To better ensure robustness, for each phenotype, we filtered variants from the meta-analytic results to variants that were present in at least three studies, had an effective sample size (sample size multiplied by imputation accuracy) to maximum sample size ratio of ≥ 0.1, and minor allele frequency (MAF) > 0.001 in the multi-ancestry and EUR-stratified meta-analysis or MAF > 0.01 for AMR-stratified, AFR-stratified and EAS-stratified meta-analysis, given the expected drop off in imputation accuracy for those ancestries. These filters reduce potential artefacts arising from sparse data or poor imputation and retain variants with reasonable statistical power.
