First, descriptive statistics were estimated for sociodemographic, alcohol-related, and comorbid psychiatric measures. Means and proportions were calculated for continuous and categorical variables, respectively, and compared across AUD severity categories. Statistical comparisons of mild vs moderate AUD and moderate vs severe AUD were conducted across correlate measures using Wilcoxon rank sum and Fisher exact tests to evaluate the validity of combining mild and moderate as mild-to-moderate AUD. Second, a 1-parameter logistic IRT model assuming unidimensional structure for the 11 AUD criteria was applied to estimate criteria severity using the mirt package (version 1.37.1)31 in R32 (eMethods in Supplement 1). Criteria were rank ordered by severity, and those with severity parameter values greater than 2 (ie, 50% endorsement probability by individuals ≥2 SD above mean AUD latent severity) were considered high risk (Figure 1 and eTable 1 in Supplement 1). Third, individuals with mild-to-moderate AUD either endorsing no high-risk criteria or at least 1 high-risk criterion were classified into low-risk (n = 2486) and high-risk (n = 993) groups, respectively. Low-risk mild-to-moderate, high-risk mild-to-moderate, and severe AUD groups were then compared across
