Long-term metabolic–adiposity burden and cardiovascular risk: the cumTyG–WHtR in Chinese adults Authors: Heping Xu, Siyuan Huang, Yan Xia, Huan Niu, Jiangyan Li, Ping He Published: 2026 ABSTRACT BackgroundEvidence on the relationship between a long-term cumulative metric that couples the triglyceride–glucose index (TyG) with the waist‒to-height ratio (WHtR) (cumTyG‒WHtR) and incident cardiovascular disease (CVD), particularly when integrated with obesity phenotypes, remains limited.MethodsWe analyzed 4,254 middle-aged and older adults from the China Health and Retirement Longitudinal Study (CHARLS), with a median follow-up of approximately 3 years. Covariates were selected via the Boruta algorithm. Cox proportional hazards models were used to assess the associations between the cumTyG-WHtR and incident CVD. Restricted cubic splines (RCSs) were used to evaluate nonlinearity and dose–response. Effect modification was examined via interaction terms and stratified analyses, and robustness was assessed via multiple sensitivity analyses. Exploratory associational decomposition using nonparametric percentile bootstrap resampling (1,000 iterations) was performed to estimate the proportion of the total association between cumTyG-WHtR and CVD that was statistically accounted for by mean arterial pressure (MAP).ResultsDuring follow-up, 587 incident CVD events occurred (cumulative incidence 13.80%), with the incidence increasing across quartiles of cumTyG-WHtR (Q1–Q4: 9.21%, 13.26%, 14.29%, 18.44%). In fully adjusted models, each 1-unit increase in cumTyG-WHtR (continuous) was associated with a 5.2% higher CVD risk (hazard ratio 1.052, 95% confidence interval 1.011–1.094). RCS revealed no evidence of nonlinearity (P for nonlinearity = 0.790), indicating a near-linear, monotonic dose–response. A significant age interaction was observed (P for interaction = 0.025), with stronger associations in participants aged 45–59 years. The findings were consistent across the sensitivity analyses. In an exploratory associational decomposition (not a causal mediation analysis), MAP statistically accounted for a significant proportion of the total association, with an indirect association estimate of 0.005 (95% CI, 0.002–0.008; p < 0.001)—a small absolute effect but in this decomposition accounting for 20.3% of the total association.ConclusionsIn this cohort of middle-aged and elderly individuals, a higher cumulative TyG - WHtR was positively associated with the incidence of CVD. This association showed an approximately linear trend, and it was more significant in the population aged 45 to 59. In an exploratory associational decomposition, MAP statistically accounted for a portion of the total association. This hypothesis-generating finding requires confirmation in future studies with stronger causal designs.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12872-026-06121-z. Introduction Cardiovascular disease (CVD), including heart disease and stroke, remains the leading cause of death and disability worldwide, substantially shortening healthy life expectancy and imposing a major burden. Although risk factor management and interventions have reduced age-standardized CVD mortality over the past two decades, population aging and growth continue to drive increases in the absolute numbers of CVD deaths, incidence, and prevalence [1–3]. These trends highlight the need, beyond the optimization of conventional management, for simple, scalable indicators that can identify high-risk individuals early and capture the effects of long-term metabolic and hemodynamic loads. Insulin resistance (IR) is a central pathophysiological basis of atherosclerosis and an important predictor of CVD [4]. The triglyceride‒glucose (TyG) index, computed from fasting triglycerides and glucose, has been widely validated as a surrogate for IR with good sensitivity and specificity; higher TyG levels are independently associated with incident CVD and adverse outcomes (5–8). Concurrently, obesity—particularly central (abdominal) adiposity—often coexists with IR and multiple metabolic disturbances, increasing CVD risk through diverse mechanisms [9–12]. The waist-to-height ratio (WHtR), which standardizes waist circumference by stature, offers greater comparability and robustness than does waist circumference alone; recent evidence indicates that the composite TyG–WHtR, which integrates information on metabolic load and fat distribution, enhances the identification of individuals at high risk of cardiovascular events and cardiovascular mortality [13–16]. Accordingly, a composite indicator combining the metabolic phenotype of IR (TyG) and central adiposity (WHtR) may provide superior discrimination of atherogenic risk relative to single measures. Hypertension is another major CVD risk factor [17, 18]. Mean arterial pressure (MAP), a hemodynamic index reflecting average arterial pressure, is independently associated with CVD risk, and greater MAP variability has been linked to adverse vascular events [19–21]. Biologically, elevated blood pressure may induce or exacerbate IR through chronic low-grade inflammation [22], renal dysfunction [23], and sympathetic nervous system activation [24, 25], thereby impairing insulin signaling and glucose metabolism. Taken together, these observations suggest that MAP may lie along the pathway from long-term metabolic load and abnormal fat distribution to incident CVD, mediating part of the effect of metabolic abnormalities on CVD. However, most existing studies rely on single baseline measurements and therefore cannot characterize the cumulative effects of metabolic and hemodynamic factors, potentially underestimating the true magnitude of exposure–outcome associations. In contrast, cumulative exposure metrics—such as cumulative TyG (cumTyG) and its combination with WHtR (cumTyG-WHtR)—better capture the average level and persistence of long-term metabolic burden and fat distribution, yet their relationships with incident CVD and potential mediation mechanisms remain insufficiently defined. Using repeated measures from the China Health and Retirement Longitudinal Study (CHARLS), we aim to (1) evaluate the association between the cumTyG-WHtR and incident CVD and examine its dose–response relationship; (2) explore, via an exploratory associational decomposition, the proportion of the total association between cumTyG-WHtR and CVD that is statistically accounted for by MAP (while acknowledging that the temporal sequence required for causal mediation cannot be strictly established); and (3) assess the robustness of the findings through a series of sensitivity and subgroup analyses. Methods Data source This study used data from the China Health and Retirement Longitudinal Study (CHARLS; http://charls.pku.edu.cn/en) [26], a nationally representative longitudinal cohort of community-dwelling adults aged ≥ 45 years in China. CHARLS employs a multistage, stratified, probability