Advanced meta-analysis 2: Performing meta-analysis in the presence of rare events

Category: 

  • Statistical methods
Date and Location

Date: 

Wednesday 6 September 2023 - 11:00 to 12:30

Location: 

Contact persons and facilitators

Contact person:

Facilitators:

Chaimani A1, Evrenoglou T1
1 Université Paris Cité, Center of Research in Epidemiology and Statistics, Inserm, Paris, France
Target audience

Target audience: 

Authors, Reviewers, or Editors of Cochrane Reviews; Researchers interested in synthesizing studies with rare events

Level of difficulty: 

Advanced
Type of workshop

Type of workshop : 

Training
Abstract

Abstract:

Background: Rare events are a common problem in meta-analysis, particularly for secondary and safety outcomes. When the events are rare, standard meta-analytical approaches have important limitations and may lead to biased and imprecise results. Alternative methods have been proposed in the literature that outperform in terms of bias and precision compared with the standard inverse-variance (IV) method. This workshop is part of a series of workshops delivered by the Cochrane Statistical Methods Group.

Objectives: The objective of this workshop is to provide guidance on handling rare events in meta-analysis. We aim to review the problems associated with the IV method and to describe several alternative methods that can be used instead.

Description: In this workshop, we will review the properties of the IV approach and explain the reasons that render this method problematic when the studied endpoints are rare. We will explain the advantages of one-stage meta-analysis models over two-stage models in such cases and go through the assumptions and the properties of some of these models. The workshop will also include a practical part in which participants will apply different models in R using data from a clinical example involving rare events. By the end of this workshop, participants will have a good understanding of the problems related to meta-analysis of rare events and will be able to fit meta-analysis models appropriate for such datasets.