The preliminary screening phase of this study involved inpatients admitted to the Department of Cardiology at The First Affiliated Hospital of Fujian Medical University, a tertiary-level medical facility located in the southeastern region of Fujian Province, as well as individuals undergoing health examinations at the Union Hospital Affiliated to Fujian Medical University during the period spanning from January 2018 to March 2024. Researchers collected demographic data (e.g. gender, age, height, weight) and lifestyle information (e.g. smoking, drinking, physical activity). Following the exclusion of participants with missing diagnoses (84 cases), missing exposure variables (1,139 cases), or abnormal exposure measurements (330 cases), final 7,419 subjects were enrolled (Supplementary Data Sheet 1). All participants or their legal representatives provided written informed consent. Prior to data extraction, the authors of the study anonymized the patient information, in accordance with the principles of the Declaration of Helsinki. The present study was approved by the Ethics Committee of the First Affiliated Hospital of Fujian Medical University.

CHD was defined as having at least one of the following conditions: (1) percutaneous coronary angiography or computed tomographic angiography (CTA) examination showed that at least one coronary artery trunk or primary branch had ≥ 50% stenosis; (2) typical exertional angina symptoms with positive stress test; (3) previously diagnosed MI; (4) previously diagnosed unstable angina pectoris (22).

Venous blood samples were collected from participants who had fasted for a minimum of 8h, including FPG, TG, HDL-C. The IR surrogates in this study include the AIP index, TyG index, and METS-IR and TG/HDL-C ratio (Equations 1–4). These indexes were calculated as formulas below (11, 23–25):

(23)

(24)

(11)

(25)

A series of covariates included as categorical or continuous variables in the analysis were as follow: demographic factors [gender (male or female), age (categorized as ≥ 60 or < 60 years), and body mass index (BMI)] and lifestyle factors [smoking status (never, current, or former), drinking (no or yes), physical activity (no or yes), and medical history (hypertension, hyperlipidemia). Specific coding details refer to Supplementary Data Sheet 1. FPG were considered as possible mediators in our mediation analysis.

The BMI was calculated using the formula: BMI (kg/m²) = body mass (kg)/height²(m²) (26). Diabetes mellitus was defined by FPG ≥ 7.0 mmol/L (126 mg/dL), a 2-h serum glucose level ≥ 11.1 mmol/L on oral glucose tolerance testing, or the current use of hypoglycemic drugs or insulin (27). The criteria for diagnosis of hypertension are Systolic Blood Pressure (SBP) ≥ 140 mmHg, diastolic blood pressure (DBP)  ≥  90 mmHg, self-reported specialist‐diagnosis history, or the use of antihypertensive medication (28). Hyperlipidemia was defined by a fasting serum total cholesterol level of more than 5.72 mmol/L (221 mg/dL) or triglyceride level of more than 1.7 mmol/L (150 mg/dL) (27). Physical activity cutoff value: Total Physical Activity MET minutes per week is < 600, that not meeting recommendations (29).

Gene targets related to insulin resistance and CHD were identified using the GeneCards (https://www.genecards.org) and OMIM (https://omim.org/) databases. We searched both databases with the keywords “insulin resistance” and “coronary heart disease”, restricting the species to Homo sapiens; all searches were performed on 31 May 2025. From GeneCards, we extracted genes together with their relevance “Score” and applied a median-based filter separately for each condition: genes with a Score ≥ 2.43 for “insulin resistance” and ≥ 6.77 for “coronary heart disease” were retained. OMIM entries corresponding to human genes were included using the same keyword terms without additional numeric thresholds. The resulting gene sets for insulin resistance and CHD were merged across GeneCards and OMIM, and their intersection was obtained using a Venn-diagram analysis. These overlapping genes were considered core candidate targets linking IR and CHD. The complete gene lists and the overlapping set are provided as files in the Supplementary Data Sheet 2.

