Aspirin and Risk of Hemorrhagic Stroke: Title and subTitle BreakA Meta-analysis of Randomized Controlled Trials FREE

ASPIRIN WAS introduced as an analgesic and antipyretic agent almost a century ago. However, it is only during the past 2 decades that attention has been focused on the therapeutic effect of aspirin on cardiovascular disease.¹ A number of large, randomized, controlled clinical trials have demonstrated that aspirin treatment reduces the risk of subsequent myocardial infarction and ischemic stroke among patients with a wide range of preexisting cardiovascular diseases.^{2 - 7} A smaller body of data suggests that aspirin treatment reduces the risk of nonfatal myocardial infarction in healthy persons.^{8 - 9} Aspirin is now being widely used for primary and secondary prevention of cardiovascular disease in the general population.^{10 - 12} For example, in 1 population-based prospective study of 15,735 middle-aged men and women drawn from 4 US communities, 30% of whites and 11% of African Americans were regularly using aspirin for prevention of cardiovascular disease.¹²

Several studies have suggested that aspirin increases the risk of hemorrhagic stroke.^{2 ,4 - 7 ,9 ,13} In the Physicians' Health Study, 22,071 male physicians aged 40 to 84 years were randomly assigned to receive 325 mg of aspirin or placebo every other day.⁹ After an average of 60.2 months of follow-up, 23 hemorrhagic stroke cases were observed in the aspirin group compared with 12 cases in the control group. Aspirin use was associated with a 2.14-fold higher risk of hemorrhagic stroke.⁹ In the Swedish Aspirin Low-dose Trial (SALT), 1360 patients with a history of a transient ischemic attack or minor stroke were randomly assigned to 75 mg/d of aspirin or placebo for 32 months. Aspirin treatment was associated with a 2.78-fold higher risk of hemorrhagic stroke.⁵ However, none of these studies had sufficient statistical power to provide definitive results regarding the effect of aspirin on risk of hemorrhagic stroke. Likewise, previous meta-analyses that have addressed this question failed to provide a conclusive answer because of limitations in statistical power.^{14 - 15}

The objectives of this meta-analysis were to quantify the relationship between aspirin use and risk of hemorrhagic stroke and examine this relationship by different participant and study design characteristics.

We performed a literature search using the MEDLINE computerized database (for studies from 1966 through June 1997) with the medical subject headings aspirin and cerebrovascular disorders, as well as the key word stroke. The search was restricted to studies that were published in English-language journals, conducted on human subjects, and classified as clinical trials in the MEDLINE database. In addition, a manual search was performed using the authors' reference files and reference lists from original communications and review articles.^{14 - 15} The contents of 137 abstracts or full-text articles that were identified during our literature search were reviewed to determine whether they met the criteria for inclusion in our meta-analysis.

The literature search and article review were conducted independently by 2 of the authors (J.H. and B.V.). Areas of disagreement or uncertainty were adjudicated by the other authors. For inclusion, a study had to meet the following criteria: (1) random allocation of study participants to aspirin or a concurrent control group; (2) no intervention difference, other than use of aspirin, between the treatment and control groups; (3) intervention duration of at least 1 month; and (4) presence of information regarding occurrence of stroke subtypes during follow-up. Sixteen trials met these criteria and were included in the meta-analysis.^{2 ,4 - 9 ,13 ,16 - 23}

Major reasons for exclusion of studies were (1) nonrandomized treatment allocation²⁴ ; (2) comparison of aspirin treatment with treatment using other antiplatelet or anticoagulant agents^{25 - 34} ; (3) comparison of higher vs lower dosages of aspirin^{28 ,35 - 36} ; (4) comparison of aspirin treatment combined with other antiplatelet or anticoagulant agents compared with a control^{25 ,37 - 41} ; (5) use of different anticoagulation therapies in the treatment and control groups^{42 - 44} ; (6) conduction of treatment trials in patients with acute complete stroke^{45 - 47} ; (7) study duration of less than 1 month^{45 - 47} ; and (8) lack of information on occurrence of stroke subtype.^{48 - 56}

All data were abstracted in duplicate using a standardized protocol and reporting form and disagreements were resolved by consensus. If a series of articles was published from a single trial, the final report was used for abstraction of data. We did not contact the primary authors to request additional information. Study characteristics recorded were as follows: (1) first author's name, year of publication, and country of origin; (2) number of participants; (3) mean age, age range, race, and sex distributions of the participants; (4) presence of preexisting disease and prevalence of hypertension, hyperlipidemia, and cigarette smoking; (5) design details, including blinding (open, single, or double) and type of control treatment (placebo or no treatment); (6) dosage of aspirin; and (7) study duration.

