ABSTRACT Genome-wide association studies (GWASs) of the Alcohol Use Disorder Identification Test (AUDIT), a ten-item screener for alcohol use disorder (AUD), have elucidated novel loci for alcohol consumption and misuse. However, these studies also revealed that GWASs can be influenced by numerous biases (e.g., measurement error, selection bias), which have led to inconsistent genetic correlations between alcohol involvement and AUD, as well as paradoxically negative genetic correlations between alcohol involvement and psychiatric disorders/medical conditions. To explore these unexpected differences in genetic correlations, we conducted the first item-level and largest GWAS of AUDIT items (N=160,824), and applied a multivariate framework to mitigate previous biases. In doing so, we identified novel patterns of similarity (and dissimilarity) among the AUDIT items, and found evidence of a correlated two-factor structure at the genetic level (Consumption and Problems, rg=.80). Moreover, by applying empirically-derived weights to each of the AUDIT items, we constructed an aggregate measure of alcohol consumption that is strongly associated with alcohol dependence (rg=.67) and several other psychiatric disorders, and no longer positively associated with health and positive socioeconomic outcomes. Lastly, by performing polygenic analyses in three independent cohorts that differed in their ascertainment and prevalence of AUD, we identified novel genetic associations between alcohol consumption, alcohol misuse, and human health. Our work further emphasizes the value of AUDIT for both clinical and genetic studies of AUD, and the importance of using multivariate methods to study genetic associations that are more closely related to AUD.