1. Selecting management strategies for patients at high risk for stroke must involve both an understanding of the best scientific evidence and a realization that other issues beyond stroke reduction should be considered. A treatment that decreases the risk for stroke may introduce the risk for serious complication or adversely affect the patient's usual activities.

We review the available literature on the medical treatment for stroke prevention in patients at elevated risk for stroke. We first review briefly the context in which medical treatment decisions are made. We highlight areas in which single studies or multiple studies reviewed simultaneously can lead to strongly supported recommendations. For areas of continued controversy, we attempt to go beyond the analytical issues to consider the clinical implications of various treatment options. Our goal is to build on previous efforts to review this literature and to understand the content and the limits of available studies [1-7].

2. Rationale

2.1 Who Is at Elevated Risk for Stroke?

2.1.1 General Factors

Therapeutic interventions directed at general modifiable risk factors should be part of any effort to reduce the risk for stroke, regardless of whether the patient receives more specific medical or surgical treatment. Potentially modifiable risk factors include hypertension (including isolated systolic hypertension), heart disease, cigarette smoking, diabetes mellitus, heavy alcohol use, and, perhaps, hypercholesterolemia.

Hypertension is the most potent contributor to the incidence of stroke. According to data from the Framingham Study [8], patients with definite hypertension (blood pressure greater than 160/95 mm Hg) have a relative risk for stroke that is approximately four times that of patients who do not have hypertension. Several randomized clinical trials of antihypertensive therapy have shown the value of blood pressure reduction in preventing stroke. A 5 mm Hg decrease in diastolic blood pressure is associated with a 42% decrease in cerebrovascular events [9, 10]. A meta-analysis of prospective observational studies showed that larger reductions in blood pressure (10 mm Hg) are associated with an even greater decrease in cerebrovascular events, over 50% fewer strokes [11]. Isolated systolic hypertension is an independent risk factor for development of stroke, and patients with this condition who receive effective therapy experience a two- to four-fold decrease in the number of strokes [8].

2.1.2 Symptom Status

For persons with asymptomatic carotid artery stenosis, the degree of carotid narrowing is the single most important factor in determining risk for stroke. A 75% or greater carotid artery cross-sectional area reduction (a diameter reduction of approximately 50%) as shown by ultrasound is associated with a 2.5% annual risk for ipsilateral stroke and a 6.5% risk for fatal myocardial infarction [12].

Approximately 10% of patients who do not have symptoms but have carotid arteriosclerotic lesions have been found on computed tomography to have lesions that are compatible with cerebral infarction, known as "silent brain infarctions" [13, 14].

Estimates of the annual risk for unheralded stroke among patients with asymptomatic carotid bruits range from 1.5% to 4%. However, carotid bruits are associated with carotid stenosis greater than 70% in no more than half of cases. Strokes are as likely to be on the contralateral as the ipsilateral side and are often heralded by a transient ischemic episode.

In general, atrial fibrillation is a predictor of stroke. This association is present whether the atrial fibrillation is persistent or paroxysmal. It exists even when no evidence suggests valvular heart disease (nonvalvular atrial fibrillation) or any type of cardiac disease (lone atrial fibrillation). The likelihood of stroke increases further if atrial fibrillation occurs with specific clinical or echocardiographic risk factors. Clinical risk factors include hypertension, previous thromboembolism (transient ischemic attack [TIA], ischemic stroke, or systemic embolism), and recent congestive heart failure [15]. Left ventricular dysfunction and a left atrial size greater than 2.5 cm/m² are echocardiographic abnormalities associated with thromboembolism [16]. Patients with none of these risk factors have a much better outlook than those with a high-risk profile. Patients younger than 60 years who have a normal echocardiogram and no risk factors have an extremely low risk for stroke (1% per year) [16].

Patients who have recently had TIA or stroke have a substantially increased annual risk for subsequent stroke, from 4.5% to 6.6% [17, 18]; the risk is even higher in symptomatic patients with high-degree carotid stenosis [19]. The risk for stroke is highest in the first several months after the initial TIA or stroke; in patients who do not have additional strokes in the next 5 to 10 years, the risk approximates that of the background population. Compared with hemispheric TIA, retinal TIA (amaurosis fugax) is associated with a substantially lower risk for ipsilateral stroke. Evidence that a difference exists in the outlook for carotid compared with vertebral basilar TIA is not convincing. Risk for stroke recurrence after an initial ischemic stroke is substantially higher in persons who have a high-risk profile or silent brain infarctions [20].

