Meta-analysis of multiple outcomes: are we using all the evidence?

Abstract: Meta-analyses are typically used to synthesise evidence from multiple studies, in order to decide which treatment is most effective or cost-effective, out of several alternatives. Often, several measures are used to assess patients’ response to treatment (outcomes) and it is not unusual for randomised trials to report more than one outcome of interest. Separate analyses of each outcome, each typically using different sets of randomised controlled trials, is inefficient and can lead to conflicting conclusions.

Using a joint model for all the outcomes, and adequately incorporating what is known about the relationships between different relative effects, is therefore essential for making decisions based on multiple outcomes and should be incorporated in the standard evidence synthesis framework.

Models incorporating within-trial correlations and outcome relationships, based on what is known from the data generating process, will be presented for some key examples. Although different applications will have different types of evidence structure, and may require bespoke models, it is often possible to incorporate known relationships in trial outcomes into a Bayesian modelling framework.

Implications of this framework for the usual process of systematic review, meta-analysis and economic modelling will also be discussed.