基于检验数据的机器学习建立2型糖尿病患者合并冠心病的风险预测模型  

Establishment of a risk prediction model for patients with type 2 diabetes and coronary heart disease based on machine learning of laboratory data

作　　者：顾智超 吴昀喆 杨帆[1] 陆怡德[1] GU Zhichao;WU Yunzhe;YANG Fan;LU Yide(Department of Laboratory Medicine,Ruijin Hospital,Shanghai Jiao Tong University School of Medicine,Shanghai 200025,China)

机构地区：[1]上海交通大学医学院附属瑞金医院检验科,上海200025

出　　处：《国际检验医学杂志》2025年第2期135-140,共6页International Journal of Laboratory Medicine

摘　　要：目的分析2型糖尿病患者的临床指标特征,利用机器学习筛选风险预测指标,建立简便、有效的2型糖尿病合并冠心病的风险预测模型。方法采用回顾性研究的方法,选取2022年1月至2023年11月该院心内科住院且诊断为2型糖尿病合并冠心病的患者217例作为合并冠心病组,同期在门诊确诊为2型糖尿病患者214例作为糖尿病组,记录两组患者实验室常规检测数据。通过最小绝对收缩和选择算法(Lasso)筛选特征,运用随机森林、决策树、支持向量机、极端梯度提升、Logistic回归、K近邻算法、人工神经网络7种机器学习算法建立模型,通过受试者工作特征曲线及曲线下面积(AUC)、校准曲线、特异度、灵敏度、F1值等指标评价不同模型的诊断效能。结果通过Lasso回归共筛选出年龄、性别、收缩压、舒张压、心率、C反应蛋白、血糖等20个关键因素。纳入模型后,支持向量机模型具有最高的灵敏度(88.37%)、阴性预测值(82.14%)及AUC(0.845),随机森林模型具有最高的准确度(76.47%)、阳性预测值(76.74%)、F1值(0.77),而极端梯度提升算法具有较好的特异度(80.95%),引入支持向量机模型及SHAP值后,归纳得出血糖对于糖尿病合并冠心病具有正向影响。结论机器学习可作为2型糖尿病患者合并冠心病风险评估的有效工具,其中支持向量机、随机森林和极端梯度提升模型均有较好的预测效能,具有一定的临床应用前景。Objective To analyze the characteristics of clinical indicators in patients with type 2 diabetes,and to establish a simple and effective risk prediction model for type 2 diabetes complicated with coronary heart disease by screening risk prediction indicators with machine learning.Methods A retrospective study was conducted,and 217 patients diagnosed with coronary artery disease combined with type 2 diabetes mellitus who were hospitalized in the Hospital from January 2022 to November 2023 were selected.Additionally,214 patients diagnosed with T2DM during the same period in the outpatient department were selected as the control group.Their routine laboratory test data were recorded.The Least Absolute Shrinkage and Selection Operator(Lasso)algorithm was used to select features,and the models were built by using seven machine learning algorithms:Random Forest,Decision Tree,Support Vector Machine,eXtreme Gradient Boosting,Logistic Regression,K-Nearest Neighbor,and Artificial Neural Network.The diagnostic efficacy of different models through receiver operating characteristic curve(ROC),area under curve(AUC),calibration curve,specificity,sensitivity,F1 value,and other indicators were evaluated.Results Twenty key factors,including age,gender,systolic blood pressure,diastolic blood pressure,heart rate,C-reactive protein and blood glucose were selected using Lasso regression.When incorporated into various models,the SVM model exhibited the highest sensitivity(88.37%),negative predictive value(82.14%),and area under curve(0.845).The Random Forest model had the highest accuracy(76.47%),positive predictive value(76.74%),and F1 score(0.77).Meanwhile,the XGBoost algorithm demonstrated relatively good specificity(80.95%).After introducing the SHAP model,it was inferred that blood glucose had a significant positive impact on the occurrence of coronary heart disease in individuals with type 2 diabetes.Conclusion Machine learning can serve as an effective tool for assessing the risk of coronary heart disease in patients with type 2

关 键 词：机器学习 2型糖尿病 冠心病 预测模型 

分 类 号：R446.1[医药卫生—诊断学] R587.2[医药卫生—临床医学]