乳腺癌患者预后评估模型的构建

Construction of prognostic evaluation model for patients with breast cancer

目的筛选影响乳腺癌患者预后的预测因素,构建评价乳腺癌患者预后的模型。方法SEER数据库获得2010—2015年间119779名乳腺癌患者的临床和随访数据,将所有病例随机分为70%(83854名)的训练集以及30%(35925名)的验证集。利用Lasso回归筛选出对预测乳腺癌患者预后有意义的预测因子,并构建Cox比例风险模型,利用受试者工作曲线(ROC曲线)分别在训练集以及验证集中评价模型的鉴别能力,并与单纯纳入TNM预后分期因素(原发肿瘤大小、区域淋巴结以及远处转移)以预测患者预后的准确性进行对比。利用校准图评价模型的准确性。结果年龄、种族、肿瘤分级、肿瘤大小、分子分型、手术治疗、淋巴结状态、远处转移是能够预测患者生存的独立预测因子。在训练集中,1年、3年、5年ROC曲线下面积分别为0.840、0.812、0.797,在验证集中为0.712、0.707、0.684。当单纯纳入TNM预后因素时为0.738、0.714、0.693。1年、3年、5年校准图显示了模型具有良好的预测准确性。结论与单纯依靠传统的TNM预后因素预测预后相比,该模型能够更准确的量化、评估患者的预后,可为医生、病人以及医疗政策制定者提供直观的信息。

Objective To screen the predictive factors affecting the prognosis of patients with breast cancer and to establish a model to evaluate the prognosis of patients with breast cancer.Methods The clinical and follow-up data of 119779 breast cancer patients from 2010 to 2015 were obtained from SEER database.All cases were randomly divided into 70%(83854)training set and 30%(35925)validation set.Lasso regression was used to screen the predictive factors for the prognosis of patients with breast cancer,and a Cox proportional hazard model was constructed.The receiver operating characteristic(ROC)curve was used to evaluate the differential ability of the model in the training set and verification set,respectively,and to compare with TNM prognostic staging factors(primary tumor size,regional lymph nodes and distant metastasis)in the accuracy to predict the prognosis of patients.The calibration map was used to evaluate the accuracy of the model.Results The age,race,tumor grade,tumor size,molecular classification,surgical treatment,status of lymph node and distant metastasis were independent predictors of survival.The areas under the ROC curve of the subjects in one year,three years and five years were 0.840,0.812 and 0.797 in the training set and 0.712,0.707 and 0.684 in the verification set,respectively.When TNM prognostic staging factors were included alone,the values were 0.738,0.714 and 0.693 respectively.One-year,three-year and five-year calibration maps showed that the model had a good prediction accuracy.Conclusion Compared with relying solely on the traditional TNM prognostic staging factors to predict prognosis,this model can quantify and evaluate the prognosis of patients more accurately and provide intuitive and rational information for doctors,patients and medical policy makers.