Balancing the Risks of Stroke and Upper Gastrointestinal Tract Bleeding in Older Patients With Atrial Fibrillation

Objective  To determine how factors that increase the risk of major upper gastrointestinal (GI) tract hemorrhage (recent upper GI tract bleeding or concurrent use of nonsteroidal anti-inflammatory drugs) influence the choice of antithrombotic therapy in older patients (those ≥65 years) with atrial fibrillation.

Methods  For older patients with atrial fibrillation and no other contraindications to antithrombotic therapy, a Markov decision-analytic model was used to determine the preferred treatment strategy (no antithrombotic therapy, long-term aspirin use, or long-term warfarin sodium use) based on their risk of major upper GI tract hemorrhage. Input data were obtained by a systematic review of MEDLINE. Outcomes were expressed as quality-adjusted life-years (QALYs).

Results  For 65-year-old patients with average risks of stroke and upper GI tract bleeding, warfarin therapy was associated with 12.1 QALYs per patient; aspirin therapy, 10.8 QALYs; and no antithrombotic therapy, 10.1 QALYs. For persons with significantly higher risks of upper GI tract bleeding and/or lower risks of stroke, warfarin was no longer clearly the optimal antithrombotic therapy (eg, for 80-year-old persons with a baseline risk of stroke of 4.3% per year who were concurrently taking a conventional nonsteroidal anti-inflammatory drug: warfarin, 7.44 QALYs; aspirin, 7.39 QALYs; and no treatment, 7.21 QALYs).

Conclusions  For older patients with atrial fibrillation and factors that place them at a higher than average risk of upper GI tract bleeding, the optimal choice of antithrombotic therapy to prevent stroke can vary according to the magnitude of this risk. Based on the risks of stroke and upper GI tract bleeding, clinicians can use the treatment recommendations of this study to provide rational stroke prevention therapy for older patients with atrial fibrillation.

A TRIAL FIBRILLATION is the most prevalent cardiac arrhythmia, occurring in approximately 5% of persons 65 years and older.¹ Persons with atrial fibrillation are at an increased risk of having a thromboembolic stroke, with an average yearly risk of 5%.² This risk is further increased in the presence of risk factors, including congestive heart failure, history of hypertension, history of stroke or transient ischemic attack, and increasing age.² Long-term antithrombotic therapy with warfarin sodium or aspirin reduces the relative chance of stroke from atrial fibrillation by approximately 65%^3,4 and 20%,^3,5 respectively. Many studies^6-8 have shown that warfarin for prophylaxis of stroke is cost-effective, especially in older persons. Thus, an expert panel recommended that all persons older than 75 years with atrial fibrillation should be considered for long-term warfarin therapy unless a contraindication exists.²

Balanced against this benefit is the risk of antithrombotic-associated life-threatening bleeding complications. Most commonly, these complications involve the upper gastrointestinal (GI) tract, but also include intracranial bleeding (subdural hematomas and intracerebral hemorrhages). The risk of these complications increases with age.⁹ Before routine Helicobacter pylori testing and treatment became the standard of care for patients with peptic ulcer bleeding, the rate of rebleeding after resolution of the initial bleed was between 10% and 20% per year.¹⁰ Also, patients who take nonsteroidal anti-inflammatory drugs (NSAIDs) are clearly at an increased risk of upper GI tract bleeding.¹¹ In fact, the concurrent use of warfarin and NSAIDs in elderly persons is estimated to cause a 13-fold increase in the risk of upper GI tract bleeding.¹² For this reason, many of the studies^13,14 evaluating the effectiveness and appropriateness of antithrombotic therapy in patients with atrial fibrillation have excluded subjects with an increased predisposition to major bleeding. In fact, one study¹⁵ found that patients with atrial fibrillation who had a history of GI tract bleeding were half as likely to receive warfarin as those who did not. Thus, many physicians are reluctant to prescribe antithrombotic therapy (warfarin or aspirin) for older patients (those ≥65 years) with atrial fibrillation whom they deem at an increased risk of major upper GI tract hemorrhage.¹⁶ This decision analysis determines how factors that increase the risk of major upper GI tract hemorrhage should influence the choice of antithrombotic therapy in older patients with atrial fibrillation.

A decision-analytic model was developed to assess how the risk of upper GI tract bleeding influences the choice of antithrombotic therapy in older persons with atrial fibrillation.

