两阶段方法预测分析美罗培南血药浓度  

Prediction and analysis of the plasma concentration of meropenem by a two-stage method

作　　者：宋婷 揭琼 张倩[2] 胡真 SONG Ting;JIE Qiong;ZHANG Qian;HU Zhen(School of Mathematics,Hohai University,Nanjing 210001,China;Department of Pharmacy,Nanjing First Hospital,Nanjing 210006,China)

机构地区：[1]河海大学数学学院,江苏南京210001 [2]南京市第一医院药学部,江苏南京210006

出　　处：《应用科技》2024年第5期107-113,共7页Applied Science and Technology

基　　金：中央高校基本科研业务费项目(B200202002)。

摘　　要：为提升美罗培南血药浓度预测的精度、提供用药指导,本文使用两阶段预测方法,基于机器学习算法建立美罗培南血药浓度预测模型。本文所用199条数据来源于南京市第一医院2019~2020年接受美罗培南治疗的重症感染患者的血药浓度以及影响因素指标。应用修改的卡方算法离散血药浓度,并在2个预测阶段均比较多种机器学习算法,选取13个显著特征建立模型。在偏好预测阶段,自适应增强(adaptive boosting,Adaboost)算法表现最优,准确度达到88.33%;在数值预测阶段,随机森林(random forest,RF)、装袋算法(bootstrap aggregaing,Bagging)表现最优;最终建立Ada-RF与Ada-Bag模型,相对误差在30%内的比例达到55%,取得较好的预测效果。两阶段方法为美罗培南血药浓度预测提供了新思路,可辅助制定个性化的用药方案。To improve the prediction accuracy of the plasma concentration of meropenem and provide medication guidance,a two-stage prediction method is applied to the establishment of a prediction model for the plasma concentration of meropenem based on machine learning algorithms.The 199 pieces of data used in this paper come from the plasma concentration and influencing factors of critically ill patients receiving meropenem treatment in Nanjing First Hospital from 2019 to 2020.The modified chi-square algorithm is applied to discretize the plasma concentration,and multiple machine learning algorithms are compared in the two prediction stages.Thirteen significant features are selected to establish a model.In the preference prediction stage,the adaptive boosting(Adaboost)algorithm performs best,with an accuracy of 88.33%.In the numerical prediction stage,random forest(RF)and bootstrap aggregating(Bagging)perform best.Finally,the Ada-RF and Ada-Bag models are established,and the proportion of relative errors within 30%reaches 55%,achieving good prediction results.Two-stage method provides a new idea for predicting the plasma concentration of meropenem and assists in formulating personalized medication plans.

关 键 词：两阶段方法 机器学习 血药浓度 美罗培南 卡方离散化 浓度预测模型 自适应增强 随机森林 装袋算法 

分 类 号：O29[理学—应用数学]