Incidence of stroke subtype during treatment was abstracted as the primary outcome. In addition, data were abstracted on the incidence of total stroke and myocardial infarction, cardiovascular disease mortality, and all-cause mortality during treatment.

Both absolute risk (AR) and relative risk (RR) were used to measure the relationship between aspirin use and clinical outcomes. Whereas AR measures the magnitude of aspirin use on risk of hemorrhagic stroke, RR measures the strength of the association. The numbers of various outcomes for both the aspirin and control groups were recorded for each study using 2 × 2 tables. Absolute risks, RRs, and their 95% confidence intervals (CIs) were calculated individually for each study.⁵⁷ In several trials, no hemorrhagic strokes were noted in the aspirin and/or control group(s).^{2 ,13 ,16 - 17 ,21 - 23} In these studies, we added 0.5 to zero cells to reduce bias and variance of these estimates.^{58 - 59} We also conducted an analysis in which trials with no hemorrhagic strokes were excluded. The overall pooled estimates using these 2 approaches were similar.

To calculate the pooled AR, each study was weighted by its sample size (N_t × N_c)/(N_t + N_c), where N_t and N_c were the sample size for aspirin therapy and control groups, respectively. Homogeneity of the ARs across studies was tested using the Woolf χ² statistic, which summed the squared deviation of each individual AR from its weighted average.⁵⁷ The reciprocal of the variance of the ARs was not used as the weight because it would bias the pooled estimate toward zero.⁶⁰ Relative risks from individual studies were transformed to their natural logarithm (ln [RRs]) before pooling to stabilize the variances and normalize the distributions.⁵⁹ To calculate the pooled ln (RR), each study was assigned a weight consisting of the reciprocal of the variance of the ln (RR) (fixed-effects model). If heterogeneity of ln (RRs) was detected, the weight consisted of the reciprocal of the sum of the variance for each study and the variance across all studies (random-effects model).⁶¹ Approximate 95% CIs were obtained on the natural log transformation scale and were reexpressed by natural antilogarithm transformation of the data. The z statistic was calculated and a 2-tailed P<.05 was used as the level of statistical significance.

A series of prestated subgroup analyses were performed to explore the influence of covariables on the relationship between aspirin use and hemorrhagic stroke. Subgroups were selected on the basis of biological plausibility and our knowledge of the literature. Continuous covariables (eg, age) were categorized into 2 groups, using the median values as a dividing cut point. The statistical significance of differences in AR between various subgroups was tested using the Woolf χ² statistic.⁵⁷ Univariate and multivariate regression analysis models were used to explore the influence of a series of prestated covariables on the ARs.

The potential for publication bias was examined using a funnel plot, in which sample size was plotted against AR.⁶² In addition, a rank correlation for the association between sample size and standardized AR was conducted using the Kendall τ method. If small studies with negative results were less likely to be published, the correlation between sample size and AR would be high. In the absence of publication bias, no significant correlation between sample size and AR would be evident.⁶²

Sixteen trials with 55,462 participants were included in our meta-analysis. Participant and study design characteristics for these trials are presented in Table 1. Placebo was used as the control in all of the trials except for the British Doctors' Study, in which participants were randomly allocated to either take or avoid aspirin.⁸

Absolute risks for all-cause mortality, cardiovascular disease mortality, myocardial infarction, and stroke associated with aspirin treatment are presented in Table 2. Overall, aspirin use was associated with an AR reduction in all-cause deaths of 120 per 10,000 persons (95% CI, 77-162) and in cardiovascular deaths of 97 per 10,000 persons (95% CI, 59-135) among the 16 trials. The AR reduction for total and fatal myocardial infarction was 137 per 10,000 persons (95% CI, 107-167) and 36 per 10,000 persons (95% CI, 16-55), respectively. Aspirin use was also associated with an AR reduction in total strokes of 31 per 10,000 persons (95% CI, 5-57).