Embolic stroke occurs in 3% to 4% of patients who have myocardial infarction. The risk for embolic stroke is highest between 3 days and 3 weeks after the myocardial infarction. After 4 weeks, cerebral emboli resulting from myocardial infarction are uncommon, and strokes that occur are most likely attributable to other causes [21]. Mural thrombi develop almost exclusively with antero-apical infarctions. Because mural thrombi are infrequent in postero-inferior and subendocardial infarctions, efforts to detect them by echocardiography are probably unnecessary [22].

2.2 Who Are Surgical Candidates?

The question of medical treatment for stroke prevention must be considered in the context of surgical options, particularly carotid endarterectomy. Three recently completed studies indicate that surgery can be beneficial when done in a low-risk setting (that is, a total perioperative stroke and death rate lower than 6%) in symptomatic patients (those with anterior circulation TIAs or nondisabling strokes with a stable deficit [minor stroke]) who have carotid lesions greater than 70% (by strict measurement criteria) [19, 23, 24]. Symptomatic patients with stenosis less than 30% have been shown to have worse outcomes after surgery [23]. Two clinical trials are continuing to enroll patients to determine the efficacy of carotid endarterectomy for symptomatic patients with mid-range stenosis (30% to 69%) [19, 23]. The efficacy of surgery for stroke reduction among asymptomatic patients is another area of continuing investigation [25-27]. We focus on the medical management of patients not judged to be candidates for carotid endarterectomy.

3. Methods

We reviewed the literature to identify and summarize the best available published evidence on the medical treatment for the prevention of stroke in persons at elevated risk. English-language studies published between 1977 and 1993 were identified from MEDLINE, a computerized bibliographic database of the National Library of Medicine, and Current Contents on Disk. The search strategy is described in Appendix Table 1. The computerized database searches were supplemented by articles cited in pertinent reviews, textbooks, and published guidelines. We also examined citations from the included articles.

Two investigators independently extracted data, and any differences were resolved by consensus. We compared the effects of various treatment regimens on three mutually exclusive end points: stroke, death from causes other than stroke, and nonfatal major complications. These end points were chosen because they capture the primary benefit of interest here, stroke reduction, as well as other benefits and harms (including changes in cardiac mortality and increases in fatal and nonfatal bleeding episodes). Transient ischemic attacks were not included as an end point. Complete definitions are provided in the Appendix.

We counted events using the intention-to-treat principle; events were attributed to the treatment group to which the patient was initially assigned. In some cases, only combined outcomes (for example, stroke or death from stroke) or summary-effect measures (relative risks) could be determined. The number of patient-years at risk (that is, the mean duration of follow-up times the number of patients) was recorded or estimated from the follow-up information given.

Studies dealing with similar patients and similar interventions were considered for meta-analysis. Where data were available to calculate a summary relative risk for several studies (that is, patient-years of follow-up and the number of outcome events for each group), the resulting relative risk estimates were tested for homogeneity. We calculated a relative risk and 95% confidence interval for each end point using FAST*PRO software (Academic Press, Inc., Boston, Massachusetts) and applying an equal-effects model or a random-effects model as appropriate, depending on whether individual study relative risk estimates were or were not reasonably homogeneous (P > 0.15). This Bayesian method uses a uniform prior distribution for the relative risk estimate, and, unlike other contingency Table methods, can accommodate "zero cells" [29].

To provide some sense of the effect of treatments on the estimates of the relative risk for outcomes, we calculated the number of outcome events avoided or induced by the use of one treatment over another. The resulting difference in outcome rates is termed the "attributable risk" [30]. For example, consider two treatments: One is associated with an annual stroke rate of 5% and the other with an annual stroke rate of 2%. The relative risk is 40% (2%/5%); the efficacy, or relative risk reduction, is 60% (100% minus the relative risk); and the attributable risk is 3% (5% – 2%). The number of events avoided annually by treating 1000 patients is obtained by multiplying the attributable risk by 1000 (3% x 1000 = 30). Unlike relative risk, attributable risk can be used to make direct comparisons across the three major outcomes. The example treatment above that is estimated to lead annually to 30 fewer strokes per 1000 treated patients might also lead to 5 more major complications in the same group. Attributable risk allows us to weigh directly the effect of these treatments on primary benefits (stroke reduction), secondary benefits (reduction in the rate of death from causes other than stroke, primarily vascular causes), and risks (major complications). A treatment is judged to be clinically useful when the benefits sufficiently outweigh the risks.

