Background: A major advantage of meta-analyses using pooled raw individual patient data (IPD-MA) of randomised trials, compared to conventional meta-analysis of published data (MAP), are the possibilities for subgroup analysis. IPD-MA have greater power to carry out informative subgroup analyses allowing a more thorough assessment as to whether differences are spurious or not. Both the IPD-MA and MAP have never been compared with respect to the methodology used.
Objective: To identify differences in the methodology of subgroup analyses between IPD-MA and MAP.
Methods: An extended literature search was performed to identify all IPD-MA of randomized controlled trials, followed by a related article search to identify MAPs on the same research question. Data were extracted with special attention to the methodology of subgroup analysis.
Results: We have identified 130 IPD-MA. To date, we have completed the search for 21 IPD-MA, these relate to 13 MAP. Fewer studies were included in the IPD-MA as compared to the MAP; median of 5 and 16 respectively. Subgroup analyses were performed in 12 IPD-MA (57%), and 9 MAPs (69%). In 8 of the 12 IPD-MA (67%) and in 4 of the 9 MAPs (44%) that performed subgroup analyses, the subgroups were defined a priori. In 7 IPD-MA (58%) and in 1 MAP (11%) an interaction test was performed, whereas 9 IPD-MA (75%) and 8 MAPs (89%) analysed their data stratified per trial.
Conclusion: In spite of common objectives both IPD-MA and MAP use distinctly different strategies to perform subgroup analysis. The reasons and implications of these differences require further study. Final results will be presented at the Cochrane Colloquium.