjustment approach is well-validated in observational studies and enables rigorous assessment of the independent association between IR-related indicators and CVD outcomes [41].

Fine-Gray competing risks model was employed in the sensitivity analysis to account for competing events such as all-cause mortality during the course of cardiovascular disease progression. These sensitivity analyses comprehensively incorporated alternative markers of baseline insulin resistance, cuIR measurements, and their corresponding clusters, and were uniformly applied across both the CHARLS and ELSA cohorts to verify the stability and consistency of the association between alternative markers of insulin resistance and the incidence of cardiovascular disease.

RCS and subgroup analysis

To assess potential nonlinear relationships between surrogate IR markers and CVD risk, we conducted three Restricted Cubic Spline (RCS) analyses describing dose-response relationships with incident CVD. Threshold effect analyses were performed for identified nonlinear relationships. Stratified subgroup analyses were performed by age (< 65 years, ≥ 65 years), sex, education level, marital status, smoking, alcohol consumption, and physical activity to explore whether the association between CVD incidence and IR surrogate measures varied by covariate status.

Assessment of the incremental value of baseline IR and cuIR

In this study, the Boruta feature selection algorithm was employed to determine the relevant covariates for the construction of the base model. Subsequently, the baseline IR substitution indicators and cuIR exposure values were incorporated into the base model to construct the incremental model.

The incremental predictive value of these models was evaluated using the C-index, Net Reclassification Improvement (NRI), and Integrated Discrimination Improvement (IDI). An IDI > 0 indicates improved discrimination ability, while NRI > 0 reflects enhanced reclassification performance of the incremental model compared with the baseline model.

Statistical analysis

All statistical analyses in this study were performed using R software (version 4.4.1); a two-sided P-value < 0.05 was considered statistically significant. The specific statistical methods and corresponding R packages used are as follows: ① Descriptive statistics: The “tableone” package was used to summarize the baseline characteristics of the study subjects. For normally distributed continuous variables, data were described as “mean±standard deviation,” and an independent samples t-test was used to assess differences between groups. For non-normally distributed continuous variables, statistical descriptions were presented as “median (interquartile range),” and the Mann-Whitney U test was used to analyze differences between groups. Categorical variables were expressed as “frequency (percentage),” and the chi-square test or Fisher’s exact test was used to compare differences between groups. ② Survival Analysis: The “survival” package was used to perform Cox proportional hazards regression to assess the association between different IR indices and new-onset CVD. Note that to avoid multicollinearity, the components of the composite IR surrogate indicator were excluded as covariates in the regression analysis. ③ Other Analyses: The “gbmt” package was used to establish trajectory models. Restricted cubic spline (RCS) analysis was performed using the “rms” package; multiple imputation for missing data was conducted using the “mice” package; data processing and visualization were performed using the “dplyr” and “ggplot2” packages, respectively.

Results

Clustering patterns of cuIR exposure

The K-means clustering method was employed to group and classify the IR surrogate indicators of of Waves 1 and 3 of the CHARLS data. Based on the elbow method, three aggregation centers were identified (Fig. 4). Ultimately, three clusters based on IR surrogate indicators were established, and participants were classified as follows: Group 1 (Low-level group): IR surrogate indicators were at low levels at baseline, followed by minor variations while remaining at low levels; Group 2 (Moderate Level Group): IR surrogate indicators remained at moderate levels across the observation period; Group 3 (High Level Group): IR surrogate indicators were at higher baseline levels, followed by minor variations while remaining at high levels, with no overlap with other groups.

Fig. 4.

Clustering patterns of cumulative exposure of TyG, TyG-WC, TyG-WWI, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, and CHG between 2012 and 2015.A: Scatter plots showing clustering patterns based on standardized values of IR-related surrogate indicators. Each cluster is represented by a unique color and shape across all graphs. Standardized values (mean = 0, standard deviation = 1) are shown for 2012 (X-axis) and 2015 (Y-axis), reflecting cumulative exposure patterns across two time points.B: Line charts displaying the average standardized values of each cluster at the two time points (2012 and 2015), illustrating differences in cumulative exposure levels across groups

Baseline characteristics of surrogate markers of IR in the study population

Based on inclusion and exclusion criteria, 7,299 participants from the CHARLS cohort were included in the baseline IR surrogate indicator analysis. This comprised 3,495 men and 3,408 women, with 1,870 CVD events occurring during the median 8-year follow-up period. Table 1 presents baseline characteristics of the study population stratified by cardiovascular disease occurrence. Compared with participants without CVD, those diagnosed with CVD were more likely to be female, older, heavier, have a larger waist circumference, be widowed/divorced/separated/ unmarried, have asthma, arthritis, or lung disease, report lower life satisfaction, experience depression or sleep problems, engage in no physical activity, and have lower income. Regarding biomarkers, the CVD cohort exhibited higher levels of HbA1c, TG, TC, LDL-C, FBG, SBP, DBP, TyG, TyG-WC, TyG-WWI, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, and CHG, while HDL-C levels were lower (P < 0.05).

Table 1.

Baseline characteristics of the study population for the association between baseline IR indicators and cardiovascular disease based on the CHARLS database

Variables	Overall(n = 7299)	non-CVD(n = 5429)	CVD(n = 1870)	P
Gender (%)				0.001
Male	3495 (47.9)	2664 (49.1)	831 (44.4)
Female	3804 (52.1)	2765 (50.9)	1039 (55.6)
Age	58.7 ± 9.2	58.2 ± 9.3	60.1 ± 8.7	< 0.001
Weight	59.0 ± 11.4	58.7 ± 11.2	60.0 ± 11.9	< 0.001
Height	1.6 ± 0.1	1.6 ± 0.1	1.6 ± 0.1	0.094
WC	85.4 ± 9.7	84.9 ± 9.4	86.7 ± 10.2	< 0.001
BMI	21.1 ± 1.2	21.1 ± 1.2	21.2 ± 1.2	0.108
Education (%)				0.768
Below high school	6506 (89.1)	4837 (89.1)	1669 (89.3)
High school or equivalent	695 ( 9.5)	516 ( 9.5)	179 ( 9.6)
College or above	98 ( 1.3)	76 ( 1.4)	22 ( 1.2)
Marital (%)				0.005
Widowed/Divorced/Separated/Unmarrie	861 (11.8)	606 (11.2)	255 (13.6)
Married/Cohabiting	6438 (88.2)	4823 (88.8)	1615 (86.4)
Drink (%)				0.124
NO	4373 (59.9)	3224 (59.4)	1149 (61.4)
Yes	2926 (40.1)	2205 (40.6)	721 (38.6)
Smoke (%)				0.001
Never smoked	4400 (60.3)	3253 (59.9)	1147 (61.3)
Used to smoke	612 ( 8.4)	426 ( 7.8)	186 ( 9.9)
Currently smoke	2287 (31.3)	1750 (32.2)	537 (28.7)
Asthmae (%)				< 0.001
NO	7001 (95.9)	5257 (96.8)	1744 (93.3)
Yes	298 ( 4.1)	172 ( 3.2)	126 ( 6.7)
Lung (%)				< 0.001
NO	6677 (91.5)	5018 (92.4)	1659 (88.7)
Yes	622 ( 8.5)	411 ( 7.6)	211 (11.3)
Arthre (%)				< 0.001
NO	4904 (67.2)	3757 (69.2)	1147 (61.3)
Yes	2395 (32.8)	1672 (30.8)	723 (38.7)
Cancer (%)				0.763
NO	7231 (99.1)	5380 (99.1)	1851 (99.0)
Yes	68 ( 0.9)	49 ( 0.9)	19 ( 1.0)
Satisfaction (%)				0.048
Not satisfied	1075 (14.7)	773 (14.2)	302 (16.1)
Satisfied	6224 (85.3)	4656 (85.8)	1568 (83.9)
Depression (%)				< 0.001
NO	5117 (70.1)	3890 (71.7)	1227 (65.6)
Yes	2182 (29.9)	1539 (28.3)	643 (34.4)
Aleep problem(%)				< 0.001
NO	4908 (67.2)	3743 (68.9)	1165 (62.3)
Yes	2391 (32.8)	1686 (31.1)	705 (37.7)
Physical activity (%)				0.002
NO	2408 (33.0)	1735 (32.0)	673 (36.0)
Yes	4891 (67.0)	3694 (68.0)	1197 (64.0)
HbA1c	5.2 ± 0.7	5.2 ± 0.7	5.3 ± 0.7	0.027
Income	26609.1 ± 41300.4	27247.8 ± 41260.2	24754.9 ± 41372.0	0.024
Hemoglobin	14.5 ± 2.2	14.4 ± 2.2	14.5 ± 2.1	0.324
TG	128.4 ± 74.1	126.6 ± 74.1	133.5 ± 73.9	0.001
TC	196.8 ± 37.0	196.1 ± 37.0	198.9 ± 37.0	0.005
CRP	2.6 ± 6.8	2.5 ± 6.6	2.8 ± 7.4	0.14
BUN	15.8 ± 4.6	15.8 ± 4.6	15.6 ± 4.5	0.057
SCR	0.8 ± 0.2	0.8 ± 0.3	0.8 ± 0.2	0.344
SUA	4.5 ± 1.2	4.5 ± 1.2	4.4 ± 1.3	0.121
LDL	120.3 ± 33.7	119.6 ± 33.5	122.3 ± 34.3	0.003
HDL	51.7 ± 15.2	52.0 ± 15.3	50.9 ± 14.7	0.007
FBG	108.5 ± 25.7	108.1 ± 25.0	109.7 ± 27.6	0.017
SBP	130.4 ± 20.0	129.2 ± 19.5	133.6 ± 20.9	< 0.001
DBP	76.1 ± 11.4	75.7 ± 11.3	77.4 ± 11.7	< 0.001
TyG	8.7 ± 0.6	8.7 ± 0.6	8.7 ± 0.6	< 0.001
TyG-WC	743.6 ± 108.5	737.9 ± 106.5	760.1 ± 112.7	< 0.001
TyG-WWI	97.2 ± 10.5	96.7 ± 10.4	98.6 ± 10.6	< 0.001
TyG-BMI	183.9 ± 16.9	183.4 ± 16.8	185.4 ± 16.9	< 0.001
TyG-ABSI	0.8 ± 0.1	0.8 ± 0.1	0.8 ± 0.1	< 0.001
TyG-BRI	37.0 ± 12.3	36.3 ± 11.9	39.0 ± 13.0	< 0.001
TyG-WHtR	4.7 ± 0.7	4.7 ± 0.7	4.8 ± 0.7	< 0.001
METS-IR	31.7 ± 3.9	31.6 ± 3.9	32.0 ± 4.0	< 0.001
AIP	0.4 ± 0.3	0.3 ± 0.3	0.4 ± 0.3	< 0.001
CTI	8.8 ± 0.8	8.7 ± 0.8	8.8 ± 0.8	< 0.001
CHG	5.3 ± 0.4	5.3 ± 0.4	5.4 ± 0.4	< 0.001

Findings from the CHARLS cohort were validated using ELSA data, which included 4,622 participants. Participants were divided into two groups based on CVD status: those with CVD (941 individuals) and those without CVD (3,681 individuals). Supplementary Table 2 presents participant characteristics, with significant differences in baseline features between groups (P < 0.05). Compared to those without CVD, individuals with CVD were more likely to be older, heavier, have a larger waist circumference, higher BMI, lower income, higher systolic blood pressure, and be married/cohabiting women. Additionally, those with lung disease, arthritis, depression, and lower physical activity exhibited significantly increased CVD prevalence. Regarding hematological indicators, the CVD group exhibited significantly higher levels of CRP, FBG, TyG-WC, TyG-WWI, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, METS-IR, CTI, and CHG compared to the non-CVD group, while HDL-C levels were lower (P < 0.05).

Baseline characteristics of study population based on cuIR and clustering patterns

This study analyzed cumulative IR surrogate indicators in 3,847 CHARLS participants, including 1,821 men and 2,026 women. Among them, 9.5% had completed secondary education or higher, and 90.7% were married or cohabiting. During the median 8-year follow-up, 773 CVD cases occurred. Table 2 presents baseline characteristics and clustering patterns of the study population. CVD patients were older, heavier, and had larger waist circumferences. They exhibited higher prevalence of asthma, lung disease, arthritis, depression, and sleep problems, alongside lower physical activity levels. Additionally, they demonstrated higher levels of HbA1c, TG, TC, LDL-C, FBG, SBP, and DBP, along with lower HDL-C. Regarding cumulative exposure characteristics, participants who experienced CVD events during follow-up exhibited significantly higher cumulative levels of IR surrogate indicators, with greater between-group differences observed in cuTyG-WC, cuTyG-WWI, cuTyG-BMI, cuTyG-BRI, and cuMETS-IR. Regarding clustering patterns based on cumulative exposure, the composition ratios of each cluster subtype exhibited distinct distribution characteristics. Participants who developed CVD during follow-up generally demonstrated poorer control of IR surrogate indicators, suggesting a strong association between these exposure-based clustering patterns and CVD occurrence.

