Log rank检验的功效

目的:L og rank检验是生存资料比较的标准方法,但无与之匹配的样本量测定方法。论述了这种检验的功效,为样本量研究提供依据。方法:由L ach in-Fou lkes法计算期望功效作为参照,回顾L og rank检验的3种形式,按M on te C arlo方法分别计算其观测功效,然后作对比分析。结果:所得观测功效在多数试验集均低于期望功效。与上半部相比,寿命表下半部期望和观测功效均较低。所得观测功效在不同终检水平或不同生存分布各不相同。结论:L ach in-Fou lkes法产生的样本量偏小,不能满足L og rank检验的预定功效。L og rank检验所需样本量因终检水平、生存时间或生存分布而异,L ach in-Fou lkes法无视这些事实,无法作出切合实际的测定,因此必须寻求与这种检验匹配的样本量测定方法。

The Power of Log Rank Test

Objective: Log rank test is the standard method in comparing survival data but there is no matched merthod for sample size drterminations.This paper concerns the power of the test so as to provide access to the relevant sample size determinations.Methods: The Lachin-Foulkes procedure is taken to calculate the expected power.The three forms of log rank test are reviewed.The Monte Carlo method is applied to calculate the observed power for the three forms.The expected power is taken as a reference for comparative analyses.Results: The observed power derived from the most sets of experiments is lower than the expected power.Both the expected power and the observed power are lower in the lower half of life tables than those in the upper half.The observed power varies with censoring levels or survival distribujtions.Conclusion: The sample sizes determinated by the Lachin-Foulkes procedure are biaed to small and not enogh to meet with the prescribed power of log rank test.The Lachin-Foulkes procedure is unable to provide a sensible determiknation since it disregards such facts that the sample sizes reqired for log rank test varies with censoring levels,survival times,or survival distributons.Thus approaches have to be explored for the relevant sample size determininations matched with the test.