Machine Learning–Based Periprocedural Prediction Model for No-Reflow Risk After Percutaneous Coronary Intervention in Patients With Acute Coronary Syndrome

Coronary no-reflow (NR) after percutaneous coronary intervention (PCI) predicts adverse prognosis in patients with acute coronary syndrome (ACS). This study aimed to establish a machine learning-based risk prediction model for peri-procedural NR in ACS patients undergoing PCI. We conducted a single-center, prospective, observational cohort study enrolling consecutive ACS patients undergoing PCI between January 2023 and March 2024. Patients were allocated to a training set (n=692) and a test set (n=297) based on admission order. Key predictive features were screened using Lasso regression and the Boruta algorithm, and the overlapping variables were utilized to construct 5 models: logistic regression (LR), support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and light gradient boosting machine (LightGBM). Model performance was validated in the test set. Among 989 included patients, 145 (14.7%) developed NR. The final predictors included cardiac troponin (cTn), aspartate aminotransferase (AST), pre-PCI TIMI flow grade, number of stents implanted, intraoperative hypotension, and ACS subtypes. The LR model showed the optimal discriminative capacity, with an area under the receiver operating characteristic curve (AUC) of 0.8799 (95% confidence interval [CI]: 0.8151-0.9447), as well as favorable accuracy (0.8822), F1-score (0.6316), G-mean (0.7983), and precision-recall curve-area under the curve (PR-AUC) (0.6249). Restricted cubic spline (RCS) analysis confirmed a nonlinear relationship between AST levels and NR risk (P-nonlinear < 0.001). In conclusion, we established a concise machine learning-based model for post-PCI NR risk stratification; the nomogram and online calculator support efficient peri-procedural risk assessment in ACS patients.