Huster W J, Louv W C
Biostatistics Department, Marion Merrell Dow Research and Development, Cincinnati, Ohio 45215.
J Biopharm Stat. 1992;2(2):219-38. doi: 10.1080/10543409208835041.
According to the Food and Drug Administration's Guidelines for the Format and Content of the Clinical and Statistical Sections of New Drug Applications, approval of a new drug "should be supported by more than one well-controlled trial and carried out by independent investigators. This interpretation is consistent with the general scientific demand for replicability." Nevius has described a four-point proposal for assessing statistical evidence in a single multicenter trial. Briefly, these four points are: (1) combined analysis shows significant results, (2) consistency over centers in terms of direction, (3) consistency over centers in terms of producing nominally significant results in centers with sufficient power, and (4) evidence of efficacy after adjustment for multiple comparisons. What is not clear from Nevius' proposal is how to quantify whether the amount of evidence in a single multicenter trial is equivalent to that from two separate trials. It is proposed that the post hoc subdivision of a multicenter trial may address this issue if the inherent multiple testing problem is accommodated. A minimax statistic is developed to test the hypothesis that the effect of the drug has been reproduced in a single multicenter trial. Monte Carlo simulation is used to generate the distribution of the minimax statistic under the null and several alternative hypotheses. Data from a multicenter trial are used to demonstrate the technique. Bootstrapping is used to determine the null distribution of the minimax statistic.
根据美国食品药品监督管理局关于新药申请临床和统计部分格式与内容的指南，新药的批准“应有不止一项严格对照试验的支持，且由独立研究者进行。这种解释与对可重复性的一般科学要求一致。”内维斯描述了一项在单一多中心试验中评估统计证据的四点提议。简而言之，这四点是：（1）合并分析显示显著结果；（2）各中心在方向上一致；（3）各中心在有足够检验效能的中心产生名义上显著结果方面一致；（4）经多重比较调整后的疗效证据。内维斯的提议中不明确的是如何量化单一多中心试验中的证据量是否等同于两项单独试验的证据量。有人提出，如果能解决固有的多重检验问题，多中心试验的事后细分可能会解决这个问题。开发了一种极小极大统计量来检验药物效应在单一多中心试验中是否重现的假设。蒙特卡罗模拟用于生成在原假设和几个备择假设下极小极大统计量的分布。来自一项多中心试验的数据用于演示该技术。使用自助法来确定极小极大统计量的原分布。
