Past-year smoking prevalence was reported for drinkers at risk of alcohol dependence, drinkers not at risk and non-drinkers alongside 95% confidence intervals (CIs). Past-year smoking prevalence was modelled as the outcome in a logistic regression model with the full range of AUDIT scores among drinkers as the predictor, adjusted for survey year. To allow for non-linear trends, AUDIT score and survey year were transformed using natural, cubic splines with five and three knots, respectively, placed at equally spaced quantiles of the data. Splines are preferred over categorisation, which neglects that outcomes are more similar in adjacent than distant AUDIT scores or years.26 Past-year smoking prevalence was described across the full range of AUDIT scores in a figure and prevalence estimates for AUDIT scores of 5, 10, 15, 20, 25, 30, 35, and 40 reported in a table.
