Tudor G, Koch G G
Quintiles Inc., Research Triangle Park, NC.
Stat Methods Med Res. 1994 Dec;3(4):345-81. doi: 10.1177/096228029400300404.
This paper reviews nonparametric methods for the analysis of crossover studies. Primary attention is given to crossover studies to compare two treatments for a response variable that has a metric measurement level. For this situation, one can often test hypotheses or obtain confidence intervals for parameters of interest by applying well known univariate nonparametric rank methods (e.g., the Wilcoxon rank sum test, or the Wilcoxon signed rank test) to appropriately specified functions of the data. Related extensions are also available, to some degree, for crossover studies to compare more than two treatments or those for which the measurement level of the response variable is ordinal or has a censored time-to-event nature. Methods for several specific situations along these lines are discussed in terms of principles with potentially broader applicability. Several examples are provided to illustrate the performance of some of the methods.
本文综述了用于交叉研究分析的非参数方法。主要关注用于比较两种针对具有度量测量水平的响应变量的治疗方法的交叉研究。对于这种情况，通过将著名的单变量非参数秩方法（例如，威尔科克森秩和检验或威尔科克森符号秩检验）应用于数据的适当指定函数，通常可以检验假设或获得感兴趣参数的置信区间。在某种程度上，对于比较两种以上治疗方法的交叉研究或响应变量的测量水平为有序或具有删失事件发生时间性质的交叉研究，也有相关扩展。沿着这些思路针对几种特定情况的方法在具有潜在更广泛适用性的原理方面进行了讨论。提供了几个例子来说明一些方法的性能。
