基于WGBDT的心衰患者半年内再入院风险预测  

WGBDT-based risk prediction for readmission within six months in heart failure patients

作　　者：徐瑞 肖海军[1] 胡琛 XU Rui;XIAO Haijun;HU Chen(School of Mathematics and Physics,China University of Geosciences,Wuhan 430074,China;Department of Pediatrics,Tongji Hospital,Tongji Medical College,Huazhong University of Science and Technology,Wuhan 430030,China)

机构地区：[1]中国地质大学数学与物理学院,武汉430074 [2]华中科技大学同济医学院附属同济医院儿科学系,武汉430030

出　　处：《中南民族大学学报（自然科学版）》2023年第3期425-432,共8页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家自然科学基金资助项目(11901543);国家级外专资助项目(华科专字G2021154019L)。

摘　　要：为了解决现有心衰患者再入院风险预测评估模型缺乏可解释性、无法满足临床应用要求的问题,提出了一种基于自适应加权的梯度提升决策树(Weighted Gradient Boosting Decision Trees,WGBDT)的心衰患者半年内再入院风险预测模型.这一模型由基于WGBDT算法的风险预测和基于可解释机器学习(SHapley Additive exPlanation,SHAP)模型的解释性框架构成.其一,WGBDT风险模型通过样本权重更新来完成基分类器的训练.对通过基分类器分类误差率更新的残差样本进行迭代训练的基分类器加权累加,可以获得泛化性和准确率更好的模型;其二,SHAP可解释性框架采用Kernel SHAP和临床医学先验知识相结合的方式,对WGBDT黑箱模型进行解释,完成该模型的可解释性.使用四川省某医院的2008例心衰患者临床数据集对模型进行训练与测试,结果显示:利用该模型获得的结论优于梯度提升决策树(GBDT)、XGBoost、支持向量机、决策树、Adaboost等主流的机器学习算法获得的结论;同时,利用SHAP框架提高了该模型的可解释性,并根据特征的重要性,识别出了影响心衰因素的重要性排序,这为医生制定更加合理的决策提供了科学的参考.In order to solve the problem that existing models for predicting and assessing the risk of readmission in heart failure patients lack interpretability and cannot meet the requirements of clinical application,a WGBDT(Weighted Gradient Boosting Decision Trees)-based model for predicting the risk of readmission in heart failure patients within six months is proposed.This model consists of a risk prediction based on the WGBDT algorithm and an interpretation framework based on the SHAP(SHapley Additive exPlanation)model.On the one hand,the WGBDT risk model completes the training of the base classifier by updating the sample weights.The weighted accumulation of the base classifier that iteratively trains the residual samples updated by classification error rates of the base classifier can obtain a model with better generalization and accuracy.On the other hand,the SHAP interpretability framework uses a combination of Kernel SHAP and clinical medicine prior knowledge to interpret the WGBDT black box model and complete the interpretability of the proposed model.By using a clinical dataset with 2008 heart failure patients from a hospital in Sichuan Province as training set and test set,the results show that the conclusions obtained by the proposed model outperform those obtained by mainstream machine learning algorithms such as GBDT,XGBoost,SVM,DT,Adaboost,et al.At the same time,the interpretability of the proposed model is enhanced by using the SHAP framework and the order of the importance of factors affecting heart failure is identified according to the importance of the features.These provide a scientific reference for doctors to formulate more reasonable treatment plans.

关 键 词：心衰 再入院 梯度提升决策树 样本权重 可解释机器学习模型 

分 类 号：R541.6[医药卫生—心血管疾病]