of death and specific death causes are the subject of a separate paper. Third, all data were from self-reports without corroborating information. A fourth caveat is inherent in the COGA dataset where the original families were selected based on a proband who was in alcohol-related treatment and had multiple relatives with AUDs. Therefore, this is not an epidemiological study of individuals with AUDs in the general population (e.g., Grant et al., 2015). These considerations indicate that the results might not be representative of all individuals with AUDs and might be different in less severely impaired individuals without alcoholic close relatives. Fifth, reflecting concerns regarding what events are likely to be most accurately remembered, details of drinking quantities and frequencies were limited to the 12 months preceding the follow-up interview, while major drinking problems were recorded regarding the prior five years. Sixth, although age differences across groups were relatively small and age was included in regression analyses, it is possible that some of the differences across outcome groups could reflect differences in age. Finally, the COGA baseline data available as predictors of outcome categories did not include other valuable variables such as the greater alcohol-related stimulation early in the blood alcohol
