COX联合XGBoost的原发睾丸弥漫大B细胞淋巴瘤预后模型构建与验证  

Construction and verification of prognostic model of primary testicular diffuse large B-cell lymphoma with COX and XG Boost

作　　者：秦佳佳 陈曦 沈子园 朱晓晓 桑威 金英良 QIN Jia-jia;CHEN Xi;SHEN Zi-yuan;ZHU Xiao-xiao;SANG Wei;JIN Ying-liang(Department of Health Statistics,School of Public Health,Xuzhou Medical University,Xuzhou,Jiangsu 221004,China;不详)

机构地区：[1]徐州医科大学公共卫生学院卫生统计学教研室,江苏徐州221004 [2]安徽医科大学公共卫生学院卫生统计学教研室 [3]徐州医科大学附属医院血液科

出　　处：《现代预防医学》2023年第13期2326-2331,共6页Modern Preventive Medicine

基　　金：江苏省科技厅社会发展重点项目(BE2019638);徐州市科技计划社会发展重点研发项目(KC22128)。

摘　　要：目的原发睾丸弥漫大B细胞淋巴瘤全球发病率逐年增长,且因为血睾屏障的存在,死亡率高,目前尚缺乏多中心的大型回顾性分析数据进行疾病的预后研究。方法从SEER数据库的22个中心机构获取2000—2019年期间,诊断为原发睾丸弥漫大B细胞淋巴瘤的患者资料进行回顾性分析,共纳入816例患者,按7:3的比例随机分为建模组和验证组。采用多变量COX逐步回归联合XGBoost机器学习模型筛选预后特征变量,将二者筛选出的共同预后因子构建列线图并对其重要度进行排序;时间依赖性ROC曲线验证模型的预测性能,校准曲线确定模型的一致性,决策曲线(decision curve analysis,DCA)评估模型的临床实用性与有效性。结果COX回归结合XGBoost筛选出的共同预后因子为:年龄、手术、放疗、双侧发病、化疗、系统治疗、Ann Arbor分期;将其按照重要度排序,由大到小依次为:年龄、系统治疗、Ann Arbor分期、手术、双侧发病、化疗、放疗。基于联合模型筛选出的共同预后因子构建列线图,ROC曲线、校准曲线和DCA曲线显示模型具有良好的预测性能和临床实用性。结论本研究构建的机器学习模型结合回归模型能准确预测原发睾丸DLBCL患者的预后,为临床的精准诊疗提供科学依据。Objective The global incidence of primary testicular diffuse large B-cell lymphoma is increasing year by year,and because of the existence of blood-testis barrier and high mortality,there is still a lack of multicenter large-scale retrospective analysis data to study the prognosis of the disease.Methods The data of patients diagnosed as primary testicular diffuse large B-cell lymphoma from 2000 to 2019 were collected from 22 central institutions in SEER database.A total of 816 patients were randomly divided into model group and verification group according to the ratio of 7:3.Multivariate COX stepwise regression combined with XG Boost machine learning model was used to screen the prognostic characteristic variables,and the common prognostic factors selected by the two were used to construct a line chart and rank their importance.Then the time-dependent ROC curve was used to verify the predictive performance of the model,the calibration curve to determine the consistency of the model,and the clinical practicability and effectiveness of the decision curve analysis(DCA)evaluation model.Results The common prognostic factors screened by COX regression and XG boost were age,operation,radiotherapy,bilateral disease,chemotherapy,systematic therapy,and Ann Arbor stage,and then they were ranked according to the degree of importance:age,systematic treatment,Ann Arbor stage,operation,bilateral disease,chemotherapy,and radiotherapy.The nomogram was constructed based on the common prognostic factors screened by the joint model,and the ROC curve,calibration curve,and DCA curve showed that the model had good predictive performance and clinical practicability.Conclusion The machine learning model combined with regression model constructed in this study can accurately predict the prognosis of patients with primary testicular diffuse large B-cell lymphoma(DLBCL)and provide scientific basis for clinical accurate diagnosis and treatment.

关 键 词：原发睾丸弥漫大B细胞淋巴瘤 机器学习模型 预后 

分 类 号：R733.4[医药卫生—肿瘤]