摘要
目的 探讨老年上尿路尿路上皮癌手术联合放疗效果和基于机器学习的生存预测的研究.方法 通过监测、流行病学和最终结果的数据库搜集数据.采用倾向性得分匹配两组信息.采用单因素和多因素Cox回归分析比较未放疗与放疗两组上尿路尿路上皮癌特异性生存和总生存.开发6种机器学习模型,建立预测放疗后5年肿瘤特异性生存(CSS)机器学习模型.结果 根据CSS单因素和多因素分析,未接受放疗的肿瘤特异性生存降低[95％置信区间(CI):1.151～1.618],年龄 80 岁以上(95％CI:1.052～1.278)、单身(95％CI:1.134～2.321)、离异(95％CI:1.255～1.877)、T3～4(95％CI:1.572～2.382)、N2～3 分期(95％CI:1.162～1.987),未接受化疗(95％CI:1.108～1.608)、浸润性尿路上皮癌(95％CI:1.082～2.185)和 N1(95％CI:1.229～2.282)特异性生存降低.相对于其他5种机器学习模型,极端梯度提升(XGBoost)模型最优,该模型的准确度为0.922,精度为0.923,灵敏度为0.931,F1得分为0.901,曲线下面积(AUC)为0.901.结论 通过筛选肿瘤特异性生存的相关因素,进一步选出术后放疗生存的最佳的机器学习预测模型.
Abstract
Objective To explore the effects and survival prediction of surgery combined with radi-ation therapy in elderly patients with upper urinary tract urothelial carcinoma,using machine learning ap-proaches.Methods Data were collected from the surveillance,epidemiology,and end results(SEER)database.Propensity score matching was utilized to balance information between groups.Univariate and multivariate Cox regression analyses compared cancer specific survival(CSS)and overall survival(OS)be-tween patients who did and did not receive radiation therapy.A total of 6 machine learning models were de-veloped,with extreme gradient boosting(XGBoost)identified as providing the best performance in predic-ting 5-year CSS for patients post-radiation therapy.Results Univariate and multivariate analyses showed decreased tumor-specific survival for patients not receiving radiation therapy[95％confidence interval(CI):1.151-1.618],with factors such as age over 80(95％CI:1.052-1.278),being single(95％CI:1.134-2.321),divorce(95％CI:1.255-1.877),T3-4 stage(95％CI:1.572-2.382),N2-3 stage(95％CI:1.162-1.987),not receiving chemotherapy(95％CI:1.108-1.608),invasive urothelial carci-noma(95％CI:1.082-2.185),and N1 stage(95％CI:1.229-2.282).The XGBoost model outper-formed 5 other machine learning models in predicting 5-year CSS,with an accuracy of 0.922,precision of 0.923,sensitivity of 0.931,Fl score of 0.901,and area under curve of 0.901.Conclusion By screen-ing factors related to tumor-specific survival,the best machine learning prediction model for postoperative radiotherapy survival is further selected.
基金项目
河南省医学科技攻关计划联合共建项目(LHGJ20230060)
河南省医学科技攻关计划联合共建项目(LHGJ20210054)
NETL
NSTL
万方数据