sampling design. The baseline household survey was conducted in 150 counties (districts) across 28 provinces, covering over 10,000 households and enrolling 17,708 participants. Follow-up interviews were carried out face-to-face every 2–3 years to collect demographic characteristics, health status, and socioeconomic information systematically. The study adhered to the Declaration of Helsinki and was approved by the Institutional Review Board of the Peking University Health Science Center (IRB00001052–11015). Written informed consent was obtained from all participants. Study population Wave 1 (2011–2012) served as the baseline. Because blood samples were collected in waves 1 and 3(2015), we used these measurements to compute the cumulative TyG-WHtR (cumTyG-WHtR) exposure; incident cardiovascular disease (CVD) was ascertained at wave 4 (2018). For the 17,708 baseline participants, we constructed the analytic sample by applying the following criteria: (1) missing follow-up information on incident CVD; (2) missing key variables required to compute cumTyG-WHtR (height, waist circumference, triglycerides [TG], or fasting blood glucose [FBG]); (3) baseline age < 45 years; and (4) prevalent CVD at Waves 1, 2, and 3. The final sample included 4,254 participants. The selection process is shown in Fig. 1. Fig. 1Flow chart of patient selection for analysis Flow chart of patient selection for analysis Definitions CVD was defined on the basis of standardized self-reports of physician diagnoses: “Have you ever been diagnosed by a doctor with stroke or heart disease (including myocardial infarction, coronary heart disease, angina, congestive heart failure, or other heart problems)?” Participants who answered “yes” and specified any of the listed conditions were classified as having prevalent CVD at baseline [27, 28]; the first report of a physician-diagnosed heart disease or stroke during follow-up was classified as incident CVD. The triglyceride‒glucose (TyG) index was calculated as TyG = ln [TG (mg/dL) × FBG (mg/dL)/2] [29]. Waist‑to‑height ratio (WHtR) was defined as waist circumference (cm) divided by height (cm), and TyG–WHtR at each wave was obtained as TyG × WHtR, reflecting the combined burden of metabolic load and central adiposity at that visit. The cumulative TyG-WHtR exposure was approximated by the time-weighted mean over two measurements, computed as the arithmetic mean of TyG-WHtR values at waves 1 and 3 multiplied by the number of years between measurements: [(TyG-WHtRwave1 + TyG-WHtRwave3)/2] × time(wave3 – wave1) [30]. MAP = 1/3 × systolic blood pressure + 2/3 × diastolic blood pressure (wave3). Because the cumulative exposure variable (cumTyG-WHtR) was fully defined only after the wave 3 (2015) measurement, we adopted a 2015 landmark analysis as the primary analytical approach. Time zero was set as the wave 3 interview date. The risk set included participants who were alive and free of CVD at wave 3 (2015). Participants who developed CVD between wave 1 (2011) and wave 3 (2015) were excluded from the primary analysis to avoid reverse causality and immortal time bias. Follow-up duration was calculated from wave 3 to wave 4 (2018), with a median follow-up of approximately 3 years. Data extraction At baseline, trained interviewers used standardized questionnaires to collect demographic and behavioral information, including age (years), sex (male/female), education level (no formal education, primary school, middle school, high school or above), residence (urban/rural), marital status (married/not married), smoking status (yes/no), and alcohol use (yes/no). Anthropometric measurements (height, weight, waist circumference) and blood pressure (systolic and diastolic) were obtained according to standard procedures. Baseline comorbidities (yes/no), including diabetes, heart disease, stroke, dyslipidemia, hypertension, and arthritis, were assessed by self‑reported physician diagnosis. In 2011 and 2015 (waves 1 and 3), certified personnel obtained fasting venous blood samples after ≥ 12 h of fasting and promptly transported them to the central laboratory in Beijing for batch testing. The following biochemical indices were measured by enzymatic colorimetry: fasting blood glucose (FBG/GLU), total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), white blood cell count (WBC), platelet count (PLT), serum creatinine (Cr), C-reactive protein (CRP), glycated hemoglobin (HbA1c), hemoglobin (Hb), uric acid (UA), and blood urea nitrogen (BUN). Statistical analysis We included 4,254 participants and grouped them into quartiles (Q1–Q4) of cum TyG-WHtR. Baseline characteristics are described as follows: approximately normally distributed continuous variables are presented as the mean ± standard error (SE) with between-group comparisons by analysis of variance (ANOVA); skewed continuous variables are presented as the median (interquartile range, IQR) with comparisons by the Kruskal–Wallis test; and categorical variables are presented as n (%) with comparisons by the chi-square test. CumTyG‑WHtR was further modeled both as a continuous variable (per 1‑SD increase) and as a categorical variable (quartiles, with Q1 as the reference) in the Cox regression analyses. For covariate selection, we adopted a hybrid approach that combined data exploration with clinical knowledge. First, we conducted a literature review to identify established risk factors for cardiovascular disease (including age, sex, education level, smoking, hypertension, diabetes, dyslipidemia, etc.). Second, we used the Boruta algorithm as an exploratory tool to assess variable importance and detect potential nonlinear relationships [31]; Boruta was used only as an exploratory check. however, it was not used as the sole criterion for inclusion. Given the overlap and potential collinearity between cumTyG-WHtR and several cardiometabolic indicators, we subsequently calculated the variance inflation factor (VIF) and excluded variables that were highly collinear with cumTyG-WHtR or its components (retaining those with VIF < 5). Finally, we performed multiple sensitivity analyses to ensure robustness (see Table 3). All retained covariates had VIFs < 5 and clear clinical relevance [32]; ultimately, 10 variables were included in the final models. For the preliminary analysis of incident CVD, we fitted four Cox proportional hazards models with progressive adjustment: Model 1, unadjusted; Model 2, adjusted for age, gender, and education level; Model 3, additionally adjusted for smoking status, dyslipidemia, hypertension, and diabetes; and Model 4, further adjusted for Hb, Cr and UA. The proportional hazards assumption for the Cox models was tested using Schoenfeld residuals. No significant violations were detected (global test p > 0.05 for all models). To examine potential nonlinearity and dose–response relationships between the cumTyG-WHtR and CVD risk, we incorporated restricted cubic splines (RCSs) within a Cox