Most/all best friends drinking compared to none/few best friends drinking between ages 12-17 was associated with a main hazardous effect in univariate and multivariate models of early drinking behaviors (Table 2). In the final interaction model set with GxE product terms, self-reported peer drinking had a robust effect on first intoxication (Hazards ratio (HR)=1.81, 95% CI 1.62–2.01) and first DSM-5 symptom (HR=2.17, 95% CI 1.88–2.50). In secondary analyses examining all four responses for best friends drinking (none, few, most, all), an increase in the number of best friends drinking was similarly related to the first intoxication (multivariate model set with none as the reference; few HR=1.72, 95% CI 1.44–2.05; most HR=2.65, 95% CI 2.20–3.18; all HR=3.69, 95% CI 2.93–4.64) and first DSM-5 symptom (multivariate model set with none as the reference; few HR=2.43, 95% CI 1.77–3.33; most HR=4.29, 95% CI 3.12–5.92; all HR=5.84, 95% CI 4.16–8.21). These results indicate a “dosage effect” where the reported proportion of best friends drinking was positively associated with higher risk for developing adolescent drinking milestones.