Summary recommendations are graded as A, B, or C according to the level of evidence supporting the recommendation [31]. Grade A recommendations are supported either by one or more level I studies or by a meta-analysis in which the lower limit of the confidence interval for the effect of treatment exceeds the minimal clinically significant benefit. Grade B recommendations are supported either by one or more level II studies or by a meta-analysis in which the estimate of treatment effect exceeds the minimal clinically significant benefit but the lower limit of the confidence interval does not. Grade C recommendations are supported by published data other than randomized trials, including secondary analyses of level I or II studies.

4. Results

4.1 Evidence of Treatment Efficacy by Risk Group

In this section, we focus on level I and II studies, which are all randomized controlled trials of treatment efficacy. Tables 1, 2, and 3 summarize details of each study according to patient type. Each study was examined individually and in the context of studies relating to the same risk group.

We identified no level I or II studies examining the relative efficacy of available medical treatment options in patients with asymptomatic carotid artery lesions.

4.1.2 Nonvalvular Atrial Fibrillation

Warfarin for patients with nonvalvular atrial fibrillation is one of the most well-studied treatments for stroke prevention. Five level I or II studies compare the efficacy of warfarin with that of control in this risk group [32-36] (Table 1). In each study, the estimated relative risk for stroke was less than 1.0 (fewer strokes among patients receiving warfarin). In all studies except one [34], this benefit was statistically significant.

When studies using meta-analysis are combined, the overall estimate of the relative risk for stroke among patients receiving warfarin compared with control is 0.33; that is, approximately one third as many strokes occurred in patients receiving warfarin as occurred in control patients. When the results of the five studies are combined, warfarin also appears to be effective in reducing the rate of death from causes other than stroke. The combined relative risk estimate of 0.57 indicates that patients receiving warfarin have about half as many deaths from causes other than stroke as the control group; this is not statistically significant (95% CI, 0.31 to 1.1). Individually or collectively, these studies do not provide evidence that warfarin leads to significantly higher rates of complications, perhaps because of low statistical power from low complication rates (approximately 1% overall).

The currently available data on aspirin treatment for nonvalvular atrial fibrillation are somewhat less compelling than those for warfarin. Aspirin and placebo have been compared in two recent large clinical trials. In the Stroke Prevention in Atrial Fibrillation study [32], patients who were candidates for anticoagulation were randomly assigned to receive warfarin, aspirin (325 mg/d), or placebo, whereas those who were not candidates for anticoagulation were randomly assigned to receive aspirin or placebo. The relative risk for stroke was 0.56 for patients receiving aspirin.

The Copenhagen study [33] also randomly assigned patients to receive warfarin, aspirin, or placebo. In this study, no benefit was seen for aspirin. However, the aspirin dose in this study was only 75 mg/d, and the lack of efficacy may be dose-related. Because these patients were older than those in the group included in the Stroke Prevention in Atrial Fibrillation study, the results imply that aspirin may be less efficacious in older patients.

Combining data across studies suggests a modest and statistically significant decrease in the relative risk for stroke (relative risk, 0.67; CI, 0.45 to 0.99), and a similar but not statistically significant reduction in the rate of death from causes other than stroke (relative risk, 0.78; CI, 0.54 to 1.13). Because major complications were relatively uncommon in all groups, no significant difference was found in the rates of major complications among groups.

The Copenhagen study [33] provides one direct comparison of warfarin and aspirin for the prevention of stroke. This study indicated that warfarin (75 mg/d) is superior to aspirin for the prevention of stroke (relative risk, 0.34; CI, 0.11 to 0.87). Significant results were also seen for death from causes other than stroke (relative risk, 0.21; CI, 0.04 to 0.78). Rates of major complications for warfarin and aspirin were indistinguishable in this study.