proportional hazards framework. Subgroup analyses were conducted by age (45–59 vs. ≥ 60 years), gender, residence, smoking status, alcohol consumption, hypertension, and diabetes. Robustness was assessed via nine sensitivity analyses: excluding participants with dyslipidemia or hypertension; using a complete-case dataset; stratifying by exposure quartiles; and using alternative TyG-WHtR exposures (mean of wave 1 and wave 3, wave 1 alone, wave 3 alone, change [wave 3 – wave 1]) as well as restricting to those with complete death data. We conducted an exploratory associational decomposition using nonparametric percentile bootstrap with 1,000 resamples to evaluate the proportion of the total association between cumTyG-WHtR and incident CVD that was statistically accounted for by MAP. This analysis does not imply a causal or temporal sequence; it is purely hypothesis-generating. Because MAP was measured concurrently with the second exposure time point (wave 3), the temporal sequence required for causal mediation cannot be established. The sampling distribution of the indirect association and direct association components was approximated via bootstrap resampling. We estimated the average indirect association, the average direct association, and the proportion of the total association statistically accounted for, and assessed statistical significance using corresponding p values and percentile-based 95% confidence intervals. All analyses were performed via R version 4.2.1 and Stata version 18.0. Two-sided tests were used, and statistical significance was defined as P < 0.05. Extreme values were winsorized at the 1st and 99th percentiles. Missing data were handled via multiple imputation by chained equations (MICE), implemented with the mice package in R. We generated five imputed datasets, and the proportion of missing data for each variable is presented in Table S1. Cox regression was performed using the coxph function (from the survival package), and exploratory associational decomposition was conducted using the mediation function (from the mediation package) with the explicit understanding that this is not a causal mediation analysis. Results Baseline characteristics This study included 4,254 participants from the China Health and Retirement Longitudinal Study (CHARLS) database (mean age 58.0 ± 8.5 years; 1974 men [46.4%]). The selection process is shown in Fig. 1. The participants were categorized into quartiles on the basis of cumTyG-WHtR: Q1 (< 12.34), Q2 (12.34–13.91), Q3 (13.91–15.53), and Q4 (> 15.53). Baseline demographic and clinical characteristics across quartiles are summarized in Table 1. There were no significant differences in age or marital status (P > 0.05). Compared with the other quartiles, Q4 had a greater proportion of women; higher levels of systolic and diastolic blood pressure; higher incidences of dyslipidemia, hypertension, diabetes, and arthritis; and higher levels of WBC, glucose, triglyceride (TG), total cholesterol (TC), C-reactive protein (CRP), glycated hemoglobin (HbA1c), uric acid (UA), and hemoglobin, along with a higher incidence of incident cardiovascular disease (CVD). Table 1Baseline characteristics of participants according to cumTyG-WHtR quartile groupVariablescumTyG-WHtRAll patientsQ1Q2Q3Q4 P Number42541064106310641063Age, years58.0 (8.5)58.5 (8.7)57.8 (8.4)57.6 (8.6)58.0 (8.2)0.139Gender< 0.001Female2280 (54.6)358 (33.6)515 (49.4)636 (59.8)771(72.5)Male1974 (46.4)706 (66.4)548 (51.6)428 (40.2)292 (27.5)Marital statusOther555 (13.0)139 (13.1)150 (14.1)128 (12.0)138 (13.0)0.565Married3699 (87.0)925 (86.9)913 (85.9)936 (88.0)925 (87.0)Education level0.005No formal education1213 (28.5)268 (25.2)295 (27.8)302 (28.4)348 (32.7)Primary school1786 (42.0)479 (45.0)454 (42.7)423 (39.8)430 (40.5)Middle school875 (20.6)218 (20.5)213 (20.0)232 (21.8)212 (19.9)High school or above380(8.9)99(9.3)101 (9.5)107 (10.1)73 (6.9)Residence< 0.001Urban561 (13.2)104 (9.8)114 (10.7)188 (17.7)155 (14.6)Rural3693 (86.8)960 (90.2)949 (89.3)876 (82.3)908 (85.4)SBP, mmHg124.5 (112.5, 139.0)119.0 (108.5, 131.5)122.5 (111.5, 136.5)125.5 (113.2, 139.2)131.5 (119.0, 147.5)< 0.001DBP, mmHg74.0 (66.5, 82.0)70.0 (63.0, 78.0)72.5 (65.5, 80.5)74.0 (68.0, 82.2)78.0 (70.5, 86.5)< 0.001MAP, mmHg91.0 (82.3, 100.3)86.6 (78.8, 95.7)89.7 (81.5, 98.7)91.8 (83.7, 100.3)96.5 (87.7, 106.7)< 0.001Smoking1630 (38.3)579 (54.4)445 (41.9)349 (32.8)257 (24.2)< 0.001Drinking1771 (41.6)567 (53.3)456 (42.9)414 (38.9)334 (31.4)< 0.001Dyslipidemia311 (7.3)28 (2.6)52 (4.9)77 (7.2)154 (14.5)< 0.001Hypertension879 (20.7)116 (10.9)146 (13.7)239 (22.5)378 (35.6)< 0.001Diabetes213 (5.0)24 (2.2)30 (2.8)46 (4.3)113 (10.6)< 0.001Arthritis1369 (32.2)331 (31.1)321 (30.2)329 (30.9)388 (36.5)0.006CVD587 (13.8)98 (9.2)141 (13.3)152 (14.3)196 (18.4)< 0.001WBC, cell/mm35.9 (4.9, 7.1)5.7 (4.8, 6.9)5.9 (4.8, 7.2)5.9 (4.9, 7.1)6.1 (5.2, 7.4)< 0.001Platelet, cell/mm3207.0 (162.0, 253.0)203.0 (163.0, 248.0)206.0 (158.0, 254.0)209.0 (167.0, 257.5)207.0 (159.0, 257.0)0.024BUN, mg/dL15.1 (12.6, 18.2)15.7 (13.0, 18.9)15.2 (12.6, 18.3)14.9 (12.5, 17.9)14.8 (12.4, 17.5)< 0.001GLU, mg/dl102.2 (94.3, 112.5)99.1 (91.4, 108.2)100.3 (93.1, 108.7)102.0 (95.0, 110.5)108.9 (99.4, 126.4)< 0.001Creatinine, mg/dL0.75 (0.64, 0.87)0.78 (0.66, 0.89)0.76 (0.66, 0.87)0.75 (0.63, 0.87)0.72 (0.63, 0.84)< 0.001TC, mg/dl193.5 (38.4)184.4 (36.1)188.4 (35.7)197.4 (36.3)203.7 (42.2)< 0.001TG, mg/dl103.5 (74.3, 151.3)73.5 (57.5, 96.5)92.0 (70.8, 126.6)112.4 (85.8, 152.2)164.6 (116.8, 245.1)< 0.001HDL-C, mg/dl51.4 (15.3)59.0 (16.4)54.1 (14.1)49.6 (12.9)42.7 (12.5)< 0.001LDL-C, mg/dl116.1 (34.4)109.7 (30.9)115.4 (31.7)122.2 (34.4)117.2 (39.1)< 0.001CRP, mg/dl0.96 (0.53, 1.95)0.74 (0.43, 1.59)0.77 (0.45, 1.56)1.03 (0.57, 1.92)1.35 (0.75, 2.59)< 0.001HbA1C, %5.1 (4.9, 5.4)5.1 (4.8, 5.3)5.1 (4.8, 5.3)5.1 (4.9, 5.4)5.3 (5.0, 5.6)< 0.001UA, mg/dL4.2 (3.5, 5.1)4.1 (3.5, 4.9)4.2 (3.5, 5.0)4.2 (3.5, 5.1)4.4 (3.6, 5.2)< 0.001Hemoglobin, mg/dL14.4 (2.2)14.2 (2.1)14.3 (2.2)14.3 (2.1)14.5 (2.2)0.029TyG8.6(8.2,9.0)8.2(8.0,8.5)8.4(8.2,8.8)8.7(8.4,9.0)9.1(8.8,9.6)< 0.001WHtR0.53(0.49,0.58)0.47(0.44,0.49)0.51(0.49,0.53)0.56(0.53,0.58)0.61(0.58,0.64)< 0.001cumTyG25.8 (24.8, 27.1)24.6 (24.0, 25.4)25.4 (24.7, 26.1)26.2 (25.4, 27.0)27.6 (26.7, 28.7)< 0.001cumWHtR1.61 (1.47, 1.74)1.39 (1.30, 1.45)1.55 (1.50, 1.60)1.68 (1.62, 1.73)1.84 (1.76, 1.92)< 0.001cumTyG_WHtR13.9 (12.3, 15.5)11.5 (10.7, 11.9)13.2 (12.7, 13.6)14.6 (14.2, 15.0)16.8 (16.1, 17.7)< 0.001Continuous variables are presented as the means (SDs) or medians (quartiles), whereas categorical variables are presented as absolute numbers (percentages). SBP, systolic blood pressure; DBP, diastolic blood pressure; MAP, mean arterial pressure, TC, total cholesterol; TG, triglycerides; UA, uric acid; BUN, blood Urea nitrogen; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; CRP, C-reactive protein; TyG, triglyceride-glucose Index Table 1Baseline characteristics of participants according