Aspirin Use and Incident Stroke in the Cardiovascular Health Study
Background and Purpose—Randomized clinical trials testing aspirin in relatively low-risk, middle-aged people have consistently shown small increases in stroke associated with aspirin use. We analyzed the relationship between the regular use of aspirin and incident ischemic and hemorrhagic stroke among people aged 65 years or older participating in the Cardiovascular Health Study.
Methods—We conducted a multivariate analysis of incident stroke rates in a prospectively assessed, observational cohort of 5011 elderly people followed for a mean of 4.2 years.
Results—Participants had a mean age of 72 years, and 58% were women. Twenty-three percent used aspirin frequently, and 17% used aspirin infrequently at study entry. Frequent aspirin use was associated with an increased rate of ischemic stroke compared with nonusers (relative risk=1.6; 95% confidence interval [CI], 1.2 to 2.2; P=0.001). After adjustment for other stroke risk factors, women who used aspirin frequently or infrequently at study entry had a 1.8-fold (95% CI, 1.2 to 2.8) and 1.6-fold (95% CI, 0.9 to 3.0) increased risk of ischemic stroke, respectively (P<0.01, test for trend), compared with nonusers. In men, aspirin use was not statistically significantly associated with stroke risk. Findings were similar when aspirin use in the years before the incident stroke was used in the modeling. Aspirin use at entry was also associated with a 4-fold (95% CI, 1.6 to 10.0) increase in risk of hemorrhagic stroke for both infrequent and frequent users of aspirin (P=0.003).
Conclusions—Aspirin use was associated with increased risks of ischemic stroke in women and hemorrhagic stroke overall in this elderly cohort, after adjustment for other stroke predictors. The possibility exists of confounding by reasons for aspirin use rather than cause and effect. Whether regular aspirin use increases stroke risk for elderly people without cardiovascular disease can only be determined by randomized clinical trials.
The regular use of aspirin reduces the risk of ischemic stroke for many patients with clinically manifest atherosclerotic vascular disease.1 In contrast, randomized clinical trials involving people at relatively low risk for stroke have shown an opposite trend. Although aggregate data are inconclusive, these trials associated aspirin use with an increased risk of stroke.2 3 4 In a previously reported analysis of the population-based Cardiovascular Health Study (CHS), regular aspirin use emerged as an independent risk factor for stroke (P<0.007) even after adjustment for other stroke risk factors.5 Here we explore this association further, considering additional follow-up of the CHS cohort with more stroke events, analyzing ischemic and hemorrhagic strokes separately, stratifying by the frequency of aspirin use, and accounting for changes in aspirin use at the start of each year of follow-up.
Subjects and Methods
CHS participants were recruited in 1989 and 1990 from a random sample of the Health Care Financing Administration Medicare eligibility lists in four communities: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Allegheny County, Pennsylvania. Details of design and methods have been reported.6 7 Participants were all aged 65 years or older at study entry; those verified as having stroke before enrollment (n=190) were excluded from this analysis. (In a previous report5 we described 184 participants with a history of stroke at entry. During the extra year of follow-up reported here, CHS investigators, in the course of adjudicating incident cardiovascular and stroke events, identified an additional six participants who had a stroke before entry in the CHS.)
Aspirin use was determined at entry interview by participant recall, recorded as the number of days on which aspirin was taken during the prior 14 days. Patients were categorized as frequent users if they reported taking any dose of aspirin on at least 10 of the 14 days before entry (only by self-report; n=1075; 92% of frequent users) or if they had a prescription for the daily use of any dose of aspirin at enrollment (only by prescription; n=25; 2% of frequent users). Seventy-two frequent users (6%) were identified by both self-report and a prescription. Aspirin dose was recorded only for those with prescriptions (n=97); only four (4%) were using less than 325 mg/d. Participants who took aspirin 2 to 9 days, on average, during the 14 days before entry (n=583) were categorized as less frequent users and the remainder as nonusers. The reasons for aspirin use and its duration before CHS entry were not specifically ascertained. This categorization of aspirin usage was made arbitrarily without reference to any results for stroke risk. At annual follow-up visits, aspirin use by the criteria above was reassessed.
