Course summary - Meta-analysis methods in health research

Meta-analysis is a statistical method used to combine estimates of effects from several studies into a single combined estimate.  While meta-analysis is commonplace in health research, the field is rapidly evolving, with extensions to standard methods that allow more useful questions to be addressed, such as which factors modify the magnitude of the intervention effects, and which treatment of a competing set of treatments is best according to their safety or effectiveness. Developments in methods also allow common, but complex scenarios, to be better addressed, such as dealing with few studies and studies with missing data.

In this first day of this two-day workshop, we will begin with an introduction to meta-analysis models (fixed effect(s), random effects), and the concept of heterogeneity, and provide participants with the skills to be able to identify, quantify, and investigate heterogeneity through subgroup analysis and meta-regression.  We will introduce the latest Cochrane risk of bias tool for randomised trials, and discuss options for incorporating risk of bias assessments in the analysis. We also present methods for visualising and detecting reporting biases. Finally, we will conclude day 1 with discussion of how to incorporate non-standard randomised designs in a meta-analysis (e.g. cluster and cross-over trials).

In day 2, we will cover challenges in meta-analysis of continuous data, options for dealing with studies with missing data, and introduce network meta-analysis (also known as mixed treatment comparisons or multiple treatment comparisons); a method that synthesises direct and indirect estimates of treatment effect across a network of treatments.

In the lectures, methods and concepts will be introduced. The computer practicals will provide participants with an opportunity to implement the methods. Our focus will be on meta-analysis of the effects of interventions from randomised trials, but many of the methods are applicable to other contexts, such as meta-analysis of effects from observational studies.

Target audience
This course is suitable for quantitative epidemiologists and applied statisticians working in health research.  It will be assumed that participants will have a working knowledge of the statistical package Stata, and have training or experience of introductory statistics and multivariable regression.  For participants enrolling in day 2 only, knowledge of the material covered in day 1 will be assumed.

Computing
Bring your own laptop (with Stata version 14 or above installed).