to cumTyG-WHtR quartile groupVariablescumTyG-WHtRAll patientsQ1Q2Q3Q4 P Number42541064106310641063Age, years58.0 (8.5)58.5 (8.7)57.8 (8.4)57.6 (8.6)58.0 (8.2)0.139Gender< 0.001Female2280 (54.6)358 (33.6)515 (49.4)636 (59.8)771(72.5)Male1974 (46.4)706 (66.4)548 (51.6)428 (40.2)292 (27.5)Marital statusOther555 (13.0)139 (13.1)150 (14.1)128 (12.0)138 (13.0)0.565Married3699 (87.0)925 (86.9)913 (85.9)936 (88.0)925 (87.0)Education level0.005No formal education1213 (28.5)268 (25.2)295 (27.8)302 (28.4)348 (32.7)Primary school1786 (42.0)479 (45.0)454 (42.7)423 (39.8)430 (40.5)Middle school875 (20.6)218 (20.5)213 (20.0)232 (21.8)212 (19.9)High school or above380(8.9)99(9.3)101 (9.5)107 (10.1)73 (6.9)Residence< 0.001Urban561 (13.2)104 (9.8)114 (10.7)188 (17.7)155 (14.6)Rural3693 (86.8)960 (90.2)949 (89.3)876 (82.3)908 (85.4)SBP, mmHg124.5 (112.5, 139.0)119.0 (108.5, 131.5)122.5 (111.5, 136.5)125.5 (113.2, 139.2)131.5 (119.0, 147.5)< 0.001DBP, mmHg74.0 (66.5, 82.0)70.0 (63.0, 78.0)72.5 (65.5, 80.5)74.0 (68.0, 82.2)78.0 (70.5, 86.5)< 0.001MAP, mmHg91.0 (82.3, 100.3)86.6 (78.8, 95.7)89.7 (81.5, 98.7)91.8 (83.7, 100.3)96.5 (87.7, 106.7)< 0.001Smoking1630 (38.3)579 (54.4)445 (41.9)349 (32.8)257 (24.2)< 0.001Drinking1771 (41.6)567 (53.3)456 (42.9)414 (38.9)334 (31.4)< 0.001Dyslipidemia311 (7.3)28 (2.6)52 (4.9)77 (7.2)154 (14.5)< 0.001Hypertension879 (20.7)116 (10.9)146 (13.7)239 (22.5)378 (35.6)< 0.001Diabetes213 (5.0)24 (2.2)30 (2.8)46 (4.3)113 (10.6)< 0.001Arthritis1369 (32.2)331 (31.1)321 (30.2)329 (30.9)388 (36.5)0.006CVD587 (13.8)98 (9.2)141 (13.3)152 (14.3)196 (18.4)< 0.001WBC, cell/mm35.9 (4.9, 7.1)5.7 (4.8, 6.9)5.9 (4.8, 7.2)5.9 (4.9, 7.1)6.1 (5.2, 7.4)< 0.001Platelet, cell/mm3207.0 (162.0, 253.0)203.0 (163.0, 248.0)206.0 (158.0, 254.0)209.0 (167.0, 257.5)207.0 (159.0, 257.0)0.024BUN, mg/dL15.1 (12.6, 18.2)15.7 (13.0, 18.9)15.2 (12.6, 18.3)14.9 (12.5, 17.9)14.8 (12.4, 17.5)< 0.001GLU, mg/dl102.2 (94.3, 112.5)99.1 (91.4, 108.2)100.3 (93.1, 108.7)102.0 (95.0, 110.5)108.9 (99.4, 126.4)< 0.001Creatinine, mg/dL0.75 (0.64, 0.87)0.78 (0.66, 0.89)0.76 (0.66, 0.87)0.75 (0.63, 0.87)0.72 (0.63, 0.84)< 0.001TC, mg/dl193.5 (38.4)184.4 (36.1)188.4 (35.7)197.4 (36.3)203.7 (42.2)< 0.001TG, mg/dl103.5 (74.3, 151.3)73.5 (57.5, 96.5)92.0 (70.8, 126.6)112.4 (85.8, 152.2)164.6 (116.8, 245.1)< 0.001HDL-C, mg/dl51.4 (15.3)59.0 (16.4)54.1 (14.1)49.6 (12.9)42.7 (12.5)< 0.001LDL-C, mg/dl116.1 (34.4)109.7 (30.9)115.4 (31.7)122.2 (34.4)117.2 (39.1)< 0.001CRP, mg/dl0.96 (0.53, 1.95)0.74 (0.43, 1.59)0.77 (0.45, 1.56)1.03 (0.57, 1.92)1.35 (0.75, 2.59)< 0.001HbA1C, %5.1 (4.9, 5.4)5.1 (4.8, 5.3)5.1 (4.8, 5.3)5.1 (4.9, 5.4)5.3 (5.0, 5.6)< 0.001UA, mg/dL4.2 (3.5, 5.1)4.1 (3.5, 4.9)4.2 (3.5, 5.0)4.2 (3.5, 5.1)4.4 (3.6, 5.2)< 0.001Hemoglobin, mg/dL14.4 (2.2)14.2 (2.1)14.3 (2.2)14.3 (2.1)14.5 (2.2)0.029TyG8.6(8.2,9.0)8.2(8.0,8.5)8.4(8.2,8.8)8.7(8.4,9.0)9.1(8.8,9.6)< 0.001WHtR0.53(0.49,0.58)0.47(0.44,0.49)0.51(0.49,0.53)0.56(0.53,0.58)0.61(0.58,0.64)< 0.001cumTyG25.8 (24.8, 27.1)24.6 (24.0, 25.4)25.4 (24.7, 26.1)26.2 (25.4, 27.0)27.6 (26.7, 28.7)< 0.001cumWHtR1.61 (1.47, 1.74)1.39 (1.30, 1.45)1.55 (1.50, 1.60)1.68 (1.62, 1.73)1.84 (1.76, 1.92)< 0.001cumTyG_WHtR13.9 (12.3, 15.5)11.5 (10.7, 11.9)13.2 (12.7, 13.6)14.6 (14.2, 15.0)16.8 (16.1, 17.7)< 0.001Continuous variables are presented as the means (SDs) or medians (quartiles), whereas categorical variables are presented as absolute numbers (percentages). SBP, systolic blood pressure; DBP, diastolic blood pressure; MAP, mean arterial pressure, TC, total cholesterol; TG, triglycerides; UA, uric acid; BUN, blood Urea nitrogen; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; CRP, C-reactive protein; TyG, triglyceride-glucose Index Baseline characteristics of participants according to cumTyG-WHtR quartile group Continuous variables are presented as the means (SDs) or medians (quartiles), whereas categorical variables are presented as absolute numbers (percentages). SBP, systolic blood pressure; DBP, diastolic blood pressure; MAP, mean arterial pressure, TC, total cholesterol; TG, triglycerides; UA, uric acid; BUN, blood Urea nitrogen; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; CRP, C-reactive protein; TyG, triglyceride-glucose Index Incidence of cardiovascular disease Over 3 years of follow-up,587 participants developed CVD (3-year cumulative incidence, 13.80%; Fig. 2). When stratified by quartiles of cumTyG-WHtR, the proportion of CVD cases increased from Q1 to Q4: 9.21%, 13.26%, 14.29%, and 18.44%, respectively. The participants in the highest quartile (Q4) consistently had a greater incidence than did those in the lowest quartile (Q1). Fig. 2Bar chart showing CVD incidence by cumTyG-WHtR quartile group Bar chart showing CVD incidence by cumTyG-WHtR quartile group Boruta algorithm We applied the Boruta feature selection algorithm to identify factors associated with incident CVD (Fig. S1). A total of 17 variables were identified as important (green in Fig. S1): cumTyG-WHtR, gender, age, education, smoking, dyslipidemia, hypertension, diabetes, systolic blood pressure, diastolic blood pressure, creatinine, total cholesterol, HDL, LDL, HbA1c, uric acid, and hemoglobin. The remaining variables—marital status, residence, drinking, arthritis, white blood cell count, and blood urea nitrogen—were classified as unimportant (red in Fig. S1). Associations between the cumTyG-WHtR and the risk of new-onset CVD We used Cox proportional hazards models (Table 2) to evaluate the associations between the cumTyG-WHtR and incident CVD. Model 1 was unadjusted; Model 2 was adjusted for age, gender, and education; Model 3 was additionally adjusted for smoking status, dyslipidemia, hypertension, and diabetes; and Model 4 was further adjusted for WBC, platelet count, hemoglobin, creatinine and UA. Modeling as a continuous variable, each 1-unit increase in cumTyG-WHtR was associated with HRs of 1.105 (95% CI, 1.067–1.143) in Model 1; 1.092 (1.054–1.132) in Model 2; 1.043 (1.004–1.084) in Model 3; and 1.052 (1.011–1.094) in Model 4. In Model 4, in the quartile analysis, compared with Q1, the fully adjusted HR values were 1.394 (1.075–1.809) for Q2, 1.453 (1.118–1.889) for Q3, and 1.522 (1.161–1.994) for Q4 (p for trend = 0.005). Table 2Associations between cum TyG-WHtR, cumWHtR and incident CVD risk across statistical modelsModel1HR (95% CI) P valueModel2HR (95% CI) P valueModel3HR (95% CI) P valueModel4HR (95% CI) P valuecumTyG-WHtR (per 1 SD)1.105 (1.067,1.143)< 0.0011.092(1.054,1.132)< 0.0011.043(1.004,1.084)0.0291.052(1.011,1.094)0.011Q1RefRefRefRefQ21.431(1.106,1.852)0.0061.413(1.090,1.831)0.0091.370(1.056,1.775)0.0181.394(1.075,1.809)0.012Q31.608(1.247,2.073)< 0.0011.564(1.207,2.025)0.0011.413(1.089,1.883)0.0091.453(1.118,1.889)0.005Q41.979(1.553,2.523)< 0.0011.868(1.451,2.404)< 0.0011.451(1.114,1.891)0.0061.522(1.161,1.994)0.002P for trend< 0.001< 0.0010.0110.005cumWHtR(per 1 SD)3.029 (2.041,4.496)< 0.0012.602(1.720,3.936)< 