Korn Edward L, Albert Paul S, McShane Lisa M
Biometric Research Branch, EPN-8128, National Cancer Institute, Bethesda, MD 20892, USA.
Stat Med. 2005 Jan 30;24(2):163-82. doi: 10.1002/sim.1779.
Having a surrogate for a definitive endpoint in a clinical trial can sometimes be useful when it is impractical, invasive or very time consuming to obtain the definitive endpoint. This paper discusses methods for assessing whether the surrogate-endpoint results of a trial can be used in place of definitive-endpoint results. It is important when examining this trial-level surrogacy to include the possibility of trial-level effects and to distinguish whether the treatment arms are naturally ordered, e.g. A vs A+B rather than A vs B. Methods using mixed models of trial-level summaries are discussed and compared to fixed-effects models and to the possibility of using models of individual-level data. We give estimators for definitive-endpoint results of a trial that are predicted from the surrogate-endpoint results of the trial and a set of results from previous trials in which both the definitive and surrogate trial results were available. Graphical displays are also suggested. Two sets of trial results previously analysed for trial-level surrogacy are used as examples.
在临床试验中，当获取确定的终点不切实际、具有侵入性或非常耗时的时候，使用替代指标来代表确定的终点有时会很有用。本文讨论了评估试验的替代终点结果是否可用于替代确定终点结果的方法。在检验这种试验水平的替代指标时，重要的是要考虑试验水平效应的可能性，并区分各治疗组是否是自然排序的，例如A与A + B对比，而不是A与B对比。文中讨论了使用试验水平汇总的混合模型的方法，并将其与固定效应模型以及使用个体水平数据模型的可能性进行了比较。我们给出了根据试验的替代终点结果以及一组之前试验（其中确定终点和替代终点结果均可用）的结果预测出的试验确定终点结果的估计量。还建议使用图形展示。以前针对试验水平替代指标进行分析的两组试验结果用作示例。