A decision-analytic model was constructed to describe the possible outcomes of 3 different treatment strategies for older persons (those ≥65 years) with atrial fibrillation in relation to their risk of upper GI tract hemorrhage and stroke: (1) warfarin therapy, with a switch to aspirin in the event of a nonfatal major upper GI tract bleeding episode; (2) aspirin therapy, with a switch to warfarin in the event of a transient ischemic attack, a reversible ischemic neurologic deficit, or stroke or a switch to no therapy in the event of a nonfatal major upper GI tract bleeding episode; and (3) no treatment, with a switch to warfarin in the event of a transient ischemic attack, a reversible ischemic neurologic deficit, or stroke. An age-specific standardized mortality table was used to model the chance of all-cause mortality. Outcomes were expressed in quality-adjusted life-years (QALYs) for persons aged 65 years at the starting point of the analysis. Quality-adjusted life-years were obtained by multiplying the appropriate life-years in the different health states with the utility estimate for each health state. All life-years were discounted at the rate of 3% per annum.¹⁷

Markov subtrees with identical structures were used to model the chance events associated with the 3 treatment strategies (Figure 1). The probability of each event was based on a systematic review of the published literature. The Markov cycle length was fixed at 3 months, with all relevant probabilities and utilities adjusted to reflect this cycle length. The results of the analysis are reported for a 1-year period. The model was constructed and analyzed using computer software (DATA 3.0; TreeAge Software, Inc, Williamstown, Mass). (Further details regarding the model are available from the authors.)

Six states related to stroke and GI tract bleeding were considered in the analysis:

• Well—the state for persons who have had no adverse event. Adverse events included stroke, intracranial hemorrhage (subdural hematoma or intracerebral hemorrhage), and major upper GI tract bleeding. The well state was the starting point for all persons.

• Minor stroke—a mild residual neurologic deficit corresponding to a modified Rankin Scale¹⁸ 0 or 1 stroke (eg, mild right arm and leg weakness but remaining essentially functionally independent).

• Moderate stroke—a moderate neurologic deficit corresponding to a modified Rankin Scale¹⁸ 2 or 3 stroke (eg, right arm and leg weakness sufficient to require assistance for some functional activities, including bathing and dressing, but having independent ambulation with a walker or a cane).

• Major stroke—a severe neurologic deficit corresponding to a modified Rankin Scale¹⁸ 4 or 5 stroke (eg, total paralysis of the right arm and leg requiring almost total care with functional activities, including help with ambulation and feeding).

• Major upper GI tract bleeding—bleeding sufficient to require a blood transfusion, an emergency procedure, surgical intervention, or hospitalization.⁴

• Dead.

Relevant data for input variables (probabilities, outcomes, and utilities) used in the model were gathered by performing a systematic literature search using the MEDLINE (January 1, 1966, to December 31, 2000) computerized database. Relevant articles were identified by using the following keywords (human only): anticoagulants, cerebral hemorrhage, subdural hematoma, gastrointestinal hemorrhage, duodenal ulcer, stomach ulcer, peptic ulcer, aspirin, warfarin, atrial fibrillation, nonsteroidal anti-inflammatory drugs, recurrent (text word), Helicobacter pylori, outcome assessment (health care), treatment outcome, prognosis, and risk factors. The bibliographies of each article were hand searched to identify additional articles. Content experts were also consulted to identify other relevant published work.

For each input variable, information was extracted from each relevant study. Data pertaining to persons older than 65 years were preferentially sought. Point estimates for input variables were determined by pooling the results from all relevant studies.

Table 1 summarizes the input data used in the decision-analytic model. For the probabilities of events and outcomes related to stroke and intracranial hemorrhage (subdural hematomas and intracerebral hemorrhages), input data from a decision analysis⁶¹ assessing how the risk of falls should influence the choice of antithrombotic therapy in older patients with atrial fibrillation were used, except for one new relevant meta-analysis³ related to the efficacy of warfarin and aspirin in preventing stroke in persons with atrial fibrillation. One hundred nineteen potentially relevant articles related to GI tract bleeding were reviewed, with 22 containing information relevant to this decision analysis. Most excluded articles contained no original data or dealt with recurrent bleeding before the initial GI tract bleeding episode had fully resolved.