4.1.3 Transient Ischemic Attack and Stroke

The efficacy of aspirin for patients with TIA has been evaluated in eight level I or II trials [37-45] (Table 2). Individually, none of these studies showed a statistically significant benefit for either relative risk for stroke or death. When the results are aggregated across studies, aspirin does appear to confer a statistically significant reduction in the rate of stroke (relative risk, 0.84; CI, 0.72 to 0.99) but not a statistically significant reduction in the rate of death from causes other than stroke (relative risk, 0.83; CI, 0.58 to 1.2). The combined evidence from these studies suggests that aspirin may be associated with more nonfatal major complications of treatment than is placebo (relative risk, 1.7; CI, 0.86 to 3.2). In combining data from several studies, we did not include groups randomly assigned to receive aspirin plus another agent (for example, dipyridamole or sulfinpyrazone). This exclusion was based on the conservative assumption that additional agents may act synergistically with aspirin. If these studies are added to the studies of aspirin alone, the magnitude of the benefit is slightly better (relative risk for stroke, 0.76; CI, 0.64 to 0.90).

A point of special controversy relates to the effect of aspirin dosage. Two randomized trials directly compared different aspirin doses: The United Kingdom TIA Aspirin Trial compared aspirin at doses of 1200 mg/d and 300 mg/d [43], and the Dutch TIA Study compared aspirin at doses of 283 mg/d and 30 mg/d [46]. These studies showed no statistically significant differences in stroke, death from causes other than stroke, or major complications. The negative result of the United Kingdom TIA Aspirin Trial may be due, in part, to the relatively low risk for stroke in the population studied (annual risk for stroke risk, 3.2% compared with approximately 7% in the other large trials).

To examine this issue further, the relative risks for stroke estimated from each of the eight included trials that compared aspirin with placebo were plotted against the target aspirin dose (Figure 1). We estimated the relative risk for stroke at the highest dose compared with the lowest aspirin dose by considering all the trials and assuming that any trend in log odds of stroke for aspirin compared with placebo groups was linear according to aspirin dose. No significant trend was observed in these studies. The relative risk estimate suggests no better efficacy at 1300 mg/d than at 325 mg/d (relative risk, 1.02; CI, 0.71 to 1.32). This absence of dose effect is also observed for major nonfatal complications. However, the confidence intervals are wide and do not exclude a clinically significant benefit relative to aspirin dose in reducing stroke or reducing nonfatal complications.

View larger version (13K): [in this window] [in a new window]   Figure 1. The relative risk for stroke estimated from eight randomized controlled trials of aspirin as compared with placebo, plotted against the target aspirin dose in that trial.

Ticlopidine is a new antiplatelet agent that has been compared with placebo in patients who have had stroke (Canadian American Ticlopidine Study) [47]. Although bedridden patients were excluded, the study patients had a wide range of stroke deficits. At a dose of 500 mg/d, this agent was shown to reduce stroke recurrence by approximately one third (relative risk, 0.63; CI, 0.45 to 0.88).

The only level I study of ticlopidine in patients with TIA and minor stroke compared this agent with aspirin (Ticlopidine Aspirin Stroke Study) [48]. Ticlopidine appears to be modestly superior to aspirin in reducing the rate of stroke (relative risk, 0.82; CI, 0.67 to 1.0) but not that of causes other than stroke (relative risk, 0.93; CI, 0.75 to 1.16). In a secondary analysis, the subgroup of patients with minor strokes was shown to have a similar benefit with this agent when compared with aspirin [49]. If the side effects of uncomplicated neutropenia are excluded, fewer major complications also occurred with this agent (relative risk, 0.34; CI, 0.14 to 0.76). However, ticlopidine has a high rate of intolerance (20.9% of patients receiving ticlopidine in this study had treatment terminated because of adverse effects as compared with 14.5% of patients receiving aspirin). In addition, ticlopidine is associated with a reversible neutropenia (absolute neutrophil count, < 1.2 x 10⁹/L) observed early in the course of treatment in 2.4% of patients [49].

Many level I or II trials in patients with TIAs and previous strokes have examined several other agents for their potential ability to reduce stroke. These agents include dipyridamole alone or with aspirin [41, 50-54]; warfarin [54, 55]; sulfinpyrazone, either alone or with aspirin [40, 56]; pentoxifylline [57]; and suloctidil [58]. None has shown efficacy in stroke reduction.

4.1.4 Myocardial Infarction

Two studies [59-61] compared warfarin with control in patients several months after myocardial infarction. When combined, these studies suggest that warfarin is effective in reducing stroke (relative risk, 0.59; CI, 0.40 to 0.88) and death (relative risk, 0.71; CI, 0.56 to 0.90). However, the major complications associated with this treatment were significantly greater (relative risk, 10.1; CI, 3.8 to 26.0) (Table 3).

