Application of Machine Learning Algorithms in Predicting Major Adverse Cardiovascular Events after Percutaneous Coronary Intervention in Patients with New-Onset ST-Segment Elevation Myocardial Infarction

Fig. 1.. Patients selection process and machine learning pipeline for acute myocardial infarction study. (a) Data inclusion and exclusion flowchart. (b) Flowchart of ML data processing. Abbreviations: CNB, Complement Naive Bayes; KNN, K-nearest neighbors; Logistic, logistic regression; SVM, support vector machine; SHAP, Shapley Additive Explanations; MACE, major adverse cardiovascular events; ML, machine learning.

Fig. 2.. Results of feature variable selection based on Boruta’s algorithm. (Note: where Tentative variables are labeled yellow, Rejected is red, Accepted is green, and Shadow is blue). Abbreviations: BUN, blood urea nitrogen; Gensini, coronary artery lesion stenosis score; CR, creatinine; GLT, glutamine transferase; PCT, plateletcrit; HR, heart rate; HB, hemoglobin; GLU, glucose; GST, glutathione S-transferase; PCW, platelet volume distribution width; RBC, red blood cell; UA, uric acid; MPV, mean platelet volume; TG, triglyceride; VLDL, very low-density lipoprotein; N, absolute neutrophil value; WBC, white blood cell; L, lymphocyte absolute value; HDL, high-density lipoprotein; M, absolute monocyte value; HP, high blood pressure; BP, diastolic blood pressure; DBIL, direct bilirubin; TC, total cholesterol; LVD, left ventricular internal diameter; IBIL, indirect bilirubin; SP, systolic blood pressure; TDM, type 2 diabetes mellitus; PLT, platelet count; LDL, low-density lipoprotein.

Fig. 3.. ROC curve analysis and forest plot of ML algorithms for predicting the occurrence of MACE after PCI in patients with acute myocardial infarction. ROC, receiver operating characteristic; ML, machine learning; MACE, major adverse cardiovascular events; PCI, percutaneous coronary intervention; KNN, K-nearest neighbors; CNB, Complement Naive Bayes; Logistic, logistic regression; SVM, support vector machine; AUC, area under the curve.

Fig. 4.. Comparison of calibration curves for each model in the validation set. KNN, K-nearest neighbors; CNB, Complement Naive Bayes; Logistic, logistic regression; SVM, support vector machine.

Fig. 5.. Analysis of clinical decision curves for each model in the validation set. KNN, K-nearest neighbors; CNB, Complement Naive Bayes; Logistic, logistic regression; SVM, support vector machine.

Fig. 6.. Analysis of the area under the ROC curves of the training and validation sets under 5-fold cross-validation within the logistic regression model. ROC, receiver operating characteristic.

Fig. 7.. Learning curve analysis of logistic regression model in training and validation sets.

Fig. 8.. Calibration curve analysis of logistic regression model. Logistic, logistic regression.

Fig. 9.. Mixed matrix analysis of logistic regression model in training and validation sets.

Fig. 10.. DCA curve analysis of logistic regression model. DCA, decision curve.

Fig. 11.. The top 7 important eigenvariables and eigenvalues screened based on Boruta’s algorithm. BUN, blood urea nitrogen; HR, heart rate; CR, creatinine; GLT, glutamine transferase; PCT, platelet count; SHAP, Shapley Additive Explanations.

Fig. 12.. Nomogram to show the risk prediction of postoperative MACE in patients with acute myocardial infarction. BUN, blood urea nitrogen; HR, heart rate; CR, creatinine; GLT, glutamine transferase; PCT, platelet count; MACE, major adverse cardiovascular events.