To improve our understanding of observed associations, the top SNPs identified in the whole SAGE dataset were examined using the same models described above in strata of the data defined by study (COGEND, COGA, FSCD), smoking status (FTND cases and FTND controls), and alcohol use disorder (DSM-5 cases and DSM-5 controls). A two-SNP haplotype analysis was run in R using the top SNP and the SNP tagging the bin previously found to be genome-wide significant with nicotine dependence (rs1451240)23. This model included the covariates age, sex, study, DSM-5 alcohol symptom count, FTND total (in the cocaine haplotypes) and DSM-5 cocaine use disorder (in the FTND haplotypes), and examined the association with each haplotype with the phenotype compared to homozygotes for the reference allele at both SNPs. Finally, we used conditional analyses to examine the extent of independence between these cocaine-associated SNPs and the previous association in the region with nicotine dependence tagged by rs1451240.
