Tang Nian-Sheng, Tang Man-Lai, Chan Ivan Siu Fung
Adult Education College, Yunnan University, No 2, North Road of Cuihu, Kunming 650091, P.R. China.
Stat Med. 2003 Apr 30;22(8):1217-33. doi: 10.1002/sim.1213.
Matched-pair design is often used in clinical trials to increase the efficiency of treatment comparison. We consider the problem of equivalence test with a relative risk endpoint in matched-pair studies with binary outcomes, and develop several score and Wald-type statistics for testing a hypothesis of non-unity relative risk. Examples from an assessment of HIV screening test and a cross-over clinical trial of soft contact lenses are used to illustrate the proposed methods. Through simulations we compare the empirical performance of these tests with the test proposed by Lachenbruch and Lynch. We show that a score test based on a reparameterized multinomial model by Tango performs best in the sense that the test satisfactorily controls the type I error rate and its empirical type I error rates are generally much closer to the prespecified nominal significance level than those of the other tests.
配对设计常用于临床试验，以提高治疗比较的效率。我们考虑在二元结局的配对研究中使用相对风险终点进行等效性检验的问题，并开发了几种得分和 Wald 型统计量，用于检验相对风险不为 1 的假设。通过评估 HIV 筛查试验和软性隐形眼镜交叉临床试验的实例来说明所提出的方法。通过模拟，我们将这些检验的实证性能与 Lachenbruch 和 Lynch 提出的检验进行比较。我们表明，基于 Tango 重新参数化多项模型的得分检验表现最佳，因为该检验能令人满意地控制 I 型错误率，并且其经验性 I 型错误率通常比其他检验更接近预先指定的名义显著性水平。
