|
A Comparison of Warfarin and Aspirin for the
Prevention of Recurrent Ischemic Stroke
J.P. Mohr, M.D., J.L.P. Thompson, Ph.D., R.M. Lazar, Ph.D.,
B. Levin, M.D., R.L. Sacco, M.D., K.L. Furie, M.D., J.P. Kistler, M.D., G.W.
Albers, M.D., L.C. Pettigrew, M.D., H.P. Adams, Jr., M.D., C.M. Jackson, M.D.,
P. Pullicino, M.D., for the Warfarin–Aspirin Recurrent Stroke Study Group
ABSTRACT
Background Despite the
use of antiplatelet agents, usually aspirin, in patients who have
had an ischemic stroke, there is still a substantial rate of
recurrence. Therefore, we investigated whether warfarin, which is
effective and superior to aspirin in the prevention of cardiogenic
embolism, would also prove superior in the prevention of recurrent
ischemic stroke in patients with a prior noncardioembolic ischemic
stroke.
Methods In a multicenter,
double-blind, randomized trial, we compared the effect of warfarin
(at a dose adjusted to produce an international normalized ratio of
1.4 to 2.8) and that of aspirin (325 mg per day) on the combined
primary end point of recurrent ischemic stroke or death from any
cause within two years.
Results The two
randomized study groups were similar with respect to base-line risk
factors. In the intention-to-treat analysis, no significant
differences were found between the treatment groups in any of the
outcomes measured. The primary end point of death or recurrent
ischemic stroke was reached by 196 of 1103 patients assigned to
warfarin (17.8 percent) and 176 of 1103 assigned to aspirin (16.0
percent; P=0.25; hazard ratio comparing warfarin with aspirin, 1.13;
95 percent confidence interval, 0.92 to 1.38). The rates of major
hemorrhage were low (2.22 per 100 patient-years in the warfarin
group and 1.49 per 100 patient-years in the aspirin group). Also,
there were no significant treatment-related differences in the
frequency of or time to the primary end point or major hemorrhage
according to the cause of the initial stroke.
Conclusions Over a
two-year period, we found no difference between aspirin and warfarin
in the prevention of recurrent ischemic stroke or death or in the
rate of major hemorrhage. Consequently, we regard both warfarin and
aspirin as reasonable therapeutic alternatives.
Long-standing doubts, expressed as late as
the 1980s, about the efficacy of warfarin for the prevention of
stroke1
were mitigated by the results of more recent clinical trials.
Recurrence rates were lower with warfarin than with placebo in
patients who had stroke after myocardial infarction.2 The
rates of first stroke in patients with atrial fibrillation were
lower with warfarin than with a range of other therapies,3
placebo,4
or aspirin.5
Also, in open-label studies, the rates of recurrent stroke were
lower with warfarin than with placebo or aspirin.6
Rates of adverse events with warfarin were acceptably low at the
ranges of the international normalized ratio (INR) used in the
studies (1.5 to 3.0).5,7
Most previous clinical trials of drugs to prevent recurrent ischemic
stroke after a noncardiogenic ischemic stroke studied one or more of
a wide variety of platelet-antiaggregant drugs, particularly
aspirin, with which the recurrence rate approximates 8 percent.8,9,10,11
The organizers of the current trial believed that a trial comparing
warfarin and aspirin in the prevention of recurrent ischemic stroke
was justified. This belief was based on the success of warfarin in
the prevention of strokes among patients with atrial fibrillation
and the inference that some ischemic strokes are due to embolism.12
Furthermore, no trial had determined whether anticoagulant agents
were superior to platelet-antiaggregant drugs in preventing other,
noncardioembolic forms of ischemic stroke.
