Meta-analysis is a statistical technique for combining sources of quantitative evidence. It is an example of secondary analysis, which is a subsequent analysis of clinical trial data. The original, or primary analysis of the data was for the purpose that the data was collected – with regards to the data presently shared on Vivli, a now-completed clinical trial.
Meta-analysis, like other forms of secondary analysis, looks back at data to answer a new question. Pairwise meta-analysis compares two sources of quantitative evidence “head-to-head.” For example, with clinical trial data a pairwise meta-analysis could compare directly the interventions in two different clinical studies; or it could compare an intervention and control arm.
Network meta-analysis evaluates more than two interventions against one another. Network meta-analysis can provide an estimate of the relative effectiveness of all interventions in the network – which can be helpful for improving health care decision-making.
Meta-analysis has traditionally been performed with aggregate data, including summary statistics (mean differences, event counts, odds ratios, hazard ratios etc.) extracted from published journal articles, conference abstracts, trial registries (e.g. clinicaltrials.gov), and unpublished documentation such as protocols, statistical analysis plans, and clinical study reports. However, the gold standard in meta-analysis involves the use of not aggregate but individual participant-level data or IPD. In IPD meta-analysis, the original participant level data is requested and re-analysed.
IPD meta-analysis allows a more flexible and complex analysis approach. The advantages are many, and include:
• Researcher can standardise or redefine outcomes
• Researcher can reinstate participants who may have been excluded
• Reduces publication, reporting and ecological biases
• Allows detailed checks of any analysis assumptions (e.g normality or proportional hazards).
• Allows for modelling heterogeneity (within and between studies)
• Consideration of covariates and treatment-covariate interactions
• Allows for modelling of prognostic and diagnostic data in synthesis
While the advantages are many, historically, IPD meta-analysis has been a difficult undertaking for even experienced researchers. However, new tools and platforms are changing the landscape of secondary analysis. If you would like to learn more about IPD meta-analysis, watch our webinar presentation on this topic. The learning objectives for our IPD meta-analysis webinar include:
• An introduction to individual patient data (IPD) meta-analysis
• Explanation of how platforms such as Vivli can facilitate IPD meta-analysis
• Exploration of the differences between IPD meta-analysis and aggregate data meta-analysis, including determining when IPD meta-analysis is the best approach to use.
• Review of examples of IPD meta-analysis.
• Discussion of practical aspects of retrieval of IPD from different sources.
• What to do when not all IPD is available for analysis.
If you have any questions about this blog or the Vivli platform, please email support@vivli.org.