Background Dabigatran (D) and rivaroxaban (R) showed better benefit-risk than vitamin K antagonists for stroke prevention in nonvalvular atrial fibrillation (NVAF), but there is no direct face-to-face randomized trial comparing D to R. However, our previous results and other studies conducted in real-life settings found similar or better results with D at either dose than R after 1 or 2 years of follow-up. This analysis aimed to estimate the real-life comparative benefit-risk of standard (D150 vs. R20) and reduced doses (D110 vs. R15) on major events over 3 years of follow-up. Methods New D or R users for NVAF in 2013 were identified and followed for 3 years in the SNDS French nationwide claims database. NVAF was defined from long-term disease registration, hospital diagnoses or procedures for atrial fibrillation without valvular disease (3-year database history). D and R patients were 1:1 matched according to dose, gender, age, drug start date and high-dimensional propensity scores (hdPS) including individual stroke and bleeding risk factors (CHA2DS2-VASc and HAS-BLED). Hazard ratios (HR) [95% confidence interval] were estimated on treatment using Cox proportional hazards or Fine and Gray models. Results Of the 10,847 D150, 15,532 D110, 18,829 R20 and 11,195 R15 new users for NVAF in 2013, 8,195 D150/R20 per arm, and 7,651 D110/R15 per arm were matched, i.e. 76% and 68% of patients of the smallest group (D150 and R15), respectively; with very good overlap of hdPS distributions, and most standardized differences < 0.02. The 3-year D vs. R HR for clinically relevant bleeding was 0.53 [0.42-0.66] for standard dose and 0.70 [0.58-0.83] for reduced dose, and respectively 0.78 [0.58-1.05] and 0.80 [0.63-1.02] for stroke and systemic embolism, 1.00 [0.73-1.36] and 0.85 [0.65-1.12] for acute coronary syndrome, 0.83 [0.65-1.07] and 0.91 [0.80-1.03] for death, 0.73 [0.64-0.84] and 0.84 [0.77-0.92] for the composite of the 4 major events. Conclusion This real-life nationwide study with long-term follow-up shows that dabigatran standard and reduced doses have similar or better effectiveness than rivaroxaban with lower bleeding risk.< Réduire