Table 2.

Baseline characteristics of the study population for the association between cuIR indicators and cardiovascular disease based on the ELSA database

Variables	Overall(n = 3847)	non-CVD(n = 3074)	CVD(n = 773)	P
Gender (%)
Male	1821 (47.3)	1473 (47.9)	348 (45.0)	0.161
Female	2026 (52.7)	1601 (52.1)	425 (55.0)
Age	58.1 ± 8.5	57.7 ± 8.5	59.6 ± 8.0	< 0.001
Weight	58.9 ± 10.6	58.6 ± 10.5	60.0 ± 11.0	0.001
Height	1.6 ± 0.1	1.6 ± 0.1	1.6 ± 0.1	0.421
WC	85.2 ± 9.3	84.8 ± 9.1	86.8 ± 9.7	< 0.001
BMI	20.6 ± 1.2	20.6 ± 1.2	20.7 ± 1.2	0.322
Education (%)
Below high school	3480 (90.5)	2778 (90.4)	702 (90.8)	0.928
High school or equivalent	335 ( 8.7)	270 ( 8.8)	65 ( 8.4)
College or above	32 ( 0.8)	26 ( 0.8)	6 ( 0.8)
Marital (%)
Widowed/Divorced/Separated/Unmarrie	359 ( 9.3)	281 ( 9.1)	78 (10.1)	0.458
Married/Cohabiting	3488 (90.7)	2793 (90.9)	695 (89.9)
Drink (%)
NO	2300 (59.8)	1833 (59.6)	467 (60.4)	0.721
Yes	1547 (40.2)	1241 (40.4)	306 (39.6)
Smoke (%)
Never smoked	2339 (60.8)	1860 (60.5)	479 (62.0)	0.027
Used to smoke	293 ( 7.6)	220 ( 7.2)	73 ( 9.4)
Currently smoke	1215 (31.6)	994 (32.3)	221 (28.6)
Asthmae (%)
NO	3707 (96.4)	2987 (97.2)	720 (93.1)	< 0.001
Yes	140 ( 3.6)	87 ( 2.8)	53 ( 6.9)
Lung (%)
NO	3552 (92.3)	2855 (92.9)	697 (90.2)	0.014
Yes	295 ( 7.7)	219 ( 7.1)	76 ( 9.8)
Arthre (%)
NO	2641 (68.7)	2163 (70.4)	478 (61.8)	< 0.001
Yes	1206 (31.3)	911 (29.6)	295 (38.2)
Cancer (%)
NO	3817 (99.2)	3050 (99.2)	767 (99.2)	1
Yes	30 ( 0.8)	24 ( 0.8)	6 ( 0.8)
Satisfaction (%)
Not satisfied	546 (14.2)	432 (14.1)	114 (14.7)	0.662
Satisfied	3301 (85.8)	2642 (85.9)	659 (85.3)
Depression (%)
NO	2724 (70.8)	2220 (72.2)	504 (65.2)	< 0.001
Yes	1123 (29.2)	854 (27.8)	269 (34.8)
Sleep problem(%)
NO	2596 (67.5)	2110 (68.6)	486 (62.9)	0.003
Yes	1251 (32.5)	964 (31.4)	287 (37.1)
Physical activity (%)
NO	1183 (30.8)	921 (30.0)	262 (33.9)	0.038
Yes	2664 (69.2)	2153 (70.0)	511 (66.1)
Income	24533.3 ± 38137.2	24862.6 ± 38270.3	23223.7 ± 37599.0	0.286
Hemoglobin	14.5 ± 2.2	14.5 ± 2.2	14.5 ± 2.2	0.713
HbA1c	5.2 ± 0.6	5.2 ± 0.6	5.3 ± 0.7	0.004
TG	125.9 ± 71.5	124.4 ± 71.2	131.9 ± 72.3	0.009
TC	196.6 ± 36.6	195.5 ± 36.5	201.0 ± 36.9	< 0.001
CRP	2.4 ± 6.3	2.3 ± 6.3	2.6 ± 6.6	0.21
BUN	15.7 ± 4.4	15.8 ± 4.4	15.5 ± 4.4	0.14
SCR	0.8 ± 0.2	0.8 ± 0.2	0.8 ± 0.2	0.795
SUA	4.4 ± 1.2	4.4 ± 1.2	4.4 ± 1.3	0.886
LDL	120.4 ± 33.3	119.3 ± 33.0	125.0 ± 34.1	< 0.001
HDL	51.8 ± 15.1	52.2 ± 15.4	50.3 ± 13.7	0.002
FBG	107.5 ± 23.4	107.0 ± 22.5	109.2 ± 26.5	0.017
SBP	129.3 ± 19.3	128.6 ± 19.0	132.2 ± 20.0	< 0.001
DBP	75.7 ± 11.3	75.4 ± 11.2	76.9 ± 11.5	0.001
cuTyG	26.0 ± 1.5	26.0 ± 1.5	26.2 ± 1.5	< 0.001
cuTyG-WC	2233.7 ± 303.8	2219.3 ± 300.1	2290.8 ± 312.1	< 0.001
cuTyG-WWI	292.0 ± 28.2	290.8 ± 27.9	296.6 ± 28.9	< 0.001
cuTyG-BMI	577.8 ± 69.7	575.0 ± 69.1	589.0 ± 70.8	< 0.001
cuTyG-ABSI	2.3 ± 0.2	2.3 ± 0.2	2.3 ± 0.2	< 0.001
cuTyG-BRI	111.9 ± 34.5	110.2 ± 33.8	118.7 ± 36.4	< 0.001
cuTyG-WHtR	14.2 ± 2.0	14.1 ± 1.9	14.6 ± 2.0	< 0.001
cuMETS-IR	99.0 ± 13.8	98.4 ± 13.7	101.1 ± 13.8	< 0.001
cuAIP	1.1 ± 0.8	1.1 ± 0.8	1.2 ± 0.7	< 0.001
cuCTI	26.3 ± 2.0	26.2 ± 2.0	26.6 ± 2.0	< 0.001
cuCHG	15.8 ± 0.9	15.7 ± 0.9	15.9 ± 0.9	< 0.001
Cluster_TyG (%)
Class 1	1345 (35.0)	1112 (36.2)	233 (30.1)	0.003
Class 3	1671 (43.4)	1324 (43.1)	347 (44.9)
Class 2	831 (21.6)	638 (20.8)	193 (25.0)
Cluster_TyG-WC (%)
Class 3	1243 (32.3)	1053 (34.3)	190 (24.6)	< 0.001
Class 2	1602 (41.6)	1273 (41.4)	329 (42.6)
Class 1	1002 (26.0)	748 (24.3)	254 (32.9)
Cluster_TyG-WWI (%)
Class 1	1162 (30.2)	972 (31.6)	190 (24.6)	< 0.001
Class 3	1705 (44.3)	1355 (44.1)	350 (45.3)
Class 2	980 (25.5)	747 (24.3)	233 (30.1)
Cluster_TyG-BMI (%)
Class 3	947 (24.6)	800 (26.0)	147 (19.0)	< 0.001
Class 1	1817 (47.2)	1448 (47.1)	369 (47.7)
Class 2	1083 (28.2)	826 (26.9)	257 (33.2)
Cluster_TyG-ABSI (%)
Class 3	1344 (34.9)	1135 (36.9)	209 (27.0)	< 0.001
Class 1	1650 (42.9)	1302 (42.4)	348 (45.0)
Class 2	853 (22.2)	637 (20.7)	216 (27.9)
Cluster_TyG-BRI(%)
Class 1	1388 (36.1)	1169 (38.0)	219 (28.3)	< 0.001
Class 2	1637 (42.6)	1291 (42.0)	346 (44.8)
Class 3	822 (21.4)	614 (20.0)	208 (26.9)
Cluster_TyG-WHtR(%)
Class 3	1271 (33.0)	1074 (34.9)	197 (25.5)	< 0.001
Class 1	1662 (43.2)	1314 (42.7)	348 (45.0)
Class 2	914 (23.8)	686 (22.3)	228 (29.5)
Cluster_METS-IR(%)
Class 3	1129 (29.3)	943 (30.7)	186 (24.1)	< 0.001
Class 1	1757 (45.7)	1399 (45.5)	358 (46.3)
Class 2	961 (25.0)	732 (23.8)	229 (29.6)
Cluster_AIP (%)
Class 3	1310 (34.1)	1091 (35.5)	219 (28.3)	0.001
Class 1	1731 (45.0)	1362 (44.3)	369 (47.7)
Class 2	806 (21.0)	621 (20.2)	185 (23.9)
Cluster_CTI (%)
Class 1	1303 (33.9)	1096 (35.7)	207 (26.8)	< 0.001
Class 3	1697 (44.1)	1325 (43.1)	372 (48.1)
Class 2	847 (22.0)	653 (21.2)	194 (25.1)
Cluster_CHG (%)
Class 3	1296 (33.7)	1085 (35.3)	211 (27.3)	< 0.001
Class 1	1798 (46.7)	1421 (46.2)	377 (48.8)
Class 2	753 (19.6)	568 (18.5)	185 (23.9)

Baseline characteristics of participants in the ELSA trajectory analysis

The trajectory analysis included a total of 2,740 eligible participants from the ELSA database, of whom 360 (13.1%) had cardiovascular disease (CVD) at baseline, while 2,380 (86.9%) did not have CVD. Compared with participants without CVD, those with CVD were older (67.5 ± 8.0 vs. 62.9 ± 7.7, P < 0.001), had lower educational attainment (P = 0.009), and had a higher proportion of widowed, divorced, or separated individuals (P = 0.021); however, there was no significant difference in gender distribution (Table 3).

Table 3.

Baseline characteristics of the study population for the association between IR trajectories and cardiovascular disease based on the ELSA database

