Department of Dental Public Health Sciences, University of Washington, 1959 NE Pacific Street, Box 357475, Seattle, WA 98195, United States.
Contemp Clin Trials. 2010 Mar;31(2):180-4. doi: 10.1016/j.cct.2009.12.005. Epub 2010 Jan 8.
Since its introduction in 1974 the use of the term Minimization has been broadened to include other algorithms. All algorithms use patient characteristics to determine the assignment that produces the best overall balance between treatment groups. They differ in whether or not they use all of the data from each previously assigned subject to assign subsequent subjects so the methods are classified as complete or partial minimization. PubMed, Citation Index and Cochrane searches determined the frequency of articles using these types of minimization and a subset was selected for detailed review regarding the adequacy of the usage and reporting of minimization. In the past 10 years usage has increased three fold over the previous decade but is still less than 2% of clinical trials. None of the studies makes maximum use of minimization and they are not following good reporting practices. Concerns about the use of minimization have involved selection bias and statistical analysis. Several modifications to minimization are suggested to reduce the possibility of selection bias so that adding randomization will rarely be required. Separating primary and secondary analyses can avoid the statistical problems that minimization poses. The two types of analyses are distinguished by opposite limiting signs, providing reliable, simplified statistical results. This will improve data utilization and make clinical trials more reproducible. Minimization should be the method of choice in assigning subjects in all clinical trials.
自 1974 年引入以来，“最小化”一词的使用范围已扩大到包括其他算法。所有算法都使用患者特征来确定最佳分组方案，以实现治疗组之间的最佳总体平衡。它们的区别在于是否使用每个先前分配的患者的所有数据来分配后续患者，因此这些方法被分类为完全或部分最小化。PubMed、引文索引和 Cochrane 搜索确定了使用这些类型最小化的文章的频率，并选择了一部分进行详细审查，以了解最小化的使用和报告是否充分。在过去的 10 年中，其使用量是前十年的三倍，但仍不到临床试验的 2%。这些研究都没有充分利用最小化，也没有遵循良好的报告实践。对最小化的使用的担忧涉及选择偏差和统计分析。提出了对最小化的几种修改，以减少选择偏差的可能性，从而很少需要添加随机化。将主要和次要分析分开可以避免最小化带来的统计问题。这两种类型的分析通过相反的极限符号来区分，提供可靠、简化的统计结果。这将提高数据利用率并使临床试验更具可重复性。最小化应成为所有临床试验中分配受试者的首选方法。