To elucidate the biological functions of these overlapping targets, we conducted a comprehensive Gene Ontology (GO) enrichment analysis, encompassing biological processes (BP), cellular components (CC), and molecular functions (MF). This approach allowed for a systematic interpretation of the biological roles underlying the shared targets. In addition, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was performed to identify critical signaling pathways through which IR may influence the development and progression of CHD. Through these integrative analyses, we aimed to provide a more precise and mechanistic understanding of the molecular interactions bridging IR and CHD.

The results are expressed as median (interquartile range [IQR]) for continuous variables, and categorical variables are presented as numbers (percentages). We treated IR surrogates as continuous variables or categorical variables (tertile interval) in the analysis. Characteristics of the participants of the study are presented as descriptive statistics stratified by CHD status. The differences among these groups were examined using the Mann-Whitney U test for continuous variables and the Chi-square test for categorical variables.

For covariates with under 10% missing data, we used multiple imputation by chained equations (MICE) under a missing-at-random assumption, implemented in the mice package (version 3.18.0) to create five imputed datasets. The imputation model included covariates: smoking, drinking, metabolic equivalent of task, systolic and diastolic blood pressure, and total cholesterol. The standard trace plots of chain-specific showed good mixing and stable trajectories over iterations, without systematic trends, indicating satisfactory convergence of the MICE algorithm. Further details of the imputation procedure and convergence diagnostics are provided in the Supplementary Data Sheet 1.

We further used logistic regression models to explore the relationship between insulin resistance indices and CHD. IR indices were expressed as a continuous variable (per one-unit increase and per tertile increment) and a categorical variable (tertile) across three sequential adjustment levels: unadjusted model1, basic adjustment model 2 (adjusted gender and age), and full adjustment model 3 (adjusted BMI, drinking, smoking, physical activity, hypertension, hyperlipidemia on the basis of model 2). Model 3 was designated as the primary Model for all substantial inferences.

To explore the exposure-response relationship and to test for linearity between IR indices levels and CHD, we plotted restricted cubic spline curves with 3 knots at the 10th, 50th and 90th percentiles with the ‘rms’ and ‘splines’ packages in R software. Moreover, subgroup and interaction analyses were conducted to investigate potential heterogeneity in the association between the insulin resistance indices and CHD.

For the enrichment analysis, GO and KEGG terms with P values less than 0.05 were initially selected for inclusion. Thereafter, statistically significant terms were identified based on a threshold of false discovery rate (FDR) < 0.05, to control for multiple comparisons and reduce the likelihood of type I errors.

To further explore the potential mediating roles of insulin resistance indicators in the pathway linking IR to CHD, multivariate causal mediation analyses were conducted using the mediation package in R. A logistic regression framework was adopted due to the binary nature of the outcome variable. Specifically, the mediation models assessed the role of FPG as a mediator in the relationship between four insulin resistance indices and CHD risk. Within this framework, the Average Direct Effect (ADE) represents the proportion of the total effect that is independent of the mediator, whereas the Average Causal Mediation Effect (ACME) quantifies the effect mediated through FPG. The mediation proportion was calculated using the formula: ACME/(ADE + ACME) × 100%, indicating the extent to which the mediator accounts for the total effect. The parameters for mediation analysis are detailed in the Supplementary Data Sheet 1.

Additionally, we carried out several sensitivity analyses to confirm the stability of the main results. Firstly, in order to observe the stability of the results of the imputation-free dataset, this study conducted a correlation analysis on the imputation-free dataset. Second, to account for potential confounding by the sampling frame, the source of participants was further adjusted on the basis of the fully adjusted Model 3. We then evaluated effect modification by adding multiplicative interaction terms between each IR index and recruitment source (index × source) to the pooled models and deriving P-values for interaction. Third, the population is re-analyzed by inverse probability weighting with weights truncated at the 1st and 99th percentiles, and the result remains unchanged. Fourthly, given the impact of diabetic to CHD, these participants were excluded to. In the last, external random subsampling was conducted under the Probit link function (500 iterations; with 80% of the original sample being randomly sampled each time). The mediating effect was consistent with the primary analysis based on the Logit model.