The baseline probability of major upper GI tract bleeding in patients with atrial fibrillation receiving no treatment (not taking aspirin or warfarin) was estimated from the Atrial Fibrillation Investigators' (AFI) data,⁴ a pooling of the results of all randomized controlled trials assessing the efficacy of warfarin in preventing stroke in patients with atrial fibrillation. In a similar meta-analysis, Hart et al³ estimated that persons taking warfarin were 2.4 times more likely to develop major GI tract bleeding than those taking no antithrombotic therapy. The AFI meta-analysis^4,5 estimated that patients taking aspirin were 1.2 times more likely to develop major GI tract bleeding than those taking no antithrombotic therapy. These analyses defined major bleeding as bleeding that required a blood transfusion, an emergency procedure, a surgical intervention, or hospitalization. For simplicity, minor bleeding episodes (external bruising or epistaxis) were not modeled.

Before the advent of routine testing and treatment for H pylori in the early 1990s, patients with GI tract bleeding due to gastric or duodenal peptic ulcer disease had a recurrent bleeding rate of approximately 30% over 5 years.^10,62 This excess risk was estimated to be as high as 13.5 times that of persons without previous peptic ulcer bleeding.⁶³ However, persons with GI tract bleeding due to peptic ulcer disease who undergo subsequent H pylori testing and treatment do not seem to be at an increased risk of recurrent bleeding.^49-52 Considering that H pylori testing and treatment is the standard of care for patients with peptic ulcer bleeding, the chance of recurrent bleeding, once the initial episode has resolved, was estimated to be the same as that for persons without a previous upper GI tract bleeding episode.

Multiple studies have assessed the excess risk of upper GI tract bleeding in persons taking NSAIDs, with a recent meta-analysis⁵³ summarizing this information. Therefore, the estimate of the excess risk of upper GI tract bleeding when taking NSAIDs (relative risk, 3.8) was derived from this study.⁵³

A strategy used to improve the GI tract safety of NSAIDs is the concurrent administration of medications to protect the gastroduodenal mucosa. Studies have shown that hydrogen receptor antagonists^64,65 and sucralfate⁶⁶ are ineffective in the prevention of NSAID-associated ulcers and cannot be recommended.¹¹ In persons taking NSAIDs, there is an approximate 50% decreased risk of upper GI tract bleeding with the concurrent use of either proton pump inhibitors⁵⁵ (PPIs) or misoprostol.⁵⁴

Recent data⁵⁶ suggest that persons taking a cyclooxygenase-2 (COX-2)–specific NSAID have a relative risk of serious upper GI tract bleeding that is 0.51 times that of those taking conventional NSAIDs. To our knowledge, no published studies have examined whether concomitant therapy with GI tract protective agents (eg, misoprostol) reduces this risk further.

According to the AFI,⁴ too few deaths occurred from major upper GI tract bleeding episodes to precisely estimate these fatality rates. For persons receiving no treatment or aspirin, data from the Antiplatelet Trialists' Collaborators⁴⁵ were used to estimate the proportion of those dying as a result of a major upper GI tract bleeding episode. For persons receiving warfarin, we estimated the fatality rate from major upper GI tract bleeding from anticoagulation clinic cohort studies.^9,24,46-48 We assumed that all patients who survived a major upper GI tract bleeding episode returned to their preexisting health state.

With no published evidence that the chance of dying from a recurrent bleed is greater than after an initial bleed, that the fatality rate associated with an NSAID-induced bleeding episode is different than after a non–NSAID-induced bleeding episode, and that the proportion of persons dying due to a COX-2 NSAID-induced bleeding episode is different from the proportion of those dying due to a bleeding episode without taking NSAIDs, it was assumed that these fatality rates were similar.

Utilities for disabilities associated with minor, moderate, and major stroke and long-term aspirin and warfarin use were assigned from a study⁵⁷ that interviewed persons with atrial fibrillation for these values. The estimate for the utility of a major upper GI tract bleeding episode was derived from 4 studies.^6,58-60 By definition, the well state and death were assigned the utilities of 1 and 0, respectively.

Figure 2 presents the optimal antithrombotic treatment thresholds for different combinations of stroke and major upper GI tract bleeding risk in older patients with atrial fibrillation.

For persons aged 65 years with atrial fibrillation who had average risks of stroke (6.0% per year) and major GI tract bleeding (1.17% per year), warfarin therapy was associated with 12.1 QALYs per patient; aspirin therapy, 10.8 QALYs; and no antithrombotic therapy, 10.1 QALYs.