Variables	Overall(n = 2740)	non-CVD(n = 2380)	CVD(n = 360)	P
Weight	76.5 (14.3)	76.3 (14.2)	77.4 (14.9)	0.176
Height	1.7 (0.1)	1.7 (0.1)	1.7 (0.1)	0.072
WC	94.4 (12.1)	94.2 (12.0)	95.9 (12.5)	0.011
BMI	27.7 (4.2)	27.6 (4.2)	28.3 (4.4)	0.002
HbA1c	5.5 (0.6)	5.5 (0.5)	5.6 (0.7)	< 0.001
Age	63.5 (7.9)	62.9 (7.7)	67.5 (8.0)	< 0.001
Gender (%)
Male	1185 (43.2)	1033 (43.4)	152 (42.2)	0.715
Female	1555 (56.8)	1347 (56.6)	208 (57.8)
Education (%)
Below high school	900 (32.8)	757 (31.8)	143 (39.7)	0.009
High school or equivalent	1382 (50.4)	1214 (51.0)	168 (46.7)
College or above	458 (16.7)	409 (17.2)	49 (13.6)
Marital (%)
Widowed/Divorced/Separated/Unmarrie	727 (26.5)	613 (25.8)	114 (31.7)	0.021
Married/Cohabiting	2013 (73.5)	1767 (74.2)	246 (68.3)
Drink (%)
NO	197 ( 7.2)	163 ( 6.8)	34 ( 9.4)	0.095
Yes	2543 (92.8)	2217 (93.2)	326 (90.6)
Smoke (%)
Never smoked	1147 (41.9)	1000 (42.0)	147 (40.8)	0.313
Used to smoke	1264 (46.1)	1087 (45.7)	177 (49.2)
Currently smoke	329 (12.0)	293 (12.3)	36 (10.0)
TG	1.8 (1.1)	1.8 (1.1)	1.8 (1.0)	0.448
TC	6.0 (1.2)	6.0 (1.2)	6.0 (1.4)	0.827
CRP	3.5 (5.8)	3.4 (5.8)	4.1 (5.4)	0.029
LDL	3.7 (1.1)	3.7 (1.1)	3.7 (1.2)	0.414
HD	1.6 (0.4)	1.6 (0.4)	1.5 (0.4)	0.015
Glucose	5.1 (0.9)	5.0 (0.9)	5.1 (0.9)	0.616
SBP	134.0 (16.7)	133.5 (16.8)	137.3 (15.6)	< 0.001
DB	76.3 (9.8)	76.4 (9.8)	75.7 (9.8)	0.241
Asthmae (%)
NO	2425 (88.5)	2114 (88.8)	311 (86.4)	0.207
Yes	315 (11.5)	266 (11.2)	49 (13.6)
Lung (%)
NO	2637 (96.2)	2303 (96.8)	334 (92.8)	< 0.001
Yes	103 ( 3.8)	77 ( 3.2)	26 ( 7.2)
Arthre (%)
NO	1895 (69.2)	1683 (70.7)	212 (58.9)	< 0.001
Yes	845 (30.8)	697 (29.3)	148 (41.1)
Cancer (%)
NO	2586 (94.4)	2243 (94.2)	343 (95.3)	0.502
Yes	154 ( 5.6)	137 ( 5.8)	17 ( 4.7)
Satisfaction (%)
Not satisfied	390 (14.2)	330 (13.9)	60 (16.7)	0.181
Satisfied	2350 (85.8)	2050 (86.1)	300 (83.3)
Depression (%)
NO	2392 (87.3)	2091 (87.9)	301 (83.6)	0.03
Yes	348 (12.7)	289 (12.1)	59 (16.4)
Sleep problem(%)
NO	1650 (60.2)	1441 (60.5)	209 (58.1)	0.4
Yes	1090 (39.8)	939 (39.5)	151 (41.9)
Physical activity (%)
NO	190 ( 6.9)	153 ( 6.4)	37 (10.3)	0.01
Yes	2550 (93.1)	2227 (93.6)	323 (89.7)
Income	23191.0 (20734.1)	23750.1 (21680.3)	19494.5 (12221.4)	< 0.001
Hemoglobin	14.4 (1.3)	14.4 (1.3)	14.4 (1.4)	0.857
TyG_group (%)
Trajectory 3	605 (22.1)	537 (22.6)	68 (18.9)	0.089
Trajectory 2	1944 (70.9)	1685 (70.8)	259 (71.9)
Trajectory 1	191 ( 7.0)	158 ( 6.6)	33 ( 9.2)
TyG-WC_group (%)
Trajectory 2	657 (24.0)	594 (25.0)	63 (17.5)	0.002
Trajectory 3	1310 (47.8)	1137 (47.8)	173 (48.1)
Trajectory 1	773 (28.2)	649 (27.3)	124 (34.4)
TyG-WWI_group (%)
Trajectory 3	340 (12.4)	309 (13.0)	31 ( 8.6)	< 0.001
Trajectory 2	1395 (50.9)	1232 (51.8)	163 (45.3)
Trajectory 1	1005 (36.7)	839 (35.3)	166 (46.1)
TyG-BMI_group (%)
Trajectory 2	778 (28.4)	695 (29.2)	83 (23.1)	0.035
Trajectory 1	1224 (44.7)	1058 (44.5)	166 (46.1)
Trajectory 3	738 (26.9)	627 (26.3)	111 (30.8)
TyG-ABSI_group (%)
Trajectory 3	412 (15.0)	371 (15.6)	41 (11.4)	0.01
Trajectory 2	1391 (50.8)	1218 (51.2)	173 (48.1)
Trajectory 1	937 (34.2)	791 (33.2)	146 (40.6)
TyG-BRI_group (%)
Trajectory 2	730 (26.6)	655 (27.5)	75 (20.8)	0.001
Trajectory 1	1268 (46.3)	1106 (46.5)	162 (45.0)
Trajectory 3	742 (27.1)	619 (26.0)	123 (34.2)
METS-IR_group (%)
Trajectory 3	789 (28.8)	706 (29.7)	83 (23.1)	0.004
Trajectory 2	1199 (43.8)	1044 (43.9)	155 (43.1)
Trajectory 1	752 (27.4)	630 (26.5)	122 (33.9)
TyG-WHtR_group (%)
Trajectory 2	642 (23.4)	580 (24.4)	62 (17.2)	< 0.001
Trajectory 1	1328 (48.5)	1165 (48.9)	163 (45.3)
Trajectory 3	770 (28.1)	635 (26.7)	135 (37.5)
AIP_group (%)
Trajectory 3	560 (20.4)	504 (21.2)	56 (15.6)	0.024
Trajectory 2	1909 (69.7)	1649 (69.3)	260 (72.2)
Trajectory 1	271 ( 9.9)	227 ( 9.5)	44 (12.2)
CTI_group (%)
Trajectory 3	451 (16.5)	416 (17.5)	35 ( 9.7)	< 0.001
Trajectory 2	1918 (70.0)	1660 (69.7)	258 (71.7)
Trajectory 1	371 (13.5)	304 (12.8)	67 (18.6)
CHG_group (%)
Trajectory 3	438 (16.0)	382 (16.1)	56 (15.6)	0.802
Trajectory 2	1340 (48.9)	1168 (49.1)	172 (47.8)
Trajectory 1	962 (35.1)	830 (34.9)	132 (36.7)

Participants in the CVD group had higher waist circumference (95.9 ± 12.5 vs. 94.2 ± 12.0, P = 0.011), BMI (28.3 ± 4.4 vs. 27.6 ± 4.2, P = 0.002), glycated hemoglobin (P < 0.001), systolic blood pressure (P < 0.001), C-reactive protein (P = 0.029), and low-density lipoprotein cholesterol (P = 0.015) were all significantly elevated, while high-density lipoprotein cholesterol levels were significantly lower (P = 0.015). In addition, this group had a higher proportion of individuals with insufficient physical activity (10.3% vs. 6.9%, P = 0.01), a higher prevalence of arthritis (41.1% vs. 29.3%, P < 0.001), and a higher prevalence of lung disease (7.2% vs. 3.2%, P < 0.001). The CVD group had a higher prevalence of depressive symptoms (16.4% vs. 12.9%, P = 0.03) and lower social engagement (P = 0.018).

The distribution of trajectory groups for most insulin resistance indices (including TyG-WC, TyGWWI, TyG-WHTR, TyG-BMI, TyG-ABSI, TyG-BRI, TyGwhtr, METS-IR, and CTI) showed significant differences between the two groups (P < 0.05); no significant differences were observed for the TyG index and CHG.

Association between baseline IR surrogate indicators and CVD risk

The results of the Schoenfeld residual test indicated that, in the CHARLS cohort, the four variables—“educational level” “marital status” “physical activity” and “arthritis”—did not meet the proportional hazards (PH) assumption. Therefore, in the Cox regression analysis, these variables were included as stratification variables in a stratified Cox proportional hazards model. In contrast, no violation of the PH assumption was found in the ELSA cohort (Supplementary Table 3), and a standard Cox proportional hazards model was used. It should be emphasized that the primary exposure variable (baseline insulin resistance index) satisfied the PH assumption in both cohorts. After the aforementioned adjustments, its association with cardiovascular disease risk remained statistically significant, and the effect size and direction remained largely consistent.

In the CHARLS cohort study, baseline surrogate markers of insulin resistance were consistently associated with an increased risk of cardiovascular disease (Table 4). When analyzed as continuous variables, all markers showed a significant positive correlation (all P < 0.001). Among these, TyG-ABSI showed the strongest effect (HR = 4.93, 95% CI 3.14–7.75), while TyG, TyG-WC, TyG-WWI, TyG-BMI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, and CHG also demonstrated significant associations.In the tertile analysis, most indicators exhibited a dose-response relationship, with an increasing trend in cardiovascular disease risk across each category. For TyG, TyG-WWI, TyG-BMI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, and CHG, both the second and third quartiles were significantly associated with increased risk, whereas TyG-WC and TyG-ABSI were significant primarily at higher levels.

Table 4.

Cox regression analysis of baseline IR markers and cardiovascular disease risk based on CHARLS data

Variables	Model1		Model2		Model3
	HR(95%CI)	P	HR(95%CI)	P	HR(95%CI)	P
TyG	1.2(1.11, 1.30)	< 0.001	1.21(1.12, 1.31)	< 0.001	1.2(1.11, 1.30)	< 0.001
TyG tertile groups
Q1	Reference		Reference		Reference
Q2	1.22(1.09, 1.37)	< 0.001	1.22(1.09, 1.37)	< 0.001	1.21(1.07, 1.35)	0.001
Q3	1.27(1.14, 1.42)	< 0.001	1.27(1.13, 1.42)	< 0.001	1.26(1.13, 1.41)	< 0.001
P for trend	1.19(1.09, 1.28)	< 0.001	1.18(1.09, 1.28)	< 0.001	1.18(1.09, 1.28)	< 0.001
TyG-WC	1(1.00, 1.00)	< 0.001	1(1.00, 1.00)	< 0.001	1(1.00, 1.00)	< 0.001
TyG-WC tertile groups
Q1	Reference		Reference		Reference
Q2	1.12(1.00, 1.26)	0.055	1.13(1.00, 1.27)	0.049	1.13(1.00, 1.27)	0.052
Q3	1.48(1.32, 1.65)	< 0.001	1.49(1.33, 1.66)	< 0.001	1.5(1.33, 1.68)	< 0.001
P for trend	1.32(1.22, 1.43)	< 0.001	1.32(1.22, 1.43)	< 0.001	1.33(1.22, 1.44)	< 0.001
TyG-WWI	1.02(1.01, 1.02)	< 0.001	1.01(1.00, 1.01)	< 0.001	1.01(1.00, 1.01)	< 0.001
TyG-WWI tertile groups
Q1	Reference		Reference		Reference
Q2	1.21(1.08, 1.36)	0.001	1.14(1.02, 1.29)	0.025	1.12(1.00, 1.27)	0.053
Q3	1.41(1.26, 1.58)	< 0.001	1.24(1.10, 1.40)	< 0.001	1.2(1.07, 1.36)	0.003
P for trend	1.28(1.18, 1.38)	< 0.001	1.16(1.07, 1.27)	< 0.001	1.14(1.05, 1.24)	0.002
TyG-BMI	1.01(1.00, 1.01)	< 0.001	1.01(1.00, 1.01)	< 0.001	1.01(1.00, 1.01)	< 0.001
TyG-BMI tertile groups
Q1	Reference		Reference		Reference
Q2	1.1(0.98, 1.23)	0.117	1.15(1.02, 1.29)	0.022	1.16(1.03, 1.30)	0.013
Q3	1.25(1.12, 1.40)	< 0.001	1.31(1.17, 1.47)	< 0.001	1.31(1.17, 1.47)	< 0.001
P for trend	1.17(1.08, 1.27)	< 0.001	1.21(1.12, 1.31)	< 0.001	1.21(1.12, 1.31)	< 0.001
TyG-ABSI	6.1(3.98,9.34)	< 0.001	5.08(3.26,7.93)	< 0.001	4.93(3.14,7.75)	< 0.001
TyG-ABSI tertile groups
Q1	Reference		Reference		Reference
Q2	1.13(1.01, 1.28)	0.035	1.09(0.96, 1.22)	0.172	1.08(0.96, 1.22)	0.189
Q3	1.47(1.31, 1.64)	< 0.001	1.39(1.24, 1.56)	< 0.001	1.38(1.23, 1.55)	< 0.001
P for trend	1.31(1.21, 1.42)	< 0.001	1.26(1.16, 1.37)	< 0.001	1.25(1.15, 1.36)	< 0.001
TyG-BRI	1.02(1.01, 1.02)	< 0.001	1.01(1.01, 1.02)	< 0.001	1.01(1.01, 1.02)	< 0.001
TyG-BRI tertile groups
Q1	Reference		Reference		Reference
Q2	1.26(1.12, 1.42)	< 0.001	1.25(1.11, 1.41)	< 0.001	1.24(1.10, 1.40)	< 0.001
Q3	1.54(1.38, 1.73)	< 0.001	1.47(1.30, 1.65)	< 0.001	1.44(1.28, 1.63)	< 0.001
P for trend	1.36(1.26, 1.47)	< 0.001	1.31(1.20, 1.43)	< 0.001	1.29(1.19, 1.41)	< 0.001
TyG-WHtR	1.3(1.22, 1.39)	< 0.001	1.27(1.19, 1.36)	< 0.001	1.26(1.18, 1.35)	< 0.001
TyG-WHtR tertile groups
Q1	Reference		Reference		Reference
Q2	1.2(1.07, 1.35)	0.002	1.19(1.05, 1.34)	0.005	1.18(1.05, 1.33)	0.006
Q3	1.52(1.36, 1.70)	< 0.001	1.46(1.30, 1.65)	< 0.001	1.45(1.28, 1.63)	< 0.001
P for trend	1.34(1.24, 1.46)	< 0.001	1.31(1.20, 1.42)	< 0.001	1.3(1.19, 1.41)	< 0.001
METS-IR	1.02(1.01, 1.03)	0.001	1.03(1.01, 1.04)	< 0.001	1.03(1.02, 1.04)	< 0.001
METS-IR tertile groups
Q1	Reference		Reference		Reference
Q2	1.14(1.02, 1.28)	0.024	1.2(1.07, 1.34)	0.002	1.21(1.07, 1.35)	0.002
Q3	1.26(1.13, 1.41)	< 0.001	1.34(1.19, 1.50)	< 0.001	1.35(1.20, 1.51)	< 0.001
P for trend	1.18(1.09, 1.28)	< 0.001	1.23(1.13, 1.33)	< 0.001	1.24(1.14, 1.34)	< 0.001
AIP	1.36(1.17, 1.57)	< 0.001	1.41(1.21, 1.63)	< 0.001	1.41(1.21, 1.64)	< 0.001
AIP tertile groups
Q1	Reference		Reference		Reference
Q2	1.19(1.06, 1.33)	0.003	1.2(1.07, 1.34)	0.002	1.18(1.05, 1.32)	0.004
Q3	1.25(1.11, 1.40)	< 0.001	1.28(1.14, 1.43)	< 0.001	1.27(1.14, 1.43)	< 0.001
P for trend	1.17(1.08, 1.27)	< 0.001	1.19(1.10, 1.29)	< 0.001	1.19(1.09, 1.29)	< 0.001
CTI	1.21(1.14, 1.29)	< 0.001	1.19(1.12, 1.26)	< 0.001	1.18(1.11, 1.25)	< 0.001
CTI tertile groups
Q1	Reference		Reference		Reference
Q2	1.23(1.09, 1.38)	< 0.001	1.19(1.06, 1.33)	0.004	1.17(1.04, 1.31)	0.008
Q3	1.38(1.23, 1.54)	< 0.001	1.33(1.19, 1.49)	< 0.001	1.3(1.16, 1.46)	< 0.001
P for trend	1.26(1.16, 1.36)	< 0.001	1.22(1.13, 1.32)	< 0.001	1.21(1.11, 1.31)	< 0.001
CHG	1.32(1.17, 1.48)	< 0.001	1.29(1.14, 1.45)	< 0.001	1.3(1.15, 1.47)	< 0.001
CHG tertile groups
Q1	Reference		Reference		Reference
Q2	1.19(1.06, 1.33)	0.004	1.18(1.05, 1.33)	0.004	1.17(1.05, 1.32)	0.007
Q3	1.26(1.13, 1.41)	< 0.001	1.24(1.11, 1.39)	< 0.001	1.25(1.11, 1.40)	< 0.001
P for trend	1.18(1.09, 1.27)	< 0.001	1.17(1.08, 1.26)	< 0.001	1.17(1.08, 1.27)	< 0.001

