Differential associations of cumulative average metabolic score for insulin resistance with heart disease and stroke: a prospective cohort study of middle-aged and older Chinese adults Authors: Chun Yang, Zeng Guo, Mengqing Ma, Shiyu Jin, Shuyi Zhang, Ling Xia, Longfeng Xia, Xianhe Lin Published: 2025 ABSTRACT AbstractBackgroundInsulin resistance (IR) is a well-established risk factor for cardiovascular disease (CVD). The metabolic score for insulin resistance (METS-IR) is a novel surrogate marker for IR, yet the association between its long-term cumulative burden and incident CVD remains unexplored, particularly in the Chinese population.MethodsThis prospective study included 5,264 participants (mean age 58.50 years; 46.6% male) from the China Health and Retirement Longitudinal Study (CHARLS). Cumulative average METS-IR was calculated as the mean of measurements from 2011 (baseline) and 2015 (wave 3). Incident CVD events were ascertained via standardized questionnaires. Multivariable logistic regression and restricted cubic spline models were employed to assess associations.ResultsDuring a 4-year follow-up, 554 (10.5%) incident CVD cases occurred. After full adjustment, each SD increase in cumulative average METS-IR was associated with a 16.1% higher risk of total CVD (OR = 1.161, 95% CI: 1.050–1.282). A significant dose-response relationship was observed (P for trend = 0.003). Crucially, we found a striking organ-specific disparity: the highest quartile of cumulative METS-IR was strongly associated with a 19.8% increased risk of heart disease (OR = 1.509, 95% CI: 1.118–2.046) but not with stroke (OR = 0.902, 95% CI: 0.440–1.865). Subgroup analyses revealed that this association was particularly pronounced among males, normotensive individuals, and current smokers or drinkers.ConclusionsCumulative exposure to insulin resistance, quantified by METS-IR, is an independent predictor of CVD, especially heart disease, in middle-aged and older Chinese adults. Our findings highlight the clinical utility of longitudinal METS-IR assessment for early risk stratification and pinpoint specific high-risk populations for targeted preventive strategies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12986-025-01049-1.Keywords: Metabolic score for insulin resistance, Cumulative exposure, Heart disease, CHARLS, Prospective cohort study Introduction Since the 20th century, cardiovascular disease (CVD) has displaced infectious diseases as the predominant cause of global morbidity and mortality, imposing escalating burdens on healthcare systems worldwide [1]. In China, accelerated population aging and rising metabolic risk factor prevalence have driven a continuous increase in CVD burden. By 2019, approximately 330 million individuals were affected by CVD, accounting for 46.7% and 44.3% of total deaths in rural and urban areas, respectively—equivalent to CVD-attributable deaths representing nearly 40% of national mortality [2]. Notably, economically developed coastal regions exhibit markedly lower CVD burdens compared to less developed central/western areas, revealing substantial regional health disparities [3]. These findings underscore an urgent imperative to refine risk identification systems for controlling CVD incidence and mortality. Insulin resistance, a pathological state characterized by impaired insulin-mediated glucose uptake in peripheral tissues, is widely recognized as a key driver in the development and progression of CVD [4–7]. Currently, the hyperinsulinemic-euglycemic clamp (HEC) technique remains the gold standard for assessing IR; however, its complexity and high cost prohibit its widespread clinical application [8].In recent years, the METS-IR has emerged as a novel composite index, providing an efficient solution for the early identification of insulin resistance [9]. Population-based cohort studies have demonstrated that METS-IR offers superior predictive performance for IR compared to traditional surrogate markers such as the triglyceride-glucose (TyG) index [10]. Existing evidence has established significant associations between METS-IR and metabolic disorders, including diabetes mellitus and non-alcoholic fatty liver disease (NAFLD) [11, 12]. However, the predictive value of longitudinal METS-IR trajectories for incident CVD risk has not been systematically elucidated in middle-aged and older Chinese adults. To address this gap, this study leverages the CHARLS database to examine the association between cumulative average METS-IR levels and CVD incidence among adults aged 45 years and older. Methods Data source and study population This study utilized data from the nationally representative, prospective CHARLS, which tracks the health trajectory of adults aged ≥ 45 years. The baseline survey (Wave 1, 2011) enrolled 17,708 participants from 10,257 households across 450 villages/districts within 150 county-level units spanning 28 Chinese provinces. Biennial follow-ups were conducted (Waves 2–4: 2013, 2015, 2018), with fasting serum samples collected during Wave 1(n = 11,847) and Wave 3 (n = 13,420). Detailed methodologies regarding CHARLS sampling design, anthropometric protocols, and biomarker assays are documented in prior publications [13, 14]. The CHARLS received ethical approval from the ethics review committee (institutional review board) of Peking University (IRB00001052-11015), with all participants providing written informed consent. Based on biomarker availability across Waves 1 and 3, we applied