Methods
Study Design
The Warfarin–Aspirin Recurrent Stroke Study (WARSS) was an
investigator-initiated, randomized, double-blind, multicenter clinical
trial conducted in 48 academic medical centers in the United States
and sponsored by the National Institute of Neurological Disorders
and Stroke. It also served as the basis for four parallel stroke
studies.13
The trial was formulated and designed by the stroke research staff
at the Neurological Institute of Columbia Presbyterian Medical
Center. Clinical data were collected and monitored by the
data-management center in the Stroke Unit at the Neurological
Institute. Management of data on anticoagulant therapy,
double-blinding procedures, and statistical analysis were conducted
by the statistical-analysis center of the Department of
Biostatistics, Mailman School of Public Health, Columbia University.
Study medications were bottled, packaged, and distributed by
Quintiles (Mount Laurel, N.J.). To eliminate variations between laboratories,14
blood samples for determination of the INR were processed centrally
by Quest Diagnostics (Teterboro, N.J.). The study protocol was
approved by the institutional review board at each participating
center. Written informed consent was obtained from each patient.
Patient recruitment began in June 1993, and follow-up ended, as
scheduled, in June 2000.
Eligibility
Eligible patients were 30 to 85 years old, were considered
acceptable candidates for warfarin therapy, had had an ischemic
stroke within the previous 30 days, and had scores of 3 or more on
the Glasgow Outcome Scale. On this scale a score of 3 indicates severe
disability, a score of 4 moderate disability, and a score of 5
minimal or no disability. Patients were ineligible if they had a
base-line INR above the normal range (more than 1.4), stroke that
was due to a procedure or that was attributed to high-grade carotid
stenosis for which surgery was planned, or stroke associated with an
inferred cardioembolic source; most of the last group had atrial
fibrillation at the time of stroke. Eligibility was verified before
randomization by telephone contact with the data-management center,
in which each criterion for eligibility or ineligibility, the dates
of stroke and randomization, magnetic resonance imaging or computed
tomography of the brain, and signing of the consent form were confirmed.
Medications and Blinding
The medications evaluated were aspirin (Bayer, Morristown, N.J.),
one 325-mg tablet daily, and warfarin (Dupont, Wilmington, Del.), one
2-mg scored tablet daily. The warfarin doses were adjusted to
achieve and maintain an INR in the range of 1.4 to 2.8. The patients
were randomly assigned to receive active aspirin and warfarin
placebo or active warfarin and aspirin placebo. Randomization was
stratified according to site. No patients received two placebos or
two active treatments. All centers and patients were informed as to
the double-blind design and the plan for the use of false INR values
in the group receiving active aspirin and warfarin placebo. All
centers followed the same schedule of visits to the clinic for
drawing of blood to measure the INR, monitoring of medication, and
adjustment of the dose of warfarin or warfarin placebo.
Blood samples for determination of the INR were sent to Quest
Diagnostics on the same day or by overnight courier service. Before
a center was admitted as a study site, we confirmed that blood
samples sent to Quest Diagnostics were viable and yielded reliable
INR determinations. All INR results were transferred electronically
to the statistical-analysis center, which sent the results to the
local centers by facsimile transmission. According to prior
agreement among the center clinicians and with the use of a method
validated early in the trial,15
the INR results sent to local centers were unmodified for the
patients receiving active warfarin, but for patients receiving
active aspirin and warfarin placebo, they were replaced by the
statistical-analysis center with fabricated values that were
plausible for the dose and duration of warfarin therapy. No INR
results were available directly to the local centers from Quest
Diagnostics. According to the guidelines of the Food and Drug
Administration, high INR values (4.5 or more) were forwarded to the
data-management center and transmitted immediately to local centers
by cellular telephone. To preserve blinding, some emergency
notifications for falsely elevated values in patients receiving
warfarin placebo were also sent by the statistical-analysis center.
The principal clinical investigator reviewed all outgoing INR
reports, writing a personal cautionary note to the local
investigator in the case of reports showing trends for values below
or above the desired ranges. All participants other than the
principal statistical investigator at the statistical-analysis
center were blinded to the patients' study-group assignments. During
the course of the trial, unblinding was required for 15 patients, in
most cases because of the need for an invasive surgical procedure.
