基于机器学习的甲状腺手术后恶心呕吐因素分析  

Construction of prediction models for postoperative nausea and vomiting in patients with thyroid surgery based on machine learning

作　　者：邱情 万重阳 路伟娜 沈徐 QIU Qing;WAN Chongyang;LU Weina;SHEN Xu(Department of Anesthesiology,Affiliated Hospital of Jiaxing University,Jiaxing 314000,China)

机构地区：[1]嘉兴市第一医院麻醉科,314000

出　　处：《浙江医学》2024年第23期2498-2502,I0006,共6页Zhejiang Medical Journal

基　　金：嘉兴市医学重点专科—麻醉学(2023-ZC-001)。

摘　　要：目的筛选甲状腺术后恶心、呕吐的相关因素,并建立预测模型评估系统。方法收集2022年9月至2023年3月在嘉兴市第一医院择期行甲状腺手术的196例患者资料,采用Python编程语言构建一个全面的模型及预测评估系统。在模型构建过程中,采用支持向量机、决策树、随机森林、逻辑回归和自适应增强(Adaboost)等5种机器学习方法。为确保模型的稳健性,从数据集中随机选取了90%的数据组成训练集,并将剩余10%的数据作为验证集进行验证。采用十折交叉验证方法对模型准确度进行评估。结果196例患者中术后恶心、呕吐组73例,无恶心、呕吐组123例。单因素分析结果显示,两组患者在性别、吸烟史、饮酒史、晕动病史或恶心、呕吐史、液体输入量、舒芬太尼及瑞芬太尼用量之间比较差异均有统计学意义(均P<0.05)。5种机器学习算法构建了预测模型,发现Adaboost算法构建的模型预测甲状腺术后恶心、呕吐准确度最佳(Mean AUC=0.74)。结论Adaboost模型预测甲状腺术后恶心、呕吐准确度最佳,可开发软件应用于临床实践,根据预测结果采取有针对性的防治措施,有效预防术后恶心、呕吐的发生。Objective To construct machine learning models for predicting postoperative nausea and vomiting(PONV)in patients undergoing thyroid surgery.Methods Clinical data of 196 patients who underwent elective thyroid surgery at Affiliated Hospital of Jiaxing University from September 2022 to March 2023 were collected,including 73 patients with postoperative nausea and vomiting(PONV group)and 123 patients without PONV(non-PONV group).A comprehensive model and predictive evaluation system were constructed using the Python programming language.Five classic machine learning methods were used,including support vector machine,decision tree,random forest,logistic regression,and adaptive boost(Adaboost).To ensure the robustness of the model,90%of the data were randomly selected as the training set,and 10%of the data as the validation set.To further evaluate the performance of the model,ten-fold cross-validation was used to comprehensively examine the accuracy of the model.Results Univariate analysis showed that there were significant differences in gender,smoking history,drinking history,motion sickness history,fluid intake,sufentanil and remifentanil dosage between the PONV amd non-PONV groups(all P<0.05).Among prediction models constructed by five algorithms of machine learning,the model based on Adaboost algorithm was the most accurate one to predict PONV in thyroid surgery with a mean AUC of 0.74.Conclusion The Adaboost model is the most accurate model for predicting postoperative nausea and vomiting in thyroid surgery,and can be developed into software for clinical practice.Based on the prediction results,targeted prevention and treatment measures can be taken to effectively prevent postoperative nausea and vomiting.

关 键 词：机器学习 甲状腺手术 术后恶心、呕吐 

分 类 号：R614[医药卫生—麻醉学]