While the first tranche of COGA GWAS data followed a case–control design, 72 , 73 all subsequent COGA analyses have used family‐based analytic approaches. The initial family‐based GWAS of COGA, 74 , 75 conducted in a second subset of the data, was analyzed using Genome‐Wide Association analyses with Family data (GWAF 76 ). Subsequent GWAS combining the case–control and family‐based resources 77 allowed for analyses using the Generalized Disequilibrium Test 78 and linear mixed effects modeling with kinship matrices (lmekin) from the coxme package 79 in R. 80 As genotyping of the COGA sample proceeded, integrated analyses of imputed GWAS data across COGA's dense and multigenerational families posed computational challenges. For instance, commonly used software such as PLINK 62 or analytic modules that accounted for fewer degrees of relatedness (e.g., siblings or trios that could be analyzed using generalized estimating equations 81 , 82 ) did not adequately model these complex family structures, and Genome‐wide Complex Traits Analysis 83 proved far too computationally burdensome for the high dimensional relatedness matrices arising in COGA. Indeed, the commonly used Ricopili pipeline 84
