基于SEER数据库的老年乳腺癌患者预后的影响因素研究

A Study of the Influencing Factors of Prognosis in Elderly Breast Cancer Patients Based on the SEER Database

摘要: 在全球老龄化加剧的背景下，我们基于生存分析的方法对老年乳腺癌患者预后的影响因素进行分析，能够更好地确定风险的分层并制定合适的治疗方式，这对预测患者的生存概率以及降低死亡风险有着现实的意义和应用价值。本文主要分析了来自美国癌症数据库SEER的数据。首先用K-M估计刻画了总体、年龄、性别、种族、肿瘤学等级、T分期、M分期、N分期、是否接收放疗、是否接受化疗的生存曲线。其次通过单因素Cox模型分析确定独立的预后因素，进而通过逐步回归的方式构建多因素Cox比例风险回归模型，最终确定的影响变量为：年龄、M分期、是否接受放疗；同样在考虑竞争事件的情况下，首先应用Nelson-Aalen累计风险曲线(CIF)分析各个变量不同组间的风险函数的差异性，其次建立单因素竞争风险模型确定影响生存时间和结局的独立预后变量，确定的独立预后因素为：年龄、M分期、是否接受放疗，据此构建多因素竞争风险模型通过比较这两个预测模型的一致性指标(C-index)、校准曲线来判断模型预测的准确性，结果显示竞争风险模型相比较好，最后通过构建列线图的方式可视化生存概率。

Abstract: In the context of increasing global aging, our survival-based method analyzes the influencing factors of prognosis of elderly breast cancer patients, which can better determine the risk stratification and formulate appropriate treatment methods, which has practical significance and application value for predicting the survival probability of patients and reducing the risk of death. This paper mainly analyzes the data from SEER, an American cancer database. First, the survival curves of population, age, sex, race, oncology grade, T stage, M stage, N stage, whether to receive radiotherapy and whether to receive chemotherapy were described by K-M estimation. Secondly, independent prognostic factors were determined by single factor Cox model analysis, and then a multi factor Cox proportional hazards regression model was constructed by stepwise regression. The final influencing variables were: Age, M stage, whether to receive radiotherapy; Similarly, in the case of competitive events, firstly, Nelson-Aalen cumulative risk curve (CIF) is used to analyze the difference of risk function between different groups of each variable. Secondly, a single factor competitive risk model is established to determine the independent prognostic variables that affect survival time and outcome. The independent prognostic factors determined are: Age, M stage, whether to receive radiotherapy or not, based on this, a multi factor competitive risk model is constructed. The accuracy of the model prediction is judged by comparing the consistency index (C-index) and calibration curve of the two prediction models. The results show that the competitive risk model is relatively good. Finally, the survival probability is visualized by constructing a nomogram.