Risk prediction and interpretability analysis of chronic coronary syndromes based on XGBoost model and SHAP values
Objective To construct an interpretable model based on XGBoost and SHAP values,which can simultaneously achieve good predictive ability and explanatory ability,for interpretable prediction in pa-tients with chronic coronary syndrome(CCS).Methods This study selected data from CCS patients who vis-ited the Third Affiliated People's Hospital of Fujian University of Traditional Chinese Medicine,Nanping Peo-ple's Hospital Affiliated to Fujian University of Traditional Chinese Medicine,and Longhua Hospital Shanghai University of Traditional Chinese Medicine between September 2019 and June 2023.The data included pa-tients'clinical baseline information,cardiovascular risk factors,history of percutaneous coronary intervention(PCI)and/or coronary artery bypass grafting(CABG),and coronary angiography results.In this study,the predictive performance of the XGBoost model was evaluated by comparing it with four other machine learning models.In addition,a visual interpreter based on SHAP values was used to provide personalized assessment and interpretation for personalized clinical decision support.Results The XGBoost model demonstrated good predictive ability for major adverse cardiovascular events(MACE)in the CCS population and this model was simpler and more effective than previous prediction models,with high predictive accuracy,and the model recall rate(RR)and the area under the receiver operating characteristic curve(AUC)were 84.85％and 98.01％,re-spectively,both higher than those of the other four models.In addition,this study conducted a feature-depend-ent analysis of the two groups of indicators and found that hypertension,LDL-C,smoking index,traditional Chinese medicine syndrome types and age could significantly affect the risk of MACE.Conclusion The inter-pretable model based on XGBoost and SHAP values may help clinicians more accurately and rapidly identify CCS patients at risk of MACE and provide better treatment for patients.In addition,the use of a visual inter-pretable framework increases model transparency and facilitates clinicians to analyze the reliability of predictive model.
万方数据
维普
