Surrogate markers of insulin resistance and coronary artery disease in type 2 diabetes: U-shaped TyG association and insights from machine learning integration.

Abstract

Background: Surrogate insulin resistance (IR) indices are simpler and more practical alternatives to insulin-based IR indicators for clinical use. This study explored the association between surrogate IR indices, including triglyceride-glucose index (TyG), triglyceride glucose-body mass index (TyG-BMI), triglyceride glucose-waist circumference (TyG-WC), triglyceride glucose-waist to height ratio (TyG-WHtR), metabolic score for insulin resistance (METS-IR), and the triglycerides/high-density lipoprotein cholesterol (TG/HDL-C) ratio, and coronary artery disease (CAD) in patients with type 2 diabetes (T2D).

Methods: Patients with T2D were enrolled in this study and divided into two groups, matched for age and diabetes duration: those with CAD and those without CAD. The association between surrogate IR indices and CAD was evaluated using restricted cubic spline (RCS) and multivariable logistic regression and their discriminative ability was assessed via Receiver operating characteristic (ROC) curve analysis. Additionally, machine learning models, including Logistic Regression, Random Forest, eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Support Vector Machine (SVM), were employed to predict CAD presence using multiple surrogate IR indices and their components.

Results: All surrogate IR indices exhibited non-linear associations with CAD. TyG demonstrated a U-shaped relationship, where both extremely low and high levels were associated with higher odds of CAD compared to intermediate levels. The surrogate IR indices showed a relatively strong discriminative ability for CAD, with AUC values exceeding 0.708 across all indices. The TG/HDL-C ratio displayed the highest AUC (0.721), accuracy (68%), and sensitivity (71%), whereas TyG-WC showed the highest specificity (78%). Machine learning algorithms (except logistic regression) demonstrated greater discriminative power than individual IR indices. Random forest and XGBoost revealed the best performance when using either multiple surrogate IR indices or their components.

Conclusions: Surrogate IR indices could be used as valuable tools for evaluating cardiometabolic risk in patients with T2D, who are at high risk for CAD. Integrating machine learning models further improved CAD prediction, underscoring their potential for better risk stratification. The observed association between these indices and CAD in T2D may help clarify the complex pathophysiology of CAD and offer insights for future research.