sequential exclusions: participants lost to Wave 3 follow-up (n = 1,399) were first excluded, followed by those with incomplete baseline demographics (n = 97). We subsequently excluded individuals missing FBG, TG, BMI, or HDL-c in either wave (n = 3,905). Further exclusions comprised participants aged < 45 years (n = 181), those with baseline CVD (n = 923), and outliers defined as BMI < 15 or >55 kg/m² (n = 73) or METS-IR beyond mean ± 3 SD (n = 5). The final analytical cohort comprised 5,264 participants (Fig. 1). Flowchart of the study population. FBG, fasting blood glucose; TG, triglyceride; BMI, body mass index; HDL-c, high-density lipoprotein cholesterol; METS-IR, metabolic score for insulin resistance; Q1, Quartile 1; Q2, Quartile 2; Q3, Quartile 3; Q4, Quartile 4 Data assessment and definitions Assessment of exposure This study employed the cumulative average METS-IR as the primary indicator of long-term homeostatic status for insulin resistance. METS-IR was computed using the standardized formula established by Bello-Chavolla et al. [9]: METS-IR = Ln[(2 × FBG (mg/dL) + TG (mg/dL)) × BMI]/Ln[HDL-C (mg/dL)]. The cumulative average METS-IR was calculated by integrating measurements from baseline (2011) and follow-up (2015) waves: (METS-IR2011 + METS-IR2015)/2. Assessment of outcomes The primary outcome of this study was incident CVD, ascertained through standardized questionnaires administered during each CHARLS follow-up wave. Participants’ disease histories were collected using two validated questions: “Have you ever been diagnosed by a physician with a stroke or heart disease (including myocardial infarction, coronary heart disease, angina, heart failure, or other cardiac conditions)?” and “Are you currently receiving traditional Chinese medicine, Western medicine, or other targeted treatments for stroke/heart disease?“. In accordance with established epidemiological criteria for cohort studies [15–18], incident CVD cases were defined as participants who either self-reported physician-diagnosed stroke/heart disease or were undergoing disease-specific treatment at the time of assessment. Data collection This study systematically collected five categories of variables: (i) Demographic characteristics: gender, age, educational attainment, residential registration type (urban/rural), and marital status; (ii) Anthropometric measurements: systolic blood pressure (SBP), diastolic blood pressure (DBP and BMI; (iii) Lifestyle factors: smoking status and alcohol consumption frequency; (iv) Medical histories: dyslipidemia, hypertension, diabetes, kidney disease, liver disease, and medication histories for lipid-lowering therapy (LLT), antihypertensive therapy (AHT), and glucose-lowering therapy (GLT); (v) Laboratory biomarkers: FBG, TG, HDL-c, glycosylated hemoglobin (HbA1c), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-c), white blood cell count (WBC), hemoglobin (Hb), mean corpuscular volume (MCV), platelet count (PLT), C-reactive protein (CRP), uric acid (UA), serum creatinine (Scr), and blood urea nitrogen (BUN). Hypertension was defined according to established criteria as meeting any one of the following [19]: self-reported physician-diagnosed hypertension, current antihypertensive treatment, baseline SBP ≥ 140 mmHg, or DBP ≥ 90 mmHg. Diabetes mellitus was defined as fulfilling ≥ 1 criterion [20]: self-reported physician-diagnosed diabetes, active glucose-lowering therapy, baseline FBG ≥ 7.0 mmol/L (126 mg/dL), or HbA1c ≥ 6.5%. Other disease states were determined based on self-reported medical history or disease-specific treatment records. Handling of missing variables The distribution patterns of missing variables are detailed in Additional file 1: Figure S1. To maximize statistical power and minimize selection bias, we implemented multiple imputation using the mice package in R software [21]. After initializing the imputation method matrix and predictor matrix, algorithm assignment was performed according to variable type: predictive mean matching (PMM) for continuous variables (e.g., BMI, SBP, DBP), logistic regression (logreg) for binary variables (e.g., hypertension, diabetes), and multinomial logistic regression (polyreg) for unordered categorical variables (e.g., smoking status, drinking status). We generated five imputed datasets through 50 iterations with a fixed random seed (set.seed = 1234), subsequently pooling the results for final analysis. This comprehensive approach ensured robust handling of missing data while maintaining methodological rigor and enhancing the robustness of statistical conclusions. Statistical analysis The analytical cohort (n = 5,264) was stratified into quartiles from Quartile 1 (Q1) to Quartile 4 (Q4) based on cumulative average METS-IR. Continuous variables were expressed as mean ± standard deviation and compared across quartiles using one-way-ANOVA test, while categorical variables were presented as frequency counts with percentages and evaluated using Pearson chi-squared test to assess between-group differences. We employed hierarchical logistic regression models to evaluate the association between cumulative average METS-IR and incident CVD, implementing sequential covariate adjustment across three progressive tiers: Model 1 adjusted for demographic factors (age and gender); Model 2 expanded adjustment to sociodemographic and lifestyle variables (residence, education level, marital status, smoking, drinking consumption) and comorbidities (kidney diseases, liver