On average, aspirin use was associated with a 15% proportional reduction in all-cause mortality (RR, 0.85; 95% CI, 0.80-0.90; P<.001) and a 16% reduction in cardiovascular mortality (RR, 0.84; 95% CI, 0.79-0.90; P<.001). Aspirin therapy was also associated with a 32% proportional reduction in total myocardial infarction (RR, 0.68; 95% CI, 0.62-0.74; P<.001) and a 22% reduction in fatal myocardial infarction (RR, 0.78; 95% CI, 0.68-0.90; P<.001). Furthermore, aspirin treatment was also associated with a 12% proportional reduction in total stroke (RR, 0.88; 95% CI, 0.76-1.02; P = .08) that was of borderline statistical significance, but not in fatal stroke (RR, 1.07; 95% CI, 0.85-1.35; P = .60).

A total of 108 hemorrhagic stroke cases occurred in 13 of the 16 trials. In the remaining 3 trials, no cases of hemorrhagic stroke were reported.^{17 ,21 ,23} In 11 of the 13 trials reporting hemorrhagic stroke, aspirin treatment was associated with an increased AR of hemorrhagic stroke (Figure 1). However, none of the ARs reached the level of statistical significance. The RR of hemorrhagic stroke was also increased in the 11 trials, varying from 1.08 in the British Doctors' Study⁸ to 4.09 in the trial conducted by Elwood and Sweetnam.² There was no significant heterogeneity in AR or RR among these studies (P = .99 for both).

Table 3 presents the ARs of stroke subtype associated with aspirin therapy. Compared with control, treatment with aspirin was associated with an increase of 12 (95% CI, 5-20) hemorrhagic strokes per 10,000 persons and a reduction of 39 (95% CI, 17-61) ischemic strokes per 10,000 persons. The results were similar in an analysis that included only the 9 trials in which at least 1 hemorrhagic stroke case was reported in each group (an AR of 11 events per 10,000 persons [95% CI, 4-18; P<.001]). Regarding RR, aspirin use was associated with an 84% increase in the risk of hemorrhagic stroke (RR, 1.84; 95% CI, 1.24-2.74; P<.001). In contrast, aspirin use was associated with an 18% decrease in the risk of ischemic stroke (RR, 0.82; 95% CI, 0.73-0.92). The number needed to treat to prevent 1 event was 73 for total myocardial infarction, 278 for fatal myocardial infarction, and 256 for ischemic stroke. The number needed to treat to cause 1 event was 833 for hemorrhagic stroke.

Absolute risks of hemorrhagic stroke did not vary significantly by type of trial participant or characteristics of the study design (Table 4). In addition, linear regression analysis, which used trial characteristics as a continuous variable, did not identify any significant association between these covariables and the AR of hemorrhagic stroke (data not shown). Likewise, RRs also did not vary significantly by study design and participant characteristics.

The Kendall τ correlation coefficient between sample size and standardized AR was 0.03 (P = .90), suggesting that there was no publication bias. However, the Kendall τ correlation coefficient between sample size and standardized-logarithm RR was 0.37 (P = .05), suggesting the possibility of publication bias. A plot of sample size vs ln (RR) indicated that this significant correlation was due to 2 trials that had a large sample size and a moderately large RR (data not shown).^{4 ,9}

Aspirin has been widely used for primary and secondary prevention of myocardial infarction and ischemic stroke.^{10 - 12} Our meta-analysis demonstrates that aspirin use is associated with a 15% proportional reduction in all-cause mortality, a 16% reduction in cardiovascular mortality, a 31% reduction in total myocardial infarction, and an 18% reduction in ischemic stroke. These estimates are comparable with results from a meta-analysis conducted by the Antiplatelet Trialists' Collaborative Group.¹⁵ In their overview, which included 145 randomized trials of antiplatelet therapy vs control, antiplatelet therapy was associated with 16%, 17%, 34%, and 20% reductions in all-cause mortality, vascular mortality, nonfatal myocardial infarction, and nonhemorrhagic stroke, respectively.¹⁵ However, our study differed from theirs in several important aspects. Our primary goal was to study the effect of aspirin on risk of hemorrhagic stroke. Therefore, we focused our attention on randomized trials in which aspirin therapy was compared with a control and stroke subtype was reported. Furthermore, we were able to take advantage of results from several trials that had been reported subsequent to publication of the Antiplatelet Trialists' Collaboration meta-analysis.^{5 - 7 ,13 ,22 - 23}

