无删失生存数据Wilcoxon秩和检验与logrank检验的比较

摘要

目的 比较Wilcoxon秩和检验与logrank检验处理无删失生存数据的统计性能.方法 采用MonteCarlo模拟方法,应用SAS9.1.3软件编程,比较两种检验方法的Ⅰ型错误率和检验效能.结果 小样本情形(n≤50)下,Wilcoxon秩和检验的Ⅰ型错误率较logrank检验更接近检验水准0.05,而检验效能与logrank检验相差不大,后者检验效能的提高是以扩大Ⅰ型错误率为代价的.结论 对于无删失生存数据的比较,尤其是小样本情形(n≤50)下,建议采用Wilcoxon秩和检验.%Objective To compare the Statistical properties of the Wilcoxon rank sum test with the logrank test on uncensored survival data. Methods By means of Monte Carlo Simulation and SAS9. 1. 3 program , contrasting type I error and Power of Wilcoxon rank sum test to that of the logrank test. Results In small sample size( n ≤50) , the type I error of the Wilcoxon rank sum test is closer to the nominal size 0. 05 than the logrank test, while the Power of two tests are out of difference. The logrank test improves the power at the cost of enlarging the ratio of Type I error. Conclusion The Wilcoxon rank sum test is recommended to analyze the uncensored survival data in small sample size(n≤50),