Construction of classification model and analysis of risk factors in patients with multi-vessel coronary artery disease.

backgroundMulti-vessel coronary artery disease (MVCD) is a severe type of coronary artery disease (CAD) with high risk of major adverse cardiovascular events (MACEs). At present, the accurate identification and risk stratification of patients with MVCD is to be solved imminently. The aim of this study was to preliminarily explore the potential risk factors of the patients with MVCD and construct a classification model with logistic regression and machine learning (ML) algorithms.

methodsThe 1708 hospitalized CAD patients who underwent percutaneous coronary intervention (PCI) in Shandong Provincial Hospital were recruited in this retrospective analysis. According to the results of coronary angiography, they were divided into single-vessel disease group and multi-vessel disease group (≥ 2 major coronary arteries have more than 50% stenosis). Except the state of coronary, the basic clinical data, laboratory test results and auxiliary examination results were collected after admission. The risk factors of patients with MVCD were studied by univariate and multivariate logistic regression analysis. Logistic regression and ML algorithms of XGBoost and Random Forest (RF) were employed to construct clinical risk prediction models for MVCD. Age, gender, hypertension, heart rate (HR), ApoB, the use of statins and nitrates, HDL-C, vaso-occlusion were included in the construction of models. Model evaluation included Calibration Curve, decision curve analysis (DCA), area under the curve (AUC), and classification metrics (Accuracy, Sensitivity, Specificity, PPV, NPV).

resultsUnivariate and multivariate regression analysis showed that gender, age, hypertension, heart rate (HR), ApoB, the use of statins and nitrates, HDL-C, TyG, vaso-occlusion were independently influential factors (all P < 0.1) of MVCD. Among the current models, the Random Forest model performed best on the training set, while the Logistic Regression model performed best on the validation set. Comprehensively considering the DeLong test, the calibration curve and the DCA curve, the Logistic model is relatively more robust among 3 models.

conclusionsThis study preliminarily explored the risk factors of MVCD and auxiliary diagnostic models in angiography. The analysis of related factors and the construction of classification models provide intraprocedural diagnostic support, thereby expecting offer some ideas for the comprehensive management, diagnosis and treatment of MVCD patients.