To control for genotype quality, we excluded SNPs with missing genotypes in > 5% of individuals in each dataset for survival analysis, and > 2% for CSF association analysis. For the EADI cohort, variants with minor allele frequency < 1%, Hardy-Weinberg P value < 1×10−6 and missingness > 2% were removed prior to imputation. Genome-wide genotype imputation was performed using IMPUTE271 with 1000 Genomes reference haplotypes. We excluded imputed SNPs with an IMPUTE2 quality score < 0.5 for survival analysis. For CSF association, we excluded SNPs with an IMPUTE2 quality score of < 0.3 since the dataset was only used for follow-up. In the ADGC, GERAD, CHARGE, and CSF datasets, we then removed SNPs that failed the Hardy-Weinberg equilibrium in controls calculated based on the imputed best-guess genotypes using a P value threshold of 1×10−6. We excluded SNPs with minor allele frequency ≤ 0.02. Finally, we excluded SNPs with available statistics in only one consortium dataset in the meta-analysis. The number of filtered samples and SNPs in each of the above steps are recorded in Supplementary Table 1.
