Table 1
Baseline characteristics of the study population according to CVD.
| VARIABLES | WITHOUT CVD (N = 2658) | WITH CVD (N = 874) | p |
|---|---|---|---|
| HbA1c (mean, SD) | 5.39 (0.89) | 5.48 (0.95) | 0.017 |
| FBG (mean, SD) | 120.24 (37.68) | 122.88 (41.46) | 0.08 |
| UA (mean, SD) | 4.47 (1.24) | 4.41 (1.27) | 0.196 |
| Creatinine (mean, SD) | 0.78 (0.19) | 0.77 (0.20) | 0.686 |
| TG (mean, SD) | 136.78 (88.28) | 143.45 (86.23) | 0.051 |
| TC (mean, SD) | 197.70 (38.98) | 199.52 (38.78) | 0.229 |
| HDL (mean, SD) | 51.31 (15.40) | 49.84 (15.15) | 0.014 |
| CRP (mean, SD) | 2.70 (7.69) | 2.93 (7.67) | 0.438 |
| WC (mean, SD) | 85.02 (11.15) | 87.66 (11.90) | <0.001 |
| Weight (mean, SD) | 59.01 (10.97) | 60.74 (12.06) | <0.001 |
| Height (mean, SD) | 1.58 (0.08) | 1.57 (0.09) | 0.072 |
| BUN (mean, SD) | 16.00 (4.56) | 15.64 (4.42) | 0.041 |
| LDL (mean, SD) | 119.24 (34.92) | 120.93 (36.74) | 0.22 |
| Gender (%) | |||
| Female | 1407 (52.9) | 523 (59.8) | <0.001 |
| Male | 1251 (47.1) | 351 (40.2) | |
| Age (mean, SD) | 58.32 (8.55) | 60.74 (8.59) | <0.001 |
| BMI (mean, SD) | 23.63 (3.70) | 24.48 (3.99) | <0.001 |
| SBP (mean, SD) | 129.08 (20.01) | 134.15 (21.56) | <0.001 |
| DBP (mean, SD) | 75.21 (11.39) | 77.26 (12.29) | <0.001 |
| Marital status (%) | |||
| Unmarried | 366 (13.8) | 152 (17.4) | 0.01 |
| Married | 2292 (86.2) | 722 (82.6) | |
| Smoking status (%) | |||
| never smoker | 1640 (61.7) | 571 (65.3) | <0.001 |
| former smoker | 201 (7.6) | 96 (11.0) | |
| current smoker | 817 (30.7) | 207 (23.7) | |
| Drinking status (%) | |||
| never drinker | 1590 (59.8) | 544 (62.2) | 0.01 |
| former drinker | 208 (7.8) | 88 (10.1) | |
| current drinker | 860 (32.4) | 242 (27.7) | |
| Education level (%) | |||
| elementary school or below | 1891 (71.1) | 624 (71.4) | 0.049 |
| middle school | 708 (26.6) | 218 (24.9) | |
| college or above | 59 (2.2) | 32 (3.7) | |
| Rural (%) | |||
| No | 398 (15.0) | 168 (19.2) | 0.004 |
| Yes | 2260 (85.0) | 706 (80.8) | |
| Hypertension (%) | |||
| No | 1647 (62.0) | 394 (45.1) | <0.001 |
| Yes | 1011 (38.0) | 480 (54.9) | |
| Dyslipidemia (%) | |||
| No | 2485 (93.5) | 723 (82.7) | <0.001 |
| Yes | 173 (6.5) | 151 (17.3) | |
| Kidney disease (%) | |||
| No | 2548 (95.9) | 822 (94.1) | 0.033 |
| Yes | 110 (4.1) | 52 (5.9) | |
| Liver disease (%) | |||
| No | 2589 (97.4) | 843 (96.5) | 0.176 |
| Yes | 69 (2.6) | 31 (3.5) | |
| Antidiabetic drugs (%) | |||
| No | 2549 (95.9) | 804 (92.0) | <0.001 |
| Yes | 109 (4.1) | 70 (8.0) | |
| Lipid-lowering agents (%) | |||
| No | 2570 (96.7) | 800 (91.5) | <0.001 |
| Yes | 88 (3.3) | 74 (8.5) | |
| Antihypertensive drugs (%) | |||
| No | 2263 (85.1) | 604 (69.1) | <0.001 |
| Yes | 395 (14.9) | 270 (30.9) | |