0.0011.685(1.100,2.583)0.0171.798(1.162,2.781)0.008Q1RefRefRefRefQ21.228(1.052,1.584)0.0021.217(1.021,1.572)0.0021.185(1.018,1.531)0.0051.205(1.003,1.557)0.003Q31.315(1.067,1.691)0.0011.288(1.028,1.664)0.0011.144(1.013,1.482)0.0381.171(1.022,1.521)0.021Q41.902(1.508,2.400)< 0.0011.775(1.389,2.269)< 0.0011.426(1.106,1.839)0.0061.478(1.141,1.913)0.003P for trend< 0.001< 0.0010.0100.005HR, hazard ratio; CI, confidence interval; Ref, reference; Model 1: Unadjusted model. Model 2: Adjusted for age, gender, education level. Model 3: Adjusted for variables included in Model 2 + smoking, dyslipidemia, hypertension, diabetes. Model 4: Adjusted for variables included in Model 3 + hemoglobin, creatinine, UA Table 2Associations between cum TyG-WHtR, cumWHtR and incident CVD risk across statistical modelsModel1HR (95% CI) P valueModel2HR (95% CI) P valueModel3HR (95% CI) P valueModel4HR (95% CI) P valuecumTyG-WHtR (per 1 SD)1.105 (1.067,1.143)< 0.0011.092(1.054,1.132)< 0.0011.043(1.004,1.084)0.0291.052(1.011,1.094)0.011Q1RefRefRefRefQ21.431(1.106,1.852)0.0061.413(1.090,1.831)0.0091.370(1.056,1.775)0.0181.394(1.075,1.809)0.012Q31.608(1.247,2.073)< 0.0011.564(1.207,2.025)0.0011.413(1.089,1.883)0.0091.453(1.118,1.889)0.005Q41.979(1.553,2.523)< 0.0011.868(1.451,2.404)< 0.0011.451(1.114,1.891)0.0061.522(1.161,1.994)0.002P for trend< 0.001< 0.0010.0110.005cumWHtR(per 1 SD)3.029 (2.041,4.496)< 0.0012.602(1.720,3.936)< 0.0011.685(1.100,2.583)0.0171.798(1.162,2.781)0.008Q1RefRefRefRefQ21.228(1.052,1.584)0.0021.217(1.021,1.572)0.0021.185(1.018,1.531)0.0051.205(1.003,1.557)0.003Q31.315(1.067,1.691)0.0011.288(1.028,1.664)0.0011.144(1.013,1.482)0.0381.171(1.022,1.521)0.021Q41.902(1.508,2.400)< 0.0011.775(1.389,2.269)< 0.0011.426(1.106,1.839)0.0061.478(1.141,1.913)0.003P for trend< 0.001< 0.0010.0100.005HR, hazard ratio; CI, confidence interval; Ref, reference; Model 1: Unadjusted model. Model 2: Adjusted for age, gender, education level. Model 3: Adjusted for variables included in Model 2 + smoking, dyslipidemia, hypertension, diabetes. Model 4: Adjusted for variables included in Model 3 + hemoglobin, creatinine, UA Associations between cum TyG-WHtR, cumWHtR and incident CVD risk across statistical models cumWHtR (per 1 SD) HR, hazard ratio; CI, confidence interval; Ref, reference; Model 1: Unadjusted model. Model 2: Adjusted for age, gender, education level. Model 3: Adjusted for variables included in Model 2 + smoking, dyslipidemia, hypertension, diabetes. Model 4: Adjusted for variables included in Model 3 + hemoglobin, creatinine, UA Further analyses of the cumulative WHtR revealed that, as a continuous exposure, the unadjusted HR was 3.029 (2.041–4.496), and the fully adjusted HR was 1.798 (1.162–2.781). In the quartile analyses, compared with those for Q1, the HRs were 1.205 (1.003–1.557) for Q2, 1.171 (1.022–1.521) for Q3, and 1.478 (1.141–1.913) for Q4 (p for trend = 0.005). Overall, the cumTyG-WHtR was independently and positively associated with a higher risk of incident CVD in middle-aged and older adults, with a dose–response pattern; the cumulative WHtR showed a similar association. We further performed subtype analyses for heart disease and stroke separately. Among the 587 incident CVD events occurring between wave 3 and wave 4 (2015–2018), 389 were classified as heart disease events, 234 as stroke events, and 36 participants developed both heart disease and stroke during follow-up. These participants were counted in both subtypes for the subtype‑specific analyses, but were considered as a single event in the analysis of the composite CVD outcome. In the subtype analyses, the association between cumTyG‑WHtR and incident cerebrovascular disease (HR = 1.103, 95% CI: 1.036–1.173) was slightly stronger than that for heart disease (HR = 1.031, 95% CI: 0.983–1.082); This difference is statistically significant (p-value < 0.001). (Table S2). Restricted cubic spline We used restricted cubic spline (RCS) regression to assess the exposure–response relationship between the cumTyG-WHtR and incident cardiovascular disease (CVD) and to explore potential thresholds. After adjustment for confounders, there was no evidence of nonlinearity (P for nonlinearity = 0.790), indicating an approximately linear, monotonic dose–response: a higher cumTyG-WHtR is associated with a greater risk of incident CVD (Fig. 3). Fig. 3Nonlinear association between cumulative TyG–WHtR and CVD incidence. The models were adjusted for age, gender, education level, smoking status, dyslipidemia, hypertension, diabetes, hemoglobin, serum creatinine and uric acid Nonlinear association between cumulative TyG–WHtR and CVD incidence. The models were adjusted for age, gender, education level, smoking status, dyslipidemia, hypertension, diabetes, hemoglobin, serum creatinine and uric acid Subgroup analyses Effect modification was evaluated by adding interaction terms to the Cox models and conducting prespecified subgroup analyses (Fig. 4), stratified by age (45–59 years vs. ≥ 60 years), gender, residence, smoking, alcohol use, and history of hypertension and diabetes. A significant interaction was observed for age (P for interaction = 0.025), with a stronger association in those aged 45–59 years. In the age-stratified models, the hazard ratio (HR) for incident CVD per 1-SD increase in the cumTyG-WHtR was 1.079 (95% CI, 1.019–1.141) in the 45–59-year group and 1.028 (0.973–1.086) in the ≥ 60-year group. The results for the other strata are shown in Fig. 5. Fig. 4Subgroup analyses of the association between cumulative TyG–WHtR and CVD risk. The models were adjusted for age, gender, education level, smoking status, dyslipidemia, hypertension, diabetes, hemoglobin, serum creatinine and uric acid Subgroup analyses of the association between cumulative TyG–WHtR and CVD risk. The models were adjusted for age, gender, education level, smoking status, dyslipidemia, hypertension, diabetes, hemoglobin, serum creatinine and uric acid Fig. 5Exploratory associational decomposition of the association between cumulative TyG-WHtR and CVD incidence, with MAP as a statistical explanatory variable. The models were adjusted for age, gender, education level, smoking status, dyslipidemia, diabetes, hemoglobin, serum creatinine and uric acid. Abbreviations: IE, indirect association; DE, direct association; TE, total association; Prop., proportion of total association statistically accounted for Exploratory associational decomposition of the association between cumulative TyG-WHtR and CVD incidence, with MAP as a statistical explanatory variable. The models were adjusted for age, gender, education level, smoking status, dyslipidemia, diabetes, hemoglobin, serum creatinine and uric acid. Abbreviations: IE, indirect association; DE, direct association; TE, total association; Prop., proportion of total association statistically accounted for Sensitivity analyses To assess robustness (Table 3), we performed several sensitivity analyses: (1) excluding