基于SEER数据库异时双原发肺癌预后模型的构建和验证

Development and internal validation of a prognostic model of metachronous dual primary lung cancer based on SEER database

摘要:

目的： 运用回顾性研究的分析方法，探究异时双原发肺癌（metachronous dual primary lung cancer，mDPLC）的独立预后因素，并建立列线图预后模型。方法： 通过美国癌症监测、流行病学和结果（Surveillance，Epidemiology and End Results，SEER）数据库获得mDPLC患者的临床数据。将数据分为训练集和验证集进行建模和验证。采用单因素和多因素Cox回归分析确定训练集患者的独立预后因素，并纳入列线图对患者生存时间进行预测。通过C-指数、校准图、受试者工作特征（receiver operating characteristic，ROC）曲线、决策曲线分析（decision curve analyses，DCA）和综合鉴别改进（integrated discrimination improvement，IDI）评分对预后模型的准确性和可靠性进行评价。结果： 共有610例mDPLC患者纳入训练集，260例mDPLC患者纳入验证集。Cox回归分析表明，年龄（59~76岁：HR=1.453，95%CI=1.007~2.096，P=0.046；≥77岁：HR=1.953，95%CI=1.303~2.926，P<0.001）、性别（女性：HR=0.669，95%CI=0.529~0.845，P=0.001）、病理类型（腺癌+鳞癌：HR=2.251，95%CI=1.583~3.200，P<0.001）、TNM 分期（Ⅲ：HR=1.611，95%CI=1.126~2.305，P=0.009；Ⅳ：HR=2.443，95%CI=1.713~3.486，P<0.001）、淋巴结转移（是：HR=1.653，95%CI=1.199 ~2.280，P=0.002）、手术（1次：HR=1.431，95%CI=1.110~1.844，P=0.006；0次：HR=1.845，95%CI=1.183~2.878，P=0.007）和化疗（是+否：HR=0.603，95%CI=0.433~0.842，P=0.003）是独立预后因素。训练集和验证集的C-指数分别为0.711（95%CI=0.696~0.726，P<0.05），0.677（95%CI=0.655~0.699，P<0.05）。训练集中ROC曲线的3年生存率和5年生存率AUC值分别为0.756（95%CI=0.713~0.800，P<0.05）和0.785（95%CI=0.732~0.838，P<0.05），验证集中ROC曲线的3年生存率和5年生存率AUC值分别是0.695（95%CI=0.621~0.769，P<0.05）和0.711（95%CI=0.618~0.804，P<0.05）。训练集中3年和5年的IDI值分别为9.0%（P<0.001）和11.9%（P<0.001），验证集中3年和5年的IDI值分别为6.6%（P<0.001）和7.8%（P<0.001）。根据独立预后因素建立的模型经验证可准确预测患者预后，并且效果优于TNM分期。结论： 本研究建立了一个mDPLC预后预测模型，内部验证表明该模型具有较好的效能，可为mDPLC患者提供准确和个性化的生存预测。

Abstract:

Objective: To explore independent prognostic factors of metachronous dual primary lung cancer（mDPLC） by using retrospective analysis，and establish a prognostic nomogram model. Methods: Clinical information of mDPLC patients was collected from Surveillance，Epidemiology，and End Results（SEER） database. The dataset was divided into training and validation sets for modeling and validation. Univariate and multivariable Cox regression analyses were used to explore independent prognostic factors of training set，and the survival time of patients was predicted by nomogram. The accuracy and reliability of the prognostic model were evaluated by C-indexes，calibration plots，receiver operating characteristic（ROC） curves，decision curve analyses（DCA） and integrated discrimination improvement（IDI） scores. Results: A total of 610 mDPLC patients were included in the training set，and 260 mDPLC patients were included in the validation set. Cox regression analysis showed that age（59-76：HR=1.453，95%CI=1.007-2.096，P=0.046；≥77：HR=1.953，95%CI=1.303-2.926，P<0.001），gender（female HR=0.669，95%CI=0.529-0.845，P=0.001），pathological type（adenocarcinoma + squamous carcinoma HR=2.251，95%CI=1.583-3.200，P<0.001），TNM staging（Ⅲ：HR=1.611，95%CI=1.126-2.305，P=0.009；Ⅳ：HR=2.443，95%CI=1.713-3.486，P<0.001），lymph node metastasis（HR=1.653，95%CI=1.199-2.280，P=0.002），surgery（once：HR=1.431，95%CI=1.110-1.844，P=0.006；0：HR=1.845，95%CI=1.183-2.878，P=0.007） and chemotherapy（yes + no：HR=0.603，95%CI=0.433-0.842，P=0.003） were independent prognostic factors. The C-index of training set and verification set were 0.711（95%CI=0.696-0.726，P<0.05） and 0.677（95%CI=0.655-0.699，P<0.05），respectively. The AUC values of 3-year survival rate and 5-year survival rate of ROC curve in training set were 0.756（95%CI=0.713-0.800，P<0.05） and 0.785（95%CI=0.732-0.838，P<0.05），respectively，and the AUC values of the 3-year survival rate and 5-year survival rate of the ROC curve in validation set were 0.695（95%CI=0.621-0.769，P<0.05） and 0.711（95%CI=0.618-0.804，P<0.05）. IDI values of 3 and 5 years were 9.0%（P<0.001） and 11.9%（P<0.001） in the training set，and 6.6%（P<0.001） and 7.8%（P<0.001） in the validation set，respectively. The internal validation showed that the predictive effect of the novel prognostic model was accurate and reliable and was better than that of the model based on TNM system. Conclusion: This study has established a prognostic prediction model of mDPLC，and the internal validation has shown good efficiency，which could provide accurate and personalized survival prediction for mDPLC patients.