PubMed2002
BACKGROUND: In alcoholism treatment clinical trials, conventional analysis of efficacy outcomes often focuses on the time to a first event, where the event may be "any drinking", "safe (or low risk) drinking", "moderate drinking" or "heavy drinking," in addition to multiple outcomes such as frequency of drinking days, percent abstinence days, etc. METHODS: We consider the multivariate failure time analytic methods. In alcoholism treatment trials, the naturalistic course of drinking behavior during treatment intervention often presents with a gradual change in drinking before the emergence of a more stable drinking or abstinence pattern. Thus, for each subject, evaluation of all drinking events, and incorporating the event times over a defined duration, may give a more comprehensive description of his/her drinking pattern. As a consequence, the efficacy of a new treatment for alcoholism may be elevated with greater statistical sensitivity. RESULTS: The utility of the multiple failure time method is demonstrated via a real case study for evaluation of alcoholism treatments. The multiple event time analyses showed that the risk of having "any drinking days" or "heavy drinking days" during the entire duration of the study was significantly lower with experimental treatment than with placebo. Further explorations showed that the treatment effect was primarily observed in the later relapse events and not the first event with respect to relapse to any drinking episodes. Such effect would have missed using the traditional time to first event analysis approach. The observed effect of treatment with respect to relapse to multiple heavy drinking episodes was shown not only in the first event but also in the later events. CONCLUSION: The multiple failure time approach may be applicable when 'drinking failure' is variously defined as a single drink, one at-risk drinking day, one heavy drinking day, or one alcohol-related social, occupational or medical problem. If "a drinking episode" is properly defined and the design gains statistical efficiency, the multiple event analytic strategy should provide improved statistical power to detect treatment effects.