All computations of odds ratios (OR) and standard errors for the sub-studies were performed using SAS (16). In each analysis, covariates representing sex and age [using quartiles, defined by 34 years and younger (reference), 35–39 years, 40–44 years, and 45 years and older] were included as categorical variables. Because use of a substance is required to develop dependence, binary covariates representing use of each of the 4 substances were also included. Meta-analyses were performed using the R package r-meta. None of the corresponding tests for heterogeneity based on Woolf’s Test were significant (range p=0.13 to p=0.95). For this reason, reported p-values for the meta-analyses are based on fixed effects models.
