Dechartres Agnes, Trinquart Ludovic, Faber Timor, Ravaud Philippe
Centre de Recherche Epidémiologie et Statistique, INSERM U1153, Team METHODS, Hôpital Hôtel-Dieu, 1 place du parvis Notre Dame, 75004 Paris, France; Centre d'Epidémiologie Clinique, Hôpital Hôtel-Dieu, Assistance Publique-Hôpitaux de Paris, 1 place du parvis Notre Dame, 75004 Paris, France; Faculté de Médecine, Université Paris Descartes, Sorbonne Paris Cité, 15 rue de l'école de médecine, 75006 Paris, France; Cochrane France, Hôpital Hôtel-Dieu, 1 place du parvis Notre Dame, 75004 Paris, France.
Centre de Recherche Epidémiologie et Statistique, INSERM U1153, Team METHODS, Hôpital Hôtel-Dieu, 1 place du parvis Notre Dame, 75004 Paris, France; Cochrane France, Hôpital Hôtel-Dieu, 1 place du parvis Notre Dame, 75004 Paris, France; Department of Epidemiology, Mailman School of Public Health, Columbia University, 722 West 168th Street, 10032 New York, NY, USA.
J Clin Epidemiol. 2016 Sep;77:24-37. doi: 10.1016/j.jclinepi.2016.04.005. Epub 2016 Apr 29.
Meta-epidemiological studies provide empirical evidence of trial characteristics associated with treatment effects. We aimed to evaluate methods used and characteristics associated with treatment effect in these studies.
For this systematic review, we searched MEDLINE, Embase, Cochrane Methodology Register, Web of Science, and PROSPERO up to April 2015. We particularly assessed four key methodological components: constitution of the collection, clustering of trials within meta-analyses, heterogeneity assessment, and adjustment on meta-confounders. We also assessed trial characteristics evaluated and their association with treatment effect.
We included 56 meta-epidemiological studies with data from 3,199 meta-analyses, 32 networks, and 21,468 trials. Thirty-two (58%) were published since 2010. Only 13 (23%) included all key methodological components. Overall, 58 trial characteristics were assessed. Allocation concealment and sequence generation were assessed in 22 (39%) and 17 (30%) meta-epidemiological studies, respectively, and trial size in 9 (16%). These characteristics were consistently associated with treatment effect estimates with larger effects in trials with inadequate sequence generation or allocation concealment or smaller trials.
Key methodological components (e.g., constitution of the collection) were frequently missing. Concerning trial characteristics evaluated, there was consistent evidence that allocation concealment, sequence generation, and trial size were associated with treatment effect.
Meta流行病学研究提供了与治疗效果相关的试验特征的实证证据。我们旨在评估这些研究中所使用的方法以及与治疗效果相关的特征。
对于这项系统评价，我们检索了截至2015年4月的MEDLINE、Embase、Cochrane方法学注册库、科学引文索引和国际系统评价前瞻性注册库。我们特别评估了四个关键的方法学组成部分：研究集合的构成、Meta分析中试验的聚类、异质性评估以及对Meta混杂因素的调整。我们还评估了所评价的试验特征及其与治疗效果的关联。
我们纳入了56项Meta流行病学研究，其数据来自3199项Meta分析、32个网络和21468项试验。其中32项（58%）于2010年以后发表。只有13项（23%）纳入了所有关键的方法学组成部分。总体而言，共评估了58个试验特征。分别在22项（39%）和17项（30%）Meta流行病学研究中评估了分配隐藏和随机序列产生，在9项（16%）研究中评估了试验规模。这些特征与治疗效果估计始终相关，在随机序列产生不当或分配隐藏不足的试验或规模较小的试验中效果更大。
关键的方法学组成部分（如研究集合的构成）常常缺失。关于所评价的试验特征，有一致的证据表明分配隐藏、随机序列产生和试验规模与治疗效果相关。