diseases); Model 3 constituted full adjustment through additional incorporation of hemodynamic parameters (SBP, DBP), metabolic biomarkers (HbA1c, TC, LDL-c), inflammatory markers (WBC, CRP), hematological indices (Hb, MCV, PLT), renal function parameters (UA, Scr, BUN), and clinical histories (hypertension, diabetes). Results were expressed as odds ratios (OR) with corresponding 95% confidence intervals (95% CI). Multicollinearity assessment via variance inflation factors [22] (VIF) confirmed absence of significant collinearity (all VIF < 5, Additional file 1: Table S1). We subsequently utilized restricted cubic splines (RCS) with knots positioned at the 5th, 35th, 65th, and 95th percentiles to characterize the dose-response relationship, maintaining full multivariate adjustment. To investigate potential effect modification, we conducted stratified analyses across clinically relevant subgroups defined by age (< 60 vs. ≥60 years), gender, residence, smoking, drinking, kidney diseases, liver diseases, hypertension, diabetes and marital status. All analyses were executed in R version 4.4.1 (https://www.r-project.org/), with statistical significance defined as two-tailed P < 0.05. Sensitivity analysis Three sequential sensitivity analyses were conducted to verify result robustness: First, complete-case analysis excluding participants with any missing variables was performed, refitting logistic regression models to assess potential impacts of missing data. Second, we excluded individuals diagnosed with CVD during Wave 2 (2013) follow-up to eliminate potential confounding from short-term (< 2 years) incident cases. Third, propensity score matching (1:1 ratio using nearest-neighbor matching without replacement, caliper width = 0.01) constructed a balanced cohort, with post-matching equilibrium confirmed by standardized mean differences (SMD < 0.10 for all covariates) when comparing Q1-Q2 vs. Q3-Q4 groups, establishing statistically acceptable intergroup comparability. Absolute standardized mean differences for all covariates before and after matching are presented in Additional file 1: Figure S2. Results Baseline characteristics of participants The participant selection flowchart is detailed in Fig. 1, with 5,264 individuals ultimately included in the analytical cohort. The study population had a mean age of 58.50 ± 8.68 years, comprising 2,453 (46.6%) males. As detailed in Table 1, baseline characteristics stratified by cumulative average METS-IR quartiles revealed progressive metabolic and demographic gradients across groups. The overall cohort exhibited a mean cumulative average METS-IR of 32.31 ± 6.82. Participants in higher quartiles (Q2-Q4) demonstrated significantly younger age, greater proportions of females and urban residents, and higher prevalence of never-smoking and alcohol abstinence compared to Q1. Conversely, these groups showed lower educational attainment (predominantly elementary level or below) and reduced marriage rates. Biomarker profiles displayed consistent patterns: Q2-Q4 had elevated BMI, SBP, DBP, FBG, HbA1c, TC, TG, LDL-c, WBC, Hb, PLT and UA levels alongside reduced HDL-c and MCV. Comorbidity burden progressively increased across quartiles with higher prevalence of dyslipidemia, hypertension, diabetes mellitus, and utilization of LLT, AHT and GLT. (all P < 0.05). No significant interquartile differences were observed for kidney disease, liver disease, CRP, or Scr (P > 0.05). Table 1Baseline characteristics of individuals classified by quartiles of the cumulative average METS-IRCharacteristicsOverallQuartiles of the cumulative average METS-IRQuartile 1Quartile 2Quartile 3Quartile 4P-valuen52641316131613161316Gender< 0.001Male2453 (46.6%)742 (56.4%)624 (47.4%)584 (44.4%)503 (38.2%)Female2811 (53.4%)574 (43.6%)692 (52.6%)732 (55.6%)813 (61.8%)Age, years58.50 ± 8.6868.50 ± 9.0258.70 ± 8.9157.64 ± 8.1756.82 ± 8.05< 0.001Residence< 0.001Urban741 (14.1%)109 (8.3%)164 (12.5%)230 (17.5%)238 (18.1%)Rural4523 (85.9%)1207 (91.7%)1152 (87.5%)1086 (82.5%)1078 (81.9%)Education level< 0.001Primary or below3738 (71.0%)1010 (76.7%)966 (73.4%)885 (67.2%)877 (66.6%)Middle school1398 (26.6%)287 (21.8%)318 (24.2%)390 (29.6%)403 (30.6%)College or above128 (2.4%)19 (1.4%)32 (2.4%)41 (3.1%)36 (2.7%)Current married4497 (85.4%)1207 (91.7%)1152 (87.5%)1086 (82.5%)1078 (81.9%)< 0.001Smoking status< 0.001Never3232 (61.6%)665 (50.6%)808 (61.5%)857 (65.3%)902 (68.7%)Former404 (7.7%)102 (7.8%)78 (5.9%)112 (8.5%)112 (8.5%)Current1615 (30.8%)546 (41.6%)427 (32.5%)343 (26.1%)299 (22.8%)Drinking status< 0.001Never3184 (60.5%)705 (53.6%)784 (59.6%)837 (63.7%)858 (65.2%)Former410 (7.8%)108 (8.2%)98 (7.4%)96 (7.3%)108 (8.2%)Current1666 (31.7%)502 (38.2%)434 (33.0%)380 (28.9%)350 (26.6%)BMI, kg/m223.42 ± 3.5519.78 ± 1.6522.25 ± 1.7024.25 ± 2.0227.39 ± 3.10< 0.001SBP, mmHg128.45 ± 20.78124.30 ± 20.34126.44 ± 20.13128.80 ± 20.24134.31 ± 21.05< 0.001DBP, mmHg74.97 ± 12.0871.52 ± 11.4273.43 ± 11.8675.78 ± 11.6479.17 ± 12.03< 0.001Dyslipidemia421 (8.2%)35 (2.7%)71 (5.5%)105 (8.1%)210 (16.4%)< 0.001Hypertension2005 (38.1%)344 (26.1%)421 (32.0%)529 (40.2%)711 (54.0%)< 0.001Diabetes825 (15.7%)119 (9.0%)152 (11.6%)204 (15.5%)350 (26.6%)< 0.001Kidney disease257 (4.9%)63 (4.8%)66 (5.0%)62 (4.7%)66 (5.0%)0.976Liver disease143 (2.7%)46 (3.5%)29 (2.2%)38 (2.9%)30 (2.3%)0.142LLT203 (3.9%)14 (1.1%)25 (1.9%)52 (4.0%)112 (8.7%)< 0.001AHT841 (16.0%)105 (8.0%)139 (10.6%)224 (17.1%)373 (28.5%)< 0.001GLT159 (3.0%)18 (1.4%)31 (2.4%)33 (2.5%)77 (5.9%)< 0.001FBG, mg/dL109.36 ± 34.84102.82 ± 24.18104.88 ± 25.88109.17 ± 35.26120.57 ± 46.60< 0.001HbA1c, %5.25 ± 0.805.13 ± 0.595.17 ± 0.685.24 ± 0.775.48 ± 1.03< 0.001TC, mg/dL193.39 ± 37.63188.73 ± 35.69191.01 ± 35.66193.81 ± 38.41199.99 ± 39.70< 0.001TG, mg/dL130.13 ± 99.8682.76 ± 38.50104.19 ± 51.30129.37 ± 67.21204.19 ± 151.23< 0.001HDL-c, mg/dL51.55 ± 15.2463.57 ± 15.8554.35 ± 12.4248.05 ± 11.0040.22 ± 10.47< 0.001LDL-c, mg/dL116.42 ± 34.35110.59 ± 31.01117.45 ± 31.63120.66 ± 35.36116.99 ± 38.20< 0.001WBC, 10³/µL6.24 ± 1.896.07 ± 1.916.10 ± 1.856.32 ± 1.916.45 ± 1.88< 0.001Hb, g/dL14.38 ± 2.1814.08 ± 2.1314.26 ± 2.1014.44 ± 2.0914.73 ± 2.36< 0.001MCV, fl.90.71 ± 8.4391.40 ± 8.8990.79 ± 8.8690.65 ± 8.1290.01 ± 7.75< 0.001PLT, 109/L211.74 ± 72.19206.77 ± 70.90210.90 ± 73.98213.67 ± 70.76215.63 ± 72.840.012CRP, mg/L2.52 ± 7.112.44 ± 8.772.44 ± 8.302.42 ± 5.452.79 ± 5.180.476UA, mg/dL4.39 ± 1.224.30 ± 1.224.25 ± 1.174.40 ± 1.224.61 ± 1.26< 0.001Scr, mg/dL0.77 ± 0.180.77 ± 0.180.76 ± 0.190.77 ± 0.170.77 ± 0.190.555BUN, mg/dL15.68 ± 4.3716.26 ± 4.7415.68 ± 4.3315.49 ± 4.2915.29 ± 4.02< 0.001Cumulative average METS-IR32.31 ± 6.8224.36 ± 2.0929.31 ± 1.2433.86 ± 1.4741.69 ± 3.97< 0.001BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; LLT, lipid-lowering therapy; AHT, antihypertensive therapy; GLT, glucose-lowering therapy; FBG, fasting blood glucose; HbA1c, glycosylated hemoglobin A1c; TC, total cholesterol; TG, triglyceride; HDL-c, high‐density lipoprotein cholesterol; LDL-c, low-density lipoprotein cholesterol; WBC, white blood cell; Hb, hemoglobin; MCV, mean corpuscular volume; PLT, platelet; CRP, c-reactive protein; UA, uric acid; Scr, serum creatinine; BUN, blood urea nitrogen; METS-IR, metabolic score for insulin resistance Baseline characteristics of individuals classified by quartiles of the cumulative average METS-IR BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; LLT, lipid-lowering therapy; AHT, antihypertensive therapy; GLT, glucose-lowering therapy; FBG, fasting blood glucose; HbA1c, glycosylated hemoglobin A1c; TC, total cholesterol; TG, triglyceride; HDL-c, high‐density lipoprotein cholesterol; LDL-c, low-density lipoprotein cholesterol; WBC, white blood cell; Hb, hemoglobin; MCV, mean corpuscular volume; PLT, platelet; CRP, c-reactive protein; UA, uric acid; Scr, serum creatinine; BUN, blood urea nitrogen; METS-IR, metabolic score for insulin resistance In addition, The distribution of cumulative average METS-IR across the study population is presented in Additional file 1: Figure S3, while Additional file 1: Table S2 summarizes baseline characteristics stratified by incident CVD status during follow-up. Association between the cumulative average METS-IR and cardiovascular disease incidence Over the 4-year follow-up period (2011 to 2015), 554 incident cardiovascular disease events occurred, representing 10.5% of the cohort. We observed a graded increase in CVD incidence with ascending cumulative average METS-IR quartiles (median values: Q1 = 24.80, Q2 = 29.33, Q3 = 33.79, Q4 = 40.67), where event rates progressively rose from 110 cases (8.4%) in Q1 to 116 cases (8.8%) in Q2, 160 cases (12.2%) in Q3, and 168 cases (12.8%) in Q4. This pattern demonstrates a significant dose-response relationship between elevated cumulative average METS-IR levels and increased CVD risk (Fig. 2; Table 2). Incidence rates of CVD categorized by quartiles of the cumulative average METS-IR. Q1, Quartile 1; Q2, Quartile 2; Q3, Quartile 3; Q4, Quartile4; METS-IR, metabolic score for insulin resistance Table 2Relationship between the cumulative average METS-IR and CVD developmentCumulative average METS-IRQuartilesP for trendContinuousPer 1 SD increaseQuartile 1Quartile 2Quartile 3Quartile 4Median24.8029.3333.7940.67––Cases, n (%)110 (8.4%)116 (8.8%)160 (12.2%)168 (12.8%)––Crude, OR (95% CI)Reference1.060 (0.807–1.393)1.517 (1.177–1.964)1.604 (1.247–2.072)< 0.0011.221 (1.121–1.330)Model 1, OR (95% CI)Reference1.127 (0.855–1.485)1.684 (1.299–2.191)1.810 (1.396–2.356)< 0.0011.275 (1.169–1.391)Model 2, OR (95% CI)Reference1.106 (0.837–1.461)1.599 (1.229–2.088)1.695 (1.302–2.215)< 0.0011.251 (1.144–1.367)Model 3, OR (95% CI)Reference1.037 (0.782–1.375)1.395 (1.062–1.838)1.366 (1.027–1.823)0.0031.161 (1.050–1.282)Model 1, adjusted for gender and age; Model 2, adjusted for gender, age, residence, education level, marital status, smoking status, drinking status, kidney disease and liver disease; Model 3, adjusted for variables included in Model 2 and SBP, DBP, hypertension, diabetes, HbA1c, TC, LDL-c, WBC, Hb, MCV, PLT, CRP, UA, Scr and BUN; METS-IR, metabolic score for insulin resistance, OR, odds ratio; CI, confidence interval; SD, standard deviation Relationship between the cumulative average METS-IR and CVD development Model 1, adjusted for gender and age; Model 2, adjusted for gender, age, residence, education level, marital status, smoking status, drinking status, kidney disease and liver disease; Model 3, adjusted for variables included in Model 2 and SBP, DBP, hypertension, diabetes, HbA1c, TC, LDL-c, WBC, Hb, MCV, PLT, CRP, UA, Scr and BUN; METS-IR, metabolic score for insulin resistance, OR, odds ratio; CI, confidence interval; SD, standard deviation Fully adjusted logistic regression models revealed a significant dose-response relationship between elevated cumulative average METS-IR and incident CVD risk. In the comprehensively adjusted Model 3, compared with the lowest quartile (Q1), the CVD risk increased by 39.5% (OR = 1.395, 95% CI: 