For the time-dependent analyses, persistent frequent aspirin users were those who were frequent aspirin users at baseline and each follow-up visit preceding the occurrence of a stroke. Never users were nonusers at baseline and each follow-up preceding the occurrence of a stroke. The remainder of the participants were classified as “sporadic/infrequent” users. A persistent frequent user could change to a sporadic/infrequent user later in the follow-up if the person reported no or infrequent aspirin use. A never user could later become a sporadic/infrequent user by reporting usage later in the follow-up. Finally, once a person was classified as a sporadic/infrequent user, he or she remained in that category throughout the remainder of the follow-up. This scheme was used to try to minimize two potential problems. First, many people take aspirin in response to acute illnesses and may not be regular frequent users. This causes misclassification in the frequent user group. In the time-dependent analysis, this misclassification can cause serious bias toward an RR of 1 for the persistent frequent category. By insisting that a persistent frequent user must be a frequent user at all follow-ups before an event, this bias is minimized. Second, a person may either stop aspirin usage because of an illness (eg, in preparation for surgery or due to gastrointestinal disease) or start aspirin because of new illness (eg, angina, TIA, atrial fibrillation). To analytically adjust for this in the time-dependent analysis, it would be necessary to know the exact date that the person started or stopped taking aspirin and the date of all events that could influence aspirin usage. Because we only determined aspirin usage at the annual follow-up, it is not possible to know whether change in aspirin usage was before or after other potentially biasing events. Thus, adjustment with the use of multiple time-dependent covariates was not possible. We are certain that persistent frequent aspirin use preceded any new reason to begin aspirin use, since a persistent frequent user was required to be a frequent aspirin user at baseline. However, any stopping of frequent aspirin use due to illness might bias our analysis toward the null. Thus, these definitions are conservative, and any bias would be toward the null hypothesis of no aspirin effect.
Participants were categorized as having clinical cardiovascular disease at entry or not, based on the presence of histories of TIA, myocardial infarction, angina pectoris, coronary or carotid revascularization procedures, atrial fibrillation, aortic aneurysm repair, venous thromboembolism, or peripheral vascular disease. Among those without clinical cardiovascular disease, subclinical cardiovascular disease status was defined by a combined index, published previously,8 based on ECG and echocardiogram abnormalities, carotid artery wall thickness and stenosis based on carotid ultrasound, decreased ankle-brachial blood pressure, and positive response to a Rose Questionnaire for angina or intermittent claudication.
To assess the association of aspirin with stroke risk, statistical adjustment was made for previously determined predictors of stroke in the CHS and other variables associated with aspirin use (Table 1⇓). Predictors of stroke in this cohort included age, diabetes, atrial fibrillation, systolic blood pressure, left ventricular hypertrophy by ECG, left ventricular mass and wall motion by echocardiography, carotid stenosis, time to walk 15 feet, serum creatinine, heart failure, and a history of frequent falls.5
Strokes were identified during annual follow-up examinations and at 6-month telephone contacts, as previously reported.5 7 Stroke cases were adjudicated by a committee of neurologists, neuroradiologists, and internists, with information from patient interviews, medical records, and brain-imaging studies; imaging studies were available in 87% of adjudicated strokes. Criteria for stroke classification and subtyping have been reported.9 In brief, ischemic strokes were diagnosed when rapid onset of focal neurological deficit occurred without evidence of hemorrhage by neuroimaging, lumbar puncture, or autopsy. Strokes were classified as hemorrhagic if there was evidence of blood in the subarachnoid space, ventricles, or parenchyma by neuroimaging that was not due to secondary hemorrhage into an infarct. Strokes were also classified as hemorrhagic if autopsy found bloody spinal fluid or evidence of hemorrhage. Participants who died less than 24 hours after stroke onset were assumed to have hemorrhage as a cause if they did not undergo lumbar puncture, neuroimaging, or autopsy. The stroke type was classified as unknown if information was insufficient to categorize as hemorrhagic or ischemic. Analyses of hemorrhagic strokes excluded 61 participants taking warfarin at entry and four participants for whom this was unknown.
