Abstract / Summary
Given the limitations of coronary angiography for widespread screening and the inherent risk of restenosis after percutaneous coronary intervention (PCI), this study aimed to develop a non-invasive diagnostic model and identify predictive biomarkers. Candidate biomarkers were screened via Olink proteomic technology, validated by ELISA, and subsequently used to construct a UA diagnostic model by applying six machine learning algorithms. A logistic regression model combining resistin with routine clinical parameters demonstrated excellent performance in both the internal validation set (AUC = 0.955) and the external validation set (AUC = 0.842). Subgroup analyses stratified by lipid, blood pressure and glycemic status further confirmed model robustness, with all AUC values exceeding 0.800. Resistin, triglycerides, white blood cell count, systolic blood pressure, and aspartate aminotransferase were identified as independent risk factors for UA, while high-density lipoprotein cholesterol and hemoglobin emerged as protective factors. In addition, resistin levels were significantly correlated with the severity of coronary artery stenosis. A marked post-PCI reduction in serum matrix metalloproteinase-3 (MMP-3) levels was observed, and its extent was associated with the risk of in-stent restenosis (ISR). This study successfully developed and validated a non-invasive UA diagnostic model and first identified decreased MMP-3 as a potential predictor of ISR after PCI, offering a new strategy for precise diagnosis of UA and postoperative risk management.