Maximizing the value of IPD when meta-analysing randomised trials
Background: Individual participant data meta-analysis (IPD MA) of randomised trials (RCTs) is considered the gold standard for evaluating whether an intervention is beneficial, and further to identify which individuals benefit most from the intervention. However, IPD requires considerable time and resources from both reviewers and trial investigators. It is therefore important that IPD is used optimally.
Methods: Our objective is to elucidate some key considerations and possibilities for optimal IPD use, throughout the entire course of an IPD MA project, using an example to demonstrate the potential benefits this brings. Firstly, it is critical that appropriate and complete data is collected, thoroughly checked and documented. Units and coding systems, precise outcome definitions, and covariate adjustment factors should be harmonized across trials. Prior to analysis, missing data, power for treatment-covariate interaction analysis, design complexities, and non-proportionality of hazards (for time-to-event outcomes) should be considered at the trial level. Considerations at the meta-analysis level may include scale of effect measure (e.g. relative vs absolute), avoidance of aggregation bias when analysing treatment-covariate interactions, and consideration of multiple interactions and/or higher-order interactions. Finally, results should be presented and described clearly so as to be understood by different stakeholder groups. All methods should be detailed in advance within protocols and analysis plans.
Results: In our example, an IPD MA of RCTs investigating the long-term benefits of adding docetaxel chemotherapy to standard of care for men with metastatic hormone-sensitive prostate cancer, we demonstrate considerable gains from maximising the use of collected IPD. This includes identification of clinically-important complex interactions which may not have been identified using simpler methodology.
Conclusions: Given the considerable effort, resource, and time it takes to collect IPD, it is important to plan, check data, and undertake analysis in a manner which fully maximises its potential. Considering detailed investigations at the trial level can greatly inform the eventual MA, bringing substantial benefits.
Patient, public and/or healthcare consumer involvement: Two patient representatives have been involved in the design, interpretation, and dissemination of our example meta-analysis. Both are supportive of the importance of maximising the value of IPD in our MAs going forward.