Given the high co-morbidity of alcohol and tobacco use, we repeated the original analysis within smokers only to ensure that the signals were not merely a reflection of an association with smoking, shown in Table 6. Each composite was created specifically for the smokers, using FBAT-PC which led to small differences in weights for each phenotypic item compared to what was derived for the entire sample. There was modest evidence for significant associations with a few of the SNPs in the smokers only, but none of these were as significant as in the full sample. A similar trend was found for the average drinks phenotype (Table 7).
