Meta-analysis with overlapping subjects: an adjusted Fisher's method

Meta-analysis is a popular approach to combine results from multiple genome-wide association studies (GWAS) without sharing individual-level data. Large consortiums can meta-analyze dozens to hundreds of individual studies. For example, the HostSeq project is a Canadian initiative in which 15 clinical and epidemiological studies were assembled with DNA samples from ∼10,000 Canadians infected by the SARS-CoV-2 virus. However, individuals might have been recruited into more than one study, which induces sample overlaps between GWAS and thus violates the independence assumptions in traditional meta-analysis approaches. To resolve this problem, we propose an adjustment to Fisher's method that explicitly corrects for the correlation between the p-values to be combined. This adjustment requires no individual level data and minimal summary statistics from the overlapping samples. We demonstrate with simulation studies that our proposed method is accurate and has comparable power with mega analysis. We further applied our method on partially overlapping studies in the HostSeq initiative.