Service de pharmacotoxicologie, hospices civils de Lyon, 16, avenue Lacassagne, 69003 Lyon, France; Laboratoire de biométrie et biologie évolutive, CNRS, UMR5558, université Lyon 1, 69008 Lyon, France.
Unité de recherche clinique innovation pharmacologie, UMR 1059 Sainbiose, bâtiment recherche, CHU de Saint-Étienne, université Jean Monnet, hôpital Nord, 42022 Saint-Étienne, France.
Therapie. 2020 Jan-Feb;75(1):21-27. doi: 10.1016/j.therap.2019.11.007. Epub 2019 Dec 16.
Single-arm studies are sometimes used as pivotal studies but they have methodological limitations which prevent them from obtaining the high level of reliability as for a randomised controlled study which remains the gold standard in the evaluation of new treatments. The objective of this roundtable was to discuss the limitations of these single-arm studies, to analyse available and acceptable solutions in order to propose guidelines for their conduct and assessment. Single-arm studies themselves are intrinsically inappropriate for demonstrating the benefit of a new treatment because it is impossible to infer the benefit from a value obtained under treatment without knowing what it would have been in the absence of the new treatment. The implication is that comparison with other data is necessary. However this comparison has limitations due to (1) the post hoc choice of the reference used for comparison, (2) the confusion bias for which an adjustment approach is imperative and, (3) the other biases, measure and attrition among others. When these limitations are taken into account this should, first and foremost, lead to the conduct of externally controlled trials instead of single-arm trials as is proposed by the latest version of ICH E10. Moreover, the external control must be formalised in the study protocol with a priori selection of both the reference control and the formal method of comparison: test in relation to a standard, adjustment on individual data, a synthetic control group or matching-adjusted indirect comparisons (MAIC). Lastly, externally controlled studies must be restricted to situations where randomisation is infeasible. To be acceptable, these studies must be able to guarantee freedom from residual confusion bias, which is only truly acceptable if the observed effect is dramatic and the usual course of the disease is highly predicable.
单臂研究有时可用作关键研究，但它们存在方法学限制，无法像随机对照研究那样获得高度可靠性，后者仍然是评估新治疗方法的金标准。本次圆桌会议的目的是讨论这些单臂研究的局限性，分析现有和可接受的解决方案，以便为其进行和评估提出指南。单臂研究本身不适合证明新治疗方法的益处，因为在不知道没有新治疗方法的情况下会是什么样的情况下，从治疗下获得的值推断益处是不可能的。这意味着需要进行比较。然而，这种比较存在局限性，原因有三：(1) 用于比较的参考值是事后选择的，(2) 存在混淆偏差，必须采用调整方法，(3) 还有其他偏差、测量和失访等。考虑到这些局限性，首先应该进行外部对照试验，而不是像 ICH E10 的最新版本所建议的那样进行单臂试验。此外，外部对照必须在研究方案中正式化，包括参考对照和正式比较方法的预先选择：与标准进行测试、对个体数据进行调整、合成对照组或匹配调整间接比较（MAIC）。最后，外部对照研究必须限于随机化不可行的情况。为了被接受，这些研究必须能够保证没有残留的混淆偏差，如果观察到的效果显著且疾病的通常过程高度可预测，则这种偏差才是真正可以接受的。
