In addition to AUD group classifications based on the 10-year follow-up, we employed a complementary approach to determine trajectory subgroups based on the growth and progression (or regression) in AUD symptom severity throughout all follow-ups (10,11,19). These were based on DSM-IV symptom count as DSM-5 was not yet developed when the follow-ups initiated in 2005. This trajectory analysis (34) included a zero-inflated Poisson mixture model with a cubic trajectory for each subgroup. The number of trajectory groups was determined by model Bayesian Information Criterion (BIC) using 2ln(B10) ≥ 10 as strong evidence for rejecting the null model. This analysis showed that a three-trajectory group model best fit the data, including a low AUD symptom subgroup (low AUD SX, n=103), an intermediate AUD symptom subgroup (intermediate AUD SX, n=62), and a high AUD symptom group (high AUD SX, n=19). The AUD symptoms across these subgroups corresponded to heavy drinking frequency over follow-up (Figure 1B). Of note, the classifications of individuals in the AUD symptom trajectory subgroup were fairly stable, relative to the classifications of initial heavy drinkers in the 5-year analysis
