<International Circulation>: Currently， do you think results form sub group analysis of clinical studies have been over interpreted?  And how would you interpret the results correctly in terms of paper writing.
Professor Pocock : There is a bad history of over interpretation of sub group analyses and I will not go into examples， that would take too long.  I think sub group analyses need to be seen as exploratory so you come up with a global result for a study.  And then it is logical as a sequel to ask yourself “Is the overall treatment finding consistent across different types of patients?”  So that is worth attempting to do. The usual problem is you have too many ideas in your head of possible sub groups and secondly the trial lacks statistical power to give reliable results， in particular sub groups.  How to interpret them? I think you tend to see them as hypothesis generating.  So I think it is valid that you would include usually a later table or figure in a trial or report displaying sub group findings and then you should have with an appropriate interaction test to say whether the treatment effect really differs between subgroups and then of course you would like to see a cautious discussion which is in the spirit of hypothesis generating.  Of course people like to go further than that， especially in a trial that was overall negative.  In the talk I gave yesterday I gave an example where they made a post hoc claim on a sub group in a secondary endpoint and squeezed that somehow into the abstract of the publication in the Lancet which really both the authors and the Lancet should have known better not to let that happen.  So the number one word is caution when it comes to sub group analyses.

　　《国际循环》：你是否认为临床试验的亚组分析被过分解读了？发表论文时如何正确解读亚组分析结果？
Pocock博士：过度解读亚组分析结果已经有不好的先例，这里我不想一一列举了。我认为亚组研究的分析应被看作是探索性的。首先你得出了研究的总体结果，然后你会问自己“总体的治疗结果是否在不同亚组患者中也是一致的？”，这是合乎逻辑的想法，也值得继续分析。通常情况下，问题是你头脑中关于可能的亚组有太多的想法，然后试验并没有足够的统计学效能来得出可靠的结果，亚组分析尤其如此。该如何解释亚组分析的结果？我觉得应当认为亚组分析是为了产生假设的。因此，我认为随后在试验的报告中附上展示亚组结果的图或表是没有问题的，然后应当附上合适的相互作用检验，以明确治疗效应在不同亚组间是否有区别。随后，当然你期望看到作者能够展开谨慎的讨论，目的是产生假设。当然，大家期望在此基础上更近一步，尤其是对总体上得出了阴性结果的试验。在我昨天的演讲中，我举了个例子，有篇论文对一个亚组的次级终点进行了事后分析，该文的摘要竟被《柳叶刀》杂志发表了，我想无论是作者还是杂志都早该知道这样的事情不应当发生。因此对于亚组分析来说，最关键的就是谨慎。

　　<International Circulation>: my last question， for conclusions with a meta analysis or RCTs， which ones do you consider more reliable?
Professor Pocock :  Oh that depends on the RCT.  I do not think one can generalize.  An individual trial has two problems.  One it may not be big enough to reach reliable conclusions by itself and secondly it is studied in one specific population so you do not know whether you can generalize your findings for patients of a slightly different type.  So individual trials may not give enough evidence.  They give reliable evidence but not enough evidence to be generalizable.  So a good quality meta analysis which includes all relevant trials which have been done in a well defined question can give more generalizable results and more reliable estimates.  The trouble with meta analyses is they vary remarkably in quality and we saw an example here today on an ARBs in cancer， a meta analyses which should have never made it into print， and they exist unfortunately.  So a good quality meta analyses has an important place if the individual trials are not big enough to reach reliable conclusions.

　　《国际循环》：你觉得荟萃分析和随机对照试验的结论哪一个更可靠？
Pocock博士：这取决于随机对照试验。我不认为单一试验的结论可以推而广之。单一的试验有两个问题。一个是样本量不够大，单靠一个试验本身无法得出可靠的结论。另一个问题是试验针对一个特定人群，所以不知道该试验的结果能否被推及到稍有不同的另一个人群。因此，单个试验无法给出足够的证据。可以给出可靠的证据，但无法给出足够的证据来推及到其他情况。因此，一个包括所有相关试验的、拟回答被定义好的问题的高质量荟萃分析能够给出更能被推及到其他情况的结果，也能给出更为可靠的估计。荟萃分析的问题是质量参差不齐，今天我看到了一项有关ARB与癌症风险的荟萃分析，这样的荟萃分析应当永远不被发表，但不幸的是还存在这样的荟萃分析。因此，如果单个试验的样本量小，不足以得出可靠结论的情况下，一个高质量的荟萃分析是具有重要地位的。