Parmar M K, Ungerleider R S, Simon R
Medical Research Council, Cancer Trials Office, Cambridge, U.K.
J Natl Cancer Inst. 1996 Nov 20;88(22):1645-51. doi: 10.1093/jnci/88.22.1645.
A confirmatory randomized clinical trial is a trial that is aimed at assessing whether a treatment effect observed in a previous randomized trial (or trials) is real and important. There is often considerable disagreement about the need for such confirmatory trials.
Our aim is to provide a general statistical framework for evaluating whether a confirmatory trial is warranted in a particular situation.
The results of two clinical trials are considered: 1) a Cancer and Leukemia Group B trial comparing induction chemotherapy plus radiotherapy with radiotherapy alone in the treatment of patients with locally advanced non-small-cell lung cancer and 2) a North Central Cancer Treatment Group trial comparing surgery plus adjuvant chemotherapy with surgery alone in the treatment of patients with advanced colon cancer. In our analysis, we argue that differences in the interpretation of results from a randomized trial are based on differences in prior beliefs about the efficacy of the treatment(s) under study. We believe that a major factor in the decision to perform a confirmatory trial is prior skepticism about the clinical worth of the treatment in question. Both the level of prior skepticism and the minimum treatment effect deemed clinically worthwhile require subjective judgment. We develop a Bayesian framework to allow differences in interpretation to be examined systematically and the need for a confirmatory trial to be assessed. Our model allows the addition of prior belief (specified in the form of a prior distribution of treatment effect) to the results of a trial to yield a posterior distribution. The interpretation of trial results is based on the posterior distribution and will vary as the prior distribution (i.e., the prior belief) varies. To aid in the interpretation of trial results, we also advocate the specification of a minimum clinically worthwhile treatment effect at the start of a trial.
Our approach acknowledges that a number of different prior beliefs are possible, giving rise to a range of interpretations of results from a clinical trial. This approach provides a formal and systematic basis for considering both the range of likely opinions and the subsequent decision to be made with regard to the need for a confirmatory trial. We recommend that this approach be considered in the discussion of future confirmatory randomized clinical trials.
确证性随机临床试验是旨在评估在先前一项（或多项）随机试验中观察到的治疗效果是否真实且重要的试验。对于此类确证性试验的必要性，常常存在相当大的分歧。
我们的目标是提供一个通用的统计框架，用于评估在特定情况下是否有必要进行确证性试验。
考虑了两项临床试验的结果：1）癌症与白血病B组试验，比较诱导化疗加放疗与单纯放疗治疗局部晚期非小细胞肺癌患者的疗效；2）北中部癌症治疗组试验，比较手术加辅助化疗与单纯手术治疗晚期结肠癌患者的疗效。在我们的分析中，我们认为对随机试验结果解释的差异基于对所研究治疗方法疗效的先验信念的差异。我们认为进行确证性试验决策的一个主要因素是对所讨论治疗方法临床价值的先验怀疑态度。先验怀疑程度以及被认为具有临床价值的最小治疗效果都需要主观判断。我们开发了一个贝叶斯框架，以便系统地研究解释上的差异，并评估进行确证性试验的必要性。我们的模型允许将先验信念（以治疗效果的先验分布形式指定）添加到试验结果中，以产生后验分布。试验结果的解释基于后验分布，并且会随着先验分布（即先验信念）的变化而变化。为了帮助解释试验结果，我们还主张在试验开始时指定最小临床有价值的治疗效果。
我们的方法承认可能存在多种不同的先验信念，从而产生对临床试验结果的一系列解释。这种方法为考虑可能的观点范围以及随后关于是否需要进行确证性试验所做的决策提供了一个正式且系统的基础。我们建议在未来确证性随机临床试验的讨论中考虑这种方法。