| eGDR (mean, SD) | 9.24 (2.20) | 8.38 (2.34) | <0.001 |
| CVAI (mean, SD) | 96.63 (38.68) | 109.43 (40.66) | <0.001 |
| TyG (mean, SD) | 8.82 (0.63) | 8.90 (0.65) | 0.002 |
| TyG-BMI (mean, SD) | 209.09 (39.27) | 218.42 (42.11) | <0.001 |
| TyG-WC (mean, SD) | 7.52 (1.24) | 7.81 (1.31) | <0.001 |
| TyG-WHtR (mean, SD) | 4.77 (0.80) | 4.98 (0.84) | <0.001 |
| METS-IR (mean, SD) | 36.19 (7.75) | 37.99 (8.36) | <0.001 |
| AIP (mean, SD) | 0.86 (0.76) | 0.95 (0.75) | 0.002 |
| TyHGB (mean, SD) | 10.23 (3.49) | 10.69 (3.80) | 0.001 |
| CTI (mean, SD) | 8.89 (0.82) | 9.00 (0.83) | 0.001 |
Figure 1
Receiver operating characteristic (ROC) curves comparing the predictive performance of ten insulin resistance surrogate indices for cardiovascular events. (A) ROC curves for predicting total CVD events. (B) ROC curves for predicting heart disease. (C) ROC curves for predicting stroke. Among all ten insulin resistance surrogate indices evaluated, eGDR consistently showed the highest discriminative ability across all cardiovascular outcomes, while CVAI demonstrated the second-best performance for total CVD and heart disease prediction.
Table 2
Associations of ten insulin resistance surrogate indexes with new-onset cardiovascular disease incidence in patients with prediabetes or diabetes.
| MODEL I OR (95%CI) | P-VALUE | MODEL II OR (95%CI) | P-VALUE | MODEL III OR (95%CI) | P-VALUE | MODEL IV OR (95%CI) | P-VALUE | |
|---|---|---|---|---|---|---|---|---|
| eGDR (per 1 SD) | 0.682(0.631–0.737) | <0.001 | 0.705(0.651–0.762) | <0.001 | 0.809(0.686–0.952) | 0.0112 | 0.822(0.696–0.969) | 0.021 |
| eGDR quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 0.636(0.519–0.779) | <0.001 | 0.646(0.526–0.792) | <0.001 | 0.767(0.604–0.972) | 0.0289 | 0.776(0.610–0.985) | 0.0378 |
| Q3 | 0.387(0.310–0.482) | <0.001 | 0.426(0.340–0.532) | <0.001 | 0.502(0.338–0.747) | 0.0006 | 0.519(0.349–0.776) | 0.0013 |
| Q4 | 0.389(0.312–0.484) | <0.001 | 0.418(0.335–0.521) | <0.001 | 0.505(0.339–0.753) | 0.0007 | 0.527(0.353–0.789) | 0.0018 |
| P for trend | <0.001 | <0.001 | 0.0031 | 0.0064 | ||||
| CVAI (per 1 SD) | 1.383(1.281–1.495) | <0.001 | 1.324(1.223–1.435) | <0.001 | 1.137(1.042–1.241) | 0.0038 | 1.124(1.028–1.229) | 0.0099 |
| CVAI quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.33(1.057–1.684) | 0.0155 | 1.225(0.968–1.552) | 0.0915 | 1.157(0.910–1.472) | 0.234 | 1.134(0.891–1.444) | 0.3062 |
| Q3 | 1.465(1.165–1.847) | 0.002 | 1.298(1.026–1.643) | 0.0296 | 1.086(0.851–1.387) | 0.509 | 1.042(0.815–1.335) | 0.7415 |
| Q4 | 2.343(1.881–2.927) | <0.001 | 2.047(1.635–2.569) | <0.001 | 1.383(1.083–1.769) | 0.0095 | 1.331(1.038–1.709) | 0.0243 |
| P for trend | <0.001 | <0.001 | 0.0196 | 0.0483 | ||||
