含结构零2×2表中风险比齐性检验的样本量研究

摘要

本文研究了在临床和流行病学中出现的K个独立含结构零的2×2列联表.利用风险比推导出基于Wald型,基于对数变换型以及基于Score型的三个检验统计量,以检验层间齐性假设,获得了分别基于这三个检验统计量的控制功效的样本量公式和数值实验结果.模拟研究表明基于Score检验和基于对数变换检验的样本量公式比基于Wald检验的样本量公式更精确,更能有效保证预先给定的功效.推广了基于列联表问题的假设检验方法.%This paper investigates K independent 2 × 2 contingency tables with structural zero in clinical and epidemiological studies.By using the measurement of risk ratio,we present three test statistics:the Wald-based,the logarithmic-transformation-based and the Score-based statistics to test the homogeneity hypothesis across strata.Power-controlled sample size formulae and corresponding numerical experiment results are obtained based on the above three tests.The simulation studies show that generally the sample sizes associated with the Score-based and the logarithmic-transformation-based tests are more accurate to guarantee the pre-specified power than the Wald-based one,which develop the hypothesis testing methods based on the contingency tables.