基于机器学习算法的利奈唑胺相关血小板减少预测  

Prediction of Linezolid-induced Thrombocytopenia Based on Machine Learning Algorithm

作　　者：廖茹 崔祎 程晓亮[1] 王丰 李厚丽 董海燕[1] LIAO Ru;CUI Yi;CHENG Xiaoliang;WANG Feng;LI Houli;DONG Haiyan(Department of Pharmacy,the First Affiliated Hospital of Xi’an Jiaotong University,Xi’an 710061,China;Department of Clinical Pharmacy,Shaanxi Provincial Cancer Hospital,Xi’an 710061,China)

机构地区：[1]西安交通大学第一附属医院药学部,西安710061 [2]陕西省肿瘤医院临床药学部,西安710061

出　　处：《医药导报》2025年第4期676-681,共6页Herald of Medicine

基　　金：国家自然科学基金资助项目(82173898)。

摘　　要：目的构建机器学习模型预测利奈唑胺相关血小板减少(LIT)。方法回顾性纳入某三甲医院2020年1月—2024年3月接受利奈唑胺治疗的患者198例。首先将患者分为LIT组和非LIT组,比较2组患者基本特征。然后选择在2组间存在显著差异的变量作为潜在危险因素构建LIT预测模型,包括Logistic回归、决策树以及随机森林模型,并对模型的预测性能进行评估和比较。结果共有52例(26.3%)发生LIT,单因素分析结果显示在发生和未发生LIT的患者中利奈唑胺谷浓度(trough concentration,C min)、患者基线血小板计数和肌酐清除率以及脑血管疾病、急性呼吸窘迫综合征和腹腔感染的发生率存在显著差异。在基于这些变量构建的3个LIT预测模型中,随机森林模型具有最高的预测准确性。最后基于随机森林模型进行变量重要性分析,结果显示C min、基线血小板计数和合并急性呼吸窘迫综合征对模型结果输出的贡献较大。结论随机森林模型能准确预测LIT的发生。利奈唑胺暴露水平较高和基线血小板较低的患者发生LIT的风险更高。Objective To construct machine learning models to predict the incidence of linezolid-induced thrombocytopenia(LIT).Methods A total of 198 patients treated with linezolid in a hospital between January 2020 and March 2024 were retrospectively included.Firstly,the patients were divided into LIT and non-LIT groups,and the basic characteristics of the two groups were compared.Then,the variables with significant differences between the two groups were selected as potential risk factors to construct models for predicting LIT,including Logistic regression,decision tree and random forest models,and the prediction performance of the models was evaluated and compared.Results There were 52(26.3%)patients developed LIT during the treatment.The univariate analysis showed significant differences in linezolid trough concentration(C min),baseline platelet counts and creatinine clearance,the incidence of cerebrovascular disease,acute respiratory distress syndrome,and abdominal infection in patients with and without LIT.Among the three models built based on these variables,the random forest model has the best predictive performance.The results of variable importance analysis based on random forest model showed that C min,baseline platelet count and combined with acute respiratory distress syndrome had higher importance scores.Conclusions The random forest model has high accuracy in predicting the occurrence of LIT,and the risk of LIT is higher in patients with higher levels of linezolid exposure and lower baseline platelets.

关 键 词：利奈唑胺 血小板减少 机器学习 决策树 随机森林 

分 类 号：R972.4[医药卫生—药品]