The use of antiplatelet agents in patients who have had a myocardial infarction has been examined in five trials in which stroke was reported as a distinct outcome [62-66] (Table 3). When the results of the three studies comparing aspirin and placebo are combined, the relative risk reduction for stroke is statistically significant and is similar in magnitude to the reduction seen for patients with TIA and minor stroke who receive aspirin (relative risk, 0.67; CI, 0.47 to 0.96). No evidence suggests that adding dipyridamole to aspirin is beneficial. In one study, sulfinpyrazone was shown to reduce the rate of stroke relative to placebo [66]. None of the trials reviewed, either alone or in combination, provides sufficient data to indicate that the efficacy of any agent for stroke reduction is affected by the timing, location, or extent of the myocardial infarction.

4.1.5 Other Risk Groups

Two studies compared aspirin with placebo in patients who had had carotid endarterectomy [67, 68]. Even when the studies were combined, the relative risk for stroke was not statistically significant (relative risk, 0.38; CI, 0.08 to 1.9). In addition, the risk for death from causes other than stroke was increased in both studies, and the combined relative risk was statistically significant (2.4; CI, 1.1 to 5.6).

Ticlopidine has been compared with placebo in patients with intermittent claudication and in patients with unstable angina [69, 70]. In the study of those with intermittent claudication, ticlopidine was not significantly effective in reducing either stroke or death from causes other than stroke [69]. The study in patients with unstable angina was too small to show either benefit or harm attributable to the use of ticlopidine [70].

4.2 Focusing the Clinical Debate on Benefits and Risks

Three agents have been shown to be beneficial in reducing the relative risk for stroke in high-risk groups: warfarin for patients with atrial fibrillation or myocardial infarction; aspirin for patients who have had TIA, minor stroke, or myocardial infarction; and ticlopidine for patients with TIA or stroke. To focus the clinical debate for nonvalvular atrial fibrillation, TIA, stroke, and previous myocardial infarction, we re-evaluated the trial data in terms of attributable risk for the three major end points (Table 4).

4.2.1 Nonvalvular Atrial Fibrillation

Warfarin for nonvalvular atrial fibrillation has a favorable clinical profile when we consider the average expected reduction in the rates of stroke or death from causes other than stroke. Not only is it effective in reducing strokes and deaths from other causes (47 fewer events per 1000 patient-years of treatment when compared with control) but it is also associated with relatively modest increases in major complications (two more major complications per 1000 patient-years of treatment). This translates to a benefit-to-risk ratio of approximately 25:1.

The relative safety of warfarin shown in the reviewed studies may be attributable to the research setting. Major complication rates for warfarin have been shown to be higher in more typical settings [71, 72]. On the other hand, standardization of thromboplastin and the use of the international normalized ratio (INR) can be expected to lead to improved patient management. In the past, problems with nonstandard thromboplastin have led to wide ranges in warfarin dosage and, in some cases, to excessive use of oral anticoagulants [73]. In addition, low-dose regimens may prove as effective as high-intensity regimens and produce fewer complications.

In this setting, warfarin not only is clinically beneficial but also appears to be cost-effective. Eckman and colleagues [7] have estimated that using warfarin rather than not using an anticoagulant costs an additional $2000 to $20 000 per quality-adjusted life-year saved (depending largely on underlying risk for stroke without treatment).

Although available data suggest that aspirin in the setting of nonvalvular atrial fibrillation is efficacious in reducing the risk for stroke, this possible benefit is not statistically significant in terms of attributable risk reduction. This may be caused by lack of effectiveness or, more likely, by insufficient numbers of patients to show a modest level of attributable risk reduction. Warfarin has been shown to be superior to aspirin in patients with nonvalvular atrial fibrillation when the two have been compared directly, providing an advantage in reducing the rate of stroke and deaths from causes other than stroke of 43 per 1000 patient-years of treatment and a 95% chance that no more than 7 major complications exist.

4.2.2 Transient Ischemic Attacks and Stroke

Aspirin may be expected to reduce the number of strokes by 6 (and by no more than 11) and the number of deaths from causes other than stroke by 5 (and by no more than 9) per 1000 patient-years of treatment. The benefit-to-risk ratio (in terms of reduced numbers of strokes and deaths per major complication) is approximately 11:3 for aspirin in patients with TIA or stroke compared with almost 25:1 for warfarin in patients with nonvalvular atrial fibrillation.

Limited evidence exists regarding the optimal aspirin dose. Lower doses are generally more convenient and better tolerated and for this reason may result in better compliance.

