Modeling longitudinal drinking data in clinical trials: an application to the COMBINE study.