Comparative evaluation of nine insulin resistance indices for 10-year coronary artery disease risk prediction in non-diabetic adults

Surrogate indices of insulin resistance (IR) have been widely used to estimate cardiometabolic risk, yet comparative evidence among multiple indices in non-diabetic individuals remains limited. We aimed to evaluate and compare nine IR surrogate indices for prediction of 10-year incident coronary artery disease (CAD) in a large prospective cohort, with additional sex-stratified analyses. This prospective analysis included 7,858 non-diabetic participants without baseline cardiovascular disease from the MASHAD cohort, who were followed up for 10 years. Nine IR surrogate indices were calculated, including the cholesterol- High-density lipoprotein-glucose (CHG) index, triglyceride-glucose (TyG) index, TyG-derived markers, metabolic score for insulin resistance (METS-IR), and triglyceride to High-density lipoprotein ratio (TG/HDL-C). Associations with incident CAD were assessed using Cox proportional hazards models with sequential multivariable adjustment. Proportionality of hazard assumptions were evaluated using scaled Schoenfeld residuals. Restricted cubic splines (RCS) were applied to explore dose-response relationships. Discriminatory performance was compared using Harrell's C-index, receiver operating characteristic (ROC) curves and DeLong's test. We quantified the incremental predictive value of each index beyond the SCORE2 risk score using categorical and continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Model calibration was examined using calibration plots and Brier scores. To evaluate the relative predictive contribution of all nine correlated surrogate indices within a unified framework, elastic net penalized regression was applied using train-test splitting and cross-validation, and predictors retained under penalization were identified based on non-zero coefficients. During follow-up, 519 participants developed CAD. In fully adjusted models, CHG showed the strongest independent association with CAD (HR per 1-SD increase 1.394, 95% CI 1.205-1.612), whereas most TyG-derived indices lost significance after adjustment. CHG demonstrated the highest discrimination (AUC 0.616 overall; 0.623 in men) and significantly improved risk reclassification (NRI 9.42% in total population; 10.54% in men). Restricted cubic splines revealed a linear dose-response association between CHG and CAD in both sexes. In elastic net analysis including all nine indices simultaneously, only CHG retained non-zero coefficients in pooled and sex-stratified models, indicating dominant predictive contribution under penalization. Among nine surrogate IR markers, CHG consistently demonstrated superior independent association, discrimination, and incremental predictive value for incident CAD in non-diabetic adults, particularly in men. CHG offers a simple, practical, and scalable adjunct that may improve cardiovascular risk stratification when used together with conventional risk scores.