In the ELSA cohort, similar associations were observed in the unadjusted analysis (Supplementary Table 4), where all surrogate markers of insulin resistance were associated with the risk of cardiovascular disease. However, after multivariate adjustment, only TyG-WC, TyG-BMI, TyG-BRI, TyG-WHtR, METS-IR, and CTI remained significantly associated with cardiovascular disease risk, while the associations between TyG, TyG-WWI, TyG-ABSI, AIP, and CHG and cardiovascular disease risk weakened. A quantile analysis of the ELSA cohort revealed a similar but weaker pattern. The highest tertile (Q3) of TyG-WC, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, and METS-IR was significantly associated with an increased risk of cardiovascular disease, whereas TyG, TyG-WWI, AIP, and CHG were not significant after adjustment. TyG-WWI and CTI were significant in partially adjusted models but not in fully adjusted models.

Overall, baseline surrogate markers of insulin resistance were positively associated with new-onset cardiovascular disease in both cohorts. The associations were more consistent in the CHARLS cohort, whereas in the ELSA cohort, TyG-WC, TyG-BMI, TyG-BRI, TyG-WHtR, METS-IR, and CTI showed relatively more stable associations after full adjustment.

Association between cuIR and CVD risk

We conducted proportional hazards (PH) tests on the Cox proportional hazards models. The results showed that the test statistics for all exposure variables and covariates were not statistically significant (all P > 0.05), indicating that these models satisfy the proportional hazards assumption (Supplementary Table 5).

In the multivariable Cox proportional hazards model, all 11 cumulative IR-related surrogate indicators included in this study (cuTyG, cuTyG-WC, cuTyG-WWI, cuTyG-BMI, cuTyG-ABSI, cuTyG-BRI, cuTyG-WHtR, cuMETS-IR, cuAIP, cuCTI, and cuCHG) were significantly associated with CVD incidence risk (Table 5). When these indicators were incorporated as continuous variables into the models, results showed that at different levels of model adjustment (Model 1, Model 2, and Model 3), all cuIR indicators significantly increased CVD incidence risk (P < 0.001). Further analysis by quantile grouping revealed that across all models, compared with the lowest quantile (Q1), both the middle (Q2) and highest (Q3) quantiles showed significantly elevated CVD risk, with Q3 exhibiting the highest risk level. Trend tests indicated a significant linear trend for all indicators (P for trend < 0.05), suggesting that CVD risk progressively increases with rising cuIR levels.In K-means clustering analysis based on cuIR exposure, each IR surrogate indicator identified distinct population groups with different exposure-based clustering patterns. ow-level reference cluster, the high-level cluster exhibited significantly elevated CVD risk. This risk differential demonstrated consistent directionality across clustering groups defined by different IR surrogate indicators and remained stable across varying levels of model adjustment. Overall, whether assessed as continuous variables, quantile-grouped variables, or exposure-based clustering patterns, all 11 cuIR-related surrogate indicators demonstrated that higher levels or high-risk clustering patterns were significantly associated with increased CVD incidence risk. This association exhibited robust stability across different levels of model adjustment, indicating an upward trend in CVD incidence risk with increasing cuIR severity.

Table 5.

Cox regression analysis of cuIR markers and cardiovascular disease risk based on CHARLS data

Variables	Model1		Model2		Model3
	HR(95%CI)	P	HR(95%CI)	P	HR(95%CI)	P
cuTyG	1.09(1.04, 1.14)	< 0.001	1.1(1.05, 1.15)	< 0.001	1.1(1.05, 1.15)	< 0.001
cuTyG4 tertile groups
Q1	Reference		Reference		Reference
Q2	1.26(1.06, 1.51)	0.011	1.27(1.06, 1.52)	0.009	1.25(1.05, 1.50)	0.013
Q3	1.31(1.09, 1.56)	0.003	1.34(1.12, 1.60)	0.002	1.31(1.10, 1.57)	0.003
P for trend	1.21(1.07, 1.37)	0.003	1.23(1.08, 1.39)	0.002	1.21(1.07, 1.38)	0.003
Cluster_TyG
Class 1	Reference		Reference		Reference
Class 3	1.23(1.04, 1.45)	0.016	1.24(1.04, 1.46)	0.013	1.21(1.02, 1.43)	0.026
Class 2	1.38(1.14, 1.67)	0.001	1.42(1.17, 1.73)	< 0.001	1.41(1.16, 1.72)	< 0.001
cuTyG-WC	1(1.00, 1.00)	< 0.001	1(1.00, 1.00)	< 0.001	1(1.00, 1.00)	< 0.001
cuTyG-WC tertile groups
Q1	Reference		Reference		Reference
Q2	1.38(1.15, 1.66)	< 0.001	1.4(1.16, 1.69)	< 0.001	1.41(1.17, 1.71)	< 0.001
Q3	1.67(1.40, 1.99)	< 0.001	1.71(1.42,2.05)	< 0.001	1.71(1.42,2.06)	< 0.001
P for trend	1.44(1.27, 1.63)	< 0.001	1.46(1.28, 1.66)	< 0.001	1.46(1.28, 1.66)	< 0.001
Cluster_TyGwc
Class 3	Reference		Reference		Reference
Class 2	1.37(1.15, 1.64)	< 0.001	1.39(1.16, 1.67)	< 0.001	1.39(1.16, 1.67)	< 0.001
Class 1	1.75(1.45,2.11)	< 0.001	1.79(1.48,2.17)	< 0.001	1.8(1.48,2.19)	< 0.001
cuTyG-WWI	1.01(1.00, 1.01)	< 0.001	1.01(1.00, 1.01)	< 0.001	1(1.00, 1.01)	< 0.001
cuTyG-WWI tertile groups
Q1	Reference		Reference		Reference
Q2	1.24(1.04, 1.50)	0.02	1.22(1.01, 1.47)	0.041	1.2(1.00, 1.45)	0.055
Q3	1.54(1.29, 1.84)	< 0.001	1.45(1.20, 1.76)	< 0.001	1.4(1.15, 1.70)	< 0.001
P for trend	1.36(1.20, 1.54)	< 0.001	1.3(1.14, 1.49)	< 0.001	1.27(1.11, 1.45)	< 0.001
Cluster_TyG-WWI
Class 1	Reference		Reference		Reference
Class 3	1.28(1.08, 1.53)	0.005	1.25(1.04, 1.50)	0.016	1.23(1.03, 1.48)	0.026
Class 2	1.52(1.25, 1.84)	< 0.001	1.4(1.13, 1.72)	0.002	1.34(1.09, 1.65)	0.006
cuTyG-BMI	1(1.00, 1.00)	< 0.001	1(1.00, 1.00)	< 0.001	1(1.00, 1.00)	< 0.001
cuTyG-BMI tertile groups
Q1	Reference		Reference		Reference
Q2	1.36(1.14, 1.64)	< 0.001	1.44(1.20, 1.74)	< 0.001	1.45(1.21, 1.75)	< 0.001
Q3	1.53(1.28, 1.83)	< 0.001	1.68(1.40,2.03)	< 0.001	1.68(1.40,2.03)	< 0.001
P for trend	1.35(1.19, 1.53)	< 0.001	1.45(1.27, 1.65)	< 0.001	1.45(1.27, 1.65)	< 0.001
Cluster_TyG-BMI
Class 3	Reference		Reference		Reference
Class 1	1.35(1.11, 1.63)	0.002	1.42(1.17, 1.72)	< 0.001	1.45(1.19, 1.76)	< 0.001
Class 2	1.59(1.30, 1.95)	< 0.001	1.75(1.42,2.16)	< 0.001	1.76(1.42,2.17)	< 0.001
cuTyG-ABSI	2.47(1.82,3.35)	< 0.001	2.28(1.66,3.13)	< 0.001	2.22(1.60,3.06)	< 0.001
cuTyG-ABSI tertile groups
Q1	Reference		Reference		Reference
Q2	1.33(1.11, 1.60)	0.002	1.3(1.08, 1.56)	0.006	1.29(1.07, 1.55)	0.008
Q3	1.55(1.30, 1.86)	< 0.001	1.48(1.23, 1.78)	< 0.001	1.45(1.20, 1.74)	< 0.001
P for trend	1.37(1.20, 1.55)	< 0.001	1.32(1.16, 1.50)	< 0.001	1.3(1.14, 1.48)	< 0.001
Cluster_TyG-ABSI
Class 3	Reference		Reference		Reference
Class 1	1.4(1.18, 1.66)	< 0.001	1.36(1.14, 1.62)	< 0.001	1.35(1.14, 1.61)	< 0.001
Class 2	1.73(1.43,2.09)	< 0.001	1.66(1.36,2.02)	< 0.001	1.63(1.34, 1.99)	< 0.001
cuTyG-BRI	1.01(1.00, 1.01)	< 0.001	1.01(1.00, 1.01)	< 0.001	1.01(1.00, 1.01)	< 0.001
cuTyG-BRI tertile groups
Q1	Reference		Reference		Reference
Q2	1.28(1.06, 1.54)	0.009	1.29(1.07, 1.56)	0.008	1.3(1.07, 1.57)	0.007
Q3	1.66(1.39, 1.98)	< 0.001	1.64(1.36, 1.98)	< 0.001	1.61(1.33, 1.95)	< 0.001
P for trend	1.43(1.26, 1.62)	< 0.001	1.42(1.24, 1.62)	< 0.001	1.4(1.22, 1.60)	< 0.001
Cluster_TyG-BRI
Class 1	Reference		Reference		Reference
Class 2	1.37(1.16, 1.63)	< 0.001	1.38(1.16, 1.64)	< 0.001	1.38(1.16, 1.64)	< 0.001
Class 3	1.7(1.40,2.05)	< 0.001	1.67(1.36,2.05)	< 0.001	1.64(1.33,2.02)	< 0.001
cuTyG-WHtR	1.11(1.08, 1.15)	< 0.001	1.12(1.07, 1.16)	< 0.001	1.11(1.07, 1.16)	< 0.001
cuTyG-WHtR tertile groups
Q1	Reference		Reference		Reference
Q2	1.35(1.13, 1.63)	0.001	1.36(1.13, 1.64)	0.001	1.37(1.13, 1.66)	0.001
Q3	1.63(1.36, 1.95)	< 0.001	1.61(1.34, 1.95)	< 0.001	1.59(1.31, 1.92)	< 0.001
P for trend	1.41(1.25, 1.60)	< 0.001	1.4(1.23, 1.60)	< 0.001	1.39(1.21, 1.59)	< 0.001
Cluster_TyG-WHtR
Class 3	Reference		Reference		Reference
Class 1	1.39(1.16, 1.65)	< 0.001	1.39(1.16, 1.66)	< 0.001	1.39(1.16, 1.67)	< 0.001
Class 2	1.7(1.40,2.05)	< 0.001	1.69(1.38,2.07)	< 0.001	1.66(1.35,2.04)	< 0.001
cuMETS-IR	1.01(1.01, 1.02)	< 0.001	1.02(1.01, 1.02)	< 0.001	1.02(1.01, 1.02)	< 0.001
cuMETS-IR tertile groups
Q1	Reference		Reference		Reference
Q2	1.37(1.14, 1.64)	< 0.001	1.45(1.20, 1.74)	< 0.001	1.46(1.22, 1.76)	< 0.001
Q3	1.5(1.25, 1.79)	< 0.001	1.64(1.37, 1.97)	< 0.001	1.65(1.37, 1.99)	< 0.001
P for trend	1.33(1.17, 1.51)	< 0.001	1.42(1.25, 1.62)	< 0.001	1.43(1.25, 1.63)	< 0.001
Cluster_METS-IR
Class 3	Reference		Reference		Reference
Class 1	1.26(1.05, 1.50)	0.011	1.33(1.11, 1.59)	0.002	1.34(1.12, 1.61)	0.001
Class 2	1.51(1.24, 1.83)	< 0.001	1.65(1.35,2.01)	< 0.001	1.67(1.36,2.04)	< 0.001
cuAIP	1.18(1.08, 1.30)	< 0.001	1.22(1.11, 1.34)	< 0.001	1.22(1.11, 1.34)	< 0.001
cuAIP tertile groups
Q1	Reference		Reference		Reference
Q2	1.31(1.10, 1.57)	0.003	1.32(1.11, 1.59)	0.002	1.31(1.09, 1.57)	0.003
Q3	1.36(1.14, 1.62)	< 0.001	1.42(1.19, 1.70)	< 0.001	1.41(1.18, 1.69)	< 0.001
P for trend	1.24(1.10, 1.41)	< 0.001	1.28(1.13, 1.46)	< 0.001	1.28(1.12, 1.45)	< 0.001
Cluster_AIP
Class 3	Reference		Reference		Reference
Class 1	1.32(1.11, 1.56)	0.001	1.34(1.13, 1.58)	< 0.001	1.32(1.11, 1.56)	0.001
Class 2	1.42(1.17, 1.72)	< 0.001	1.5(1.23, 1.83)	< 0.001	1.5(1.23, 1.83)	< 0.001
cuCTI	1.09(1.05, 1.13)	< 0.001	1.09(1.05, 1.13)	< 0.001	1.08(1.05, 1.12)	< 0.001
cuCTI tertile groups
Q1	Reference		Reference		Reference
Q2	1.43(1.19, 1.71)	< 0.001	1.39(1.16, 1.67)	< 0.001	1.38(1.16, 1.66)	< 0.001
Q3	1.42(1.18, 1.70)	< 0.001	1.41(1.18, 1.69)	< 0.001	1.35(1.13, 1.62)	0.001
P for trend	1.28(1.13, 1.45)	< 0.001	1.27(1.12, 1.45)	< 0.001	1.24(1.09, 1.41)	0.001
Cluster_CTI	Reference		Reference		Reference
Class 1	Reference		Reference		Reference
Class 3	1.44(1.21, 1.70)	< 0.001	1.41(1.19, 1.67)	< 0.001	1.39(1.17, 1.65)	< 0.001
Class 2	1.49(1.23, 1.82)	< 0.001	1.5(1.23, 1.82)	< 0.001	1.47(1.20, 1.79)	< 0.001
cuCHG	1.21(1.12, 1.30)	< 0.001	1.2(1.12, 1.30)	< 0.001	1.21(1.12, 1.31)	< 0.001
cuCHG tertile groups
Q1	Reference		Reference		Reference
Q2	1.29(1.08, 1.55)	0.005	1.26(1.05, 1.52)	0.012	1.29(1.07, 1.55)	0.007
Q3	1.45(1.21, 1.73)	< 0.001	1.44(1.20, 1.72)	< 0.001	1.45(1.21, 1.74)	< 0.001
P for trend	1.3(1.15, 1.47)	< 0.001	1.29(1.14, 1.46)	< 0.001	1.3(1.15, 1.48)	< 0.001
Cluster_CHG
Class 3	Reference		Reference		Reference
Class 1	1.31(1.11, 1.55)	0.002	1.29(1.09, 1.53)	0.003	1.3(1.10, 1.54)	0.002
Class 2	1.57(1.29, 1.92)	< 0.001	1.56(1.28, 1.90)	< 0.001	1.59(1.30, 1.94)	< 0.001