For older persons with an average risk of stroke from atrial fibrillation (6.0% per year), warfarin was the preferred therapy if their risk of a major upper GI tract bleeding episode was less than 10.4% per year. For those with major upper GI tract bleeding risks between 10.4% and 30.0% per year, aspirin was the preferred therapy. For those with GI tract bleeding risks of greater than 30.0% per year, no antithrombotic therapy was favored.

For persons with an average risk of having a major upper GI tract bleeding episode (1.17% per year), warfarin was the preferred therapy if their chance of stroke was greater than 2.4% per year. For persons with a risk of stroke between 1.2% and 2.4% per year, aspirin was the preferred therapy. At a yearly stroke risk of 1.2% or less, no antithrombotic therapy was the preferred strategy. In the era of routine H pylori testing and treatment, patients with a recent resolved non–NSAID-related upper GI tract bleeding episode do not appear to be at increased risk of recurrent bleeding. Therefore, the treatment thresholds were the same for these persons.

To receive overall benefit from taking warfarin, persons concurrently taking conventional NSAIDs (relative risk of upper GI tract bleeding: warfarin use, 2.4; NSAID use, 3.8) must have a 4.0% or greater chance of stroke from atrial fibrillation. If their stroke risk was between 1.6% and 4.0% per year, aspirin was the preferred treatment strategy. For those with stroke risks below 1.6% per year, no antithrombotic therapy was indicated.

The use of misoprostol or PPIs will reduce the chance of NSAID-induced major upper GI tract bleeding by approximately 50%.^54,55 Therefore, persons taking NSAIDs in combination with misoprostol or PPIs have a 2.3% chance per year of experiencing a major upper GI tract hemorrhage. The threshold level for treatment with warfarin in these persons was a stroke risk of 3.2% per year. If their stroke risk was between 1.2% and 3.2% per year, aspirin was the preferred treatment strategy. For persons with stroke risks below 1.2% per year, no antithrombotic therapy was indicated. The rate of GI tract bleeding for persons who take COX-2–specific NSAIDs is also approximately 50% of the rate of those taking conventional NSAIDs.⁵⁶ Thus, their optimal stroke prophylaxis thresholds were the same as those of persons concurrently taking NSAIDs and misoprostol or PPIs.

One-way sensitivity analyses (Table 2) were performed to test the robustness of the results to changes in the values of pertinent variables. We examined the influence of each throughout its entire reasonable range.

For elderly persons with an average risk of stroke from atrial fibrillation (6.0% per year) and no additional risk factors for upper GI tract bleeding or a recently resolved upper GI tract bleed for which assessment and treatment for H pylori infection was performed, warfarin was the preferred strategy under all conditions, except when the relative efficacy of warfarin was reduced from 62% to 29%, the relative efficacy of aspirin was increased from 22% to 47%, the relative risk of upper GI tract bleeding when taking warfarin was increased from 2.4 to 11.0 times, or the utility of taking warfarin was less than 0.84. The probability of death from upper GI tract bleeding, the relative risk of aspirin to cause GI tract bleeding, and the utility values for upper GI tract bleeding and major, moderate, and minor stroke did not influence the choice of optimal antithrombotic therapy.

For elderly persons with an average risk of stroke from atrial fibrillation (6.0% per year) who were concurrently taking a COX-2–specific NSAID or a conventional NSAID with misoprostol or PPIs, warfarin was the preferred strategy under all conditions, except when the relative efficacy of warfarin was reduced from 62% to 33%, the relative efficacy of aspirin was increased from 22% to 42%, the relative risk of upper GI tract bleeding when taking warfarin was increased from 2.4 to 6.4 times, or the utility of taking warfarin was less than 0.85. The probability of death from GI tract bleeding, the relative risk of aspirin to cause GI tract bleeding, and the utility values for upper GI tract bleeding and major, moderate, and minor stroke did not influence the choice of antithrombotic therapy.

For elderly persons with an average risk of stroke from atrial fibrillation (6.0% per year) who were concurrently taking conventional NSAIDs without misoprostol or PPIs, warfarin was the preferred strategy under all conditions, except when the relative efficacy of warfarin was reduced from 62% to 42%, the relative efficacy of aspirin was increased from 22% to 33%, the relative risk of upper GI tract bleeding when taking warfarin was increased from 2.4 to 3.8 times, the probability of death from a GI tract bleed was increased from 14% to 31%, or the utility of taking warfarin was less than 0.88. The relative risk of aspirin to cause GI tract bleeding and the utility values for upper GI tract bleeding and major, moderate, and minor stroke did not influence the choice of antithrombotic therapy.

