Objective This study aimed to evaluate the predictive value of the triglyceride-glucose (TyG) index for coronary in-stent restenosis (ISR) and develop a machine learning model integrating TyG with multidimensional clinical features.

Methods Two hundred and seventy patients who underwent coronary stent implantation in Department of Cardiovascular Medicine, Longyan First Affiliated Hospital of Fujian Medical University from January 2021 to December 2024 were enrolled. They were categorized into the ISR (n = 62) and non-ISR (n = 208) groups based on follow-up angiography. The TyG index was calculated. Risk factors for ISR were identified using univariate and multivariate Logistic regression, and the predictive performance of TyG was assessed using ROC curves. Five machine learning algorithms Logistic regression(LR), random Forest(RF）,extreme gradient boosting(XGBoost）,light gradient boosting machine(LightGBM), support vector machine(SVM）were trained and validated using 5-fold cross-validation. Model interpretability was achieved via SHAP analysis.

Results The ISR incidence was 23.0% (62/270). Patients with ISR exhibited a significantly higher TyG index compared to controls (9.17±0.42 vs. 8.82±0.45, P ＜ 0.001). Multivariable analysis identified the TyG index as an independent predictor of ISR (OR = 5.641, 95% CI: 2.428 ～ 13.108, P ＜ 0.001). The TyG index yielded an AUC of 0.708 (cutoff: 9.111; sensitivity: 62.9%; specificity: 73.1%). Among the machine learning models, LR demonstrated superior performance (AUC = 0.841±0.041), with SHAP analysis highlighting the TyG index as a critical predictive variable.

Conclusions The TyG index serves as an independent risk factor and predictor for ISR following coronary stenting. Integrating TyG into machine learning models based on multidimensional clinical data significantly improves ISR prediction accuracy.