OBJECTIVES: In order to indirectly compare new oral anticoagulants (NOACs) for patients with atrial fibrillation (AF), several network meta-analyses (NMAs) have compared the number of patients with stroke or systemic embolism (SE) at study end. The aim of the present analysis was to assess the comparative efficacy of NOACs over the entire duration of the studies, including changes in comparative efficacy over time, using the published cumulative hazard rates. METHODS: A Bayesian NMA was performed using a fractional polynomial model synthesizing data from three pivotal randomized controlled trials: ARISTOTLE, RE-LY, and ROCKET-AF, which evaluated apixaban, dabigatran, and rivaroxaban, respectively, versus warfarin. Parametric survival functions were used to model the hazard rate over time for the compared interventions and the difference in the shape and scale parameters of these functions was synthesized and indirectly compared. The efficacy of NOACs was evaluated and compared to constant HRs from previous NMAs.  RESULTS: The time-varying hazard ratios (HRs) versus warfarin suggest that each NOAC is at least as efficacious as warfarin with respect to stroke and SE. The HR for dabigatran 110mg was fairly constant over time (range: 0.92-0.90). The HR for dabigatran 150mg decreased slightly over time (range: 0.78-0.56), whereas the HRs increased slightly over time for rivaroxaban (range: 0.59-1.17) and apixaban (range: 0.55-1.11). The HRs for each treatment comparison versus warfarin were transformed into cumulative hazard rates per treatment.  CONCLUSIONS: Based on the NMA of stroke or SE among the intention to treat population with AF, dabigatran 110mg is expected to be comparable to warfarin; dabigatran 150mg is expected to be comparable to warfarin for the first 5 months and more efficacious up until 30 months; rivaroxaban and apixaban are expected to more efficacious than warfarin for the first 11 and 12 months, respectively, and comparable to warfarin thereafter.