Background The use of anticoagulation (AC) in AF is challenging because stroke risk reduction and increased bleeding risk must be considered simultaneously. We developed a decision model (DA) integrating stroke and bleeding risks to guide optimal use of AC and compared actual warfarin utilization in a real-world database with model recommendations. Methods A Markov model simulated health state transitions for newly diagnosed AF patients (N=64,946; mean age=71, 45% females) from the Thomson Reuters Marketscan database. The model evaluated occurrence and disability associated with stroke, intracranial and extracranial hemorrhage, and mortality with warfarin or aspirin treatment. Probabilistic transitions were based on patient demographics, stroke and bleeding risk assessed by CHADS 2 (Gage 2001), and ATRIA index scores (Fang 2010), respectively, and AC treatment effects from published clinical trials and other sources. The model recommended warfarin if estimated quality adjusted life expectancy was higher than for aspirin. Model recommendations were contrasted with warfarin prescription claims in the Marketscan AF cohort. Results Overall, 74.8% (n=48,548) of the cohort had CHADS 2 ≥1, of which 85.7%, 4.6%, and 9.7% had low (0-3), moderate (4), and high (≥5) ATRIA bleeding risk, respectively. Almost all (91.9% of 48,548) patients with CHADS 2 ≥1 were recommended to receive warfarin. The AC decision was less certain in patients with moderate stroke and moderate/high bleeding risks. Concordance of warfarin recommendation (Rec. W) and utilization (Rx W) was low and decreased with higher bleeding risk. Stroke risk N Rec. W Rx W N Rec. W Rx W N Rec. W Rx W CHADS 2 ≥4 (high risk) 2,558 100% (n=2,558) 58.7% (n=1,501) 339 100% (n=339) 51.9% (n=176) 900 97.1% (n=874) 50.8% (n=444) CHADS 2 = 2 or 3 (moderate risk) 18,093 99.7% (n=18,043) 64.1% (n=11,574) 1,437 76.7% (n=1,102) 59.0% (n=650) 2,873 46.3% (n=1,330) 48.1% (n=640) CHADS 2 = 1 (low risk) 20,930 97.3% (n=20,365) 65.3% (n=13,304) 468 0% 0% 950 0% 0% ATRIA = 0 to 3 (low risk) ATRIA = 4 (moderate risk) ATRIA ≥5 (high risk) Bleeding risk Conclusions This analysis is distinctive in its use of explicit bleed risk (by ATRIA index) and comparison of DA with clinical care. High discordance between actual warfarin use and model recommendation suggests the AC decision is often not based on systematic evaluation of stroke and bleeding risks. Model-based clinical decision aids may improve oral AC decisions.