Hafner K B, Koch G G, Canada A T
Department of Biostatistics, University of North Carolina, Chapel Hill 27514.
Stat Med. 1988 Apr;7(4):471-81. doi: 10.1002/sim.4780070404.
Alternative analysis strategies for the three-period crossover design with two treatments are discussed in this paper. One analysis strategy involves a parametric model that incorporates the effects of interest. To implement this method, one usually assumes that the covariance structure for the data has the sphericity, or circularity, property. Alternative approaches that do not require this assumption are described. They are based on the parametric and non-parametric analysis of appropriate within-subject linear functions of the data. The advantage of these methods is that one only needs the assumption that the resulting linear functions for the respective subjects are independent and have a common distribution. The parametric approach also requires normality of the resulting within-subject linear functions for small sample situations. An extension of the non-parametric method is considered for cases in which the treatment sequences are randomly assigned within strata. The various methods are illustrated for a three-period crossover design involving two strata with randomly assigned treatment sequences of the form A:B:A and B:A:B.
本文讨论了两种治疗方法的三阶段交叉设计的替代分析策略。一种分析策略涉及一个纳入感兴趣效应的参数模型。为实施该方法，通常假定数据的协方差结构具有球对称性或循环性。还描述了不需要此假设的替代方法。它们基于对数据的适当受试者内线性函数进行参数和非参数分析。这些方法的优点是，只需要假定各受试者的所得线性函数是独立的且具有共同分布。对于小样本情况，参数方法还要求所得受试者内线性函数呈正态分布。对于治疗序列在各层内随机分配的情况，考虑了非参数方法的扩展。针对一个三阶段交叉设计示例说明了各种方法，该设计涉及两层，具有形式为A:B:A和B:A:B的随机分配治疗序列。
