| Baseline drink-year quartile | Unadjusted model* | Adjusted model† |
| RR (95% CI), p value | RR (95% CI), p value | |
| Non-drinkers | Reference | Reference |
| Lowest quartile (⩽40.7) | 1.13 (0.55 to 2.33), p = 0.73 | 1.43 (0.68 to 2.98), p = 0.34 |
| 2nd quartile (40.8 to 110.0) | 1.31 (0.68 to 2.52), p = 0.42 | 1.75 (0.89 to 3.42), p = 0.10 |
| 3rd quartile (110.1 to 138.0) | 1.27 (0.63 to 2.54), p = 0.50 | 1.82 (0.89 to 3.74), p = 0.10 |
| Highest quartile (>138.0) | 1.49 (0.77 to 2.86), p = 0.24 | 2.29 (1.15 to 4.54), p = 0.018 |
The reference category was non-drinkers. In analyses stratified by gender, comparing highest baseline drink-year quartile with non-drinkers, adjusted RRmen = 2.93 (95% CI 0.79 to 10.81, p = 0.11) and adjusted RRwomen = 1.66 (95% CI 0.50 to 5.51, p = 0.41). We created two education strata (high education = primary or secondary studies vs low education = can read or write or illiterate). In analyses stratified by education, comparing highest baseline drink-year quartile with non-drinkers, adjusted RRhigh education = 2.19 (95% CI 0.81 to 5.94, p = 0.13) and adjusted RRlow education = 2.25 (95% CI 0.86 to 5.85, p = 0.097). In an analysis of daily number of drinks at baseline (rather than drink-years), when those in the highest category (>5 drinks/day) were compared with non-drinkers, RR = 1.94 (95% CI 0.26 to 14.67, p = 0.52). When those who had 3 to 5 drinks/day were compared with non-drinkers, RR = 2.19 (95% CI 0.66 to 7.31, p = 0.20). When those who had 1 to 2 drinks/day were compared with non-drinkers, RR = 1.30 (95% CI 0.78 to 2.17, p = 0.31).
*Unadjusted Cox proportional hazards model.
†Cox proportional hazards model adjusting for cigarette pack-years, depressive symptoms and community.