The risk of warfarin and NSAID-related upper GI tract bleeding increases as elderly persons age.^9,53 Therefore, the analysis was repeated with the start age increased from 65 to 75 years. To estimate the risk of stroke in this population when taking different therapies, we used AFI data pertaining exclusively to subjects 75 years and older (the stroke rate when receiving no antithrombotic therapy is 8% per patient-year). For this age group, we estimated the relative risk of upper GI tract bleeding due to NSAID use as 4.5 times the baseline rate.⁵³ Again, substitution of these values into the decision model did not substantially affect its results (detailed results are available from the authors).

Compared with lower doses, there is some evidence that higher doses of NSAIDs increase the risk of upper GI tract bleeding. Users of high doses of NSAIDs may have a 6.9 times greater chance of GI tract bleeding than nonusers.⁶³ At this level of GI tract bleeding risk, the stroke risk threshold for warfarin to be the optimal antithrombotic therapy was 5.2% per year, which approached the average risk of stroke from atrial fibrillation (6.0% per year) in elderly persons.

Also, as part of the sensitivity analysis, the highest published estimate of age-related adjustments in the bleeding risk for those older than 65 years (Stroke Prevention in Atrial Fibrillation II Study⁶⁷ data for those aged <75 years compared with those aged ≥75 years: warfarin-related major bleeding, 2.6 times the relative risk increase; and aspirin-related major bleeding, 1.8 times the relative risk increase) was incorporated into the model. Under these conditions, all combinations of start age (65, 70, and 75 years), upper GI tract bleeding risk (recent GI tract bleeding, concomitant use of conventional NSAIDs with and without misoprostol or PPIs, and concomitant use of COX-2–specific NSAIDs), and stroke risks were tested. The results of these analyses demonstrated that warfarin was the optimal treatment strategy in all situations except when persons were 75 years at the start age, had a lower than average risk of stroke (<6.0% per year), and were also taking conventional NSAIDs without cytoprotection (warfarin, 7.44 QALYs; aspirin, 7.51 QALYs; and no treatment, 7.42 QALYs).

The base-case and sensitivity analyses suggest that for elderly persons with an average risk of stroke from atrial fibrillation (6.0% per year) and upper GI tract bleeding (1.17% per year), warfarin is the optimal treatment strategy. To not benefit from warfarin therapy, persons must have significantly higher than average risks of upper GI tract bleeding (>10.4% per year) or baseline risks of stroke that are lower than the rates found clinically (<2.4% per year, with persons ≥65 years who had atrial fibrillation having a minimum of a 4% yearly risk of stroke). Considering that persons concurrently taking warfarin and NSAIDs may have a relative risk of major upper GI tract bleeding as high as 13 times that of the baseline risk,⁶³ the therapeutic margin of safety for benefits vs risks for warfarin use may be uncomfortably thin in patients with an average risk of stroke. With these issues in mind, based on patients' risk of upper GI tract bleeding, antithrombotic treatment recommendations for older patients with atrial fibrillation were developed (Table 3).

With H pylori screening and treatment, these patients appear to have no increased incremental risk of a recurrent episode of upper GI tract bleeding. Therefore, the recommendations for antithrombotic therapy in these patients are similar to those for patients with an average risk of upper GI tract bleeding.

In these persons, the benefits of warfarin therapy may be outweighed by its risks. Therefore, all such patients should have a reassessment of the need for the use of NSAIDs, with a switch to another agent (eg, a COX-2 NSAID or acetaminophen), or the addition of misoprostol or PPIs to reduce the risk of GI tract bleeding. If these are not possibilities, warfarin is still the preferred therapy in patients with a higher than average risk of stroke (eg, start age of 70 years and risk of stroke of 8.1% per year: warfarin, 9.62 QALYs; aspirin, 9.25 QALYs; and no treatment, 8.82 QALYs).

