Using IPD Meta‐Analysis to Examine Interactions between Treatment Effect and Participant‐level Covariates

A key component of precision medicine research is exploring whether particular participant-level characteristics are associated with a differential treatment effect. This chapter outlines the importance of individual participant data (IPD) meta-analysis projects for examining treatment-covariate interactions at the participant level, and describes why a traditional meta-regression based on published aggregate data is severely flawed. The aim of meta-regression is to examine whether the value of such aggregated participant-level covariates is associated with the size of the treatment effect across trials. The meta-regression problems of low power and aggregation bias can be addressed by estimating treatment-covariate interactions using participant-level information. The chapter describes one-stage Cox regression models for a time-to-event outcome. It discusses how to handle non-IPD trials when summarising treatment-covariate interactions in an IPD meta-analysis project. It explains why categorisation, and in particular dichotomisation, of continuous covariates is inappropriate, and encourage modelling of non-linear treatment-covariate interactions.