Background and Objectives: Treatments may be more effective in some patients than others, and IPD meta-analysis provides perhaps the best method of investigating treatment-covariate interactions. We aimed to review the various methods currently used in practice, and provide a comprehensive critique of these, to provide guidance for systematic reviewers collecting and analysing IPD. Methods: We searched MEDLINE to identify all frequentist approaches either proposed in theory or used in practice and appraised them according to ease of use and interpretation, risk of bias, and power. Existing IPD datasets were re-analysed. Results: Four categories of method were identiﬁed: A: pooling of within-trial covariate interactions; B: mixed-effects model with a treatment-covariate interaction term; C: calculating pooled treatment effects within covariate subgroups and testing for differences between these; and D: augmenting the power of A by combining its effect estimate with that of a meta-regression. Distinguishing between across-and within-trial information is important, as across-trials information may be subject to ecological bias. A strategy is proposed to help reviewers select the most appropriate approach in different circumstances. We propose that most practitioners should start with method A or D due to their ease of use and interpretation. Method B allows a variety of more complex analyses to be carried out; of which A and D are special cases. Method C, though presently common, should be avoided. Our re-analysis of existing datasets shows that the choice of method can lead to substantively different ﬁndings. Conclusions: Patient-level treatment-covariate interactions are important and worthy of investigation. Different methods are available, some of which are straightforward. The choice of method will be driven mainly by whether across-trial information is considered for inclusion in the analysis, a decision which is largely subjective and requires balancing issues of bias and power.
A critique and guide of methods to assess interactions in individual patient data (IPD) meta-analysis
More like this
- A demonstration and guide to SCHARP, a software application for the analysis and plotting of individual patient data (IPD) meta-analysis
- Individual participant data (IPD) meta-analysis of prognostic studies: combining aggregate data and IPD
- Individual patient data meta-analysis of high-frequency oscillatory ventilation trials in preterm infants: addressing questions left unanswered by aggregate data meta-analysis
- Individual Participant Data (IPD) meta-analysis of diagnostic and prognostic modeling studies: a practical introduction to their rationale and conduct
- Complex interventions and composite outcomes in Individual Participant Data (IPD) meta-analysis: double challenge in i-WIP IPD