Ticlopidine has been shown to be effective for stroke reduction both in patients with either minor or major stroke and in patients with TIA or minor stroke. When compared with placebo, 31 fewer strokes can be expected for each major complication on ticlopidine treatment, a better estimated benefit-to-risk ratio than aspirin treatment. Furthermore, clinical trials have suggested that the risk for major complications is lower with ticlopidine than with aspirin. However, the estimates for benefit and risk for ticlopidine are based on a much smaller body of evidence than those for aspirin.

Other drawbacks to ticlopidine include its cost and that it occasionally produces neutropenia. Typical annual drug charges alone (for 500 mg/d) are approximately $1000 (based on a survey of pharmacies in central North Carolina). To this must be added the costs associated with increased medical surveillance. Although in clinical studies neutropenia only rarely has led to major complications, this could be a problem in the community if lower levels of surveillance are observed.

4.2.3 Previous Myocardial Infarction

Warfarin and aspirin both appear to reduce the risk for stroke in patients who have had a myocardial infarction. For warfarin, the attributable risk reduction for stroke is statistically significant (attributable risk reduction, –11;CI, –19 to –3).However, the risk for major nonfatal complications in these studies, although not statistically significant, was high (attributable risk reduction, 20; CI, –27 to 68), leading to a benefit-to-risk ratio of approximately 3:2. This high complication rate may be due to the relatively high doses of warfarin (INR, 2.5 to 4.8). If this rate could be reduced to that seen in more recent studies of warfarin for nonvalvular atrial fibrillation, the benefit-to-risk ratio would improve to approximately 15:1.

In patients who have had a myocardial infarction, support for the use of aspirin in reducing stroke is weak. The modest reduction in the number of strokes per 1000 patient-years of treatment is small and of marginal statistical significance (attributable risk reduction, –2;CI, –4 to 0). Any benefit is almost entirely counterbalanced by the risk for major nonfatal complications. These findings suggest that the decision to use aspirin in patients who have had a myocardial infarction should be considered for its effect on all vascular events rather than for stroke reduction specifically.

4.3 Importance of Individual Patient Factors

By necessity, randomized controlled trials focus on patients in aggregate. However, several secondary analyses have highlighted the probable importance of individual patient features. For example, in nonvalvular atrial fibrillation, patient age has been proposed as a useful guide to management. Persons younger than 60 years with no evidence of clinical or echocardiographic risk factors have a low risk for stroke and should probably not be considered for anticoagulation therapy [15, 16, 74]. Combined evidence from the Stroke Prevention in Atrial Fibrillation and the Copenhagen studies suggests that patients older than 75 years do not benefit from aspirin for stroke reduction.

Although there has been much discussion about differences between men and women regarding the efficacy of aspirin and ticlopidine for stroke reduction, the data do not support using sex as a factor influencing choice of therapy. The purported reduced efficacy of aspirin in women is almost certainly related to the small number of patients studied and the lower event rate in women. Claims of better effectiveness of ticlopidine in women are based on subgroup analysis that has not been reproduced [75].

Ticlopidine has been postulated to be better for patients with vertebrobasilar symptoms, symptoms while on aspirin or anticoagulant therapy, elevated creatinine levels, or hypertension or diabetes requiring treatment [76]. Again, these suggestions are based on unvalidated secondary analysis of trial data.

In selecting stroke prevention treatments, patient preferences for treatments must be paramount. A recent decision analysis examining the use of warfarin or aspirin for patients with nonvalvular atrial fibrillation suggests that a minor decrement in quality of life on treatment may be sufficient to counterbalance any benefit in terms of reduction in the risk for stroke [7]. Patient preferences are an especially important concern when considering agents that dictate changes in lifestyle and involve frequent blood drawing and more intense medical follow-up.