| TyHGB (per 1 SD) | 1.130(1.050–1.215) | 0.0009 | 1.153(1.070–1.240) | 0.0002 | 1.044(0.962–1.130) | 0.2981 | 1.049(0.955–1.151) | 0.3169 |
| TyHGB quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.324(1.061–1.653) | 0.0133 | 1.330(1.064–1.666) | 0.0126 | 1.234(0.982–1.553) | 0.0721 | 1.203(0.955–1.518) | 0.1171 |
| Q3 | 1.231(0.984–1.541) | 0.0686 | 1.261(1.005–1.583) | 0.0452 | 1.065(0.842–1.347) | 0.5993 | 1.027(0.810–1.304) | 0.8214 |
| Q4 | 1.502(1.207–1.872) | 0.0003 | 1.556(1.247–1.945) | <0.001 | 1.144(0.903–1.449) | 0.2658 | 1.100(0.859–1.410) | 0.4484 |
| P for trend | 0.0012 | 0.0004 | 0.529 | 0.753 | ||||
| TyG (per 1 SD) | 1.125(1.043–1.213) | 0.0023 | 1.131(1.046–1.222) | 0.00189 | 1.031(0.949–1.119) | 0.4689 | 1.029(0.933–1.134) | 0.5672 |
| TyG quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.228(0.983–1.534) | 0.0706 | 1.215(0.971–1.522) | 0.0888 | 1.130(0.898–1.422) | 0.2983 | 1.107(0.878–1.396) | 0.389 |
| Q3 | 1.303(1.046–1.626) | 0.0187 | 1.269(1.014–1.588) | 0.0375 | 1.114(0.885–1.404) | 0.3576 | 1.085(0.858–1.373) | 0.4956 |
| Q4 | 1.381(1.110–1.721) | 0.0039 | 1.397(1.119–1.746) | 0.0033 | 1.091(0.864–1.378) | 0.4627 | 1.053(0.812–1.364) | 0.6987 |
| P for trend | 0.0038 | 0.00378 | 0.5271 | 0.7182 | ||||
| TyG-BMI (per 1 SD) | 1.254(1.163–1.352) | <0.001 | 1.305(1.208–1.411) | <0.001 | 1.125(1.033–1.225) | 0.0067 | 1.119(1.024–1.222) | 0.0127 |
| TyG-BMI quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.021(0.812–1.283) | 0.861 | 1.100(0.872–1.389) | 0.4195 | 1.009(0.795–1.279) | 0.9434 | 1.002(0.789–1.273) | 0.9849 |
| Q3 | 1.337(1.072–1.668) | 0.01 | 1.440(1.149–1.807) | 0.0016 | 1.156(0.913–1.465) | 0.2277 | 1.133(0.892–1.441) | 0.3056 |
| Q4 | 1.662(1.340–2.065) | <0.001 | 1.901(1.518–2.384) | <0.001 | 1.269(0.993–1.624) | 0.0566 | 1.240(0.963–1.598) | 0.0959 |
| P for trend | <0.001 | <0.001 | 0.0301 | 0.0606 | ||||
| TyG-WC (per 1 SD) | 1.274(1.178–1.379) | <0.001 | 1.287(1.189–1.395) | <0.001 | 1.113(1.023–1.213) | 0.0135 | 1.108(1.014–1.212) | 0.0246 |
| TyG-WC quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.113(0.887–1.398) | 0.3534 | 1.114(0.885–1.401) | 0.3586 | 1.057(0.836–1.335) | 0.6418 | 1.043(0.824–1.319) | 0.7281 |
| Q3 | 1.233(0.985–1.544) | 0.0676 | 1.251(0.997–1.571) | 0.0534 | 1.039(0.821–1.316) | 0.7481 | 1.005(0.792–1.277) | 0.9649 |
| Q4 | 1.829(1.475–2.273) | <0.001 | 1.851(1.488–2.306) | <0.001 | 1.279(1.009–1.623) | 0.042 | 1.247(0.976–1.596) | 0.078 |
| P for trend | <0.001 | <0.001 | 0.0557 | 0.1131 | ||||
| TyG-WHtR (per 1 SD) | 1.298(1.199–1.404) | <0.001 | 1.257(1.159–1.364) | <0.001 | 1.093(1.002–1.193) | 0.0443 | 1.087(0.993–1.191) | 0.0723 |
