论文部分内容阅读
目的:提出非虚假设综合渐近正态法。方法:将渐近正态法加以扩展以覆盖非虚假设和分层设计,从而推导出单样本和两样本基本关系式,功效函数,样本量,及检验统计量。结果:当最小感知差别取零时,它还原对应经典方法,包括经典Cochran检验和Mantel-Haenszel检验。当层数取1时,它还原为非虚假设渐近正态法,包括Dunnett-Gent检验。当最小感知差别取零且层数取1时,它还原为经典渐近正态法,包括单样本和两样本比例检验。结论:这种方法可用于分层设计的有效对照临床试验,以建立临床优效性,非劣效性,或等效性。
Objective: To propose non-hypothetical synthetic asymptotic normal method. Methods: The asymptotic normal method is extended to cover the non-hypothetical and hierarchical design, thus deriving the basic relationship between single sample and two samples, power function, sample size, and test statistic. Results: When the minimum perceived difference obtained zero, it restored the corresponding classical methods, including the classical Cochran test and Mantel-Haenszel test. When the number of layers is taken as 1, it returns to non-hypothetical asymptotic normality, including the Dunnett-Gent test. When the minimum perceived difference is zero and the number of layers is taken as 1, it returns to classical asymptotic normality, including single-sample and two-sample proportional test. Conclusion: This method can be used in effective controlled clinical trials of stratified design to establish clinical superiority, noninferiority, or equivalence.