Association between IR index trajectories and CVD risk

The Schoenfeld residual tests for all 11 IR trajectory indicators included in this study showed that none of the indicators violated the PH assumption (Supplementary Table 6).

Following a longitudinal trajectory analysis of 11 insulin resistance-related indicators, we used the low-risk trajectory group as a reference and employed a stepwise adjusted Cox proportional hazards model to investigate their association with new-onset CVD (Table 6). The results showed that the high-risk trajectory groups for TyG-WC, TyG-BMI, TyG-BRI, METS-IR, and CTI remained significantly positively associated with the risk of CVD in the fully adjusted multivariable model, and exhibited a clear dose–response relationship. Among these, the high-risk trajectory group for CTI showed the most significant increase in CVD risk (HR = 1.90, 95% CI 1.25–2.89, P = 0.003). In contrast, high-risk trajectories for the TyG index, TyG-WHtR, TyG-WWI, and TyG-ABSI were associated with increased CVD risk only in unadjusted models (all P < 0.05); these associations were no longer statistically significant after adjusting for confounding factors. No statistically significant association with CVD risk was observed for any trajectory groups of AIP and CHG in any of the models (all P > 0.05). Overall, long-term high-level trajectories of insulin resistance-related indicators, particularly TyG-derived composite indices, METS-IR, and CTI, were independently associated with increased CVD risk in middle-aged and older adults.

Table 6.

Cox regression analysis of IR trajectory markers and cardiovascular disease risk based on ELSA data

Variables	Model1		Model2		Model3
	HR(95%CI)	P	HR(95%CI)	P	HR(95%CI)	P
TyG_group
Trajectory 3	Reference		Reference		Reference
Trajectory 2	1.19(0.91, 1.56)	0.194	1.08(0.83, 1.42)	0.557	1.07(0.82, 1.41)	0.601
Trajectory 1	1.56(1.03, 2.37)	0.035	1.46(0.96, 2.22)	0.08	1.37(0.90, 2.09)	0.146
P for trend	1.3(1.02, 1.84)	0.035	1.31(0.97, 1.76)	0.08	1.25(0.93, 1.68)	0.15
TyG-WC_group
Trajectory 2	Reference		Reference		Reference
Trajectory 3	1.39(1.04, 1.86)	0.024	1.37(1.02, 1.85)	0.035	1.33(0.99, 1.79)	0.059
Trajectory 1	1.73(1.27, 2.34)	< 0.001	1.79(1.30, 2.48)	< 0.001	1.67(1.21, 2.32)	0.002
P for trend	1.47(1.19, 1.82)	< 0.001	1.51(1.20, 1.90)	< 0.001	1.44(1.14, 1.81)	0.002
TyG-WWI_group
Trajectory 3	Reference		Reference		Reference
Trajectory 2	1.3(0.88, 1.91)	0.183	1.12(0.76, 1.66)	0.565	1.11(0.75, 1.65)	0.602
Trajectory 1	1.88(1.28, 2.76)	0.001	1.36(0.91, 2.04)	0.129	1.31(0.87, 1.96)	0.191
P for trend	1.56(1.19, 2.05)	0.001	1.25(0.94, 1.65)	0.13	1.21(0.91, 1.61)	0.2
TyG-BMI_group
Trajectory 2	Reference		Reference		Reference
Trajectory 1	1.29(0.99, 1.68)	0.056	1.31(1.00, 1.70)	0.048	1.29(0.99, 1.69)	0.062
Trajectory 3	1.44(1.08, 1.91)	0.012	1.49(1.12, 2.00)	0.007	1.43(1.07, 1.92)	0.017
P for trend	1.29(1.06, 1.58)	0.012	1.33(1.08, 1.63)	0.007	1.29(1.05, 1.59)	0.017
TyG-ABSI_group
Trajectory 3	Reference		Reference		Reference
Trajectory 2	1.26(0.90, 1.78)	0.179	1.12(0.78, 1.58)	0.543	1.09(0.77, 1.55)	0.623
Trajectory 1	1.61(1.14, 2.28)	0.007	1.27(0.87, 1.85)	0.218	1.22(0.84, 1.78)	0.299
P for trend	1.14(1.10, 1.79)	0.007	1.18(0.91, 1.55)	0.2	1.15(0.88, 1.51)	0.3
TyG-BRI_group
Trajectory 2	Reference		Reference		Reference
Trajectory 1	1.26(0.96, 1.66)	0.099	1.09(0.82, 1.44)	0.542	1.06(0.80, 1.40)	0.684
Trajectory 3	1.67(1.25, 2.22)	< 0.001	1.39(1.04, 1.87)	0.028	1.3(0.97, 1.76)	0.082
P for trend	1.44(1.17, 1.76)	< 0.001	1.26(1.03, 1.56)	0.028	1.21(0.98, 1.49)	0.082
TyG-WHtR_group
Trajectory 2	Reference		Reference		Reference
Trajectory 1	1.26(0.90, 1.78)	0.179	1.12(0.78, 1.58)	0.543	1.09(0.77, 1.55)	0.623
Trajectory 3	1.61(1.14, 2.28)	0.007	1.27(0.87, 1.85)	0.218	1.22(0.84, 1.78)	0.299
P for trend	1.4(1.10, 1.79)	0.007	1.18(0.91, 1.55)	0.2	1.15(0.88, 1.51)	0.3
METS-IR_group
Trajectory 3	Reference		Reference		Reference
Trajectory 2	1.24(0.95, 1.62)	0.114	1.21(0.92, 1.58)	0.175	1.19(0.91, 1.56)	0.205
Trajectory 1	1.59(1.20, 2.10)	0.001	1.6(1.20, 2.14)	0.001	1.54(1.15, 2.06)	0.003
P for trend	1.39(1.14, 1.69)	0.001	1.4(1.14, 1.71)	0.001	1.36(1.11, 1.67)	0.003
AIP_group
Trajectory 3	Reference		Reference		Reference
Trajectory 2	1.39(1.04, 1.85)	0.026	1.28(0.96, 1.71)	0.099	1.28(0.95, 1.71)	0.1
Trajectory 1	1.67(1.13, 2.49)	0.011	1.6(1.07, 2.40)	0.021	1.56(1.04, 2.33)	0.031
P for trend	1.44(1.09, 1.90)	0.011	1.4(1.05, 1.86)	0.021	1.37(1.03, 1.82)	0.031
CTI_group
Trajectory 3	Reference		Reference		Reference
Trajectory 2	1.8(1.26, 2.56)	0.001	1.53(1.07, 2.18)	0.021	1.47(1.03, 2.11)	0.034
Trajectory 1	2.45(1.63, 3.69)	< 0.001	2(1.31, 3.04)	0.001	1.9(1.25, 2.89)	0.003
P for trend	1.89(1.41, 2.52)	< 0.001	1.63(1.21, 2.20)	0.001	1.57(1.17, 2.12)	0.003
CHG_group
Trajectory 3	Reference		Reference		Reference
Trajectory 2	1(0.74, 1.35)	0.988	0.97(0.71, 1.31)	0.818	0.98(0.73, 1.33)	0.912
Trajectory 1	1.08(0.79, 1.47)	0.647	1.03(0.75, 1.42)	0.855	1.06(0.77, 1.46)	0.721
P for trend	1.05(0.84, 1.31)	0.6	1.02(0.82, 1.28)	0.9	1.04(0.83, 1.31)	0.7

Dose-response relationship between baseline IR surrogate indicators and CVD risk

To further explore the dose-response relationship between baseline IR-related surrogate indicators and CVD risk, RCS functions were applied to analyze each indicator (Fig. 5). In the multivariable-adjusted Model 3, results revealed distinct association patterns between different IR-related surrogate indicators and CVD risk. Specifically, TyG, TyG-WWI, TyG-BMI, METS-IR, and CTI exhibited near-linear positive correlations with CVD risk. As indicator levels increased, CVD risk gradually rose, with overall associations being statistically significant (P for overall < 0.05). No significant nonlinear relationships were observed (P for non-linear > 0.05). This suggests a predominantly linear dose-response relationship between these indicators and CVD risk. In contrast, the RCS curves for TyG-WC, TyG-ABSI, TyG-BRI, TyG-WHtR, AIP, and CHG exhibited fluctuations across different value ranges. Their association with CVD risk was relatively flat at lower levels but showed more pronounced increases at higher levels. However, nonlinearity tests failed to reach statistical significance, suggesting their dose-response relationship remains predominantly linear overall. Overall, RCS analysis further supports the positive association between multiple IR-related surrogate indicators and CVD risk. No apparent threshold effects or significant nonlinear relationships were identified, indicating that these indicators primarily exhibit linear dose-response relationships with CVD risk.

