The Prediction Model of Adverse Outcomes after PCI in Patients with Coronary Heart Disease based on Multi-label Deep Forest Algorithm
10.11783/j.issn.1002-3674.2025.03.007
- VernacularTitle:多标签深度森林算法在构建冠心病患者PCI术后不良结局预测模型中的应用研究
- Author:
Ruiyan ZHANG
1
;
Weichang ZHANG
;
Hong YANG
Author Information
1. 山西医科大学公共卫生学院流行病与卫生统计学教研室(030001)
- Publication Type:Journal Article
- Keywords:
Coronary heart disease;
Percutaneous coronary intervention;
Multi-label imbalance;
Multi-Label Deep Forest
- From:
Chinese Journal of Health Statistics
2025;42(3):355-359
- CountryChina
- Language:Chinese
-
Abstract:
Objective Based on the multi-dimensional postoperative outcomes of patients with percutaneous coronary intervention(PCI)the prediction model of adverse outcomes in patients with coronary heart disease after PCI was constructed by combining multi-label deep forest algorithm.Methods 521 patients diagnosed with coronary heart disease and undergoing PCI from the Second Hospital of Shanxi Medical University were collected.Multi-label ReliefF algorithm was used to filter features,and multi-label-random oversampling algorithm was used to deal with data imbalance.Finally,multi-label deep forest algorithm(MLDF)was used to build a prediction model.Results Multi-label ReliefF was used to filter the characteristics.The results showed that B-type natriuretic peptide,creatine kinase isoenzyme,hemoglobin,homocysteine,C-reactive protein and serum indirect bilirubin were important factors affecting the prognosis of PCI.The MLROS algorithm improved the imbalance of multi-tag data to a certain extent,and the mean IR of the whole tag was reduced from 3.937 to 2.668.Conclusion In this study,the multi-label deep forest algorithm was combined with the adverse outcome of patients after PCI.At the same time,considering the problem of multi-tag feature selection and data imbalance,and fully considering the actual clinical situation,patients may have multiple outcomes at the same time after PCI,which is more in line with the requirements of modern medicine.