Our analysis indicates that aspirin treatment was associated with a significant increase in risk of hemorrhagic stroke (an increase in AR of 12 events per 10,000 persons or an increase of 84% in the RR). This relationship was consistent across trials conducted in different population subgroups and using different study designs. Our findings have important clinical and public health implications. Aspirin might be used with caution in subgroups who are at high risk of hemorrhagic stroke. This would include, for example, hypertensive patients with a low level of serum cholesterol or Asians.^{63 - 64} Although aspirin therapy has been well demonstrated to reduce the incidence of myocardial infarction in those who are at relatively high risk, its benefits have not been well documented in healthy persons who are younger than 50 years.^{8 - 9} Thus, aspirin treatment may not be recommended to such persons for the purpose of primary prevention of cardiovascular disease. Further studies are necessary to determine the risk-benefit ratio of aspirin use in subgroups with varying risks of hemorrhagic stroke, ischemic stroke, and myocardial infarction.

Our findings must be interpreted with caution. Most trials included in this meta-analysis were conducted in patients with ischemic heart disease or stroke. There is overwhelming evidence that aspirin reduces cardiovascular disease morbidity and mortality in such populations. For example, aspirin use was associated with an AR reduction of 97 cardiovascular deaths per 10,000 persons, 137 myocardial infarction events per 10,000 persons, and 39 ischemic stroke events per 10,000 persons, but associated with an increase of only 12 hemorrhagic stroke events per 10,000 persons among the 16 trials included in our meta-analysis. Even in healthy populations older than 50 years in western countries, the AR of myocardial infarction and ischemic stroke is much higher than that of hemorrhagic stroke.⁶⁵ Therefore, the overall benefit of aspirin use on myocardial infarction and ischemic stroke almost certainly overcomes the potential risk of hemorrhagic stroke in such groups.

There are several possible mechanisms by which aspirin could increase the risk of hemorrhagic stroke. Aspirin selectively acetylates the hydroxyl group of a single serine residue at position 529 within the polypeptide chain of platelet prostaglandin G/H synthase 1, causing irreversible loss of its cyclooxygenase activity.^{1 ,66} This results in decreased conversion of arachidonate to prostaglandin G₂ and, ultimately, of prostaglandin H₂ and thromboxane A₂, which are important mediators in platelet aggregation and thrombi formation. Platelets are exquisitely sensitive to aspirin. A dosage of only 30 mg/d effectively eliminates the synthesis of thromboxane A₂.¹ The minimum dosage of aspirin in the trials that were included in our meta-analysis, 75 mg/d, is higher than this threshold. Use of effective dosages in all trials probably explains why risk of hemorrhagic stroke was similar in studies using less than 413 mg/d compared with those administering at least 413 mg/d of aspirin. In other meta-analyses, it has also not been possible to identify a dose-response relationship between aspirin therapy and the risk of myocardial infarction or ischemic stroke.^{15 ,67}

Several trials could not be included in this analysis because data on stroke subtype were not reported.^{48 - 56} It is possible that studies that did not publish data on stroke subtypes may have yielded different risk estimates compared with those that did report stroke subtype data, resulting in bias. Indeed, this possibility was suggested by our finding of a borderline significant correlation between sample size and standardized ln (RR) of hemorrhagic stroke among the trials we were able to evaluate. Our search strategy was limited to published studies in the English language, which could also result in publication bias. However, comparison with the Antiplatelet Trialists' Collaboration meta-analysis of 145 trials that used much less restrictive entry criteria demonstrated that we did not miss any trials that met our inclusion criteria.¹⁵ Moreover, our estimates of risk reduction for all-cause mortality, cardiovascular disease mortality, myocardial infarction, and ischemic stroke were almost identical to those identified in their meta-analysis.¹⁵

Some of these trials were conducted in the late 1970s or early 1980s, when computed tomography might not have been routinely used for diagnosis of stroke. However, the RR of hemorrhagic stroke associated with aspirin use was 2.21 (95% CI, 1.33-3.66) among trials published in 1989 and later compared with 1.37 (95% CI, 0.72-2.61) among trials published in 1988 and earlier. These data indicate that the association between aspirin use and risk of hemorrhagic stroke was unlikely to have been caused by misclassification bias due to lesser use of computed tomography studies in the earlier trials.

In summary, our meta-analysis indicates that aspirin therapy reduces the risks of myocardial infarction and ischemic stroke more than it increases the risk of hemorrhagic stroke. The risk-benefit ratio of aspirin treatment for prevention of cardiovascular disease in populations with a high risk of hemorrhagic stroke should be addressed in subsequent studies. A meta-analysis using individual data may provide the best opportunity to answer this question.