COX回归联合决策树建立宫颈癌预后的预测模型  

Prediction model of prognosis of cervical cancer based on COX regression and decision tree

作　　者：韦苑 饶群仙[3] 冯小燕 陈勍[3] 韩雪 沈天然 刘新霞 卢淮武[3] 陈青松 WEI Yuan;RAO Qun-xian;FENG Xiao-yan;CHEN Qing;HAN Xue;SHEN Tian-ran;LIU Xin-xia;LU Huai-wu;CHEN Qing-song(School of Public Health,Guangdong Pharmaceutical University,Guangzhou,Guangdong 510310,China;不详)

机构地区：[1]广东药科大学公共卫生学院,广东广州510310 [2]广东省公共卫生检测与评估工程技术研究中心,广东广州510310 [3]中山大学孙逸仙纪念医院妇科肿瘤专科,广东广州510120 [4]中山市第三人民医院,广东中山528403

出　　处：《现代预防医学》2022年第2期206-212,共7页Modern Preventive Medicine

基　　金：广东省教育厅基金项目(2019GCZX012)。

摘　　要：目的运用COX回归与决策树探讨宫颈癌患者预后的影响因素并建立预测模型。方法收集1 075例自2013—2019年入院治疗的宫颈癌患者的临床资料和随访数据。采用检验、COX回归模型探讨预后的影响因素,并借助决策树建立预测模型。结果多因素COX回归模型显示更晚的FIGO分期、非鳞癌或腺癌的病理类型、深层肌层浸润、赘生物直径≥4 cm均是宫颈癌预后的独立影响因素,宫旁阳性、阴道穹隆受累是肿瘤进展的独立影响因素,而脉管侵犯、尖锐湿疣是死亡的独立影响因素(P<0.05)。决策树结果显示肌层浸润和赘生物直径的影响最显著。进展及死亡模型AUC分别为0.698,0.745,正确分类预测百分比为89.9%、93.6%。结论肌层浸润、赘生物直径、宫旁阳性、阴道穹窿受累、病理类型、脉管侵犯、尖锐湿疣是宫颈癌患者预后的独立影响因素。COX回归联合决策树建立预测模型的方法可以联合两种模型优势,结果可视化,为临床评估提供参考。Objective To explore the influencing factors of the prognosis of cervical cancer patients using COX regression and decision tree to establish a prediction model.Methods Clinical data and follow-up data of 1 075 patients with cervical cancer who were hospitalized from 2013 to 2019 were collected.Chi-square test and COX regression were used to explore the influencing factors of prognosis,and the prediction model was established by decision tree.Results Multivariate COX regression model showed that the later Internation Federation of Gynecology and Obstetrics(FIGO) staging,pathologic type of non-squamous carcinoma or adenocarcinoma,deep muscular infiltrating,neoplasm ’s diameter ≥ 4(cm) were independent influencing factors for prognosis of cervical cancer.Positive parauterine,vaginal Fornix involvement were independent influencing factors of tumor progression,while vascular invasion and condyloma acuminatum were independent influencing factors of death(P<0.05).The results of decision tree showed that muscular infiltration and neoplasm’s diameter had the most significant effects.AUC of the progression and death models were 0.698 and 0.745,respectively,and the correct classification prediction percentages were 89.9% and 93.6%.Conclusion Muscular infiltrating,neoplasm’s diameter,positive parauterine,vaginal Fornix involvement,pathological type,vascular invasion,and condyloma acuminatum were independent factors affecting the prognosis of cervical cancer.The method of COX regression combined with decision tree to establish a prediction model can combine the advantages of two models,and the results can be visualized to provide a reference for clinical evaluation.

关 键 词：宫颈癌 决策树 COX回归 预测模型 

分 类 号：R737.33[医药卫生—肿瘤]