All analyses were conducted using R v.4.5.0 or GraphPad Prism v.8.4, unless otherwise specified. A two-sided P < 0.05 was considered statistically significant.

Table 1 provides a concise overview of the characteristics of the study population. The total number of participants included in the study was 7,419, comprising 2,349 cases of CHD (31.7%). The demographic profile of the participants revealed a preponderance of males, constituting 54.9% of the sample. A significant proportion of the participants belonged to the elderly demographic, accounting for 62.8% of the sample, with ages 60 years and above constituting the majority. Statistically significant differences were identified between the two subgroups in most baseline characteristics, including gender, age, smoking, physical activity levels, and the prevalence of hypertension, diabetes mellitus, and hyperlipidemia. In addition, the median values for four insulin resistance surrogate markers (AIP index, TyG index, METS-IR, TG/HDL-C ratio) and FPG were higher in the case group than in the control group. Furthermore, statistically significant differences were also observed between subgroups (P < 0.05). Moreover, Supplementary Data Sheet 1 showed the number of patients from the cardiology inpatient department and the health examination center were 4,788 and 2,631 respectively. several demographic and cardiometabolic variables differed significantly between the two groups (P < 0.05).

As demonstrated in Supplementary Data Sheet 1, the kernel density curves for all four standardized indicators demonstrate unimodal distributions, with a peak near zero. Of particular note are the curves for AIP index and TyG index, which exhibit approximate symmetry and proximity to normality. METS-IR demonstrates a slight right skew, while TG/HDL-C ratio displays a pronounced right skew with a protracted right tail, suggesting the presence of extreme high values. In view of this distributional characteristic, the natural logarithm of the TG/HDL-C ratio was applied prior to its incorporation into subsequent analyses. Supplementary Data Sheet 1 illustrates the shape of the ln (TG/HDL-C) ratio following natural logarithmic transformation.

The associations between alternative indices of IR and CHD are presented in Table 2. Odds ratios expressed ‘per one-unit increase’ represent the change in CHD risk associated with a one-unit increment in the corresponding index value. In analyses using tertiles of each index, a ‘per tertile increase’ corresponds to an average increment of 0.508 units for AIP index, 1.230 units for TyG index, 13.449 units for METS-IR and 1.169 units for ln(TG/HDL-C) on their original scales. All alternative indices of IR demonstrated a significant positive correlation with the risk of CHD. In the multivariate logistic regression analysis, this association remained significant after adjusting for age, gender, BMI, drinking, smoking, physical activity, hypertension, and hyperlipidemia in model 3. When insulin resistance indices were treated as continuous variables, the AIP index (OR: 1.612, 95% CI: 1.333, 1.951), TyG index (OR: 1.208, 95% CI: 1.101, 1.325), METS-IR (OR: 1.039, 95% CI: 1.030, 1.048), and ln(TG/HDL-C) (OR: 1.230, 95% CI: 1.133, 1.337) were all found to be significant. Furthermore, higher levels of the surrogate marker for IR, whether expressed as a tertile or per tertile increase, were associated with a greater likelihood of CHD. The robustness of this association was consistent across various adjustment models.

The exposure-response relationship between surrogate markers of IR and CHD is illustrated in Figure 1. A notable shift in surrogate markers of IR was observed at values of 0.391 (AIP index), 8.498 (TyG index), 35.897 (METS-IR), and 0.900 (ln (TG/HDL-C)). There was no evidence of nonlinearity (P-nonlinear>0.05) with four indexes, consistent with an approximately linear, modest increase across the observed range.

