Einarson T R, Leeder J S, Koren G
Department of Pediatrics, Hospital for Sick Children, Toronto, ON.
Drug Intell Clin Pharm. 1988 Oct;22(10):813-24. doi: 10.1177/106002808802201021.
This article presents a stepwise approach for conducting a meta-analysis of epidemiological studies based on proposed guidelines. This systematic method is recommended for practitioners evaluating epidemiological studies in the literature to arrive at an overall quantitative estimate of the impact of a treatment. Bendectin is used as an illustrative example. Meta-analysts should establish a priori the purpose of the analysis and a complete protocol. This protocol should be adhered to, and all steps performed should be recorded in detail. To aid in developing such a protocol, we present methods the researcher can use to perform each of 22 steps in six major areas. The illustrative meta-analysis confirmed previous traditional narrative literature reviews that Bendectin is not related to teratogenic outcomes in humans. The overall summary odds ratio was 1.01 (chi 2 = 0.05, p = 0.815) with a 95 percent confidence interval of 0.66-1.55. When the studies were separated according to study type, the summary odds ratio for cohort studies was 0.95 with a 95 percent confidence interval of 0.62-1.45. For case-control studies, the summary odds ratio was 1.27 with a 95 percent confidence interval of 0.83-1.94. The corresponding chi-square values were not statistically significant at the p = 0.05 level.
本文基于拟议的指南，介绍了一种对流行病学研究进行荟萃分析的逐步方法。这种系统方法推荐给评估文献中流行病学研究的从业者，以便对一种治疗方法的影响得出总体定量估计。以“反应停”作为示例。荟萃分析者应事先确定分析目的并制定完整方案。应严格遵守该方案，并且对执行的所有步骤都要详细记录。为帮助制定这样一个方案，我们介绍了研究人员可用于在六个主要领域执行22个步骤中每一步骤的方法。该示例荟萃分析证实了先前传统的叙述性文献综述，即“反应停”与人类致畸结局无关。总体汇总比值比为1.01（卡方 = 0.05，p = 0.815），95%置信区间为0.66 - 1.55。当根据研究类型对研究进行分类时，队列研究的汇总比值比为0.95，95%置信区间为0.62 - 1.45。病例对照研究的汇总比值比为1.27，95%置信区间为0.83 - 1.94。相应的卡方值在p = 0.05水平上无统计学意义。
