基于机器学习预测患者地高辛血药浓度超出治疗窗上限的风险  

Predicting the risk of digoxin concentration exceeding the upper limit of treatment window based on machine learning

作　　者：杨妙[1] 张靖悦 禹洁[1] 毕重文 顾芃 胡祎明[1] 袁恒杰[1] YANG Miao;ZHANG Jingyue;YU Jie;BI Chongwen;GU Peng;HU Yiming;YUAN Hengjie(Department of Pharmacy,General Hospital of Tianjin Medical University,Tianjin 300052,China)

机构地区：[1]天津医科大学总医院药剂科,天津300052

出　　处：《中国医院药学杂志》2023年第13期1425-1429,共5页Chinese Journal of Hospital Pharmacy

摘　　要：目的:建立人血清中地高辛血药浓度超出治疗窗上限的风险预测模型,为地高辛的安全使用提供参考。方法:回顾收集2019年1月至2021年12月期间天津医科大学总医院地高辛血药浓度监测数据。371名患者数据按7∶3随机分为训练集和测试集,训练集用于构建预测模型,测试集用于评价模型效果。采用极端梯度提升(XGBoost)特征重要性分析法选择最佳特征,Shapley加法解释(SHAP)算法分析特征与结果变量的相关性和趋势。使用逻辑回归(LR)、朴素贝叶斯(NB)、随机森林(RF)、XGBoost、梯度提升树(GBDT)5种机器学习算法用于模型构建。结果:默认参数下LR、NB、RF、XGBoost和GBDT模型的AUC分别为0.5199,0.6278,0.6951,0.5625和0.6558。优化后的RF模型AUC为0.7112。结论:机器学习模型预测人血清中地高辛血药浓度超出治疗窗上限风险的表现良好,并创新地利用SHAP算法对其进行解释。RF模型预测性能最佳,为识别地高辛中毒高风险患者提供参考。OBJECTIVE To establish a model to predict the probability of exceeding the treatment window upper limit of digoxin valley concentration in human serum and to provide reference for clinical safe medication.METHODS The clinical digoxin blood concentration data of 371 patients in General Hospital of Tianjin Medical University from January 2019 to December 2021 were retrospectively collected,and randomly divided into training set and test set at the ratio of 7∶3.The training set was used to build the prediction model,while the test set was used to evaluate the effect of the model.Extreme gradient boosting(XGBoost)feature importance analysis was used to select the best feature,and five machine learning algorithms,including logistic regression(LR),naive bayes(NB),random forest(RF),XGBoost and gradient boosting decision tree(GBDT),were used to construct the model.RESULTS Under the default parameters,the AUC of LR,NB,RF,XGBoost and GBDT models was 0.5199,0.6278,0.6951,0.5625 and 0.6558,respectively.The AUC of the optimized RF model was 0.7112.CONCLUSION Machine learning models predict the risk of digoxin blood concentration in human serum beyond the upper limit of the treatment window,and innovatively interpret it using the SHAP algorithm.The RF model has the best prediction performance,which provides reference for identifying patients with high risk of digoxin poisoning.

关 键 词：地高辛 治疗药物浓度监测 随机森林模型 机器学习 

分 类 号：R972.1[医药卫生—药品]