1.062–1.838) in Q3 and 36.6% (OR = 1.366, 95% CI: 1.027–1.823) in Q4, with a statistically significant trend (P = 0.003). When analyzed as a continuous variable, each 1-SD increase in cumulative average METS-IR was associated with a 16.1% higher CVD risk (OR = 1.161, 95% CI: 1.050–1.282) (Table 2). This association demonstrated greater magnitude for heart disease subtypes, where Q4 exhibited 50.9% increased risk versus Q1 (OR = 1.509, 95% CI: 1.118–2.046) and each 1-SD increment conferred 19.8% higher risk (OR = 1.198, 95% CI: 1.079–1.331), with a strong trend significance (P < 0.001) (Additional file 1: Table S3). Notably, while stroke risk showed marginal associations in intermediate models (Models 1–2), fully adjusted analyses (Model 3) revealed non-significant OR across all quartiles (Q4 OR = 0.902, 95% CI: 0.440–1.865; P for trend = 0.917) and non-significant continuous associations (OR = 0.987, 95% CI: 0.769–1.260) (Additional file 1: Table S3), indicating a disease-specific pattern of METS-IR’s influence on CVD risk. Fully adjusted RCS regression confirmed a significant linear dose-response relationship between cumulative average METS-IR and CVD risk (P for overall = 0.033, P for nonlinear = 0.956), with CVD risk progressively increasing across higher cumulative average METS-IR levels (Fig. 3). This linear pattern was more pronounced for heart disease subtypes (P for overall = 0.007; P for nonlinear = 0.737), exhibiting accelerated risk escalation in higher cumulative average METS-IR ranges (Fig. 4). Notably, no significant association emerged between cumulative average METS-IR and stroke risk (P for overall P = 0.890; P for nonlinear = 0.737), as visualized in Fig. 5. Association between the cumulative average METS-IR and incident CVD. The model was adjusted for gender , age, residence, education level, marital status, smoking status, drinking status, kidney disease, liver disease, SBP, DBP, hypertension, diabetes, HbA1c, TC, LDL-c, WBC, Hb, MCV, PLT, CRP, UA, Scr and BUN. SBP, systolic blood pressure; DBP, diastolic blood pressure; HbA1c, glycosylated hemoglobin A1c; TC, total cholesterol; LDL-c, low-density lipoprotein cholesterol; WBC, white blood cell; Hb, hemoglobin; MCV, mean corpuscular volume; PLT, platelet; CRP, c-reactive protein; UA, uric acid; Scr, serum creatinine; BUN, blood urea nitrogen; METS-IR, metabolic score for insulin resistance, OR, odds ratio; CI, confidence interval Association between the cumulative average METS-IR and incident heart condition. The model was adjusted for gender , age, residence, education level, marital status, smoking status, drinking status, kidney disease, liver disease, SBP, DBP, hypertension, diabetes, HbA1c, TC, LDL-c, WBC, Hb, MCV, PLT, CRP, UA, Scr and BUN. SBP, systolic blood pressure; DBP, diastolic blood pressure; HbA1c, glycosylated hemoglobin A1c; TC, total cholesterol; LDL-c, low-density lipoprotein cholesterol; WBC, white blood cell; Hb, hemoglobin; MCV, mean corpuscular volume; PLT, platelet; CRP, c-reactive protein; UA, uric acid; Scr, serum creatinine; BUN, blood urea nitrogen; METS-IR, metabolic score for insulin resistance, OR, odds ratio; CI, confidence interval Association between the cumulative average METS-IR and incident stroke. The model was adjusted for gender , age, residence, education level, marital status, smoking status, drinking status, kidney disease, liver disease, SBP, DBP, hypertension, diabetes, HbA1c, TC, LDL-c, WBC, Hb, MCV, PLT, CRP, UA, Scr and BUN. SBP, systolic blood pressure; DBP, diastolic blood pressure; HbA1c, glycosylated hemoglobin A1c; TC, total cholesterol; LDL-c, low-density lipoprotein cholesterol; WBC, white blood cell; Hb, hemoglobin; MCV, mean corpuscular volume; PLT, platelet; CRP, c-reactive protein; UA, uric acid; Scr, serum creatinine; BUN, blood urea nitrogen; METS-IR, metabolic score for insulin resistance, OR, odds ratio; CI, confidence interval Subgroup analysis To illustrate potential modification of the cumulative average METS-IR and CVD association, we conducted systematic subgroup analyses. No significant effect modification was observed for age stratification (P for interaction = 0.441), residence (P for interaction = 0.709), liver disease (P for interaction = 0.968), or kidney disease (P for interaction = 0.849). However, several clinically relevant differences emerged: Males exhibited 16% greater risk elevation per SD increase (OR = 1.24, 95%CI: 1.08–1.42) compared to females (OR = 1.08, 95%CI: 0.96–1.20). Normotensive individuals demonstrated 17% amplified risk increase (OR = 1.23, 95%CI: 1.09–1.39) versus hypertensive participants (OR = 1.06, 95%CI: 0.94–1.19). Current drinkers showed 23% additional risk elevation (OR = 1.31, 95%CI: 1.10–1.55) relative to non-drinkers (OR = 1.08, 95%CI: 0.98–1.20), while current smokers exhibited 15% greater risk increase (OR = 1.25, 95%CI: 1.06–1.48) compared to non-smokers (OR = 1.10, 95%CI: 0.99–1.22). Collectively, these patterns indicate that the detrimental impact of cumulative average METS-IR on CVD risk is differentially accentuated among middle-aged and older Chinese males, individuals without underlying metabolic disorders, and those with tobacco or alcohol exposure, as detailed in Fig. 6. Subgroup analyses of the association between the cumulative average METS-IR and CVD incidence. The model was adjusted for gender , age, residence, education level, marital status, smoking status, drinking status, kidney disease, liver