Persons aged 65 years who have no other risk factors for stroke from atrial fibrillation (ie, no history of hypertension, left ventricular dysfunction, or transient ischemic attack or stroke) have an approximately 4% yearly chance of stroke. For those who are concurrently taking conventional NSAIDs, the incremental benefit of warfarin over aspirin is offset by the increased risk of upper GI tract bleeding caused by the NSAID use. Therefore, depending on patients' preferences for therapy, warfarin, aspirin, and possibly no therapy are reasonable alternatives (eg, start age of 80 years and risk of stroke of 4.3% per year: warfarin, 7.44 QALYs; aspirin, 7.39 QALYs; and no treatment, 7.21 QALYs).

The choice of antithrombotic therapy (aspirin or warfarin) for older patients with atrial fibrillation balances the advantages of stroke prevention with the disadvantages of the adverse effects (eg, upper GI tract bleeding), costs, and inconvenience of the therapy. Many physicians are reluctant to prescribe antithrombotic therapy (especially warfarin) to patients whom they believe are at an increased risk of upper GI tract bleeding. This decision analysis was performed to quantify the benefits and risks in the relationship between the risk of upper GI tract bleeding and stroke in older persons with atrial fibrillation.

Long-term warfarin therapy is the recommended treatment strategy for most patient profiles assessed in this analysis (Table 3). This general recommendation is consistent with that of the Sixth American College of Chest Physicians Consensus Conference on antithrombotic therapy,² which also recommends warfarin therapy for almost all older patients unless a contraindication exists. For one of these contraindications (the increased risk of upper GI tract bleeding), this analysis helps delineate the magnitude of risk necessary to consider alternatives to warfarin therapy. For example, the QALYs associated with the use of warfarin, aspirin, and no therapy were similar for older persons who have no risk factors that increase their chance of stroke (4.3% yearly chance of stroke) from atrial fibrillation but are concurrently taking conventional NSAIDs without cytoprotection (4.5% yearly chance of upper GI tract bleeding). Thus, clinicians need to consider the risks of stroke and upper GI tract bleeding when deciding on the appropriate antithrombotic therapy for older patients.

The focus of this analysis is on treatment recommendations for older persons (those ≥65 years) with atrial fibrillation because we believed that clinicians are more concerned about the potential risk of GI tract bleeding in this age group compared with younger persons. However, it also gives insight into preferred treatment strategies in younger populations. For example, younger persons with no risk factors for stroke have a 1.0% to 2.0% yearly chance of stroke.² At this low level of stroke risk, aspirin is the preferred therapy unless patients are taking conventional NSAIDs without cytoprotection, at which time no therapy becomes a viable option (Figure 2). Again, these results and recommendations are consistent with those of the American College of Chest Physicians Consensus Conference.²

A major threat to the validity of the recommendations developed (Table 3) is the accuracy of the values used for the input variables. For example, changes in the value for the baseline rate of major bleeding may materially affect the results of the decision analysis. An alternative method of estimating the rate of major bleeding in persons receiving warfarin was derived using data from cohort studies^9,24,46-48 of patients attending anticoagulation clinics. Pooling of the results from these studies produced a rate of 0.018 major bleeding events per patient-year, which is similar to the AFI rate (0.012 events per patient-year). Substitution of this value into the decision model did not substantially affect its results. Also, sensitivity analyses demonstrated that the model was insensitive to a wide range of values for many variables, including utilities associated with the different health states (including GI tract bleeding and stroke), the efficacy of warfarin and aspirin in preventing stroke, and the fatality rates for upper GI tract bleeding when taking NSAIDs or COX-2–specific agents. Moreover, the variables to which the analysis was sensitive, those related to upper GI tract bleeding and stroke risk, have values that can be estimated quite precisely from the literature. For example, the 95% confidence interval for the relative risk of upper GI tract bleeding in persons who take NSAIDs is from 3.6 to 4.1.⁵³ Within this value range, the treatment recommendations in Table 3 did not change.

