Replication and Meta‐analysis of Genome‐Wide Association Studies

Replication and meta-analysis are routinely conducted when following up genome-wide association studies. Replication is a mandatory step in confirming the validity of associations. Technical replication verifies the integrity of the original data, while direct replication reproduces the association as closely as possible in independent data. Indirect replication, while not providing full support for the original association, provides evidence for its generalisability. A statistical effect that bears on the power of replication studies is the winner's curse, whereby the effect size seen in a discovery study tends to be more extreme than in truth. Several methods are available to correct for the winner's curse, depending on how associations are first identified and whether replication data are already available. Meta-analysis, in which results from several studies are combined, is increasingly important as individual studies have limited power to detect genetic associations. Although standard fixed and random effects models can be used, more powerful methods are available for detecting effects when there is heterogeneity across studies. Heterogeneity may arise from variation in linkage disequilibrium patterns, case ascertainment and interaction effects. Imputation allows the combination of studies that have genotyped different markers. Increasingly, consortia are being formed in advance of initial data analysis, allowing harmonisation of quality control criteria and sharing of individual-level data.