OBJECTIVES: Warfarin anticoagulation has been shown to effectively decrease the rate of thromboembolism (TE), with an associated increased rate of bleeding. Maximizing the benefit and minimizing the risk depends on maintaining patients in a narrow therapeutic range for prothrombin time (measured as INR); increased percentage of time spent in target range (TTR) is predictive of better clinical outcomes. However, strategies to improve TTR can be costly. To allow local clinical policy makers to determine whether these strategies can be justified, we developed a general approach to estimating bleeding and TE rates associated with a given population distribution of INR. METHODS: We used data from a large cohort (Cannegieter, 1995) to derive a logistic equation for baseline absolute risk of bleeding as a function of INR. Based upon a separate large cohort of patients (Fihn, 1993) with various time since initiation of therapy we derived a risk ratio that modifies the above baseline relationship in order to take into account time since initiation of therapy. A similar strategy was used to estimate TE, using INR and indication (mechanical valve versus atrial fibrillation) as predictors. RESULTS: The model for bleeding was: logit(p)=-8.84 + .83 INR + .64 EARLY, where EARLY=1 if the patient is in the first 3 months of anticoagulation, and 0 otherwise. The final model for TE was: logit(p)=-.73 -1.17 INR - .64(AF), where AF=1 if the indication is atrial fibrillation and 0 if the indication is mechanical heart valve. The integral of these logistic equations, weighted by a given population distribution of INR, provides population estimates of annual bleed and TE rates. CONCLUSION: Existing epidemiological data can provide tailored estimates of concrete benefits resulting from improving the quality of anticoagulation.