Fig. 5.

Restricted cubic spline curve of the association between baseline IR indices and cardiovascular diseases risk based on CHARLS database

Validation of these dose-response relationships using ELSA data yielded consistent results (Supplementary Fig. 2). Specifically, TyG-WC, TyG-BMI, TyG-BRI, TyG-WHtR, METS-IR, and CTI showed statistically significant overall associations with CVD risk (P for overall < 0.05), with their RCS curves demonstrating a progressive increase in risk with rising indicator levels. while the overall association for TyG, TyG-WWI, TyG-ABSI, AIP, and CHG did not reach statistical significance (P for overall > 0.05). Furthermore, nonlinearity tests for all indicators showed no statistical significance (P for non-linearity > 0.05), suggesting that the overall relationship with CVD risk is predominantly linear or near-linear.

Dose-response relationship between cuIR and CVD risk

As shown in Fig. 6, RCS analysis was employed to evaluate the dose-response relationship between the cuIR series indicators and CVD occurrence risk. After multivariable adjustment, the overall association between each indicator and CVD occurrence risk remained statistically significant (P for overall < 0.05). Nonlinearity tests revealed no significant nonlinear effects for any indicator except cuAIP (P for non-linear > 0.05), suggesting an overall linear relationship between these indicators and CVD risk, with CVD risk progressively increasing as indicator levels rise. Notably, a significant nonlinear association existed between cuAIP and CVD risk (P for non-linear = 0.039). The RCS curve indicated an inflection point at cuAIP = 1.041, beyond which the curve slope decreased and became relatively flatter. Overall, the RCS results further confirmed that there was a positive correlation between multiple cuIR indicators and the risk of cardiovascular diseases. Except for cuAIP, no significant nonlinear relationship was observed for the other indicators.

Fig. 6.

Restricted cubic spline curve of the association between cuIR indices and cardiovascular diseases risk based on CHARLS database

Subgroup analysis

After stratification by age (< 60 vs. ≥60 years), sex, education level, marital status, smoking, alcohol consumption, and physical activity, the associations between baseline IR-related surrogate indicators and CVD risk were generally consistent across subgroups, with most interactions not reaching statistical significance (P for interaction > 0.05; Fig. 7). Notably, marital status and physical activity modified several associations. The relationships of TyG-WWI, TyG-BRI, and TyG-WHtR with CVD were significant only among married or cohabiting individuals, but not among those who were widowed, divorced, separated, or unmarried (all P for interaction < 0.05). Consistently, in the ELSA dataset, these associations were observed only in participants with physical activity (Supplementary Fig. 3). For TyG-ABSI, both marital status (P for interaction = 0.022) and physical activity (P for interaction = 0.023) showed significant effect modification, with associations present only among married/cohabiting and physically active individuals. TyG-WC also showed a marginal interaction with marital status (P for interaction = 0.049), with significant associations confined to married/cohabiting individuals.

Fig. 7.

Subgroup and interaction analysis of the association between baseline IR and cardiovascular diseases risk based on CHARLS data

Similar patterns were observed for cumulative IR indicators (Fig. 8). Most subgroup analyses showed no significant interactions. However, age modified the association between cuTyG and CVD (P for interaction = 0.016), with a stronger effect in participants aged ≥ 60 years. Marital status also modified most cuIR indicators (all P for interaction < 0.05), again with stronger associations among married or cohabiting individuals. In addition, physical activity modified several cumulative indicators (cuTyG-WC, cuTyG-BMI, cuTyG-ABSI, cuTyG-BRI, cuTyG-WHtR, and cuMETS-IR; all P for interaction < 0.05), with stronger associations observed among physically inactive individuals.

Fig. 8.

Subgroup and interaction analysis of the association between cuIR and cardiovascular diseases risk based on CHARLS data

We further performed subgroup analyses of trajectory groups using stratified Cox proportional hazards models (Fig. 9). Using the low-level group as the reference, higher levels of TyG-related indices (TyG-BRI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BMI), as well as AIP, CTI, METS-IR, and TyG, were significantly associated with increased CVD risk (all P < 0.05), whereas CHG was not (P > 0.05). In interaction analyses, only a few indicators showed significant effect modification. TyG-WC and CHG were modified by alcohol consumption status (P for interaction = 0.030 and 0.026, respectively). The association between TyG and CVD varied across education levels (P for interaction = 0.019), and that of TyG-ABSI differed by age group (P for interaction = 0.029). No significant interactions were observed for the remaining indicators, indicating that the overall associations were largely robust across subgroups.

Fig. 9.

Subgroup and interaction analysis of the association between IR trajectories and cardiovascular diseases risk based on ELSA data

Incremental predictive value of baseline IR surrogate indicators

Variables selected by the Boruta algorithm (age, sex, marital status, lung disease, hemoglobin, education level, smoking status, asthma history, and weight) were used to construct the baseline model (Supplementary Fig. 4). Compared with the baseline model (C-index = 0.592; AIC = 23086.3), sequential addition of the 11 IR-related surrogate indicators modestly improved model discrimination (C-index: 0.595–0.607) and model fit (decreased AIC) (Table 7).

Table 7.

Incremental predictive value of adding baseline IR to risk prediction models for incident CVD

Model	C-index (95%CI)	P	AIC	IDI (95%CI)	P	NRI (95%CI)	P
Basic model	0.592 (0.579,0.609)	Ref	23086.3	Ref	Ref	Ref	Ref
Basic model + TyG	0.597 (0.583,0.614)	0.001	23076.4	0.0093 (0.0066, 0.0119)	< 0.001	0.0408 (0.0124, 0.0706)	0.007
Basic model + TyG-WC	0.607 (0.593,0.624)	< 0.001	23044.9	0.0307 (0.0259, 0.0356)	< 0.001	0.0637 (0.0252, 0.1041)	0.002
Basic model + TyG-WWI	0.595 (0.581,0.612)	0.006	23080.7	0.0067 (0.0045, 0.0086)	< 0.001	0.0142 (0.0092, 0.0417)	0.275
Basic model + TyG-BMI	0.597 (0.584,0.614)	< 0.001	23075.6	0.0092 (0.0066, 0.0118)	< 0.001	0.0269 (0.0019, 0.0606)	0.08
Basic model + TyG-ABSI	0.605 (0.592,0.622)	< 0.001	23050.7	0.0278 (0.0232, 0.0321)	< 0.001	0.0533 (0.0186, 0.0943)	0.006
Basic model + TyG-BRI	0.604 (0.591,0.62)	< 0.001	23054.2	0.025 (0.0207, 0.0292)	< 0.001	0.0457 (0.0067, 0.0831)	0.019
Basic model + TyG-WHtR	0.604(0.591,0.621)	< 0.001	23,056	0.0242 (0.0199, 0.0283)	< 0.001	0.0506 (0.0163, 0.0917)	0.008
Basic model+METS-IR	0.6 (0.586,0.617)	< 0.001	23071.6	0.0106 (0.0076, 0.0136)	< 0.001	0.0356 (0.0051, 0.0673)	0.021
Basic model + AIP	0.599 (0.586,0.617)	< 0.001	23071.7	0.0118 (0.0085, 0.0149)	< 0.001	0.0396 (0.012, 0.0757)	0.013
Basic model + CTI	0.601 (0.587,0.618)	< 0.001	23066.5	0.0164 (0.0127, 0.0201)	< 0.001	0.0511 (0.022, 0.0854)	0.002
Basic model + CHG	0.6 (0.586,0.617)	< 0.001	23069.6	0.0126 (0.0094, 0.0156)	< 0.001	0.0484 (0.0182, 0.0798)	0.002

Among these, TyG-WC showed the greatest incremental predictive value, increasing the C-index to 0.607 (P < 0.001) and reducing the AIC to 23044.9, with significant improvements in both discrimination (IDI = 0.0307, P < 0.001) and reclassification (NRI = 0.0637, P = 0.002). In addition, TyG, TyG-ABSI, TyG-BRI, TyG-WHtR, CTI, CHG, METS-IR, and AIP significantly improved discrimination (all P < 0.001) and provided additional predictive information (IDI: all P < 0.001; NRI: all P < 0.05). By contrast, TyG-WWI and TyG-BMI improved discrimination and IDI (both P < 0.001), but did not significantly enhance reclassification (NRI: P = 0.275 and 0.08, respectively).

In the ELSA dataset (Supplementary Table 7), improvements were more limited. Compared with the baseline model (C-index = 0.662; AIC = 14943.5), addition of the IR indicators resulted in only marginal increases in discrimination (C-index: 0.662–0.666). TyG-WC, TyG-BMI, and METS-IR yielded the highest C-index (0.666, all P < 0.001) and improved model fit (AIC: 14932.3, 14931.8, and 14932.3, respectively). All indicators significantly improved IDI (all P < 0.001), with the largest gains observed for TyG-BMI (IDI = 0.0216) and TyG-WC (IDI = 0.0212). However, none significantly improved risk reclassification (all NRI P > 0.05), although METS-IR showed a borderline effect (NRI = 0.0289, P = 0.069).

Incremental predictive value of cuIR

Educational level, arthritis, hemoglobin, history of asthma, smoking status, gender and age were used to construct the basic model (Supplementary Fig. 3). As shown in Table 8, the baseline model demonstrated a C-index of 0.596 and an AIC of 12499.5. Incorporating each cuIR indicator sequentially into the baseline model improved model discrimination while significantly increasing the IDI for each model, indicating that cuIR indicators provide additional information for CVD risk prediction. Among these, cuTyG-WC demonstrated the most pronounced incremental predictive value: incorporating cuTyG-WC elevated the C-index to 0.614, reduced AIC to 12465.2, and yielded IDI = 0.0426 and NRI = 0.0729, indicating substantial improvements in both model discrimination and reclassification capability. Regarding NRI: Beyond cuTyG-WC, cuTyG-BMI, cuTyG-WHtR, cuMETS-IR, cuAIP, cuCTI, and cuCHG also demonstrated significant reclassification improvements. In contrast, NRI did not reach statistical significance for cuTyG (P = 0.109), cuTyG-WWI (P = 0.339), cuTyG-ABSI (P = 0.276), and cuTyG-BRI (P = 0.058).

Table 8.

Incremental predictive value of adding cuIR to risk prediction models for incident CVD

Model	C-index (95%CI)	P	AIC	IDI (95%CI)	P	NRI (95%CI)	P
Basic model	0.596 (0.578, 0.619)	Ref	12499.5	Ref	Ref	Ref	Ref
Basic model+cuTyG	0.603 (0.584, 0.627)	< 0.001	12488.1	0.0174 (0.0135, 0.0217)	< 0.001	0.0326 (− 0.0024, 0.0756)	0.109
Basic model+cuTyG-WC	0.614 (0.596, 0.638)	< 0.001	12465.2	0.0426 (0.0358, 0.0492)	< 0.001	0.0729 (0.0214, 0.1189)	0.003
Basic model+cuTyG-WWI	0.601 (0.583, 0.625)	< 0.001	12488.9	0.0171 (0.0134, 0.0212)	< 0.001	0.0195 (− 0.0176, 0.0599)	0.339
Basic model+cuTyG-BMI	0.613 (0.595, 0.637)	< 0.001	12467.7	0.0397 (0.033, 0.0461)	< 0.001	0.0661 (0.014, 0.1097)	0.007
Basic model+cuTyG-ABSI	0.608 (0.588, 0.631)	< 0.001	12477.9	0.0298 (0.0246, 0.0353)	< 0.001	0.0256 (− 0.0196, 0.0722)	0.276
Basic model+cuTyG-BRI	0.609 (0.591, 0.634)	< 0.001	12,472	0.0353 (0.0291, 0.0411)	< 0.001	0.0467 (− 0.0032, 0.0918)	0.058
Basic model+cuTyG-WHtR	0.61 (0.592, 0.634)	< 0.001	12471.6	0.0367 (0.0304, 0.0427)	< 0.001	0.051 (4e-04, 0.0981)	0.037
Basic model+cuMETS-IR	0.613 (0.595, 0.636)	< 0.001	12,469	0.0357 (0.0294, 0.0423)	< 0.001	0.057 (0.0076, 0.1055)	0.022
Basic model+cuAIP	0.605 (0.586, 0.628)	< 0.001	12484.8	0.0196 (0.0152, 0.0245)	< 0.001	0.0445 (0.0059, 0.0889)	0.036
Basic model+cuCTI	0.606 (0.588, 0.629)	< 0.001	12481.6	0.0256 (0.0206, 0.0305)	< 0.001	0.0475 (6e-04, 0.0899)	0.036
Basic model+cuCHG	0.61 (0.592, 0.633)	< 0.001	12477.9	0.0254 (0.0203, 0.0311)	< 0.001	0.0547 (0.0095, 0.0977)	0.012

Sensitivity analysis

In the Fine–Gray competing risk model, treating death as a competing event, baseline IR-related surrogate indicators remained significantly associated with incident CVD in the CHARLS cohort (Supplementary Table 8). Continuous measures of IR indicators were positively associated with CVD risk. In quantile analyses, CVD risk increased progressively with higher levels of insulin resistance, with both Q2 and Q3 groups showing significantly higher risks compared with the reference group (Q1).

