基于SEER数据库构建Ⅳ期肝细胞癌患者癌症特异性生存期的网络预测列线图模型

Construction of web-based prediction nomogram models for cancer-specific survival in patients at stage IV of hepatocellular carcinoma depending on SEER database

目的:肝细胞癌(hepatocellular carcinoma,HCC)的预后涉及多个临床因素。尽管目前针对多个临床因素的列线图模型在早期及局部晚期HCC中已有报道,但是鲜有完整有效的IV期HCC患者预后列线图模型的报道。本研究旨在创建预测IV期HCC患者癌症特异性生存期(cancer-specific survival,CSS)的列线图,开发网络预测列线图模型,用于预测患者预后及指导个体化治疗。方法:从监测、流行病学和最终结果(Surveillance,Epidemiology,and End Results,SEER)数据库中收集2010年1月至2015年12月IV期HCC患者的临床病理信息,根据有无远处转移将IV期HCC患者分为IVA(无远处转移)和IVB(有远处转移)期2个亚组,然后将IVA和IVB期患者均按照7:3的比例随机分配到训练队列或验证队列。采用单因素和多因素Cox回归分析训练队列中显著影响CSS的独立危险因素,并根据相关的独立危险因素分别构建针对IVA期和IVB期HCC患者的列线图。通过受试者操作特征(receiveroperator characteristic,ROC)曲线和校准曲线来评估2个列线图的准确性和辨别能力。此外,利用R软件开发分别针对IVA期和IVB期HCC患者的网络列线图模型。采用决策分析曲线(decision analysis curve,DCA)评估网络列线图的临床预测效果。结果:本研究共纳入3060例患者,其中IVA期883例,IVB期2177例。多因素分析结果显示:肿瘤大小、甲胚蛋白(alpha-fetoprotein,AFP)、T分期、组织学分级、手术、放射治疗、化学治疗是IVA期HCC患者的独立预后因素;肿瘤大小、AFP、T分期、N分期、组织学分级、肺转移、手术、放射治疗和化学治疗是IVB期HCC患者的独立预后因素。在IVA期患者中,训练队列的3、6、9、12、15和18个月ROC曲线下面积分别为0.823、0.800、0.772、0.784、0.784和0.786;验证队列的3、6、9、12、15和18个月ROC曲线下面积分别为0.793、0.764、0.739、0.773、0.798和0.799。在IVB期患者中,训练队列的3、6、9和12个月ROC曲线下面积分别为0.756、0.750、0.755和0.743;验证队列的3、6、9和12个月ROC曲线下面积分别为0.744、0.747、0.775和0.779;表明列线图具有出色的预测能力。校准曲线显示预测结果与实际观察结果吻合良好。结论:本研究开发并验证的基于SEER数据库的IVA和IVB期HCC患者CSS的预测列线图模型,可以被临床医生用来预测这些患者的预后、实施个体化治疗和随访管理。

Objective:Hepatocellular carcinoma(HCC)prognosis involves multiple clinical factors.Although nomogram models targeting various clinical factors have been reported in early and locally advanced HCC,there are currently few studies on complete and effective prognostic nomogram models for stage IV HCC patients.This study aims to creat nomograms for cancer-specific survival(CSS)in patients at stage IV of HCC and developing a web predictive nomogram model to predict patient prognosis and guide individualized treatment.Methods:Clinicopathological information on stage IV of HCC between January,2010 and December,2015 was collected from the Surveillance,Epidemiology,and End Results(SEER)database.The patients at stage IV of HCC were categorized into IVA(without distant metastases)and IVB(with distant metastases)subgroups based on the presence of distant metastasis,and then the patients from both IVA and IVB subgroups were randomly divided into the training and validation cohorts in a 7꞉3 ratio.Univariate and multivariate Cox regression analyses were used to analyze the independent risk factors that significantly affected CSS in the training cohort,and constructed nomogram models separately for stage IVA and stage IVB patients based on relevant independent risk factors.Two nomogram’s accuracy and discrimination were evaluated by receiver operator characteristic(ROC)curves and calibration curves.Furthermore,web-based nomogram models were developed specifically for stage IVA and stage IVB HCC patients by R software.A decision analysis curve(DCA)was used to evaluate the clinical utility of the web-based nomogram models.Results:A total of 3060 patients were included in this study,of which 883 were in stage IVA,and 2177 were in stage IVB.Based on multivariate analysis results,tumor size,alpha-fetoprotein(AFP),T stage,histological grade,surgery,radiotherapy,and chemotherapy were independent prognostic factors for patients with stage IVA of HCC;and tumor size,AFP,T stage,N stage,histological grade,lung metastasis,surgery,radiotherapy,and chemotherapy were independent prognostic factors for patients with stage IVB HCC.In stage IVA patients,the 3-,6-,9-,12-,15-,and 18-month areas under the ROC curves for the training cohort were 0.823,0.800,0.772,0.784,0.784,and 0.786,respectively;and the 3-,6-,9-,12-,15-,and 18-month areas under the ROC curves for the validation cohort were 0.793,0.764,0.739,0.773,0.798,and 0.799,respectively.In stage IVB patients,the 3-,6-,9-,and 12-month areas under the ROC curves for the training cohort were 0.756,0.750,0.755,and 0.743,respectively;and the 3-,6-,9-,and 12-month areas under the ROC curves for the validation cohort were 0.744,0.747,0.775,and 0.779,respectively;showing that the nomograms had an excellent predictive ability.The calibration curves showed a good consistency between the predictions and actual observations.Conclusion:Predictive nomogram models for CSS in stage IVA and IVB HCC patients are developed and validated based on the SEER database,which might be used for clinicians to predict the prognosis,implement individualized treatment,and follow up those patients.