analyses that adjust for other baseline characteristics, which we acknowledge is an important step in later analyses with more complete samples. Readers should view the current findings as provisional, given that they are based on a subset of this sample. Our outcomes may also have been biased in the direction of less severity from deceased subjects, because they included some of the most affected individuals. Selective attrition due to death is inherent in many follow-up studies of chronic disease, and a limitation of this study is that we did not statistically take this into account. The numerical increase in the proportion of males interviewed across the age groups was not anticipated and might reflect, in part, cohort increases in alcohol dependence among younger women. Of the 789 located living subjects, 89% expressed a willingness to be evaluated again in the future, supporting Hypothesis 2.