To test for differences between variables by aspirin use in descriptive tables, either the χ2 test for proportions or the two-sample t test for means was used as appropriate. Kaplan-Meier survival curves were computed and tested for equality with the use of the log-rank test. Proportional hazards models were used to adjust for statistically significant stroke risk factors. In these models, the nonuser comparison group consisted of participants who reported using aspirin less than 2 days or not at all in the prior 2 weeks. Variables shown in Table 1⇑ were allowed to enter the regression model in a stepwise fashion if they were statistically significant (P<0.05). All interactions of aspirin use with variables that entered the model were also tested. Hazard ratios are referred to throughout as RRs. For the aspirin variable, a test for trend was computed. Separate analyses were done for those with and without cardiovascular disease by sex. The group without cardiovascular disease was further stratified into those with and without subclinical cardiovascular disease. Time-dependent Cox proportional hazards regression analyses were also performed in a similar fashion using the definitions for the time-dependent aspirin usage given previously. Only aspirin usage was treated in a time-dependent manner. Analyses were performed with the use of the SPSS/Windows computer package. All P values and CIs should be interpreted with caution because this is an observational study and because of the exploratory nature of the analyses.
The 5011 CHS participants (mean age, 72 years; 58% women) were followed for a mean of 4.2 years (range, 4 to 5 years). Follow-up was more than 98% complete. Aspirin was frequently used at entry by 22% of participants and was used less frequently by 12%. Participants using aspirin frequently were more often men, were slightly older, were more likely to have histories of hypertension, diabetes, TIA, heart failure, atrial fibrillation, and arthritis, and had more signs and symptoms of cardiovascular disease than nonusers (all P<0.001 compared with nonusers) (Table 1⇑).
During follow-up, 249 verified strokes occurred (an overall rate of 12 per 1000 person-years) (Table 2⇓). Of these strokes, 85% were ischemic, 10% hemorrhagic, and 4% of unknown cause; in subsequent analyses, strokes of unknown cause (n=10) were combined with ischemic strokes. The rate of ischemic stroke was higher in those with cardiovascular disease (P<0.0001) than in those without cardiovascular disease.
Figure 1⇓ shows the distribution of aspirin use in the three categories of frequent, infrequent, and nonuse at entry and at each follow-up examination. These rates are further divided for follow-up years 1 to 4 by aspirin use in the previous year. At entry, 66% were nonusers. The rate of nonuse declined to 59% at the year 4 follow-up. The infrequent use category was fairly constant, at approximately 10%. The frequent user category showed an increase from 22% at entry level to 28% at year 4. Most participants who were nonusers or frequent users in the previous year stayed in the same category in the next year. However, for the infrequent category there was considerable switching from year to year.
Figures 2⇓ and 3⇓ show the cumulative ischemic stroke-free event curves by aspirin usage at baseline for women and men, respectively. The difference in the survival curves was statistically significant for women (P<0.00001) but not for men.
The rate of ischemic stroke increased with increasing frequency of aspirin use at rates of 9.3, 10.4, and 15.2 per 1000 person-years for the nonusers, infrequent users, and frequent users of aspirin, respectively. The proportional hazards regression, adjusted for other stroke risk factors, revealed a significant (P<0.05) interaction of sex with the relationship of aspirin and ischemic stroke, and thus all subsequent analyses considered men and women separately. As shown in Table 3⇓, there is a clear pattern of increasing rates for women without cardiovascular disease (5.2, 9.2, and 12.0 per 1000 person-years for 0 to 2, 3 to 9, and ≥10, respectively) and a suggestion of increased rates for those with cardiovascular disease. For men, no clear trend is shown in the rates for those either with or without vascular disease.
Table 3⇑ shows the results for the Cox proportional hazards regression for all ischemic and hemorrhagic strokes for all participants by sex and further subdivided by cardiovascular disease, with adjustment for other significant predictors of stroke risk. Women who used aspirin frequently or infrequently showed an increased risk for ischemic stroke (RRs of 1.82 [95% CI, 1.18 to 2.81] and 1.57 [CI, 0.87 to 2.84], respectively; test for trend, P<0.01, adjusted for stroke risk factors), compared with nonusers, while men did not.