| TyG-WHtR quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.246(0.990–1.570) | 0.0611 | 1.228(0.973–1.551) | 0.0846 | 1.117(0.881–1.419) | 0.3601 | 1.102(0.867–1.402) | 0.4252 |
| Q3 | 1.404(1.119–1.763) | 0.0034 | 1.384(1.098–1.746) | 0.0059 | 1.131(0.889–1.439) | 0.3148 | 1.102(0.865–1.406) | 0.4323 |
| Q4 | 2.049(1.647–2.555) | <0.001 | 1.904(1.515–2.396) | <0.001 | 1.313(1.026–1.682) | 0.0306 | 1.285(0.996–1.663) | 0.0544 |
| P for trend | <0.001 | <0.001 | 0.0372 | 0.0696 | ||||
| MetS-IR (per 1 SD) | 1.246(1.156–1.343) | <0.001 | 1.304(1.207–1.409)) | <0.001 | 1.124(1.033–1.223) | 0.0068 | 1.112(1.020–1.213) | 0.0158 |
| MetS-IR quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.034(0.824–1.298) | 0.772 | 1.089(0.865–1.371) | 0.4675 | 1.027(0.813–1.298) | 0.8211 | 1.007(0.795–1.275) | 0.955 |
| Q3 | 1.250(1.001–1.561) | 0.0486 | 1.367(1.091–1.717) | 0.0067 | 1.130(0.893–1.431) | 0.3087 | 1.097(0.864–1.394) | 0.4459 |
| Q4 | 1.665(1.343–2.068) | <0.001 | 1.879(1.505–2.349) | <0.001 | 1.278(1.005–1.628) | 0.0456 | 1.233(0.965–1.577) | 0.0936 |
| P for trend | <0.001 | <0.001 | 0.0334 | 0.0728 | ||||
| AIP (per 1 SD) | 1.126(1.043–1.215) | 0.0023 | 1.144(1.059–1.238) | 0.0007 | 1.046(0.964–1.136) | 0.2774 | 1.038(0.949–1.136) | 0.4106 |
| AIP quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.131(0.904–1.414) | 0.2798 | 1.139(0.909–1.427) | 0.2579 | 1.097(0.872–1.382) | 0.4296 | 1.082(0.856–1.367) | 0.5095 |
| Q3 | 1.416(1.140–1.762) | 0.0017 | 1.433(1.150–1.788) | 0.0014 | 1.235(0.984–1.553) | 0.0684 | 1.200(0.952–1.514) | 0.1235 |
| Q4 | 1.270(1.019–1.583) | 0.0336 | 1.327(1.062–1.661) | 0.013 | 1.054(0.834–1.331) | 0.6595 | 1.001(0.783–1.279) | 0.9937 |
| P for trend | 0.0081 | 0.0028 | 0.4782 | 0.7089 | ||||
| CTI (per 1 SD) | 1.144(1.060–1.234) | 0.0005 | 1.133(1.049–1.223) | 0.0015 | 1.016(0.936–1.103) | 0.7001 | 1.004(0.917–1.099) | 0.9281 |
| CTI quartile | ||||||||
| Q1 | 1(Reference) | 1(Reference) | 1(Reference) | 1(Reference) | ||||
| Q2 | 1.221(0.977–1.528) | 0.0785 | 1.155(0.922–1.447) | 0.2109 | 1.079(0.857–1.359) | 0.5175 | 1.057(0.838–1.333) | 0.6409 |
| Q3 | 1.290(1.034–1.611) | 0.0243 | 1.243(0.994–1.556) | 0.0567 | 1.031(0.817–1.300) | 0.7979 | 0.999(0.789–1.265) | 0.9978 |
| Q4 | 1.474(1.184–1.835) | 0.0005 | 1.416(1.135–1.767) | 0.002 | 1.079(0.855–1.361) | 0.5231 | 1.037(0.808–1.332) | 0.7741 |
| P for trend | 0.0006 | 0.0017 | 0.6352 | 0.9036 |
[i] Model I was unadjusted; Model II included adjustments for gender and age; and Model III contained additional adjustments for marital status, smoking status, drinking status, education level, residence, hypertension, dyslipidemia, kidney disease, liver disease, antidiabetic drugs, lipid-lowering agents, antihypertensive drugs. Model IV further adjusted for TC, LDL, BUN, Creatinine, SBP, and DBP.
