AIMS: Direct oral anticoagulants (DOACs) are recommended for stroke and systemic embolism prevention in patients with non-valvular atrial fibrillation (NVAF) in preference to vitamin K antagonists (VKAs, e.g. warfarin). This study assessed the cost-effectiveness of apixaban versus warfarin, dabigatran, edoxaban, and rivaroxaban for patients with NVAF from the Belgian payer perspective. METHODS: Based on a previously-validated model, a lifetime cohort-level Markov model was built in Microsoft Excel for the Belgian setting, incorporating costs (2025) representative of a DOAC market following patent expiry. The cohort (n = 1,000) included patients with NVAF eligible for oral anticoagulation. Different study types provided clinical inputs for event risk in the base case analysis (based on a network meta-analysis of randomised controlled trials) and in a scenario analysis (based on real-world evidence). The primary model outcome was the incremental cost-effectiveness ratio (ICER), measured by the cost per quality-adjusted life year (QALY) gained. Cost-effectiveness was determined by a willingness-to-pay threshold of €30,000/QALY gained. RESULTS: Driven by a reduction in clinical event risk, apixaban treatment was associated with QALY gains of 0.306, 0.068, 0.115 and 0.034 compared with warfarin, dabigatran, edoxaban, and rivaroxaban, respectively. In the base case analysis, apixaban treatment was associated with lower total costs due to a reduction in acute event costs and long-term management costs, resulting in apixaban dominating (i.e. more effective and less costly) all comparators from a Belgian payer perspective. Similar conclusions were observed in a scenario analysis using clinical inputs derived from real-world evidence. CONCLUSIONS: Apixaban is associated with fewer clinical events and lower total costs versus warfarin, dabigatran, edoxaban, and rivaroxaban for patients with NVAF, dominating (i.e. more effective and less costly) all comparators from a Belgian payer perspective under post-patent-expiry assumptions. The consistency of model outcomes between base case and scenario analyses suggests a level of robustness to these findings.