BACKGROUND: Randomized trials have shown a beneficial effect of anticoagulation with warfarin to prevent stroke in atrial fibrillation. It is not known whether the same effect will be obtained in actual practice. The authors conducted a prospective observational study to evaluate the effect of preventive anticoagulation in patients with atrial fibrillation in 2 practice settings in Montreal. METHODS: Of the 1725 outpatients screened between October 1990 and September 1993 at a community hospital and a university-affiliated hospital, 221 with documented atrial fibrillation were enrolled and followed up for a mean of 27 months. Most (75%) of the patients excluded did not meet the inclusion criteria (because of, for example, an artificial heart valve, mitral stenosis, cardiac transplantation or transient atrial fibrillation); the remainder had not completed enrollment before the end of the study. Following the baseline visit, patients were interviewed by telephone every 6 months, and reported events were confirmed through review of the patients' charts. Hazards for stroke and for stroke and transient ischemic attack (TIA) combined were calculated for each of 4 treatment groups: ASA, warfarin, blended treatment and no treatment, based on the type of anticoagulation therapy patients received during the entire observation period. The blended-treatment group consisted of patients who started on one active therapy and switched to the other or who switched treatments more than once. Corresponding rate ratios (RRs) and 95% confidence intervals (CIs) were calculated with reference to the no-treatment group. Cox proportional hazards analysis was used to adjust for differences in patient characteristics. The rates of bleeding episodes were also analysed. RESULTS: On average, the study patients were older (71.6 [standard deviation 9.3] years) and had a higher prevalence of underlying heart disease (52.0%) than those in the randomized trials. Nineteen patients had a first stroke: 4 in the ASA group, 4 in the warfarin group, 4 in the blended-treatment group and 7 in the no-treatment group, for rates of 5.2, 1.8, 5.3 and 5.9 per 100 person-years, respectively. Only warfarin was associated with a significantly lower risk of stroke compared with no anticoagulant therapy (RR 0.31, 95% CI 0.09-1.00). A similar protective effect of warfarin was found for stroke and TIA combined (2.3 v. 6.7 per 100 person-years; RR 0.34, 95% CI 0.12-0.99); the effect of ASA and blended treatment was not significantly different from no treatment. The rate per 100 person-years of any bleeding was not significantly higher for any treatment group (ASA 2.5, warfarin 3.4 and blended treatment 3.5) compared with the no-treatment group (1.9). Patients receiving warfarin had a significantly greater risk of any bleeding event than patients not receiving anticoagulant therapy (RR 1.79, 95% CI 1.07-3.00). INTERPRETATION: The relative effect of anticoagulant therapy with warfarin in preventing stroke in these practice settings was equivalent to that in the randomized trials, although these patients were older and sicker. This preventive treatment is likely to confer additional benefit as it is more widely prescribed.