McCandless Lawrence C
Simon Fraser University.
Int J Biostat. 2012 Jan 6;8(2):/j/ijb.2012.8.issue-2/1557-4679.1350/1557-4679.1350.xml. doi: 10.2202/1557-4679.1350.
Meta-analysis of observational studies is an exciting new area of innovation in statistical science. Unlike randomized controlled trials, which are the gold standard for proving causation, observational studies are prone to biases including confounding. In this article, we describe a novel Bayesian procedure to control for a confounder that is missing across the sequence of studies in a meta-analysis. We motivate the discussion with the example of a meta-analysis of cohort, case-control and cross-sectional studies examining the relationship between oral contraceptives and endometriosis. An important unmeasured confounder is dysmennoreah, which is an indication for oral contraceptive use. To adjust for unmeasured confounding, we combine random effects models with probabilistic sensitivity analysis techniques. Information about the unmeasured confounder is incorporated into the analysis via prior distributions, and we use MCMC to sample from posterior.
观察性研究的荟萃分析是统计科学中一个令人兴奋的新创新领域。与作为证明因果关系金标准的随机对照试验不同，观察性研究容易出现包括混杂在内的偏差。在本文中，我们描述了一种新颖的贝叶斯方法，用于控制荟萃分析中一系列研究中缺失的混杂因素。我们以一项对队列研究、病例对照研究和横断面研究进行的荟萃分析为例展开讨论，该分析旨在研究口服避孕药与子宫内膜异位症之间的关系。一个重要的未测量混杂因素是痛经，它是使用口服避孕药的一个指征。为了调整未测量的混杂因素，我们将随机效应模型与概率敏感性分析技术相结合。关于未测量混杂因素的信息通过先验分布纳入分析，并且我们使用马尔可夫链蒙特卡罗方法从后验分布中抽样。
