Strategies for subgroup analysis in clinical trials.
Strategies for subgroup analysis in clinical trials.
复制标题
临床试验中亚组分析的策略。
DOI：
10.1007/978-3-642-83419-6_6
复制
发表时间：
1988
期刊：
影响因子：
--
通讯作者：
Caplan,RJ
中科院分区：
文献类型：
--
作者：
Rockette,HE;Caplan,RJ
在进行多重比较时，分配 p 值以量化错误拒绝原假设的概率的标准方法常常会被误解。检验统计量的 p 值通常适用于假设先验选择的个体比较，并且错误地拒绝至少一个比较的概率（通常称为实验错误率）可能远高于个体（或比较）p 值。在临床试验中，当在多个时间点重复分析、在两个以上的治疗组之间进行成对比较、测试多个终点或在不同的患者亚组中测试相同的假设时，就会发生多重比较。本文的主要重点是讨论临床试验中亚组分析中出现的问题。具体来说，我们计划：（a）证明大型随机临床试验的子集分析中可能出现的实验错误率的大小，（b）讨论控制实验错误的一些方法，以及（c）讨论决定子集分析策略时的一些一般问题。尽管进行亚组分析时实验错误率的增加是众所周知的（Pocock 1984），但增加的幅度取决于许多因素，包括 亚组内的样本量、用于定义患者亚组的变量之间的相关性以及亚组比较的数量。实验误差率的上限可以通过应用 Bonferroni 不等式获得，但考虑到用于定义患者亚组的变量之间经常存在的依赖性，这可能过于保守。可以使用计算机模拟来表明问题的严重程度。由于我们的目标是对分析中可能出现的实验错误率的大小进行合理的估计
Standard methods of assigning p values to quantify the probability of falsely rejecting a null hypothesis are often misinterpreted when multiple comparisons are made. The p value of a test statistic is usually applicable to an individual comparison which is assumed to have been selected a priori, and the probability of falsely rejecting at least one of several comparisons (often called the experimentwise error rate) may be much higher than the individual (or comparisonwise) p value. In a clinical trial multiple comparisons occur when analyses are repeated at several points in time, when pairwise comparisons are made among more than two treatment groups, when there is more than one end point being tested, or when the same hypothesis is tested on different subgroups of patients. The major focus of this paper will be to discuss problems arising from subgroup analysis in clinical trials. Specifically we plan to:(a) demonstrate the magnitude of the experimentwise error rate that may occur in subset analyses of a large randomized clinical trial,(b) discuss some methods of controlling the experimentwise error, and (c) discuss some general problems in deciding upon strategies for subset analysis.Although the increase in the experimentwise error rate when one does subgroup analysis is well recognized (Pocock 1984), the magnitude of the increase depends on many factors including the sample size within the subgroups, the correlations among the variables used to define patient subgroups, and the number of subgroup comparisons. An upper bound on the experimentwise error rate could be obtained by applying Bonferroni's inequality but this may be overly conservative given the dependence that often exists among variables used to define patient subgroups. Some indication of the magnitude of the problem can be made using computer simulation. Since our objective is to obtain a reasonable estimate of the magnitude of the experimentwise error rate that might occur in the analyses of
影响因子：
2
作者：
BYAR, DP
通讯作者：
BYAR, DP
在多重比较的情况下评估比例死亡率指数。
影响因子：
2
作者：
H. Rockette;V. Arena
通讯作者：
V. Arena