Insulin resistance assessed by estimated glucose disposal rate predicts cardiovascular disease in stages 0–3 of cardiovascular-kidney-metabolic syndrome: a UK biobank cohort study | Cardiovascular Diabetology

The analytical cohort comprised 325,312 participants with complete data, including 180,833 males (55.6%) and 144,479 females (44.4%), with a mean age (SD) of 56.23 (8.09) years. During the follow-up period, 48,433 participants developed incident CVD, comprising 27,060 cases of CHD, 7430 strokes, 12,742 AF events, 9263 HF diagnoses, and 9739 PAD cases.

Through systematic screening of multiple non-insulin-based surrogate markers for IR from existing literature (formulae for IR indices provided in Supplementary Table S1), ROC curve analysis demonstrated eGDR’s superior predictive performance for CVD events compared to alternative IR indicators, achieving an AUC of 0.716 (95% CI 0.714–0.719) (see Supplementary Figure S1).

Baseline characteristics of participants

Participants with lower eGDR quartiles which indicate greater IR, demonstrated distinct sociodemographic, lifestyle, and clinical profiles: they were predominantly female, older, and socioeconomically disadvantaged (lower income and educational attainment); exhibited higher rates of smoking, alcohol consumption, physical inactivity, and poor dietary quality; displayed adverse metabolic characteristics including elevated BMI, triglycerides, and cholesterol levels, accompanied by chronic inflammation, impaired glycemic control, and renal dysfunction; and manifested advanced clinical profiles with higher comorbidity burdens, complex medication regimens, and later-stage CKM classifications (Table 1). Supplementary Figure S2 illustrates the distribution trend of eGDR, showing a non-normal distribution, with a higher mean value in males compared to females. Further analysis across CKM stages revealed progressive phenotypic divergence: Stage 3 patients were characterized by older age, male predominance, lower educational attainment, and clustering of modifiable risk factors, including smoking, insomnia, alcohol misuse, poor diet, and physical inactivity. Notably, this advanced-stage subgroup exhibited the most pronounced IR signatures, underscoring the metabolic-renal-cardiovascular interplay in CKM progression (Supplementary Table S10).

The relationship between the eGDR and the incidence of CVD

We employed RCS analysis to systematically evaluate the dose–response relationship between eGDR and CVD risk, including its component events, within the CKM syndrome stages 0–3 population (Fig. 2). The results revealed a significant inverse association between eGDR and overall CVD risk, as well as risks of composite events encompassing CHD, AF, HF, and stroke. A progressive decline in CVD risk was observed with increasing eGDR levels. Notably, the association between eGDR and PAD exhibited a distinct L-shaped curve pattern: PAD risk decreased markedly with rising eGDR levels below a threshold of 9.91 mg/kg/min, beyond which the risk reduction plateaued and demonstrated a non-significant upward trend at higher eGDR values.

Multivariable-adjusted Cox regression analysis demonstrated a progressive reduction in CVD risk across increasing eGDR quartiles, the corresponding adjusted hazard compared to Q1 were 0.994 (95% CI 0.965–1.023) for Q2, 0.920 (95% CI 0.871–0.971) for Q3, and 0.883 (95% CI 0.827–0.942) for Q4. A similar graded inverse association was observed for CHD, with 0.880 (95% CI 0.817–0.948) for Q3 and 0.730 (95% CI 0.667–0.800) for Q4 showing statistically significant risk reductions (Fig. 3 and Supplementary Table S12). While analogous trends were noted for stroke, AF, HF, and PAD, statistical significance was inconsistent across these endpoints. Cumulative risk curves stratified by eGDR quartiles corroborated this dose-dependent protective relationship, with progressively lower CVD and CHD risks observed at higher eGDR levels (Fig. 4).

Adjusted hazard ratios for cardiovascular disease incidence in individuals with CKM syndrome stages 0–3, stratified by quartiles of eGDR. Adjustments were made for variables including age, sex, physical activity level, ethnicity, education, insomnia, Townsend deprivation index, alcohol consumption status, smoking status, dietary score, diabetes, hypertension, dyslipidemia, body mass index, antihyperlipidemic medications, antihypertensive medications, and antidiabetic medications. eGDR, Estimated Glucose Disposal Rate; CKM, Cardiovascular-Kidney-Metabolic

Cumulative risk curves for cardiovascular disease (CVD) (A) and coronary heart disease (CHD) (B) risk across different eGDR quartiles in individuals with CKM syndrome stages 0–3. eGDR, Estimated Glucose Disposal Rate; CKM, Cardiovascular-Kidney-Metabolic

To elucidate the stage-specific impact of eGDR on CVD risk, subgroup analyses stratified by CKM stages (0–3) demonstrated that higher eGDR quartiles were associated with reduced CVD risk across all stages, where the strongest protective effect occurred in stage 0 (HR 0.906, 95% CI 0.833–0.985), followed by stage 1 (HR 0.910, 95% CI 0.852–0.971), stage 2 (HR 0.931, 95% CI 0.916–0.946), and stage 3 (HR 0.907, 95% CI 0.862–0.954) (p for interaction < 0.001) (Supplementary Figure S3).