For women without cardiovascular disease, an increase in RR of 2.11 (95% CI, 1.13 to 3.92; P=0.02) in ischemic stroke risk was associated with frequent aspirin use compared with nonusers after adjustment for other stroke risk factors (Table 3⇑). For men without cardiovascular disease, while the estimates of RR for aspirin users were greater than 1, none reached statistical significance, nor was the test for trend significant (Table 3⇑).
For women with clinical vascular disease, the adjusted RR of ischemic stroke for frequent users was 1.62 (95% CI, 0.86 to 3.07) compared with nonusers (Table 3⇑). In contrast, for men with clinical vascular disease the observed RRs were less than 1 for both frequent and infrequent users (Table 3⇑). None of the RRS (or trends) for women or men with clinical vascular disease reached statistical significance.
When the subgroup with no cardiovascular disease was further subdivided into those with (n=1684) and without (n=1626) subclinical cardiovascular disease, the results of the regression modeling were also similar, except possibly for women without subclinical disease. For women without subclinical disease (n=1083, 25 ischemic strokes), the RRs of ischemic stroke were 1.16 (95% CI, 0.32 to 4.24) for infrequent users and 3.57 (95% CI, 1.45 to 8.76) for frequent users (P<0.05, test for trend). For women with subclinical disease (n=926, 33 ischemic strokes), the RRs were 2.29 (95% CI, 0.96 to 5.43) and 1.08 (95% CI, 0.42 to 2.79) for infrequent and frequent users, respectively (P>0.80, test for trend). For men, the RRs were not significantly different from 1, but the number of ischemic strokes was small (38 strokes in 758 participants with no subclinical disease and 54 strokes in 1301 participants with subclinical disease), and therefore power is limited (results not shown).
The results from the time-dependent proportional hazards regression (Table 4⇓) also showed a statistically significant relationship between aspirin use and ischemic stroke risk for women but not for men. In the time-dependent proportional hazards regression, women who were persistent frequent users had an adjusted RR of 1.93 (95% CI, 1.15 to 3.23) compared with never users. In contrast, the RRs for men were not significantly different from 1. When the analysis was restricted to those without cardiovascular disease, the adjusted RR for women who were persistent users compared with never users was 1.67 (95% CI, 0.75 to 3.68), while for men in this group there was no relationship between aspirin use and ischemic stroke risk. For women with cardiovascular disease the RR for persistent frequent users compared with never users was 2.22 (95% CI, 1.06 to 4.66). For men with cardiovascular disease the RRs were below 1 and not statistically significant. In all analyses, the RRs for the sporadic and/or infrequent users were not significantly different from 1.
For hemorrhagic stroke, the small number of events precluded sex-stratified analyses of cardiovascular disease or use of a time-dependent aspirin variable. Aspirin use at entry was associated with an increased risk of hemorrhagic stroke, with RRs of 4.07 (95% CI, 1.60 to 10.31) for frequent users and 2.77 (95% CI, 0.83 to 9.24) for infrequent users compared with nonusers (P<0.005, test for trend) after adjustment for other risk factors (Table 3⇑).
Frequent aspirin use was associated with a paradoxically increased rate of ischemic stroke among participants in the CHS. This excess risk was primarily due to increased risk in women. Aspirin use was not randomly assigned in this observational study, and while we adjusted for coexisting factors predictive of stroke risk in this data set, the definite possibility remains of confounding by indication (ie, reasons for aspirin use were also related to a higher stroke risk).