In the ELSA cohort, the direction of associations was generally consistent, although effect sizes were smaller and overall consistency was reduced (Supplementary Table 8). AIP was not significantly associated with CVD risk in continuous analysis (sHR = 1.15, P = 0.18), but showed a significant increase in risk in the Q2 group (sHR = 1.23, P = 0.009), while the Q3 group approached statistical significance (P = 0.054). CHG was not significantly associated with CVD in either continuous or quantile analyses (P > 0.05), although the Q3 group approached statistical significance (P = 0.052). CTI was significantly associated with increased CVD risk in continuous analysis (sHR = 1.29, P = 0.005), but its quantile analysis did not reach statistical significance. METS-IR showed a significant association in continuous analysis (sHR = 1.01, P = 0.0004), with significantly elevated risk in the Q3 group (sHR = 1.28, P = 0.0021), while the Q2 group showed marginal significance (P = 0.058). For TyG and its derived indices, most continuous analyses were not statistically significant. However, several Q3 groups, including TyG-BSI, TyG-BMI, TyG-BRI, TyG-WC, and TyG-WHtR, remained significantly associated with increased CVD risk. Notably, TyG-BMI and TyG-WC showed consistent statistical significance in continuous analyses (P < 0.001).

Additionally, in the CHARLS cohort, cuIR and its derived classifications remained significantly associated with incident CVD in the competing risk model (Supplementary Table 9). Continuous cuIR was positively associated with CVD risk, and quantile analyses showed a stepwise increase in risk across higher levels. K-means clustering–based IR groups showed significantly higher CVD risk in the moderate and high groups compared with the low group.

Overall, although associations in the ELSA cohort were weaker and less consistent than those in the CHARLS cohort, multiple IR-related indicators, including METS-IR, TyG-derived indices, and cuIR-based classifications, remained significantly associated with CVD risk, supporting the robustness of the main findings under a competing risk framework.

Discussion

Overview of study design and main findings

Based on two large-scale prospective cohorts, this study systematically evaluated the association between baseline levels,, cumulative exposure in the CHARLS cohort, and longitudinal trajectories in the ELSA cohort of 11 surrogate markers of IR and the risk of CVD in patients with subclinical metabolic disorders. It is important to note that these markers should not be interpreted as direct or universally applicable measures of insulin resistance. Rather, they represent a heterogeneous set of surrogate or composite markers derived from routine metabolic and anthropometric parameters, with each marker reflecting different aspects of metabolic and inflammatory changes associated with insulin resistance. These indices differ significantly in their underlying components and biological focus: some primarily reflect lipid metabolism, others represent the degree of obesity or body fat distribution, and still others incorporate glucose or inflammatory features. Consequently, these markers do not reflect a single, biologically specific construct of insulin resistance but rather represent partially overlapping dimensions of metabolic dysfunction. Accordingly, these findings are better interpreted as reflecting the integrated cardiometabolic burden captured by IR-related surrogate indices, rather than the isolated contribution of insulin resistance itself. Within this framework, our study conducted a multidimensional assessment of the association between insulin resistance-related metabolic changes and cardiovascular disease risk across three dimensions: baseline status, cumulative exposure, and longitudinal trajectories. The main findings are as follows: (1) During a median follow-up of 8 years, elevated baseline levels of TyG, TyG-WC, TyG-WWI, TyG-BMI, TyG-ABSI, TyG-BRI, TyG-WHtR, METS-IR, AIP, CTI, and CHG were all significantly associated with increased risk of new-onset cardiovascular disease, and these results remained consistent after comprehensive adjustment and partial validation in the ELSA cohort. (2) CVD risk increased progressively among individuals with higher cuIR and among those classified into higher cuIR-derived clustering groups. (3) In the ELSA cohort, Longitudinal trajectory analysis showed that individuals who remained in the high-level group had a significantly higher risk of cardiovascular disease compared to those in the low-level group. (4) RCS analysis supported an overall linear dose-response relationship, indicating a cumulative effect rather than a threshold effect. (5) Incorporating IR as a surrogate marker into traditional risk models improves predictive performance, highlighting its potential value for risk stratification and early identification in clinical and population settings.

Baseline IR-related surrogate markers and CVD risk

The TyG index, calculated from fasting glucose and triglyceride levels, is widely used as a surrogate marker associated with insulin resistance [42]. Given the close interplay between IR and obesity, recent studies increasingly combine the TyG index with anthropometric indicators to construct composite markers for evaluating CVD risk. In the present study, all seven baseline obesity-integrated IR surrogate indicators showed significant positive associations with CVD risk in the CHARLS cohort, consistent with previous evidence. For example, Chen et al. [43] reported that higher TyG levels were positively associated with incident CVD, with elevated TyG significantly increasing the risks of CVD and stroke by 15% and 34%, respectively. Bai et al. developed the TyG-BRI index by combining TyG and BRI, and confirmed its significant and independent association with CVD risk, with higher TyG-BRI levels associated with a 59.1% increase in CVD events [44]. A longitudinal study based on the CHARLS database involving 7,115 participants also demonstrated that TyG, TyG-BMI, TyG-WC, TyG-ABSI, and TyG-WHtR were all significantly associated with new-onset CVD [45]. Notably, TyG-ABSI showed the strongest association with CVD risk in that study (HR = 4.92).

Together, these findings suggest that composite IR surrogate indicators incorporating body shape characteristics may outperform single glyco-lipid markers in identifying CVD risk. They not only reflect core metabolic features of IR, including dysregulation of glucose and lipid metabolism, but also capture abnormal fat distribution, particularly abdominal adiposity, in individuals with “normal-weight obesity” who may be overlooked by BMI-based assessments. This interpretation is consistent with previous studies indicating that abnormal fat distribution has independent predictive value for CVD beyond overall obesity [46]. Recent evidence further suggests that individuals with concurrently elevated ABSI and TyG have the highest risk of CVD mortality, supporting a synergistic effect between abnormal fat distribution and metabolic dysfunction [24].

Beyond TyG-related indices, the associations observed for other IR-related surrogate markers were also broadly consistent with previous studies. A cohort study of 7,708 participants reported a CVD risk ratio of 1.22 for baseline AIP, which is directionally consistent with our findings [47]. Yang et al. [48] showed in the CHARLS cohort that during a 4-year follow-up, participants in the highest METS-IR quartile had a 36.6% higher cardiovascular risk after adjustment for sociodemographic characteristics, lifestyle factors, and conventional cardiovascular risk factors. Similarly, Sun et al. [49] reported a linear positive association between CTI and CVD risk in NHANES, with significant associations across multiple cardiovascular subtypes.

External validation in the ELSA cohort provided partial replication of the CHARLS findings. After multivariable adjustment, several obesity-integrated IR surrogate indicators, particularly TyG-WC, TyG-BMI, TyG-BRI, and TyG-WHtR, as well as METS-IR, remained significantly associated with CVD risk, suggesting that markers incorporating abdominal adiposity or broader metabolic burden may have relatively stronger cross-population robustness. In contrast, the associations for several other indicators, including TyG, TyG-WWI, AIP, CTI, and CHG, were attenuated or lost statistical significance in ELSA after full adjustment. These inconsistencies likely reflect cohort-specific factors, including differences in demographics, adiposity patterns, lifestyle, comorbidities, healthcare context, baseline metabolic characteristics, sample structure, covariate distributions, event ascertainment, follow-up processes, and statistical power. Taken together, these findings suggest that while some IR-related surrogate indicators—especially those integrating waist circumference or broader adiposity-related information—may have relatively better cross-cohort applicability, the predictive value of other markers may be more context- or population-dependent.

Restricted cubic spline analyses further showed that the associations between baseline IR-related surrogate indicators and CVD risk were predominantly linear or near-linear overall, with no stable threshold effects observed. This pattern suggests that cardiovascular risk may begin to increase progressively even at relatively low levels of metabolic dysfunction, supporting a prevention strategy centered on earlier identification and earlier intervention.Longitudinal trajectory analysis in the ELSA cohort further confirmed that sustained high-level IR-related metabolic profiles are relevant to subsequent CVD risk.

Cumulative IR exposure and CVD risk

TThis study systematically leveraged longitudinal CHARLS data to explore the cumulative exposure patterns of IR-related surrogate indices and their association with CVD risk. Compared to previous studies predominantly using cross-sectional designs, we integrated repeated measurements across multiple time points to capture the cumulative effects of IR-related surrogate indicators over time. This approach helps elucidate how persistent IR-related metabolic abnormalities influence CVD risk development. Such continuous metabolic monitoring is particularly important in high-risk populations, as IR levels are dynamically influenced by aging, obesity, and lifestyle factors [50].

Previous studies have indicated that cumulative insulin resistance exposure or clustering patterns are associated with cardiovascular diseases [51–54]. However, most studies only focused on a single biomarker or a combination of two insulin resistance indicators. There is still a lack of systematic comparison of multiple classic/combined insulin resistance indicators. Our research results show that all 11 cumulative insulin resistance (cuIR) markers are significantly associated with the risk of cardiovascular diseases. Cluster analysis revealed that individuals with insulin resistance surrogate indicators remaining at high levels had a significantly higher risk of cardiovascular diseases than those with low-level changes. This indicates that even in the subclinical metabolic disorder stage, persistent elevated insulin resistance significantly increases the risk of cardiovascular diseases, highlighting the crucial role of long-term exposure to risk accumulation.

These results align with multiple studies. Zhu et al. [55] and Wang et al. [56] demonstrated that individuals with higher cumulative exposure levels or exposure control patterns for indicators such as TyG, TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, and TyG-BRI exhibited significantly increased CVD risk compared to those with lower cumulative exposure or better trajectory control. Yang et al.‘s study of 4,364 middle-aged and elderly individuals using CHARLS data also found [57] that cuCTI and its dynamic changes correlate with stroke risk, particularly among those with higher cumulative CTI exposure or persistently elevated levels, where stroke risk significantly increases. Additionally, Li et al. noted [58] that high-level AIP clustering and cumulative AIP are both significantly associated with stroke risk.

Overall, these findings underscore the value of long-term dynamic monitoring of IR. Compared to single-point screening, continuously tracking cumulative exposure patterns of IR surrogate indicators and intervening early for those persistently at high levels may more effectively delay or halt the onset and progression of CVD. Notably, only cuAIP exhibited significant nonlinearity, indicating an inflection point near cuAIP = 1.041 where the rate of risk increase moderates. This “saturation effect” [59] may reflect a diminishing marginal contribution of atherosclerotic lipid burden to CVD risk after reaching a cumulative threshold, providing insights for nonlinear threshold modeling and mechanistic investigations of AIP-related lipid metabolism.

Longitudinal trajectories of IR-related markers and CVD risk

In the ELSA cohort, longitudinal trajectory analysis provided additional insights into the temporal patterns of IR-related metabolic alterations and their association with CVD risk. Individuals classified into persistently high-level trajectory groups exhibited a significantly increased risk of incident CVD compared with those in low-level trajectory groups, even after multivariable adjustment. This finding highlights the importance of sustained metabolic burden over time, rather than single-point measurements, in shaping cardiovascular risk profiles.