Another variable that may have affected the validity of the results is the incremental increased risk of upper GI tract bleeding as one ages. Many studies^9,47,68,69 have confirmed that there is an increased risk of antithrombotic-related upper GI tract bleeding in persons 65 years and older compared with those younger than 65 years. However, is it unclear whether there continues to be a gradient of increased risk as those 65 years and older continue to age. Supporting this notion is a study from Beyth and Landefeld⁷⁰ who found that the odds ratio for warfarin-related bleeding was 1.7 (95% confidence interval, 1.0-2.8) for patients aged 65 to 74 years and 3.0 (95% confidence interval, 1.7-5.1) for those 75 years and older compared with a referent population of those 64 years and younger. Also, as noted previously, the Stroke Prevention in Atrial Fibrillation II Study investigators⁶⁷ found a gradient of warfarin- and aspirin-related major bleeding risk independently and significantly related to age. The results of this study have been questioned because of the use of prothrombin time ratios, corresponding to relatively high target international normalized ratios (2.0-4.5), and the relatively small number of bleeding events.⁷¹ Subsequent trials^72-74 have found bleeding rates more in keeping with the earlier trials. Other studies^75,76 have found no association between warfarin-related bleeding and age, although they have also been criticized for possible biases, including the use of noninception cohorts (ie, patients who had been taking warfarin before the study were enrolled, so those at the highest risk of bleeding were selected out). In any case, the sensitivity analysis showed that this possible incremental age-related increase in bleeding risk had little influence on the results and recommendations of the study.

Another threat to validity is that a multiplicative approach was taken to determine the risk of upper GI tract bleeding in persons with 2 or more risk factors for bleeding. For example, in persons concurrently taking warfarin and a conventional NSAID, the relative risk of upper GI tract bleeding was calculated at 9.1 times the baseline risk (relative risk: warfarin use, 2.4; NSAID use, 3.8; and total relative risk, 9.1 [2.4 × 3.8]). This risk estimate is compatible with the estimate of Shorr et al¹² of 12.7 times (95% confidence interval, 6.3-25.7) the baseline risk, especially because their data were collected in the mid-1980s, when higher-intensity anticoagulation was the norm. Also, some clinicians may wonder about the wisdom of prescribing aspirin for stroke prophylaxis to patients who are also taking NSAIDs. However, the results of this decision analysis do use a multiplicative approach to determine the risk of upper GI tract bleeding in persons who are taking aspirin and NSAIDs concurrently (relative risk of upper GI tract bleeding: aspirin use, 1.2; NSAID use, 3.8; and concurrent aspirin and NSAID use, 4.6 [1.2 × 3.8]). Also, a previous analysis⁶³ found that current aspirin use did not substantially affect the relative risk of upper GI tract bleeding associated with NSAID use.

A limitation of this study is that the chance of lower GI tract bleeding was not incorporated into the model. With little data available to determine the effect of NSAIDs with or without misoprostol or PPI use on the rate of lower GI tract bleeding, it was not possible to reliably model treatment strategies with this variable included. A further limitation is that the analysis did not take into account other factors that may further increase the risk for upper GI tract bleeding, including cigarette smoking, alcoholism, and corticosteroid use.¹¹ For example, while there is evidence that the concomitant use of oral corticosteroids and NSAIDs further increases the risk of upper GI tract bleeding,⁷⁷ it is unclear whether corticosteroid use alone does⁶³ or does not⁷⁷ increase the risk of upper GI tract bleeding. Also, factors that increase the chance of warfarin-related serious bleeding, such as medication noncompliance and improper monitoring of patients' anticoagulation status, were not considered. These issues must be taken into account, along with patients' preferences for treatment,^60,78 when determining whether to prescribe warfarin therapy to individuals with atrial fibrillation. Finally, a recent study⁷⁹ found that a significant percentage of patients continue to prefer specific therapeutic alternatives despite being told that this choice does not maximize their QALYs. Thus, determining optimal therapy for individual patients with atrial fibrillation clearly requires the guidance of expert clinicians.

In conclusion, this study demonstrates that the risk of upper GI tract bleeding can be an important factor in the decision-making process regarding the choice of antithrombotic therapy for older patients with atrial fibrillation. Many older patients with atrial fibrillation, including many who are concurrently taking NSAIDs, derive overall benefit from long-term warfarin therapy. The results of this study can help clinicians weigh the benefits of stroke prevention and the risks of upper GI tract bleeding in older patients with atrial fibrillation.

Accepted for publication July 16, 2001.

We thank Rebecca Beyth, MD, for her helpful comments; Graham Nichol, MD, for his original development of the decision-analytic model; the AFI for providing their unpublished data; and Ruth McBride for tabulating the unpublished AFI data.

Corresponding author: Malcolm Man-Son-Hing, MD, MSc, FRCPC, Geriatric Assessment Unit, Ottawa Hospital (Civic Campus), 1053 Carling Ave, Ottawa, Ontario, Canada K1Y 4E9 (e-mail: mhing@ottawahospital.on.ca).