Other studies support that regular aspirin use may be associated with increased rates of ischemic stroke in low-risk populations (ie, those without cardiovascular disease).2 3 4 10 In the large US Physicians Health Study,2 although the 12% increased risk of ischemic stroke among middle-aged, relatively healthy participants assigned aspirin was not statistically significant, it differed from the effect of aspirin defined by meta-analysis of clinical trials involving patients with cardiovascular disease (a 26% reduction in ischemic stroke [95% CI, 19% to 37%]).1
The large RR associated with regular, frequent aspirin use for women in the CHS cohort (adjusted RR=1.82 based on entry usage and 1.93 in the time-dependent analysis) compared with other studies could reflect the older patient population, daily aspirin dose (likely higher in the CHS), the duration of aspirin use, and/or a component of confounding by indication. Randomized clinical trials to date for primary prevention have been largely restricted to middle-aged men.2 3 Among women at high risk for vascular events, antiplatelet therapy appears to reduce the occurrence of the constellation of stroke, myocardial infarction, or vascular death.1 However, the effect of aspirin alone on ischemic stroke among elderly women at lower risk of vascular events has not been fully elucidated in previous studies. Our results suggest that this issue may merit special attention. Whether women are at particular risk, as suggested by our results, might be answered when the results of the Women’s Health Study11 are available.
If the relationship between aspirin use and the increased rate of ischemic stroke in women is actually due to confounding by indication, then major nontraditional risk factors for ischemic stroke must exist that are undefined or poorly characterized in the CHS. Such potential unidentified risk factors would have to be powerful stroke predictors because the reasons for taking nonprescription aspirin are so varied, and hence only a fraction of aspirin users would harbor them. Furthermore, arthritis, probably the most frequent indication for aspirin use in the elderly, was not associated with stroke risk in the CHS (data not shown) and thus cannot be a confounder of the aspirin association.
Both the analysis of aspirin use based on self-report of participants at enrollment and the time-dependent analysis that used both the enrollment and follow-up reports are prone to bias due to misclassification of aspirin usage and crossover during follow-up. A person reporting frequent aspirin use in the 2 weeks before entry or the follow-up examination may have been using aspirin for an acute condition (eg, the flu) and may rarely have used it during the 5 years of follow-up. On the other hand, some participants who were nonusers or infrequent users at baseline (or at a particular follow-up visit) may actually be frequent users who happened to not take aspirin during those 2 weeks or may later have become frequent users. However, any error in measuring the aspirin exposure in this study will tend to cause an underestimation of the RR because the bias induced tends to dampen the RR estimates toward 1.
Among those with no clinical cardiovascular disease, an association between aspirin and the presence of subclinical cardiovascular disease did not explain the findings. In fact, the RRs associated with aspirin use in women were larger in those with no subclinical cardiovascular disease than in those with indications of the presence of unsuspected cardiovascular disease.
The potential mechanisms by which frequent aspirin use could increase ischemic stroke are not well defined. Aspirin has been variably reported to increase systolic blood pressure and to antagonize the effect of certain antihypertensive drugs (I.B. Puddey, C.D. Furberg, R.A. Kronmal, R.H. McDonald, B. Psaty, J.D. Williamson, unpublished data, 1997); hypertension is a strong, prevalent risk factor for stroke.5 Thrombogenic effects of aspirin have been demonstrated experimentally, particularly at high doses, and possibly relate to inhibition of endothelially derived prostacyclin synthesis or, in some patients, increase in platelet adhesiveness.13 14 15 16 Hypothetically, a separate and competing effect of aspirin to enhance thrombosis would be detectable in low-risk patients, in contrast to those with manifest vascular disease, in which the antiplatelet effect dominates. Aspirin in doses that reduce systemic prostacyclin appear to inhibit intrinsic thrombolytic mechanisms,17 18 possibly by interference with nitric oxide synthase.19 20 21 If aspirin-accentuated stroke is mediated by hypertension, then the adjustments for blood pressure in our model would tend to underestimate the risk associated with aspirin use.
The duration of aspirin use before CHS entry was not recorded. In a large meta-analysis of the effect of antiplatelet agents on vascular events, the protective effect diminished over time and disappeared after 4 years (although this was tentatively attributed to methodological reasons).1 Depending on the mechanism of aspirin-potentiated ischemic stroke, the duration of use could be an important factor (ie, chronic elevation of blood pressure versus acute effects on prostacyclin).