disease, SBP, DBP, hypertension, diabetes, HbA1c, TC, LDL-c, WBC, Hb, MCV, PLT, CRP, UA, Scr and BUN (excluding the variable for subgroup stratification). SBP, systolic blood pressure; DBP, diastolic blood pressure; HbA1c, glycosylated hemoglobin A1c; TC, total cholesterol; LDL-c, low-density lipoprotein cholesterol; WBC, white blood cell; Hb, hemoglobin; MCV, mean corpuscular volume; PLT, platelet; CRP, c-reactive protein; UA, uric acid; Scr, serum creatinine; BUN, blood urea nitrogen; OR, odds ratio;CI, confidence interval Sensitivity analysis Robustness of the cumulative average METS-IR and CVD association was consistently demonstrated through three complementary sensitivity approaches: Complete-case analysis excluding participants with missing covariates revealed 49.0% elevated CVD risk in the Q4 (OR = 1.490, 95%CI: 1.104–2.020) and 12.5% increased risk per 1-SD increment (OR = 1.125, 95%CI: 1.081–1.177) (Additional file 1: Table S5). After excluding individuals with incident CVD during the 2013 follow-up (Wave 2), significant associations persisted with 46.8% higher risk in Q4 (OR = 1.468, 95%CI: 1.022–2.122) and 14.6% risk elevation per SD (OR = 1.146, 95%CI: 1.011–1.298) (Additional file 1: Table S6). Propensity score matching (1:1 nearest-neighbor, caliper = 0.01) generated 1,705 balanced pairs between Q1-Q2 and Q3-Q4 groups, confirming 36.5% greater CVD risk in higher METS-IR groups (OR = 1.365, 95%CI: 1.094–1.707) and 19.5% increased risk per SD (OR = 1.195, 95%CI: 1.073–1.328) (Fig. 6, Additional file 1: Table S7 and S8). These concordant findings substantiate the causal robustness of cumulative average METS-IR as a predictor of incident CVD risk. Discussion This prospective nationwide cohort study, encompassing 5,264 middle-aged and older adults (≥ 45 years) from the China Health and Retirement Longitudinal Study (CHARLS), provides the first evidence that cumulative insulin resistance burden—quantified by cumulative average METS-IR—serves as an independent predictor of incident cardiovascular disease. Our primary analysis demonstrated that individuals in the highest cumulative average METS-IR quartile (Q4) had 36.6% significantly elevated CVD risk after comprehensive adjustment for sociodemographic characteristics, lifestyle factors, and traditional cardiovascular risk factors (OR = 1.366, 95% CI: 1.027–1.823) during the 4-year follow-up period. Crucially, this association remained robust in propensity score-matched analyses balancing multiple potential confounders (OR = 1.365, 95% CI: 1.094–1.707), exhibiting a clear dose-response relationship whereby each 1-SD increase in cumulative average METS-IR corresponded to 19.5% higher CVD risk (OR = 1.195, 95% CI: 1.073–1.328). These findings not only enhance understanding of insulin resistance’s evolving role in cardiovascular pathophysiology but also establish cumulative average METS-IR as a novel biomarker for clinical risk stratification. The observed associations align mechanistically with the multidimensional pathophysiology of insulin resistance. Elevated cumulative average METS-IR represents persistent insulin signaling impairment that collectively fosters an atherogenic microenvironment through concomitant dysfunction of endothelial cells, cardiomyocytes, and vascular smooth muscle [4, 7, 23, 24]. diminished nitric oxide (NO) bioavailability impairs vasodilation [25], increased endothelial permeability accelerates lipid infiltration [26], and upregulated leukocyte adhesion molecules propagate chronic inflammation [27, 28]. Concurrently, activated adipose tissue lipolysis drives pathological free fatty acid flux to the liver, exacerbating atherogenic dyslipidemia [29, 30]—a metabolic pattern corroborated by our baseline data showing significantly elevated TG and reduced HDL-c in higher cumulative average METS-IR quartiles. Furthermore, compensatory hyperinsulinemia in insulin-resistant states overactivates the renin-angiotensin-aldosterone system (RAAS), enhancing renal tubular sodium reabsorption [31, 32], thereby explaining the robust association between cumulative average METS-IR and hypertension prevalence. Critically, our cumulative exposure model captures these sustained pathophysiological processes more accurately than cross-sectional approaches with single-timepoint measurements, enabling detection of associations independent of traditional risk factors. This study represents a breakthrough in the field of insulin resistance surrogate markers. Although previous studies have demonstrated the predictive value of surrogate markers such as the TyG [33], triglyceride glucose-body mass index (TyG-BMI) [34], triglyceride glucose-waist height ratio index (TyG-WHtR) [35], and the triglyceride-to-high-density lipoprotein cholesterol ratio (TG/HDL-c) [36] for CVD incidence or cardiovascular mortality [37, 38], direct comparisons reveal that the cumulative average METS-IR has unique clinical advantages. Its calculation formula, optimized by nonlinear averaging transformation, enhances sensitivity to metabolic dysregulation and is particularly adept at capturing the dynamic changes of insulin resistance over the long term. A more enlightening finding in this study is the stark contrast between the strong association of cumulative average METS-IR with heart disease risk (OR = 1.509, 95% CI: 1.118–2.046) and the null association with stroke (OR = 0.902, 95% CI: 0.440–1.865), suggesting that insulin resistance may preferentially target coronary microcirculation and myocardial energy metabolism