Since the time our statistical plan was designed (2003), responder analyses based on definitions that predict clinical benefit have been proposed by the FDA as preferable to analyses of group means.31–33 The FDA's rationale for this change is that mean differences are difficult to interpret with regard to clinical relevance.31–33 Thus we modified our original analysis of mean abstinence duration to be a responder analysis based on the rate of complete abstinence over the 12-week study. We also included the rate of no heavy drinking over the 12-week trial as a co-primary outcome, as this has become a standard outcome in alcoholism clinical trials.32 We used a mixed effect model of drinking quantity (number of drinks per week) and frequency (number of heavy drinking days per week) over the 12-week study period, as supportive primary outcomes.34 We also report change in GGT, a widely accepted and validated biomarker of drinking reduction, as a supportive primary outcome.35