The association of aspirin with hemorrhagic stroke in this analysis parallels a literature consistently relating aspirin use to intracranial hemorrhage (RR ≈1.75 with no apparent relationship to aspirin dose).1 22 23 For most people, the absolute risk of hemorrhagic stroke is low, and the incremental increase by aspirin may be of marginal clinical importance.1 2 However, among those at substantial risk for intracerebral hemorrhage, such as elderly patients receiving oral anticoagulants, the addition of aspirin may importantly increase the risk of hemorrhagic stroke.24
In summary, these analyses of the CHS cohort in which aspirin use was largely self-determined suggest that the regular use of aspirin may be associated with increased risk of stroke, both ischemic and hemorrhagic, in older women. Among men, those with recognized cardiovascular disease who used aspirin had lower rates of stroke than nonusers of aspirin, while those who were aspirin users without cardiovascular disease had slightly higher rates, but none of these findings were statistically significant.
Millions of elderly people without cardiovascular disease regularly consume aspirin.25 26 Thus, the question of whether regular aspirin use increases the risk of stroke in elderly people without clinical vascular disease is important to answer. This question can only truly be settled by randomized trials with adequate sample size to assess stroke in low-risk, older adults taking aspirin.
Selected Abbreviations and Acronyms
|CHS||=||Cardiovascular Health Study|
|TIA||=||transient ischemic attack|
This study was supported by contracts N01-HC-85079 through N01-HC-85086 from the National Heart, Lung, and Blood Institute.
- Received October 2, 1997.
- Revision received February 25, 1998.
- Accepted February 25, 1998.
- Copyright © 1998 by American Heart Association
Antiplatelet Trialists Collaboration. Collaborative overview of randomized trials of antiplatelet therapy (part I). BMJ. 1994;308:81–106.
Peto R, Gray R, Collins R, Wheatley K, Hennekens C, Jamrozik K, Warlow C, Hafner B, Thompson E, Norton S, Gilliland J, Doll R. A randomized trial of prophylactic daily aspirin in British male doctors. BMJ. 1988;296:313–317.
Manolio TA, Kronmal RA, Burke GL, O’Leary DH, Price TR, for the CHS Collaborative Research Group. Short-term predictors of incident stroke in older adults: the Cardiovascular Health Study. Stroke. 1996;27:1479–1486.
Kuller LH, Shemanski L, Psaty BM, Borhani NO, Gardin J, Haan MN, O’Leary DH, Savage PJ, Tell GS, Tracy R. Subclinical disease as an independent risk factor for cardiovascular disease. Circulation. 1995;92:720–726.
Pagnani-Hill A, Chao A, Ross R, Henderson BE. Aspirin use in chronic diseases: a cohort study of the elderly. BMJ. 1989;299:1917–1950.
Deleted in proof.
Kelton JG, Hirsh J, Carter CJ, Buchanan MR. Thrombogenic effect of high-dose aspirin in rabbits. J Clin Invest. 1978;62:892–895.
Buchanan MR, DeJana E, Gent M, Mustard JF, Hirsh J. Enhanced platelet accumulation onto injured carotid arteries in rabbits following aspirin treatment. J Clin Invest. 1981;67:503–508.
Levin RI, Harpel PC, Harpel JG, Recht PA. Inhibition of tissue plasminogen activator activity by aspirin in vivo and its relationship to levels of tissue plasminogen activator inhibitor antigen, plasminogen activator and their complexes. Blood. 1989;74:571–580.
Amin AR, Vyas P, Attur M, Leszczynska-Piziak J, Patel IR, Weissmanm G, Abramson SB. The mode of action of aspirin-like drugs: effect on inducible nitric oxide synthase. Proc Natl Acad Sci U S A. 1995;92:7926–7930.
Thrift AG, McNeil JJ, Forbes A, Donnan GA. Risk factors for cerebral hemorrhage in the era of well-controlled hypertension. Stroke. 1996;27:2020–2025.
Hart RG, Boop BS, Anderson DC. Oral anticoagulants and intracranial hemorrhage: facts and hypotheses. Stroke. 1995;26:1471–1477.