participants with dyslipidemia; (2) excluding those with hypertension; (3) repeating the primary analyses in a complete-case dataset (removing records with missing data); (4) conducting stratified analyses by exposure quartiles; (5) using the mean values of TyG-WHtR across two time points (wave 1 and wave 3) as the exposure variable; (6) using baseline TyG-WHtR alone (wave 1) as the exposure variable; (7) using TyG-WHtR at wave 3 as the exposure variable; (8) using change in TyG-WHtR (Δ = wave 3 – wave 1) as the exposure variable. The magnitude, direction, and statistical significance of the associations were largely unchanged, supporting the robustness of the findings. Table 3Sensitivity analysisModel1HR (95% CI) P valueModel2HR (95% CI) P valueModel3HR (95% CI) P valueModel4HR (95% CI) P valueSensitivity analysis 1excluding participants with dyslipidemia (N = 3943)Increase 1 SD1.091 (1.050,1.134)< 0.0011.076(1.034,1.120)< 0.0011.054(1.011,1.098)0.0131.063(1.018,1.109)0.005Q1RefRefRefRefQ21.167(1.080,1.521)0.0081.144(1.062,1.494)0.0131.118(1.031,1.461)0.0251.140(1.042,1.490)0.020Q31.440(1.113,1.861)0.0051.385(1.065,1.801)0.0051.310(1.048,1.706)0.0011.344(1.030,1.753)0.029Q41.670(1.299,2.148)< 0.00011.540(1.182,2.007)0.0011.364(1.139,1.792)0.0011.435(1.084,1.899)0.012P for trend< 0.001< 0.0010.0080.006Sensitivity analysis 2excluding participants with hypertension(N = 3375)Increase 1 SD1.061 (1.032,1.109)< 0.0011.048(1.022,1.098)0.0011.039(1.013,1.089)0.0051.040(1.020,1.092)0.003Q1RefRefRefRefQ21.179(1.039,1.562)0.0041.159(1.027,1.540)0.0061.145(1.006,1.522)0.0381.153(1.017,1.534)0.018Q31.420(1.074,1.876)0.0011.374(1.033,1.829)0.0011.339(1.016,1.784)0.0211.353(1.013,1.807)0.008Q41.455(1.089,1.944)< 0.00011.364(1.055,1.844)0.0011.301(1.027,1.762)0.0111.320(1.039,1.807)0.005P for trend< 0.0010.0010.0080.004Sensitivity analysis 3excluding participants with missing data (N = 3973)Increase 1 SD1.107 (1.069,1.147)< 0.0011.095(1.056,1.137)< 0.0011.050(1.009,1.092)0.0151.057(1.015,1.101)0.007Q1RefRefRefRefQ21.226(1.045,1.590)0.0011.211(1.032,1.574)0.0021.169(1.017,1.520)0.0071.190(1.022,1.691)0.002Q31.443(1.120,1.858)0.0051.404(1.084,1.817)0.0011.287(1.033,1.669)0.0061.321(1.017,1.716)0.001Q41.895(1.492,2.407)< 0.0011.786(1.391,2.293)< 0.00011.411(1.085,1.836)0.0011.475(1.126,1.931)< 0.001P for trend< 0.001< 0.0010.0080.004Sensitivity analysis 4baseline value of TyG-WHtR (wave 1) (N = 4254)Increase 1 SD1.290 (1.167,1.426)< 0.0011.240 (1.118,1.376)< 0.0011.091(1.002,1.215)0.0451.117(1.008,1.249)0.024Q1RefRefRefRefQ21.357(1.053,1.749)0.0081.328(1.029,1.713)0.0291.271(1.014,1.641)0.0361.287(0.996,1.664)0.053Q31.504(1.172,1.930)0.0011.443(1.121,1.857)0.0041.278(1.066,1.649)0.0371.304(1.068,1.685)0.006Q41.836(1.446,2.332)< 0.0011.699(1.326,2.176)< 0.0011.326(1.023,1,719)0.0031.392(1.066,1,817)0.015P for trend< 0.001< 0.0010.0350.025Sensitivity analysis 5average value of TyG-WHtR (in wave 1 and wave 3) (N = 4254)Increase 1 SD1.347 (1.215,1.494)< 0.0011.303(1.170,1.452)< 0.0011.136 (1.013,1.273)0.0291.164 (1.035,1.309)0.011Q1RefRefRefRefQ21.252(1.060,1.615)0.0081.238(1.58,1.600)0.0121.194(1.024,1.543)0.0351.216(1.041,1.573)0.005Q31.524(1.191,1.905)0.0011.485(1.155,1.909)0.0021.345(1.045,1.732)0.0211.383(1.072,1.784)0.013Q41.871(1.479,2.367)< 0.0011.762(1.378,2.253)< 0.0011.364(1.053,1.766)0.0191.430(1.098,1.864)0.008P for trend< 0.001< 0.0010.0140.006Sensitivity analysis 6value of TyG-WHtR (wave 3) (N = 4254)Increase 1 SD1.283 (1.168,1.408)< 0.0011.251 (1.136,1.377)< 0.0011.120 (1.015,1.236)0.0241.136 (1.027,1.256)0.013Q1RefRefRefRefQ21.404(1.083,1.819)0.0101.401(1.079,1.819)0.0111.343(1.034,1.744)0.0271.369(1.053,1.779)0.004Q31.811(1.412,2.322)< 0.0011.792(1.389,2.311)< 0.0011.598(1.236,2.065)< 0.0011.633(1.262,2.114)< 0.001Q41.886(1.475,2.411)< 0.0011.810(1.403,2.336)< 0.0011.425(1.094,1.858)0.0091.477(1.130,1.932)0.004P for trend< 0.001< 0.0010.0080.004Sensitivity analysis 7change values of TyG-WHtR ((Δ = wave 3 – wave 1)) (N = 4254)Increase 1 SD1.026 (1.016,1.056)< 0.0011.025 (1.015,1.055)< 0.0011.022 (1.012,1.053)0.0031.021 (1.012,1.051)0.003Q1RefRefRefRefQ21.066(1.024,1.285)< 0.0011.051(1.012,1.267)< 0.0011.021(1.009,1.230)0.0041.014(1.007,1.210)0.006Q31.119(1.028,1.349)< 0.0011.098(1.018,1.323)< 0.0011.056(1.017,1.272)< 0.0011.025(1.012,1.234)0.001Q41.121(1.047,1.350)< 0.0011.107(1.043,1.332)< 0.0011.099(1.028,1.322)< 0.0011.069(1.013,1.322)0.001P for trend< 0.001< 0.0010.0030.007HR, hazard ratio; CI, confidence interval; Ref, reference; Model 1: Unadjusted model. Model 2: Adjusted for age, gender, education level. Model 3: Adjusted for variables included in Model 2 + smoking, dyslipidemia, hypertension, diabetes. Model 4: Adjusted for variables included in Model 3 + hemoglobin, creatinine, UA. Among the sensitivity analyses, using the change in TyG‑WHtR (Δ = wave3 – wave1) as the exposure yielded a smaller HR, indicating that cumulative exposure captures long‑term risk better than short‑term changes Table 3Sensitivity analysisModel1HR (95% CI) P valueModel2HR (95% CI) P valueModel3HR (95% CI) P valueModel4HR (95% CI) P valueSensitivity analysis 1excluding participants with dyslipidemia (N = 3943)Increase 1 SD1.091 (1.050,1.134)< 0.0011.076(1.034,1.120)< 0.0011.054(1.011,1.098)0.0131.063(1.018,1.109)0.005Q1RefRefRefRefQ21.167(1.080,1.521)0.0081.144(1.062,1.494)0.0131.118(1.031,1.461)0.0251.140(1.042,1.490)0.020Q31.440(1.113,1.861)0.0051.385(1.065,1.801)0.0051.310(1.048,1.706)0.0011.344(1.030,1.753)0.029Q41.670(1.299,2.148)< 0.00011.540(1.182,2.007)0.0011.364(1.139,1.792)0.0011.435(1.084,1.899)0.012P for trend< 0.001< 0.0010.0080.006Sensitivity analysis 2excluding participants with hypertension(N = 3375)Increase 1 SD1.061 (1.032,1.109)< 0.0011.048(1.022,1.098)0.0011.039(1.013,1.089)0.0051.040(1.020,1.092)0.003Q1RefRefRefRefQ21.179(1.039,1.562)0.0041.159(1.027,1.540)0.0061.145(1.006,1.522)0.0381.153(1.017,1.534)0.018Q31.420(1.074,1.876)0.0011.374(1.033,1.829)0.0011.339(1.016,1.784)0.0211.353(1.013,1.807)0.008Q41.455(1.089,1.944)< 0.00011.364(1.055,1.844)0.0011.301(1.027,1.762)0.0111.320(1.039,1.807)0.005P for trend< 0.0010.0010.0080.004Sensitivity analysis 3excluding participants with missing data (N = 3973)Increase 1 SD1.107 (1.069,1.147)< 0.0011.095(1.056,1.137)< 0.0011.050(1.009,1.092)0.0151.057(1.015,1.101)0.007Q1RefRefRefRefQ21.226(1.045,1.590)0.0011.211(1.032,1.574)0.0021.169(1.017,1.520)0.0071.190(1.022,1.691)0.002Q31.443(1.120,1.858)0.0051.404(1.084,1.817)0.0011.287(1.033,1.669)0.0061.321(1.017,1.716)0.001Q41.895(1.492,2.407)< 0.0011.786(1.391,2.293)< 0.00011.411(1.085,1.836)0.0011.475(1.126,1.931)< 0.001P for trend< 0.001< 0.0010.0080.004Sensitivity analysis 4baseline value of TyG-WHtR (wave 1) (N = 4254)Increase 1 SD1.290 (1.167,1.426)< 0.0011.240 (1.118,1.376)< 0.0011.091(1.002,1.215)0.0451.117(1.008,1.249)0.024Q1RefRefRefRefQ21.357(1.053,1.749)0.0081.328(1.029,1.713)0.0291.271(1.014,1.641)0.0361.287(0.996,1.664)0.053Q31.504(1.172,1.930)0.0011.443(1.121,1.857)0.0041.278(1.066,1.649)0.0371.304(1.068,1.685)0.006Q41.836(1.446,2.332)< 