Lachin John M, Younes Naji
The Biostatistics Center, Department of Epidemiology and Biostatistics, The George Washington University, 6110 Executive Boulevard, Suite 750, Rockville, MD 20852, USA.
Stat Med. 2007 Nov 30;26(27):5014-32. doi: 10.1002/sim.2963.
In drug development, a sequence of studies are conducted to evaluate effectiveness (or efficacy) and safety, such as a Phase II study to assess pharmacological activity or safety that is then followed by a definitive Phase III study to assess clinical effectiveness. Rather than conducting separate successive studies, we describe a design in which the patients enrolled in a preliminary (e.g. Phase II) study are continued into a subsequent full-scale (e.g. Phase III) study. This design also applies to a study that uses an internal pilot with a preliminary assessment of efficacy or safety. The combined preliminary to full-scale design potentially reduces the total numbers of patients required and the total duration of the program. The design allows a futility or safety stopping boundary for the preliminary study result that is specified in terms of a lower critical Z-value (Z(L)) and the pursuant type II error probability incurred under a specified alternative hypothesis of a beneficial effect or no toxicity at that stage. This boundary also leads to a deflation of the type I error probability for the final test at the completion of the full-scale study, such that a critical value for the final test (Z(F)) can be determined that provides the desired level of the type I error probability exactly. Thus, it is possible to determine sample sizes at the two stages, and critical values Z(L) and Z(F) that provide specified type I and II error probabilities for the combined study. We describe the design using large-sample normally distributed Z-tests at the two phases, including a test for means, or proportions or survival times, or combinations thereof, such as a test for means at Phase II followed by a test for proportions at Phase III. We compare the properties of this design versus the conduct of two successive studies, and explore the factors that influence the operating characteristics of the design. We also discuss the practical considerations in the implementation of the design.
在药物研发过程中，会开展一系列研究以评估有效性（或效能）和安全性，比如进行一项II期研究来评估药理活性或安全性，随后进行确定性的III期研究以评估临床有效性。我们描述的一种设计，并非进行单独的连续研究，而是让参加初步（如II期）研究的患者继续参与后续的全面（如III期）研究。这种设计也适用于使用内部预试验对有效性或安全性进行初步评估的研究。从初步研究到全面研究的联合设计有可能减少所需患者的总数以及整个项目的持续时间。该设计针对初步研究结果设定了一个无效性或安全性终止边界，这个边界是根据一个较低的临界Z值（Z(L)）以及在该阶段特定的有益效应或无毒性替代假设下产生的相应II类错误概率来确定的。这个边界还会导致在全面研究完成时最终检验的I类错误概率降低，这样就可以确定最终检验的临界值（Z(F)），使其恰好提供所需的I类错误概率水平。因此，有可能确定两个阶段的样本量以及临界值Z(L)和Z(F)，从而为联合研究提供特定的I类和II类错误概率。我们在两个阶段使用大样本正态分布Z检验来描述该设计，包括均值检验、比例检验、生存时间检验或它们的组合，例如在II期进行均值检验，在III期进行比例检验。我们比较这种设计与进行两项连续研究的特性，并探讨影响该设计操作特征的因素。我们还讨论了该设计实施过程中的实际考虑因素。
