基于机器学习的冠心病稳定型心绞痛痰浊闭阻证诊断模型研究  

Study on the Diagnosis Model of Phlegm-Dampness Obstruction Syndrome in Patients withStable Angina Pectoris Due to Coronary Heart Disease Based on Machine Learning

作　　者：陈浩然 姜童 郑一 王伟蔚 刘莹 王科军[1] CHEN Haoran;JIANG Tong;ZHENG Yi;WANG Weiwei;LIU Ying;WANG Kejun(School of Traditional Chinese Medicine,Binzhou Medical University,Yantai 264000,China;Yantai Yuhuangding Hospital,Yantai 264003,China;Dongying Hospital Affiliated to Shandong University of Traditional Chinese Medicine,Dongying 257000,China)

机构地区：[1]滨州医学院中医学院,山东烟台264000 [2]烟台毓璜顶医院,山东烟台264003 [3]山东中医药大学附属东营医院,山东东营257000

出　　处：《中国中医药信息杂志》2024年第12期142-150,共9页Chinese Journal of Information on Traditional Chinese Medicine

基　　金：山东省中医药科技项目(Z-2022020)。

摘　　要：目的构建冠心病稳定型心绞痛(CSAP)痰浊闭阻证诊断模型,为临床辨证提供参考。方法收集2022年5月-2024年1月于山东中医药大学附属东营医院心内科就诊的305例CSAP患者临床资料,运用最小绝对收缩和选择算法(LASSO)筛选特征,通过机器学习(ML)算法构建多个模型并进行比较,筛选出最优ML模型进行训练、验证与测试。通过沙普利加性解释(SHAP)的方法对最优模型的运行逻辑进行解释,并提供2个典型示例,帮助使用者理解模型的运行逻辑。结果LASSO回归显示胸部闷痛、体质量指数(BMI)、肢体困重、饮酒史、年龄、三酰甘油(TG)、总胆固醇(TC)和低密度脂蛋白胆固醇(LDL-C)为纳入模型中的特征。经过多模型比较,高斯朴素贝叶斯(GNB)模型展现出的效能最为优异。最终构建的GNB模型在训练集的平均AUC为0.938(95%CI:0.903~0.972),验证集的平均AUC为0.927(95%CI:0.851~0.992),测试集的AUC为0.856(95%CI:0.751~0.961)。学习曲线显示,模型中训练集和验证集之间的误差随训练样本数量增加而收敛,校准曲线显示模型对观察到的痰浊闭阻证患者的预测概率具有较好的一致性,临床决策曲线(DCA)显示模型在<0.7的决策阈值下能为患者提供临床获益。SHAP重要性排名特征依次为胸部闷痛、BMI、LDL-C、TG、肢体困重、TC、饮酒史和年龄。结论本研究构建的CSAP痰浊闭阻证诊断模型能够辅助医师对患者进行辨证,从而制定中西医结合的临床治疗方案,帮助患者获得更好的疗效。Objective To construct a diagnostic model for the phlegm-dampness obstruction syndrome in patients with coronary heart disease stable angina pectoris(CSAP);To provide a reference for clinical syndrome differentiation.Methods Totally 305 patients’clinical data were collected from the Department of Cardiology,Dongying Hospital Affiliated to Shandong University of Traditional Chinese Medicine,from May 2022 to January 2024.The least absolute shrinkage and selection operator(LASSO)was used to select features,and multiple models were constructed and compared using machine learning(ML)algorithms.The optimal ML model was selected for training,validation,and testing.Finally,the operational logic of the optimal model was explained using Shapley Additive Explanations(SHAP),and two typical examples were provided to help users understand the model’s operational logic. Results LASSO regression identified chest pain, body mass index (BMI), limb heaviness, drinkinghistory, age, triglycerides (TG), total cholesterol (TC), and low-density lipoprotein cholesterol (LDL-C) as featuresincluded in the model. After comparing multiple models, the Gaussian Naive Bayes (GNB) model demonstrated thebest performance. The final constructed GNB model achieved an average AUC of 0.938 (95% CI: 0.903-0.972) in thetraining set, an average AUC of 0.927 (95%CI: 0.851-0.992) in the validation set, and an AUC of 0.856 (95% CI:0.751-0.961) in the test set. The learning curve showed that the error between the training and validation sets in themodel converged as the number of training samples increased. The calibration curve showed that the model had goodconsistency in predicting the probability of observed phlegm-dampness obstruction syndrome patients. The clinicaldecision curve (DCA) showed that the model could provide clinical benefits for patients at a decision threshold below0.7. The features ranked by SHAP importance in order were chest pain, BMI, LDL-C, TG, limb heaviness, TC,drinking history and age. Conclusion The diagnostic model for CSAP p

关 键 词：冠心病 机器学习 临床预测模型 辨证论治 四诊客观化 

分 类 号：R259.414[医药卫生—中西医结合]