rather than cerebrovascular pathways. This separation of physiological effects may stem from the multi-layered organ-specific mechanisms of insulin resistance. In the energy metabolism, cardiomyocytes rely on insulin-sensitive glucose transporter 4 (GLUT4) for glucose uptake [39]. In states of insulin resistance, the shift toward fatty acid oxidation generates lipotoxic byproducts (such as diacylglycerol and ceramides) that impair mitochondrial function [40] and cause myocardial energy metabolism and contractility dysfunction [41]. In contrast, brain tissue, dominated by glucose transporter 1/3 (GLUT1/3) [42], can compensate with ketone utilization and has lower insulin signaling dependence [43], thus maintaining more stable energy supply and delaying corresponding pathology. In intracellular signaling pathways, the myocardial energy metabolism pathway IRS-PI3K-Akt signaling pathway (where Akt, also termed PKB, is a serine/threonine kinase) is vulnerable to inhibition by inflammatory cytokines such as tumor necrosis factor alpha (TNF-α), interleukin-6 (IL-6), and interleukin-8 (IL-8). leading to serine phosphorylation that blocks GLUT4 translocation [44, 45]. Meanwhile, peroxisome proliferator-activated receptor alpha (PPARα) activation through fatty acid oxidation further inhibits glucose utilization [46]. In the brain, insulin receptors are widely distributed, with compensatory mechanisms such as insulin-like growth factor receptor 1 (IGF-1R) cross-activation [47], mammalian target of rapamycin 2 (mTORC2) maintained basal glucose uptake [48] and insulin-independent GLUT3 dominance [49] that buffer the impact of insulin resistance. In oxidative stress and inflammatory response, insulin resistance increases myocardial fatty acid oxidation, leading to excessive reactive oxygen species (ROS) generation [50]. Cardiac antioxidant enzymes such as glutamate-cysteine ligase (GCL), glutathione peroxidase (GPx), and superoxide dismutase (SOD) have lower activity compared to brain tissue [51]. In contrast, the brain’s nuclear factor erythroid 2-related factor 2/heme oxygenase 1 (Nrf2/HO-1) pathway effectively clears ROS [52], microglia control neuroinflammation through interleukin-10 (IL-10) and TNF-α [53], and the blood-brain barrier restricts the entry of peripheral inflammatory factors in the early stages of disease, thereby effectively reducing neural damage [54]. In vascular function and perfusion, insulin resistance reduces endothelial NO synthesis, impairs coronary dilation [4], and causes capillary rarefaction, increasing myocardial ischemia-reperfusion injury susceptibility [41, 55]. In contrast, the brain maintains perfusion through neurogenic regulation such as NO and calcitonin gene-related peptide (CGRP) release and collateral circulation (Circle of Willis) [56–58]. During insulin resistance, brain vascular endothelium retains endothelial nitric oxide synthase (eNOS) phosphorylation [59], and astrocytes protect blood-brain barrier integrity through vascular endothelial growth factor (VEGF) paracrine signaling [60], reducing neurological metabolic damage. Furthermore, the divergent association between cumulative METS-IR and heart disease versus stroke may also be attributed to fundamental differences in the primary pathophysiology underlying these conditions. Myocardial infarction is predominantly an atherothrombotic event, a process intrinsically linked to endothelial dysfunction and advanced atherosclerosis [61], both of which are strongly potentiated by the pro-inflammatory and pro-oxidant state of chronic insulin resistance. In contrast, stroke encompasses a more heterogeneous set of etiologies. A significant proportion of ischemic strokes are cardioembolic, often secondary to atrial fibrillation—a mechanism less directly driven by insulin resistance [62]. Similarly, hemorrhagic strokes frequently result from the rupture of vascular anomalies like aneurysms or hypertensive vasculopathy [63], pathways where the role of insulin resistance is less pronounced. The aggregation of these diverse stroke subtypes in our analysis, while necessary for statistical power, likely diluted any specific signal related to insulin resistance, potentially contributing to the observed null association. These mechanisms collectively attenuate the pathological impact of insulin resistance on the cerebrovascular system. Despite the significant association between cumulative average METS-IR and CVD, the lack of statistical difference in diabetes prevalence between cases and controls (16.6% vs. 15.6%, P = 0.564) indicates that cardiovascular damage is primarily driven through non-glycemic pathways, including inflammation activation, oxidative stress, and myocardial energy metabolism disruption [64–66]. This study leveraged a nationwide cohort established by the CHARLS. The high-quality medical data and extended follow-up period (mean 4 years) provide a robust empirical foundation for the conclusions. Our findings extend the existing literature on insulin resistance in several significant aspects. While prior studies have predominantly established cross-sectional associations between single-timepoint METS-IR and metabolic derangements, our study is the first to demonstrate that the cumulative burden of insulin resistance, quantified by repeated METS-IR measurements, is an independent predictor of incident cardiovascular disease in a general population of middle-aged and older Chinese adults. This longitudinal assessment captures the