Identifying who benefits most from treatments: How to analyse, present and interpret interactions and subgroup effects in meta-analysis
Peter Godolphin, Meta-analysis Programme, MRC Clinical Trials Unit at UCL
Background: Researchers often wish to identify which individuals benefit more (or less) from interventions; this idea underpins the concept of stratified medicine. As single studies are typically underpowered for exploring whether participant characteristics determine an individual’s response to treatment, meta-analysis can provide a solution. Whilst individual participant data provide the most power and analytical flexibility to investigate interactions between such characteristics and the intervention effect, aggregate data (AD) can also often be used. However, approaches to the analysis, presentation and interpretation of interactions vary widely.
Objectives: In this workshop we aim to demystify interactions in meta-analysis and show how they can be explored using AD. Participants will (i) understand the purpose of subgroup and interaction analysis in trials and meta-analysis, as well as its strengths and limitations; (ii) explore the concept of aggregation bias and its consequences for interaction testing; (iii) extract and calculate a simple “within-trial” interaction effect using AD; (iv) understand how to use AD trial data to calculate subgroup effects compatible with “within-trial” interaction effects; and (v) explore how to present these interactions and subgroup effects clearly using novel implementations of forest plots. Material and examples are taken from Fisher et al., BMJ 2017 doi:10.1136/bmj.j573, and Godolphin et al., Research Synthesis Methods 2023 doi:10.1002/jrsm.1590.
Description: The 90-minute workshop will consist of short slide presentations, group discussion and practical activity. We will begin by considering subgroups and interactions within a single randomised trial. Using real examples, participants will discuss interpretation of results and what can and cannot be concluded from the given data. Participants will learn how interactions from multiple trials may be pooled using meta-analysis. We demonstrate how this can be done with AD, using examples from the literature. Working in small groups, participants will perform a simple AD interaction analysis and discuss results and interpretation. For clinical decision-making, subgroup effects are key; therefore, we will also show how to use the entirety of available trial data to estimate subgroup effects consistent with the ‘within-trial’ interaction. Finally, participants will explore how to present interactions and subgroup effects on forest plots and learn how to perform these analyses using Stata.