Although the randomized trials of warfarin in atrial fibrillation to which Brodsky and colleagues refer did report "all-cause" mortality, we only abstracted rates of stroke, major bleeding events, and fatalities to estimate these probabilities for our decision model. For the hypothetical cohorts of patients assigned to the "warfarin" or "no treatment" strategies, the "die from other causes" arm of the model represents a baseline population-based mortality rate, not all-cause mortality from the randomized trials. Similarly, for patients taking quinidine, all-cause mortality was not used. We calculated an "excess" mortality rate by subtracting the mortality for patients in the placebo arms of the randomized trials from that of patients taking quinidine. We also reviewed each reported death to determine whether it could have been attributable to quinidine.

As we have acknowledged, data on the use of low-dose amiodarone for atrial fibrillation are limited by the absence of randomized trials. Therefore, estimating these probabilities was somewhat more difficult. We felt the most valid method by which to estimate both toxicity and excess mortality was to pool the results of all available cohort studies using low-dose amiodarone rather than simply to choose one or two studies. In calculating their own incidence of fatal pulmonary toxicity, Brodsky and colleagues have taken the cumulative incidence of pulmonary toxicity from the study by Herre and associates [1] (in which no cases of fatal pulmonary toxicity were reported and in which many patients received high doses) and multiplied it by an arbitrarily chosen case fatality rate (10%). We feel this to be somewhat less valid. Even so, their 2% rate (or 0.4% per year over 5 years) is still lower than the threshold rate we reported (0.52%) for the warfarin strategy to have been favored over low-dose amiodarone.

The reported incidence of amiodarone-induced pulmonary toxicity varies widely but is generally lower with lower doses [2]. Of note, most patients in the review referenced by Brodsky and colleagues (23% mortality rate) received more than 400 mg per day [3]. In the largest case series of amiodarone-induced pulmonary toxicity of which we are aware, Dusman and colleagues [4] followed 573 patients over 10 years. No cases occurred in patients receiving less than 305 mg per day. The Cardiac Arrest in Seattle: Conventional Amiodarone Drug Evaluation (CASCADE) study recently found a 10% cumulative incidence of pulmonary toxicity at 3 years for patients randomly assigned to receive low-dose amiodarone, with no associated fatalities [5].

We chose not to incorporate the estimated costs of each strategy into our decision model. Although we agree with Brodsky and colleagues that these strategies should be compared on the basis of cost, this is a separate question and is beyond the scope of the present study. A formal cost-effectiveness analysis is presently under way.

Although our decision model did favor the use of low-dose amiodarone for maintenance of normal sinus rhythm over a strategy of anticoagulation with warfarin, warfarin was in fact favored over an antiarrhythmic strategy that used quinidine. Dr. Matchar and colleagues are correct, however, in stating that no randomized trials have specifically evaluated the effectiveness of any antiarrhythmic strategy for preventing strokes in patients with atrial fibrillation.

Our conclusions hinge on data derived from published studies, as well as several assumptions. The two to which Dr. Matchar and colleagues have taken exception both address an important question: What is the incidence of stroke in patients diagnosed with atrial fibrillation in whom an attempt is made to maintain normal sinus rhythm with antiarrhythmic therapy? Unfortunately, no existing data specifically address this question. As an initial estimate of this variable, we used population-based rates from numerous population-based stroke registries. Although these rates were remarkably similar (0.09% to 0.14% per year), we felt these populations were not necessarily representative of those that develop atrial fibrillation in that the latter are more prone to have independent risk factors for stroke, including preexisting hypertension. Therefore, a wide range for sensitivity analysis was necessary. A threshold analysis (unpublished data) showed that for the warfarin strategy to be favored over low-dose amiodarone, the risk for stroke while patients received antiarrhythmic agents and experienced normal sinus rhythm would have to exceed 1.62% per year. Although not inconceivable, this would seem unlikely given the annual rate of 1.35% per year in patients who persistently experience atrial fibrillation while receiving warfarin. Perhaps future randomized trials of antiarrhythmic therapy for chronic atrial fibrillation will allow more accurate determination of this important variable.