0.0011.699(1.326,2.176)< 0.0011.326(1.023,1,719)0.0031.392(1.066,1,817)0.015P for trend< 0.001< 0.0010.0350.025Sensitivity analysis 5average value of TyG-WHtR (in wave 1 and wave 3) (N = 4254)Increase 1 SD1.347 (1.215,1.494)< 0.0011.303(1.170,1.452)< 0.0011.136 (1.013,1.273)0.0291.164 (1.035,1.309)0.011Q1RefRefRefRefQ21.252(1.060,1.615)0.0081.238(1.58,1.600)0.0121.194(1.024,1.543)0.0351.216(1.041,1.573)0.005Q31.524(1.191,1.905)0.0011.485(1.155,1.909)0.0021.345(1.045,1.732)0.0211.383(1.072,1.784)0.013Q41.871(1.479,2.367)< 0.0011.762(1.378,2.253)< 0.0011.364(1.053,1.766)0.0191.430(1.098,1.864)0.008P for trend< 0.001< 0.0010.0140.006Sensitivity analysis 6value of TyG-WHtR (wave 3) (N = 4254)Increase 1 SD1.283 (1.168,1.408)< 0.0011.251 (1.136,1.377)< 0.0011.120 (1.015,1.236)0.0241.136 (1.027,1.256)0.013Q1RefRefRefRefQ21.404(1.083,1.819)0.0101.401(1.079,1.819)0.0111.343(1.034,1.744)0.0271.369(1.053,1.779)0.004Q31.811(1.412,2.322)< 0.0011.792(1.389,2.311)< 0.0011.598(1.236,2.065)< 0.0011.633(1.262,2.114)< 0.001Q41.886(1.475,2.411)< 0.0011.810(1.403,2.336)< 0.0011.425(1.094,1.858)0.0091.477(1.130,1.932)0.004P for trend< 0.001< 0.0010.0080.004Sensitivity analysis 7change values of TyG-WHtR ((Δ = wave 3 – wave 1)) (N = 4254)Increase 1 SD1.026 (1.016,1.056)< 0.0011.025 (1.015,1.055)< 0.0011.022 (1.012,1.053)0.0031.021 (1.012,1.051)0.003Q1RefRefRefRefQ21.066(1.024,1.285)< 0.0011.051(1.012,1.267)< 0.0011.021(1.009,1.230)0.0041.014(1.007,1.210)0.006Q31.119(1.028,1.349)< 0.0011.098(1.018,1.323)< 0.0011.056(1.017,1.272)< 0.0011.025(1.012,1.234)0.001Q41.121(1.047,1.350)< 0.0011.107(1.043,1.332)< 0.0011.099(1.028,1.322)< 0.0011.069(1.013,1.322)0.001P for trend< 0.001< 0.0010.0030.007HR, hazard ratio; CI, confidence interval; Ref, reference; Model 1: Unadjusted model. Model 2: Adjusted for age, gender, education level. Model 3: Adjusted for variables included in Model 2 + smoking, dyslipidemia, hypertension, diabetes. Model 4: Adjusted for variables included in Model 3 + hemoglobin, creatinine, UA. Among the sensitivity analyses, using the change in TyG‑WHtR (Δ = wave3 – wave1) as the exposure yielded a smaller HR, indicating that cumulative exposure captures long‑term risk better than short‑term changes Sensitivity analysis HR, hazard ratio; CI, confidence interval; Ref, reference; Model 1: Unadjusted model. Model 2: Adjusted for age, gender, education level. Model 3: Adjusted for variables included in Model 2 + smoking, dyslipidemia, hypertension, diabetes. Model 4: Adjusted for variables included in Model 3 + hemoglobin, creatinine, UA. Among the sensitivity analyses, using the change in TyG‑WHtR (Δ = wave3 – wave1) as the exposure yielded a smaller HR, indicating that cumulative exposure captures long‑term risk better than short‑term changes Exploratory associational decomposition Using a nonparametric percentile bootstrap (1,000 resamples), we conducted an exploratory associational decomposition (not a causal mediation analysis) to evaluate the proportion of the total association between cumTyG‑WHtR and incident CVD that was statistically accounted for by MAP (Fig. 5). Because MAP was measured concurrently with the second exposure time point (wave 3), the temporal sequence required for causal mediation cannot be established; thus, this analysis is hypothesis-generating. The estimated average indirect association was 0.005 (95% CI, 0.002–0.008; p < 0.001), the average direct association was 0.02 (95% CI, 0.006–0.032; p < 0.001), and the total association was 0.025 (95% CI, 0.011–0.038; p < 0.001). The proportion of the total association statistically accounted for by MAP was 20.3% (95% CI, 7.8%–49.4%). This finding is hypothesis-generating and does not imply causation. Discussion Leveraging the nationally representative CHARLS cohort, we found that higher cumTyG-WHtR was positively associated with incident CVD in a near-linear, monotonic dose–response manner, with stronger associations among adults aged 45–59 years. The exploratory associational decomposition revealed that MAP statistically accounted for approximately 20.3% of the total association between cumTyG-WHtR and incident CVD. This hypothesis‑generating finding does not establish causation and requires confirmation in future studies with stronger causal designs. Our study extends prior evidence in several important ways. The literature has established TyG as a surrogate for insulin resistance (IR), which predicts CVD incidence and prognosis and is associated with stroke, hypertension, and carotid atherosclerosis [4, 33–36]. Baseline TyG-related indices also predict CVD and coronary heart disease (CHD) risk [37–39]. However, most prior studies rely on a single baseline measurement and cannot adequately capture the dynamic and cumulative features of metabolic risk. By emphasizing cumulative exposure, our study integrates long-term metabolic load (TyG) and central obesity (WHtR) into the cumTyG-WHtR index, providing a more comprehensive and stable representation of the joint influence of IR and abdominal adiposity and demonstrating stronger risk discrimination than standalone metrics do. The clear dose–response gradient and robust positive association support the predictive potential and generalizability of this composite index. Mechanistically, blood pressure serves as a critical nexus linking long-term metabolic load to cardiovascular events. Elevated blood pressure increases mechanical stress on the arterial wall, induces vascular remodeling and fibrosis, impairs endothelial function, and promotes plaque formation; it frequently coexists with low-grade chronic inflammation and sympathetic activation, driving arterial stiffening and adverse hemodynamics [40–44]. Elevated blood pressure can further exacerbate IR via inflammatory and sympathoadrenergic pathways; conversely, IR elevates blood pressure by attenuating PI3K-Akt-eNOS signaling and NO bioavailability, enhancing SGK1-ENaC-mediated sodium influx, and promoting renal microvascular remodeling [45–50]. IR and central obesity also synergize across metabolic, vascular, inflammatory, and thrombotic axes: IR elevates free fatty acids, impairs endothelial-dependent vasodilation, increases arterial stiffness, and activates platelets [7, 51–56]. These coupled pathways explain the graded increase in CVD risk with higher cumulative exposure and underscore the importance of blood pressure management to interrupt the progression from metabolic abnormalities to adverse cardiovascular outcomes. Overall, cumTyG-WHtR, a comprehensive indicator of long-term metabolic load and fat distribution, was positively associated with CVD risk in middle-aged and elderly adults. A portion of this association was statistically accounted for by MAP. These findings highlight that blood pressure control and fat management may be complementary components of comprehensive CVD prevention, though further