Subgroup and interaction analyses stratified by age, gender, BMI, drinking, smoking, physical activity, hypertension, and hyperlipidemia as shown in Figure 2. Across most strata, higher levels of each IR surrogate were consistently associated with a higher prevalence of CHD. After applying the FDR procedure to the interaction tests, all physical activity × IR interactions remained statistically significant (adjusted P-interaction < 0.05). For age, FDR-adjusted interaction P-values remained below 0.05 for the AIP index, METS-IR and ln(TG/HDL-C), with larger effect estimates in participants aged < 60 years, whereas the age × TyG interaction was attenuated and no longer reached conventional statistical significance after FDR correction(adjusted P-interaction > 0.05).

In order to further investigate the potential mechanisms underlying this association, 1,370 potential targets for IR and 3,630 potential targets for CHD were identified, with 869 genes present in both diseases (Figures 3A–D). Enrichment analysis of these common targets revealed significant enrichment in pathways related to hormone response, inflammatory regulation, and membrane transport (Figure 3E). The pathway analysis indicated significant involvement of IR, the advanced glycation end-product (AGE)-receptor signaling pathway, lipid metabolism and atherosclerosis, FoxO, and adipokine pathways (Figure 3F). The full KEGG and GO enrichment output (including term, gene count, P and FDR) as Supplementary Data Sheets 3, 4. String diagrams revealed substantial overlap among IR, AGE-receptor, FoxO, and adipokine signaling pathways, with functional enrichment observed in genes associated with fasting blood glucose across these pathways (Figure 3G).

Pathway enrichment analysis indicates FoxO as one of the potential pathways. Previous studies suggest FoxO primarily influences fasting blood glucose by directly inhibiting hepatic gluconeogenesis; suppression of hepatic FoxO reduces fasting blood glucose levels (30, 31). The glycated end-product receptor activator pathway interacts with inflammatory gene networks. Randomized controlled trials indicate that limiting glycated end-products can reduce inflammation, improve glucose metabolism, and lower fasting blood glucose (32, 33). Adipokine signaling enhances insulin sensitivity and glucose uptake, consistent with reduced FPG outcomes in human studies (34, 35). In summary, these pathways collectively influence FPG regulation; consequently, we selected FPG as the mediating indicator for assessing IR’s impact on CHD in subsequent analyses. The mediation analysis in Figure 4 results indicated that FPG mediated 8.15%, 55.33%, 9.71%, and 8.17% of the association effects between AIP index, TyG index, METS-IR, ln(TG/HDL-C) and CHD, respectively. Absolute ACME, ADE and total effect values with CIs for each IR index were shown in Supplementary Table 3 of Supplementary Data Sheet 1.

In order to assess the robustness of the model, several sensitivity analyses were conducted. Re-analysis of the pre-imputation dataset revealed that the associations between the four indicators and CHD remained stable after full covariate adjustment (Supplementary Data Sheet 1). Further adjusting the source of participants on the basis of the fully adjusted Model 3, the associations between each IR surrogate and prevalent CHD were unchanged, and all associations remained statistically significant (P < 0.05) (Supplementary Data Sheet 1). In analyses stratified by recruitment source, all four IR indices showed positive associations with CHD among cardiology inpatients, whereas effect estimates in the health-examination subgroup were close to null with wide confidence intervals that all included 1. Interaction P-values were <0.05 for AIP index and ln(TG/HDL-C), but not for TyG index or METS-IR, indicating that any heterogeneity by recruitment source should be interpreted with caution given the limited power in the health-examination subgroup. (Supplementary Data Sheet 1). Subsequently, we employed the inverse probability weighting method and managed extreme weight values by truncating the distribution at the 1st and 99th percentiles. The results indicated that there were no substantial alterations in any associations (Supplementary Data Sheet 1). Following the exclusion of individuals diagnosed with diabetes, the results for all indicators remained stable, with the exception of TyG index, as illustrated in the Supplementary Data Sheet 1. In order to assess the robustness of the mediation analysis findings, outer-layer random subsampling was performed (500 iterations, with 80% of the sample being drawn each time) under the probit link function. The mediation proportions were consistent with the main findings (Supplementary Data Sheet 1).