persistent nature of metabolic dyshomeostasis more accurately than a single measurement, thereby offering superior risk stratification. More importantly, we unveil a striking organ-specific susceptibility, revealing a strong association with heart disease but not with stroke. This critical distinction moves beyond the conventional paradigm of insulin resistance as a uniform risk factor for all cardiovascular diseases and underscores the necessity for developing targeted, organ-specific preventive strategies. Consequently, the cumulative average METS-IR emerges not merely as a novel biomarker but as a practical and refined tool for enhancing early prevention paradigms in clinical practice, particularly for identified high-risk subgroups. Several limitations of this study must be acknowledged. First, although extensive covariates were adjusted for, the potential influence of unmeasured confounding factors (such as dietary patterns, life stress, and sleep quality) on the true association cannot be excluded. Second, the availability of blood biomarker data at only two time points (2011 and 2015) limits our ability to explore more complex, non-linear longitudinal metabolic trajectories. Third, the ascertainment of CVD outcomes relied solely on self-reported physician diagnoses, consistent with methodologies in similar cohort studies, yet lacking more definitive diagnostic evidence, such as angiography, this introduces a potential for outcome misclassification bias, a limitation inherent in many large-scale epidemiological cohorts. Fourth, the generalizability of our findings may be limited, as the METS-IR index was developed in Chinese cohorts. Its predictive performance for other ethnic groups may vary due to genetic and lifestyle differences, necessitating external validation. Finally, the observational nature of this study design precludes the establishment of causality; future Mendelian randomization studies and intervention trials are needed to elucidate the causal pathways linking genetic susceptibility to insulin resistance with cardiovascular outcomes. Conclusions The cumulative average METS-IR transcends traditional, cross-sectional risk assessment by capturing the dynamic and persistent nature of insulin resistance. Our study advocates for the integration of longitudinal METS-IR tracking into clinical practice to enhance early risk stratification and to guide tailored intervention strategies for primary cardiovascular disease prevention, especially in the identified high-risk populations. Future research should focus on validating these findings in diverse ethnic cohorts and elucidating the underlying mechanisms driving the observed organ-specific risk patterns. Supplementary Information Supplementary Material 1. Acknowledgements This investigation employs data sourced from the CHARLS. The researchers express appreciation to the CHARLS investigators for their dedicated work and to all study participants for their valuable contributions. Additional study details are accessible at https://g2aging.org/. Abbreviations Insulin resistance Cardiovascular disease Metabolic score for insulin resistance China health and retirement longitudinal study Fasting blood glucose Triglycerides Body mass index High-density lipoprotein cholesterol Restricted cubic spline Standard deviation Odds ratio Confidence interval Hyperinsulinemic-euglycemic clamp Triglyceride-glucose Non-alcoholic fatty liver disease Lipid-lowering therapy Antihypertensive therapy Glucose-lowering therapy Glycosylated hemoglobin Total cholesterol Low-density lipoprotein cholesterol White blood cell count Hemoglobin Mean corpuscular volume Platelet count C-reactive protein Uric acid Serum creatinine Blood urea nitrogen Quartile 1 Quartile 2 Quartile 3 Quartile 4 Variance inflation factor Standardized mean difference Renin-angiotensin-aldosterone system Triglyceride glucose-waist height ratio index Glucose transporter 4 Glucose transporter 1 Glucose transporter 3 Tumor necrosis factor alpha Interleukin-6 Interleukin-8 Interleukin-10 Peroxisome proliferator-activated receptor alpha Insulin-like growth factor receptor 1 Mammalian target of rapamycin 2 Reactive oxygen species Glutamate-cysteine ligase Glutathione peroxidase Superoxide dismutase Calcitonin gene-related peptide Endothelial nitric oxide synthase Vascular endothelial growth factor Author contributions YC, GZ and MMQ were jointly responsible for the conception and design of the study. YC, JSY and ZSY performed data extraction and collation. YC, XL and XLF undertook the statistical analysis and interpretation of the results. YC drafted the manuscript and completed the literature search. GZ and LXH assisted with manuscript revisions. LXH oversaw the research and provided critical review. All authors reviewed the manuscript and approved the final submitted version. Funding No funding. Data availability The datasets generated and analysed during the current study are available in the China Health and Retirement Longitudinal Study repository [http://charls.pku.edu.cn]. Declarations Ethics approval and consent to participate The study was approved by the ethics review committee (institutional review board) of Peking University. All participants granted formal written consent for their involvement. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The datasets generated and analysed during the current study are available in the China Health and Retirement Longitudinal Study repository [http://charls.pku.edu.cn].