Baseline demographic and clinical characteristics were compared by χ2 and ANOVA as appropriate. Outcome analyses were intention-to-treat and involved all subjects who were randomly assigned (n=150). All tests were 2-tailed, and an alpha <0.05 was considered statistically significant. Linear dose effects for rates of complete abstinence and no heavy drinking over the 12-week study were assessed using the extended Mantel-Haenszel χ2 test for linear association.38, 39 This test uses a single contingency table where both row (0, 900, 1800 mg) and column (responder, non responder) variables are ordinal values and at least one variable has more than 2 levels. It is used to specifically assess dose-effect, with df=1, and multiple comparisons are not required. To facilitate clinical interpretation of primary outcomes, the number needed to treat (NNT) and odds ratio (OR) were calculated as estimates of effect size for each drug group relative to placebo. Reasons for early termination were coded at time of termination under double-blind conditions and served as the basis for the following assumptions: 4 subjects who were verified as abstinent for their entire study participation and