studies with proper temporal ordering are needed to establish causality. Age-related effect modification has clear clinical and public health implications. The stronger association between the cumTyG-WHtR and CVD risk in adults aged 45–59 years suggests that midlife represents a critical window for prevention and intervention. In this population, priority should be given to lifestyle strategies centered on weight management, improving cardiorespiratory fitness, and optimizing dietary patterns, alongside early identification and control of metabolic risk factors and blood pressure, to maximize cardiovascular benefits. Our results also support incorporating a “dynamic–cumulative” monitoring paradigm into population management: beyond single-time assessments of TyG and WHtR, greater emphasis should be placed on cumulative levels over time (e.g., cumTyG, cumWHtR, and cumTyG-WHtR) to enhance early identification and risk stratification of high-risk individuals. Subtype analyses suggested a stronger association of cumTyG‑WHtR with stroke than with heart disease (p for difference < 0.001), potentially due to greater cerebrovascular vulnerability to insulin resistance [57, 58] and the high metabolic demand of cerebral vessels [59]. Given the modest number of stroke events (n = 234) and the self‑reported nature of the outcomes, these findings require confirmation using clinically adjudicated endpoints. This study has several methodological strengths. First, it is based on a large, nationally representative cohort with standardized follow-up, and data were collected by trained personnel via uniform protocols, ensuring high measurement and data quality. Second, we employed a systematic analytical strategy: feature selection and collinearity diagnostics; stratified and interaction analyses to identify effect modifiers; exploration of dose‒response patterns; multilevel adjusted models; multiple sensitivity analyses to verify robustness; and exploratory associational decomposition to explore potential pathways. Several limitations warrant consideration. First, the observational design precludes definitive causal inference, and despite extensive adjustments, residual confounding cannot be fully excluded. Second, CVD outcomes were based on self-reported physician diagnoses, which may introduce underdiagnosis (particularly in rural or less-educated populations) and recall bias, and these self-reports were not validated against medical insurance records or mortality registry data. If underdiagnosis or misclassification of CVD (particularly stroke) is non‑differential with respect to cumTyG‑WHtR levels, the observed hazard ratios would be biased toward the null, implying that our estimates may be conservative. Differential misclassification is unlikely in this prospective design because outcome ascertainment preceded knowledge of exposure status. Future studies should validate self-reports against objective clinical records. Third, laboratory measurements were obtained only at waves 1 and 3, limiting finer-grained characterization of exposure dynamics. More importantly, because MAP was measured concurrently with the second exposure time point (wave 3), while CVD outcomes were ascertained in wave3, the temporal sequence of “exposure → MAP → outcome” cannot be strictly established. Therefore, our analysis should be interpreted as an exploratory decomposition of the total association between cumTyG-WHtR and CVD into indirect and direct components, rather than a causal mediation analysis. The reported proportions of association explained are hypothesis-generating and require further validation in future studies with appropriate temporal spacing. Additionally, because the cumulative exposure variable required complete blood specimen data from both 2011 and 2015, participants who died or were lost to follow-up before 2015 were excluded by design. This may introduce survivor bias, potentially selecting a healthier subgroup and attenuating the true association toward the null. Thus, our effect estimates may be conservative. Future studies with more frequent assessments or different designs are needed to further evaluate this bias. Fourth, because CHARLS does not provide exact dates of CVD onset and events are only ascertained at discrete follow-up waves, we used the wave 4 interview date as the event time in standard Cox models, which do not explicitly account for interval censoring. This may introduce bias (likely toward the null) or alter precision. Future analyses using interval-censored models would be preferable. Fifth, competing risk. Death before CVD is a competing event. Fine-Gray analysis was not feasible due to 30.3% missing death data (1,291/4,254). In our Cox models, death was treated as censored. If higher cumTyG-WHtR is associated with higher mortality (plausible), this approach may overestimate the HR. Future studies with complete mortality follow-up are needed. Finally, our analysis sample consists of a group of middle-aged and elderly Chinese adults selected from CHARLS after excluding those with missing data and those who did not meet the inclusion criteria, which may limit the representativeness and external validity of the sample. Conclusion In this nationally representative CHARLS cohort, higher cumTyG-WHtR was positively associated with a higher risk of CVD. This association was more pronounced among adults aged 45–59 years. In an exploratory associational decomposition, MAP statistically accounted for approximately 20.3% of the total association. This hypothesis‑generating finding does not establish causation and requires confirmation in future studies with stronger causal designs. Supplementary Information Supplementary Material 1. Fig.S1. Feature selection for predicting CVD incidence risk via the Boruta algorithm. The horizontal axis represents the name of each variable, and the vertical axis represents the Z value of each variable. The box plot shows the Z value of each variable during model calculation. The green boxes represent important variables, and the red boxes represent unimportant variables. Supplementary Material 1. Fig.S1. Feature selection for predicting CVD incidence risk via the Boruta algorithm. The horizontal axis represents the name of each variable, and the vertical axis represents the Z value of each variable. The box plot shows the Z value of each variable during model calculation. The green boxes represent important variables, and the red boxes represent unimportant variables. Supplementary Material 2. Table S1 Number and proportion of missing values by variable. Supplementary Material 2. Table S1 Number and proportion of missing values by variable. Supplementary Material 3. Table S2. Associations between cum TyG-WHtR and incident heart disease or stroke risk across statistical models. Supplementary Material 3. Table S2. Associations between cum TyG-WHtR and incident heart disease or stroke risk across statistical models.