Figure 2
Subgroup analyses of the associations between insulin resistance indices and incident cardiovascular disease. (A) Forest plot showing the association between Chinese Visceral Adiposity Index (CVAI) and CVD risk across different demographic and clinical subgroups. (B) Forest plot illustrating the association between estimated glucose disposal rate (eGDR) and CVD risk across subgroups. No significant interactions were observed between either index and demographic or clinical characteristics (all P for interaction > 0.05), suggesting consistent associations across different subgroups.
Figure 3
Dose-response relationships between insulin resistance indices and risk of incident cardiovascular disease. (A) Restricted cubic spline analysis demonstrating the relationship between Chinese Visceral Adiposity Index (CVAI) and the odds ratio of new-onset cardiovascular disease. (B) Restricted cubic spline analysis illustrating the relationship between estimated glucose disposal rate (eGDR) and the odds ratio of new-onset cardiovascular disease.
Figure 4
Feature selection for cardiovascular disease risk prediction using machine learning approaches. (A) The Boruta algorithm identified 13 features as important predictors of CVD risk. (B) Feature selection using LASSO (Least Absolute Shrinkage and Selection Operator) regression with ten-fold cross-validation. The plot shows the binomial deviance against log(λ), where λ is the regularization parameter. The numbers above the plot indicate the number of non-zero coefficients retained in the model at each value of λ. The optimal model using the minimum lambda criterion identified 15 key variables. (C) Venn diagram showing the intersection of features selected by both Boruta and LASSO methods, resulting in 10 final variables (gender, education, dyslipidemia, hypertension, UA, WC, BUN, DBP, age, and BMI) that were subsequently used in the machine learning models for predicting CVD incidence.
Figure 5
Performance comparison of nine machine learning algorithms for predicting cardiovascular disease incidence. The receiver operating characteristic (ROC) curves illustrate the discriminative ability of each model in identifying individuals at risk of developing cardiovascular disease. The K-Nearest Neighbors (KNN) algorithm demonstrated markedly superior performance with the highest area under the curve (AUC) of 0.9315 (95% CI: 0.924–0.939), indicating excellent discrimination. All models were trained using the final set of 10 selected features identified through the combined Boruta and LASSO feature selection process.
Figure 6
Incremental predictive value of insulin resistance indices when added to the baseline model for cardiovascular disease prediction. This forest plot demonstrates the improvement in discriminative ability when eGDR and CVAI were incorporated into the K-Nearest Neighbors (KNN) algorithm, which showed superior performance among all machine learning approaches.
Figure 7
Comprehensive evaluation of performance metrics across nine machine learning models incorporating eGDR and CVAI for cardiovascular disease prediction. This multi-panel figure displays six key performance metrics for each model: (A) Accuracy; (B) Precision; (C) Recall; (D) F1 score; (E) Specificity; and (F) Sensitivity.