5. Summary

Therapeutic interventions directed at general, modifiable risk factors for stroke (including hypertension, cigarette smoking, diabetes mellitus, heavy alcohol use, and hypercholesterolemia) should be part of any effort to reduce the risk for stroke, regardless of whether the patient receives more specific medical or surgical treatment. In addition, the following specific recommendations can be supported by the evidence cited:

1. Nonvalvular atrial fibrillation: Warfarin is the drug of choice for patients who are candidates for anticoagulation (grade A recommendation). However, patients younger than 60 years without specific clinical or echocardiographic risk factors (hypertension, previous thromboembolism [TIA, ischemic stroke, or systemic embolism], recent congestive heart failure, left ventricular dysfunction, or left atrial enlargement) have a low risk for stroke and need not be treated with warfarin (grade C recommendation). To achieve an acceptable benefit-to-risk ratio, clinicians should carefully monitor the intensity of anticoagulation based on the INR, aiming for a ratio approximately between 2 and 3 (grade C recommendation). For patients unwilling or unable to take warfarin, aspirin is an appropriate alternative (grade A recommendation). The dose of 325 mg/d is supported by available evidence (grade A recommendation). For patients older than 75 years, aspirin may not be effective (grade C recommendation). Data are currently unavailable on the effectiveness of aspirin in patients at lower risk, such as those younger than 60 years who have no other risk factors.

2. Transient ischemic attack and stroke: Aspirin is effective in reducing the risk for stroke in patients with TIA and minor stroke (grade A recommendation); however, the benefit-to-risk ratio is approximately 5:2. All aspirin doses studied have been found to be similarly effective, suggesting that decisions about dose should be based on patient tolerance (grade C recommendation). No evidence suggests that aspirin reduces the risk for stroke in patients who have had major stroke, but aspirin therapy is a reasonable option (grade C recommendation). Ticlopidine has also been shown to be effective in reducing the risk for stroke in patients with TIA and stroke (grade A recommendation). Because of the extensive scientific evidence supporting the efficacy of aspirin and the cost of and risk for ticlopidine, aspirin is recommended as first-line therapy (grade C recommendation). Patients who do not respond to aspirin, cannot tolerate aspirin, or have had a major stroke are candidates for ticlopidine (grade C recommendation). Patients offered this agent must be willing to accept the supervision and expense associated with ticlopidine.

3. Previous myocardial infarction: Warfarin has been shown to reduce the rate of stroke in patients who have had myocardial infarction (grade A recommendation). However, this recommendation is tempered by the high rate of major complications reported in several studies and the resulting estimated benefit-to-risk ratio of 3:2. These high complication rates may be due to the use of relatively high-intensity anticoagulation in the included studies (INR, 2.5 to 4.8). Aspirin is an alternative strategy for stroke reduction in patients who have had myocardial infarction (grade B recommendation); however, the benefits for these patients are likely to be small. The decision to use aspirin in these patients should be based on such considerations as reduction in all vascular events rather than the expectation of dramatic reductions in the risk for stroke.

6. Discussion

The choice of stroke prevention strategy represents a weighing of benefit, risk, and cost. To promote good decision making, it is essential that the often conflicting characteristics of treatments be described in terms that are statistically appropriate as well as clinically meaningful.

As the evidence presented here is considered, it is important to recognize that we limited our evaluation to randomized controlled trials that satisfied strict methodologic standards. Although this approach may be justified on scientific grounds [28], it precluded discussion of various potentially useful but as yet insufficiently studied agents.

In the absence of a single convincing study, some may find the combination of studies using the techniques of meta-analysis to be less than compelling. However, in some cases, this approach offers the only option for drawing statistically meaningful inferences from which to make clinically important decisions. For those who might prefer to make their own judgments by examining the studies individually, Tables 1, 2, and 3 provide a useful summary. Further research is needed to resolve the ongoing debate on alternative management strategies for reducing the risk for stroke.

In a prevention setting, it is important not only that a treatment be efficacious but also that it be safe and relatively inexpensive. By this criterion, available medical treatments for stroke prevention are far from optimal. However, several treatments have been shown to be effective in specific high-risk patient groups and have reasonable benefit-to-risk ratios. This evidence is encouraging and should lead to more extensive yet deliberate use of these agents.

Appendix

Standard Definitions of Outcome Events

Transient Ischemic Attack

A transient ischemic attack (modified from special report from the National Institute of Neurological Disorders and Stroke) [77] is an episode of focal loss of brain function attributed to ischemia that lasts less than 24 hours, is localized to a portion of the brain supplied by one vascular system, has no persistent deficit, and is not attributable to any other cause.

The following symptoms are not characteristic of TIA: unconsciousness without other signs of vertebrobasilar symptoms, tonic or clonic activity, or both, a prolonged march of symptoms over several areas of the body, and scintillating scotoma. The following symptoms are not to be interpreted as TIAs: a march of a sensory deficit, vertigo alone, dysphagia alone, diplopia alone, bowel or bladder incontinence, loss of vision associated with an altered level of consciousness, focal symptoms associated with migraine, confusion alone, amnesia alone, and drop attacks alone.

