Abstract
ObjectiveMeta-analysing studies with low event rates is challenging as some of the standard methods for meta-analysis are not well suited to handle rare outcomes. This is more evident when some studies have zero events in one or both treatment groups. In this article, we discuss why rare events require special attention in meta-analysis, we present an overview of some approaches suitable for meta-analysing rare events and we provide practical recommendations for their use.MethodsWe go through several models suggested in the literature for performing a rare events meta-analysis, highlighting their respective advantages and limitations. We illustrate these models using a published example from mental health. We provide the software code needed to perform all analyses in the appendix.ResultsDifferent methods may give different results, and using a suboptimal approach may lead to erroneous conclusions. When data are very sparse, the choice between the available methods may have a large impact on the results. Methods that use the so-called continuity correction (eg, adding 0.5 to the number of events and non-events in studies with zero events in one treatment group) may lead to biased estimates.ConclusionsResearchers should define the primary analysis a priori, in order to avoid selective reporting. A sensitivity analysis using a range of methods should be used to assess the robustness of results. Suboptimal methods such as using a continuity correction should be avoided.
Funder
Swiss National Science Foundation
Subject
Psychiatry and Mental health
Reference30 articles.
1. Higgins J , Deeks JJ , Atman DG . Chapter 16: Special topics in statistics. In: Higgins JPT , Green S , eds. Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 (updated March 2011): The Cochrane Collaboration, 2011. www.cochrane-handbook.org
2. Meta-analyses of safety data: a comparison of exact versus asymptotic methods
3. Meta-analysis in clinical trials
4. Demystifying fixed and random effects meta-analysis
5. Much ado about nothing: a comparison of the performance of meta-analytical methods with rare events
Cited by
149 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献