In the CKM stages 0–3 cohort, overall incidence rates per 1000 person-years were 11.55 (95% CI 11.44–11.65) for CVD, 6.24 (95% CI 6.17–6.32) for CHD, 1.66 (95% CI 1.62–1.70) for stroke, 2.88 (95% CI 2.83–2.93) for AF, 2.07 (95% CI 2.03–2.11) for HF, and 2.18 (95% CI 2.14–2.23) for PAD (Table 2). Moreover, with increasing CKM staging, the adjusted CVD incidence rates rose progressively: 3.02 (95% CI 2.89–3.15) for Q1, 3.10 (95% CI 2.97–3.23) for Q2, 8.33 (95% CI 8.12–8.54) for Q3, and 11.41 (95% CI 10.85–11.97) for Q4, per 1000 person-years (Supplementary Table S9). After stratification by eGDR quartiles, adjusted CVD incidence rates decreased sequentially: 3.84 (95% CI 3.62–4.07) for Q1, 3.82 (95% CI 3.66–3.98) for Q2, 3.53 (95% CI 3.41–3.65) for Q3, and 3.37 (95% CI 3.25–3.50) for Q4 per 1000 person-years. Similarly, adjusted CHD incidence declined from 1.68 (95% CI 1.55–1.81) in Q1 to 1.22 (95% CI 1.15–1.29) in Q4 per 1000 person-years. Compared to Q1, higher eGDR quartiles demonstrated progressive risk reductions, with absolute CVD risk differences of 0.31 (95% CI 0.06–0.57) for Q3 and 0.47 (95% CI 0.22–0.72) for Q4. Corresponding CHD risk differences were 0.20 (95% CI 0.05–0.35) for Q3 and 0.46 (95% CI 0.31–0.60) for Q4. Stroke, AF, and HF showed weaker inverse correlations, though consistent with the overall trend. Notably, PAD exhibited a distinct risk profile, with an elevated adjusted absolute risk of 0.11 (95% CI 0.01–0.21) observed in Q4. (Table 2).

Incremental predictive performance of eGDR in the risk assessment of CVD

ROC analysis revealed that integrating eGDR with the PREVENT risk model significantly enhanced CVD prediction (AUC 0.743 vs. 0.716 for PREVENT alone, P < 0.001), with stable discriminative capacity across time points (1-year AUC 0.724, 3-year 0.725, 5-year 0.721, 10-year 0.719, 15-year 0.718) (Supplementary Figure S4). Risk reclassification analysis demonstrated clinically meaningful improvements (NRI = 0.268, 95% CI 0.264–0.274; P < 0.001), with 8.9% of high-risk (PREVENT ≥ 20%) and 23.3% of intermediate-risk (PREVENT 5–20%) subjects appropriately reclassified to lower-risk categories. These results, supported by significant discrimination improvement (IDI = 0.015, P < 0.001), establish eGDR as providing incremental predictive value beyond conventional CVD risk assessment tools (Supplementary Figure S5).

Subgroup and sensitivity analyses

Subgroup analyses revealed significant heterogeneity in the eGDR-CVD associations in certain subgroups (Bonferroni correction P for interaction < 0.0015) (Supplementary Table S11). Protective effects were more pronounced in younger participants (< 65 years: Q4 HR 0.740; 95% CI 0.685–0.800 vs. ≥ 65 years: HR 0.843, 95% CI 0.748–0.950), female (Q4 HR 0.806; 95% CI 0.727–0.893), and those with preserved renal function (eGFR ≥ 90 mL/min/1.73m²: Q4 HR 0.842; 95% CI 0.770–0.920). Additionally, interaction effects were observed in subgroups defined by education level and ethnicity, whereas no interaction effects were detected in subgroups based on other covariates. The main results remained robust across various sensitivity analyses. Additional covariate adjustment (Supplementary Table S13), excluding participants who had a CVD event within 3 years of follow-up (see Supplementary Table S14), accounting for competing risks from all-cause death (see Supplementary Figures S6), and potential selection bias from excluding participants with incomplete data(see Supplementary Figures S7), resulted in slightly attenuated associations.