A. Transient ischemic attacks of the left carotid system are characterized by one or more of the following symptoms:

1. Motor dysfunction (such as dysarthria, weakness, paralysis, or clumsiness of the right extremities or face or both)

2. Loss of vision in the left eye (amaurosis fugax) or, rarely, the right field of vision (homonymous hemianopia)

3. Sensory symptoms (such as numbness, including the loss of sensation or paresthesia involving the right upper or lower extremity [or both] or the face)

4. Aphasia (language disturbance)

B. Transient ischemic attacks of the right carotid system are characterized by one or more of the following symptoms:

1. Motor dysfunction of the left extremities

2. Loss of vision in the right eye or, rarely, the left field of vision

3. Sensory symptoms involving the left side of the body

4. Aphasia, but only when the right hemisphere is dominant for speech

C. Transient ischemic attacks of the vertebrobasilar system are characterized by one or more of the following symptoms:

1. Motor dysfunction (such as weakness, paralysis, or clumsiness) of any combination of upper and lower extremities and the left or right side of the face, or both

2. Sensory symptoms (such as loss of feeling, numbness, or paresthesia) involving the left, right, or both sides of the body

3. Loss of vision of one or both homonymous visual fields

4. Loss of balance, vertigo, unsteadiness, disequilibrium, diplopia, dysphagia, or dysarthria are characteristic of but not criteria for diagnosing TIA when any one of these symptoms occurs alone

Cerebral Infarction

A cerebral infarction is the persistent loss of brain function attributed to cerebral ischemia by the evaluating physician that is localized to a portion of the brain supplied by one vascular system and for which no other cause can be found on brain imaging studies (computed tomographic or magnetic resonance imaging). Symptoms and findings referable to ischemia in the left carotid, right carotid, and vertebrobasilar circulation systems (see sections A, B, and C above) persist after 24 hours. Minor strokes are those in which complete neurologic recovery occurs (a resolution of neurologic deficits after more than 24 hours) or persisting nondisabling neurologic deficits occur.

A. Minor strokes of the left carotid system are characterized by one or more of the following symptoms:

1. Motor dysfunction (such as dysarthria, weakness, paresis, or clumsiness of the right extremities or face, or both)

2. Loss of vision in the left eye (amaurosis fugax) or, rarely, the right field of vision (homonymous hemianopia)

3. Sensory symptoms (such as numbness, including the loss of sensation or paresthesia involving the right upper or lower extremity [or both] or the face, or both)

4. Aphasia (language disturbance)

B. Minor strokes of the right carotid system are characterized by one or more of the following symptoms:

1. Motor dysfunction of the left extremities

2. Loss of vision in the right eye or, rarely, the left field of vision

3. Sensory symptoms involving the left side of the body

4. Aphasia, but only if the right hemisphere is dominant for speech

C. Minor strokes of the vertebrobasilar system are characterized by one or more of the following symptoms:

1. Motor dysfunction (such as weakness, paralysis, or clumsiness) of any combination of upper and lower extremities and the left or right side of the face, or both

2. Sensory symptoms (such as loss of feeling, numbness, or paresthesia) involving the left, right, or both sides of the body

3. Loss of vision of one or both homonymous visual fields

4. Loss of balance, vertigo, unsteadiness, disequilibrium, diplopia, dysphagia, or dysarthria are characteristic of but not criteria for diagnosing small stroke when any one of these symptoms occurs alone

Complications of Therapy

Complications of therapy are morbid events probably attributable to the treatment drug in either arm of a randomized control trial. They may include hemorrhagic events or other known or suspected side effects of the drug. A major complication must be severe enough to be life-threatening or to require hospitalization or blood transfusion.

A. Aspirin

1. Clinical or endoscopic diagnosis of gastritis

2. Gastrointestinal bleeding requiring hospitalization or transfusion

3. Allergic reaction requiring hospitalization or ventilatory support

4. Intracranial hemorrhage

5. Any noncerebral or nongastrointestinal hemorrhage requiring hospitalization or transfusion

B. Warfarin

1. Gastrointestinal bleeding requiring hospitalization or transfusion

2. Intracranial hemorrhage

3. Any noncerebral or gastrointestinal hemorrhage requiring hospitalization or transfusion

C. Ticlopidine

1. Neutropenia associated with severe infection

2. Neutropenia requiring hospitalization