不同删失比例下AFT模型与Cox模型表现比较的模拟研究

发布时间：2018-02-27 10:46

  本文关键词： AFT模型 Cox模型 删失比例 表现 模拟研究　出处：《中国卫生统计》2017年04期 　论文类型：期刊论文

【摘要】：目的评价当生存数据出现不同删失比例的条件下,AFT模型和Cox模型的表现优劣。方法采用自编SAS宏程序模拟不同参数设置、不同比例均匀删失的Weibull分布,分别采用AFT模型和Cox模型对模拟产生的数据进行拟合,并从偏倚、准确性和覆盖程度三个指标对两个模型的表现进行评价。结果在分布参数和自变量效应值固定的情况下,AFT模型在参数估计偏倚和准确性两方面的表现却始终优于Cox模型。删失比例越大,AFT模型的表现相对而言越优异。而在覆盖程度方面,对于相同参数设置及删失比例的模拟数据,两个模型的表现相似。结论 AFT模型的总体表现优于Cox模型,当删失生存数据同时满足两类模型的应用条件时,应该优先选择偏倚更小、准确性更高的AFT模型进行分析。
[Abstract]:Objective to evaluate the performance of the Cox model and the Cox model under the condition of different censored ratios of survival data. Methods SAS macro program was used to simulate the distribution of Weibull with different parameters and different proportion evenly censored. AFT model and Cox model were used to fit the simulated data, and the bias was analyzed. The performance of the two models was evaluated by three indexes of accuracy and coverage. Results the performance of the AFT model was always better than that of the parameter estimation bias and accuracy when the distribution parameters and the effect values of independent variables were fixed. Cox model. The larger the censored ratio is, the better the performance of the model is. However, in terms of coverage, For the simulated data of the same parameter setting and deletion ratio, the performance of the two models is similar. Conclusion the overall performance of the AFT model is better than that of the Cox model. When the censored survival data satisfy the application conditions of the two models, AFT models with less bias and higher accuracy should be preferred for analysis.
【作者单位】： 昆明医科大学公共卫生学院流行病与卫生统计学系;
【基金】：国家自然科学基金(81460519) 云南省自然科学基金(2013FZ064)
【分类号】：O212.1;R181
，

本文编号：1542344