The trajectory-based approach captures not only the absolute level of metabolic dysfunction but also its stability and persistence across time. Individuals maintaining high IR-related profiles may represent a subgroup with prolonged exposure to adverse metabolic states, including chronic low-grade inflammation, dysregulated lipid metabolism, and endothelial dysfunction, all of which are key contributors to atherosclerosis development. In contrast, individuals with lower or more stable trajectories may experience less cumulative metabolic stress, potentially explaining their relatively lower cardiovascular risk.

Our findings are consistent with previous longitudinal studies demonstrating that persistent or worsening metabolic trajectories are more strongly associated with cardiovascular outcomes than baseline measurements alone [60–40]. For example, studies focusing on TyG-based indices and other metabolic markers have reported that individuals with sustained high trajectories or unfavorable temporal patterns exhibit significantly elevated risks of CVD and mortality compared to those with stable low-level profiles [61].

Importantly, trajectory analysis complements the cumulative exposure framework observed in the CHARLS cohort. While cuIR exposure reflects the aggregated burden of metabolic dysfunction, trajectory modeling further characterizes the dynamic evolution of this burden over time. Together, these approaches provide a more comprehensive understanding of how long-term metabolic disturbances contribute to cardiovascular risk.

From a clinical and public health perspective, these findings underscore the value of repeated metabolic assessments and longitudinal monitoring. Identifying individuals with persistently high-risk trajectories may enable earlier and more targeted interventions, potentially improving risk stratification and prevention strategies for CVD in middle-aged and older populations.

Potential mechanisms linking IR to CVD

Insulin resistance contributes to the development of cardiovascular disease (CVD) through multiple interconnected pathophysiological pathways, including dysregulated lipid metabolism, endothelial dysfunction, oxidative stress, and chronic low-grade inflammation [62–63]. Impaired insulin signaling under IR may promote excessive free fatty acid release and hepatic glucose overproduction, leading to glucolipotoxicity and increased oxidative stress, which in turn contribute to endothelial injury and early atherosclerotic changes [64–65].

The multi-dimensional IR surrogate indicators adopted in our study capture distinct aspects of these pathological processes. Specifically, the TyG index and its derived indices reflect combined disturbances in glucose and lipid metabolism, while AIP characterizes a pro-atherogenic lipid profile. As an integrated metabolic- inflammatory marker, CTI incorporates both IR status and systemic inflammation, reflecting the potential synergistic effects of metabolic and inflammatory dysregulation on vascular injury. In addition, the CHG index captures glucose variability beyond average glycemic levels, which may further exacerbate oxidative stress and endothelial dysfunction.

Notably, the present study is an epidemiological analysis focusing on the association between IR-related surrogate indicators and CVD risk, and does not directly assess intermediate mechanistic pathways such as renin–angiotensin–aldosterone system (RAAS) activation [66]. Nevertheless, the observed associations between multiple IR-related surrogate indices and CVD risk provide epidemiological support for the role of IR in CVD pathogenesis. Collectively, these findings suggest that IR may act as a central link connecting multiple cardiovascular risk factors, including dyslipidemia, chronic inflammation, and endothelial dysfunction.

Clinical Implications

This study also found that certain sociodemographic and behavioral factors modified the association between IR surrogate indicators and CVD. In the baseline analysis, marital status and physical activity level showed significant interaction effects across multiple IR surrogate indicators. Specifically, the associations of TyG-WWI, TyG-BRI, and TyG-WHtR with CVD were significant only among married or cohabiting individuals, which may be related to differences in social support, health behaviors, and psychological status associated with marital status. Married individuals generally have better access to healthcare resources and greater treatment adherence, such that their CVD risk may be more strongly driven by biological factors such as IR. In contrast, unmarried individuals may be more vulnerable to social stress, psychological burden, and adverse behavioral exposures, which could attenuate the relative contribution of IR to CVD risk [48].

In the cumulative exposure analysis, the modifying effect of physical activity was more prominent, with stronger associations observed among individuals without regular physical activity. This may reflect the beneficial role of physical activity in improving insulin sensitivity and metabolic health, thereby partially offsetting the adverse cardiovascular effects of long-term IR exposure [67]. The modifying effect of age on certain cumulative indicators also deserves attention. For example, the association between cuTyG and CVD was stronger in individuals aged 60 years and older, suggesting that older adults may be more vulnerable to the long-term metabolic burden associated with IR. Taken together, these subgroup findings indicate that when applying IR-related surrogate markers in CVD risk stratification, individual characteristics such as marital status, physical activity, and age should be considered to facilitate more precise and personalized management.

At the same time, although several IR-related surrogate indices yielded statistically significant improvements in model performance, the magnitude of the C-index increase was modest. This suggests that these markers may provide incremental rather than substantial predictive information. Their clinical utility may therefore lie more in refining risk stratification and capturing metabolic dysfunction through simple and accessible surrogate measures, rather than serving as stand-alone replacements for established prediction models. Moreover, because the baseline model had already incorporated conventional clinical predictors, large improvements in discrimination were not expected. In this context, modest increases in C-index, together with significant improvements in IDI and NRI for several indices, still support their incremental clinical value at the population level. Therefore, the practical value of these IR-related surrogate indicators may be greatest when used as low-cost tools for risk enrichment and subgroup identification, particularly in settings where direct measurement of insulin resistance is not feasible.

Strengths and limitations

This study has several notable strengths. First, using two large-scale prospective cohorts, this study combined static baseline assessments, cumulative exposure assessment in CHARLS, and trajectory modeling in ELSA, thereby better capturing the long-term metabolic burden of insulin resistance-related surrogate indicators and overcoming the limitations of cross-sectional studies based solely on single-point measurements. Second, we systematically compared the performance of various composite IR surrogate indices in identifying CVD risk and rigorously evaluated predictive gains using multidimensional statistical measures such as the C-index, AIC, IDI, and NRI, thereby enhancing the interpretability and clinical translational value of the study results. Through RCS, subgroup analysis, and stratified analysis, we further revealed the dose-response relationships and population heterogeneity between insulin resistance indicators and CVD risk, enabling refined risk stratification.

In terms of clinical translational value, the incremental prediction analysis revealed that both baseline IR surrogate indicators and cumulative IR surrogate indicators could improve model discrimination and reclassification when incorporated into traditional cardiovascular risk models. It is worth noting that the strength of association and predictive gain were not entirely consistent. TyG-ABSI showed the largest HR in the CHARLS analysis and may be more sensitive in capturing high-risk metabolic phenotypes related to abnormal body shape or fat distribution. In contrast, TyG-WC showed the most pronounced improvement in incremental prediction, suggesting that it may have advantages in measurement stability, sample distribution, and complementarity with traditional risk factors. Similarly, cuTyG-WC showed the most prominent incremental predictive value among cuIR indicators, further emphasizing the role of long-term cumulative exposure to abdominal obesity combined with glycolipid disorder in CVD risk prediction. Because these indicators are derived from routine biochemical tests and simple anthropometric measurements, they are low-cost, accessible, and repeatable, making them suitable for primary care and large-scale health screening.

However, this study also has several limitations that require cautious interpretation: (1) First, a potential concern of conceptual circularity should be acknowledged. Many of the evaluated IR surrogate markers are constructed from conventional cardiometabolic components, such as triglycerides, HDL cholesterol, fasting glucose, and anthropometric measures, which are themselves well-established risk factors for CVD. Therefore, the observed associations with incident CVD may substantially reflect the predictive contributions of these embedded components, rather than a distinct independent effect of insulin resistance per se, rather than a distinct effect of insulin resistance per se. Although multivariable models were adjusted for major confounders, residual overlap between exposure definitions and outcome-related risk factors cannot be fully excluded. Future studies incorporating direct measures of insulin resistance or more refined analytical strategies are warranted to further disentangle these relationships. Nevertheless, these composite markers may still capture integrated metabolic and inflammatory alterations and provide practical value for risk stratification in large-scale and primary care settings.(2) Although we controlled for known confounding factors to the greatest extent possible, observational studies cannot completely rule out the influence of residual confounders (such as dietary patterns, medication use, genetic susceptibility, and infection/chronic inflammation status) on the results, nor can they establish causal relationships. (3) CVD outcomes were defined based on self-reported physician diagnoses and were not subject to independent medical record adjudication, which may introduce potential recall bias and non-differential disease misclassification. Although the validity of self-reported CVD has been well established in large elderly cohort studies with a low false-positive rate, such measurement bias cannot be completely excluded [68]. Importantly, this non-differential misclassification tends to attenuate the true association toward the null, yielding conservative risk estimates rather than false-positive results. Future studies incorporating hospital medical record verification or objectively adjudicated clinical endpoints are needed to further validate our conclusions. (4) Measurements of insulin resistance in the CHARLS dataset were only collected during two survey waves (2011 and 2015), which limits our access to long-term continuous metabolic data; furthermore, batch-to-batch variability may exist in the test results. Although cumulative exposure mitigates measurement error to some extent, it still cannot fully reflect an individual’s true exposure levels over their entire lifespan. (5) The research subjects were mainly middle-aged and elderly individuals, which might limit the applicability of the results to younger populations or other racial groups. Additionally, the exact dates of cardiovascular disease onset are lacking, which may affect the accuracy of the time-to-event estimation in the Cox model.

Future directions

Future studies with larger sample sizes, longer follow-up durations, direct measurements of insulin resistance, and objectively adjudicated cardiovascular endpoints are warranted to further validate these findings and clarify the underlying mechanisms. In addition, integrating mechanistic investigations with longitudinal epidemiological analyses may provide deeper insights into the complex pathways linking insulin resistance-related metabolic dysfunction and cardiovascular disease. Such efforts will help refine risk stratification strategies and support the development of more targeted prevention approaches.

Conclusion

In conclusion, this study demonstrates that among middle-aged and older adults with subclinical metabolic disorders, baseline levels, cumulative exposure, and longitudinal trajectories of various surrogate markers of insulin resistance are significantly associated with the risk of new-onset cardiovascular disease (CVD) and provide incremental predictive value beyond traditional risk prediction models. Specifically, the baseline TyG-ABSI index showed the strongest association with CVD risk, while TyG-WC and cuTyG-WC demonstrated the most significant predictive value. In the CHARLS cohort, individuals with elevated cuIR levels, those classified into higher cuIR-derived cluster groups, and those in the ELSA cohort whose trajectories remained at high levels all exhibited a significantly increased risk of CVD.

Therefore, the early identification and long-term monitoring of alternative markers of insulin resistance can help identify potential high-risk populations and facilitate stratified interventions, thereby improving early warning and precision prevention strategies for cardiovascular disease.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AIP

Atherogenic index of plasma

ABSI

A body shape index

AIC

Akaike information criterion

BMI

Body mass index

CHG

Cholesterol high density lipoprotein glucose

CHARLS

China Health and Retirement Longitudinal Study

CI

Confidence intervals

CRP

C-reactive protein

CTI

C-reactive protein-triglyceride glucose index

CVD

Cardiovascular disease

cuIR

Cumulative IR

DBP

Diastolic blood pressure

ELSA

English longitudinal study of ageing

FPG

Fasting plasma glucose

HB

Hemoglobin

HbA1c

Haemoglobin A1c

HDL-C

High density lipoprotein cholesterol

HR

Hazard ratio

IDI

Integrated discrimination improvement

IR

Insulin resistance

LDL-C

Low density lipoprotein cholesterol

METS-IR

Metabolic score for IR

NRI

Net reclassification improvement

RCS

Restricted Cubic Spline

SBP

Systolic Blood Pressure

TC

Total cholesterol

TG

Triglyceride

TyG

Triglyceride glucose

WC

Waist circumference

WHtR

Waist-to-height ratio

WWI

Weight adjusted waist index

Author contributions

The data analysis was completed by Wei Zhao. The initial draft was jointly produced by Wei Zhao and Dengfeng Gao. Rongjie Tang and Jing Tian participated in the revision and layout of the manuscript. All authors participated in the discussions and revision of the previous manuscript version. All authors read and approved the final manuscript.

Funding

The research was funded by the National Natural Science Foundation of China (Grant No. 82160090) and the Xianyang Key Research and Development Project (Grant No. L2024-ZDYF-ZDYF -SF-0020; Grant No. L2024-ZDYF-ZDYF -SF-0021).

Data availability

The data supporting the results of this study can be found on the CHARLS (http://charls.pku.edu.cn) and ELSA (https://www.elsa-project.ac.uk/) websites.

Declarations

Ethics approval and consent to participate

The China Health and Retirement Longitudinal Study was approved by the Ethics Review Committee of Peking University (IRB00001052-11015). The UK English Longitudinal Study of Ageing was approved by the London Multicenter Research Ethics Committee (MREC/01/2/91). Informed consent was obtained from each subject in these 2 cohorts

Competing interests

The authors declare no competing interests.