OBJECTIVES: A large proportion of patients with nonvalvular atrial fibrillation (NVAF) remain untreated with warfarin. We aimed to compare survival and adverse outcome in patients with NVAF treated with or without warfarin. METHODS: A number of data sources were linked from a large population in the United Kingdom from 1995 to 2000. Adverse events were characterised from inpatient and death records. Survival was estimated using Kaplan-Meier and Cox proportional hazards models, and comparisons between event rates using logistic or Poisson regression and the Sign test. Patients were assumed to be receiving warfarin if they had more than four INR measurements. RESULTS: We identified 6108 patients with NVAF, of whom 36.4% received warfarin. Mean survival in warfarin and non-warfarin groups was 52.0 months and 38.2 months, respectively (p <0.001), and 14.4 months (p <0.001) after adjustment for confounding factors. Warfarin patients in the upper quartile of INR control had significantly longer survival (57.5 months) compared with those in the lowest quartile of control (38.1 months; p <0.001). The risk of stroke in the warfarin group when on treatment was lower than that in the non-warfarin group (Relative Rate=0.74, p <0.001). The risk of death from ischaemic stroke was lower (RR=0.43; p <0.001) and there was a lower risk of all ischaemic and embolic events (RR=0.74; p <0.001) in the warfarin group when on treatment compared to the non-warfarin group. Risk of bleeding in the warfarin group on treatment was greater compared to the non-warfarin group (RR=1.78; p=0.001). CONCLUSIONS: These data suggest that failure to treat according to guidelines in routine clinical practice, both in terms of patients selected for anticoagulation and difficulty maintaining those selected patients within the recommended target INR range of 2.0-3.0, is associated with additional mortality and morbidity in patients with NVAF.