摘要
目的探讨多因素Logistic回归模型和XGBoost模型对舌癌患者放疗期间发生口腔感染的预测价值。方法选取2003年1月—2022年12月在新疆医科大学第一附属医院就诊的431例放疗的舌癌患者,随机分为训练组288例和预测组143例。比较多因素Logistic回归模型和XGBoost模型对舌癌患者放疗期间发生口腔感染的预测效能。结果多因素Logistic回归模型结果显示,年龄[O^R=3.250(95%CI:1.476,7.634)],肿瘤分期[O^R=2.941(95%CI:1.248,7.613)],口腔环境[O^R=0.210(95%CI:0.079,0.502)],是否手术[O^R=0.285(95%CI:0.113,0.663)],血红蛋白[O^R=0.323(95%CI:0.139,0.712)],血清白蛋白[O^R=0.353(95%CI:0.148,0.851)]是放疗期间发生口腔感染的独立预测因素。XGBoost模型结果显示,口腔环境、手术、肿瘤分期、血清白蛋白、年龄、同步化疗、红细胞计数、血红蛋白、中性粒细胞计数为重要性指标。多因素Logistic回归模型和XGBoost模型受试者工作特征曲线下面积分别为0.830、0.835,两者比较,差异无统计学意义(P>0.05);敏感性分别为88.24%(95%CI:0.729,1.000)、82.35%(95%CI:0.642,1.000);特异性分别为68.25%(95%CI:0.601,0.764)、69.84%(95%CI:0.627,0.786)。结论多因素Logistic回归模型和XGBoost模型对舌癌患者放疗期间发生口腔感染的预测均有意义,两者预测效能相当。建立模型有助于筛选出口腔感染的高危人群,及早采取预防措施,降低口腔感染发生风险。
Objective This study aimed to explore the predictive value of multifactor logistic regression and XGBoost models for oral infections during radiotherapy in tongue cancer patients.Methods A total of 431 tongue cancer patients receiving radiotherapy at the First Affiliated Hospital of Xinjiang Medical University from January 2003 to December 2022 were randomly divided into a training group(n=288)and a prediction group(n=143).The predictive performance of multifactor logistic regression and XGBoost models for oral infections during radiotherapy in tongue cancer patients was compared.Results The results of the multifactor logistic regression model showed that age[O^R=3.250(95%CI:1.476,7.634)],tumor stage[O^R=2.941(95%CI:1.248,7.613)],oral environment[O^R=0.210(95%CI:0.079,0.502)],surgery status[O^R=0.285(95%CI:0.113,0.663)],hemoglobin[O^R=0.323(95%CI:0.139,0.712)],and serum albumin[O^R=0.353(95%CI:0.148,0.851)]were independent predictors of oral infections during radiotherapy.The XGBoost model identified oral environment,surgery status,tumor stage,serum albumin,age,concurrent chemotherapy,red blood cell count,hemoglobin,and neutrophil count as important predictors.The area under the receiver operating characteristic curve(AUC)for the multifactor logistic regression model and the XGBoost model were 0.830 and 0.835,respectively,with no statistically significant difference between them(P>0.05).Sensitivity was 88.24%(95%CI:0.729,1.000)and 82.35%(95%CI:0.642,1.000),while specificity was 68.25%(95%CI:0.601,0.764)and 69.84%(95%CI:0.627,0.786)for the multifactor logistic regression and XGBoost models,respectively.Conclusion Both multifactor logistic regression and XGBoost models have significant predictive value for oral infections during radiotherapy in tongue cancer patients,with comparable predictive performance.Establishing these models can help identify high-risk individuals for oral infections,enabling early preventive measures and reducing the risk of oral infections.
作者
张洋
赵化荣
刘攀
宿伟鹏
龚忠诚
Zhang Yang;Zhao Hua-rong;Liu Pan;Su Wei-peng;Gong Zhong-cheng(Oncology Center,The First Affiliated Hospital of Xinjiang Medical University,Urumqi,Xinjiang 830054,China;Oncology Department of Oral&Maxillofacil Surgery,The First Affiliated Hospital of Xinjiang Medical University,Urumqi,Xinjiang 830054,China)
出处
《中国现代医学杂志》
CAS
北大核心
2023年第19期66-73,共8页
China Journal of Modern Medicine
基金
省部共建国家重点实验室(No:SKL-HIDCA 2021-47,SKL-HIDCA 2021-6,SKL-HIDCA 2022-44)。
