基于机器学习的顺铂化疗引起肿瘤患者持续期恶心呕吐预测  

Prediction of Nausea and Vomiting Duration in Patients with Tumours Induced by Cisplatin Chemotherapy Based on Machine Learning

作　　者：张靖悦 蓝高爽[1] 杨翀[3] 孙银娟 钟殿胜[2] 张琳琳[2] 袁恒杰[2] ZHANG Jing-yue;LAN Gao-shuang;YANG Chong;SUN Yin-juan;ZHONG Dian-sheng;ZHANG Lin-lin;YUAN Heng-jie(Department of Pharmacy,Tianjin Medical University General Hospital,Tianjin 300052,China;Department of Medical Oncology,Tianjin Medical University General Hospital,Tianjin 300052,China;Department of Pharmacy,Tianjin Huanhu Hospital,Tianjin 300350,China)

机构地区：[1]天津医科大学总医院,药剂科,天津300052 [2]天津医科大学总医院肿瘤科,天津300052 [3]天津市环湖医院药剂科,天津300350

出　　处：《中国药学杂志》2023年第11期1031-1036,共6页Chinese Pharmaceutical Journal

基　　金：国家自然科学基金青年项目资助(81102447)。

摘　　要：目的基于机器学习建立顺铂持续期化疗引起的恶心呕吐(chemotherapy-induced nausea and vomiting,CINV)预测模型,为临床止吐药物的选择提供用药依据。方法回顾收集2018年7月~2022年10月天津医科大学总医院肿瘤科接收以顺铂为基础化疗方案的74名患者的临床信息作为变量,采用主成分分析(principal component analysis,PCA)降低变量维度,应用交叉验证重复训练,分别建立支持向量机(support vector machine,SVM)、随机森林(random forest,RF)、K最近邻(K near neighbor,KNN)、朴素贝叶斯(naive bayes,NB)及决策树(decision tree,DT)预测模型,比较5种模型对顺铂持续期CINV预测的准确性。结果采用PCA将41个变量降维获得4个主成分,NB的准确度、AUC值和灵敏度最大,分别为84.21%、0.9167%、100%。RF的F1值最大,NB次之,分别为86.96%、85.71%。结论基于机器学习建立的5种预测模型,NB模型性能最佳,可为顺铂CINV的预防提供参考。OBJECTIVE To provide a dosing basis for clinical antiemetic drug selection by establishing cisplatin chemotherapy-induced nausea and vomiting(CINV)duration prediction model based on machine learning.METHODS The cisplatin CINV duration model was established using support vector machine(SVM),decision tree(DT),naive bayes(NB),random forest(RF),and K near neighbor(KNN)machine learning algorithms with the clinical information of 74 patients who underwent cisplatin-based chemotherapy in our hospital from July 2018 to October 2022 served as its variables detected by principal component analysis(PCA).RESULTS The PCA reduced the dimension of 41 variables,and finally obtained 4 principal components.NB had the highest accuracy,AUC value and sensitivity of 84.21%,0.9167%and 100%respectively.RF had the largest F1 value,followed by NB with 86.96%and 85.71%respectively.CONCLUSION The five predictive models established by machine learning,with the NB model performing best,can inform the prevention of cisplatin CINV.

关 键 词：顺铂 机器学习 恶心 呕吐 主成分分析 

分 类 号：R911[医药卫生—药学]