risk for transitioning. Offspring from intact families were right-censored on this variable at age at interview if younger than 18 years: these individuals were not assessed throughout childhood (defined as birth through age 18), and thus contribute to prediction through their age at interview only. In the case of maternal death (n=23), individuals from intact families were right-censored at offspring age when their mother died. Control variables with available ages of onset were also modeled as time-varying covariates. In models predicting timing of first AD symptom and timing of AD diagnosis, offspring entered the analysis (risk-set) at onset of alcohol use. To adjust for the temporal proximity from alcohol initiation to onset of any AD symptom and AD diagnosis, age at first use was included as an additional covariate in the fully adjusted models. In these analyses, age at initiation was modeled as a series of 7 dummy-variables with first standard drink occurring: before age 13, at ages 13, 14, 15, 17, 18, and at age 19+, with the modal age of initiation (16 years) used as the referent group. For all survival analyses, the Efron approximation (Efron, 1977) was used for survival ties. The Grambsch and Therneau test of