All 15 patients stopped treatment with study drugs, but their data
were included in the intention-to-treat analysis.
Follow-up
Patients were followed for 2 years ±1 month, up to a maximum
of 761 days. Follow-up was conducted monthly by telephone or in person
at the time of drawing of blood for the determination of the INR to
assess compliance and to regulate INR values, quarterly in person
for clinical evaluation, and annually for detailed examination; the
occurrence of end points was also ascertained at each contact.
Personnel at the data-management center also conducted site visits
to audit the records of all patients at each center for end points
and adverse events.
Assessment of End Points and Major Adverse
Events
The primary end point was death from any cause or recurrent ischemic
stroke, whichever occurred first. Recurrent ischemic stroke was
defined as a new lesion detected by computed tomography or magnetic
resonance imaging or, in the absence of a new lesion, clinical
findings consistent with the occurrence of stroke that lasted for
more than 24 hours. Local centers reported potential outcome events
to the events coordinator at the data-management center and
submitted clinical summaries, study forms documenting clinical
details, and brain imaging studies. An independent, treatment-blinded
neuroradiologist reviewed the images. Five treatment-blinded
neurologists adjudicated all clinical events using a majority
verdict for decisions about outcomes.
Major hemorrhage was defined as intracranial, intraspinal,
intracerebral, subarachnoid, subdural, or epidural hemorrhage or any
other bleeding event requiring transfusion. Minor hemorrhage, which
did not require transfusion, included gastrointestinal, genitourinary,
retroperitoneal, joint, subcutaneous or muscular, gingival or oral,
and conjunctival hemorrhage; epistaxis; hemoptysis; ecchymoses; and
hemorrhage after trauma or from multiple sites. A treatment-blinded adjudicator
classified hemorrhagic events as major or minor, reviewed data on
death due to any reported hemorrhage, and determined the relation of
the hemorrhage to treatment.
Statistical Analysis
The primary null hypothesis was that there would be no difference
between patients receiving warfarin and those receiving aspirin in
the time to or rate of death from any cause or recurrent ischemic
stroke. Secondary null hypotheses of major clinical interest were
that there would be no differences in the time to either component
of the primary end point or to major hemorrhage according to sex,
race or ethnic group, or cause of prior stroke.
The original target sample size was 1920 patients, which provided
the study with 80 percent power and a 5 percent two-sided probability
of a type I error for a test of the primary null hypothesis according
to the intention to treat, allowing for a 30 percent reduction in
the event rate for one therapy from a 16 percent event rate over two
years for the other, and an overall dropout and discontinuation rate
of 20 percent at two years for both therapies combined. In 1995,
while still blinded to event rates according to treatment group, the
performance and safety monitoring board appointed by the National
Institute of Neurological Disorders and Stroke increased the target
sample size to 2200 to adjust for the possible effects of
interruption of therapy. In 1996, they revised the original stopping
rule based on a single interim analysis by adopting a modified
repeated significance test16
procedure that called for three scheduled interim analyses and allowed
for additional interim analyses. The trial proceeded to its planned
completion and final analysis without crossing the efficacy or
safety boundaries.
All the major study hypotheses were prespecified and tested on
an intention-to-treat basis with a two-tailed alpha of 0.05. The
Kaplan–Meier method17
was used to estimate curves for the length of time to the event, and
the log-rank test18
was used to compare the cumulative incidence curves in the treatment
groups. The primary analysis was adjusted for loss to follow-up by
a prespecified stratified imputation procedure that distinguishes different
types of loss to follow-up and incorporates assumptions appropriate
to each. The reported P values and confidence intervals have not
been adjusted for interim analyses.
Results
A total of 2206 patients were randomly assigned to treatment groups
at a steady rate during the recruitment phase. Their clinical and
demographic features are shown in Table 1. Of
these, 1302 (59 percent) were over the age of 60 years, 1309 (59
percent) were male, 1499 (68 percent) had hypertension, 705 (32
percent) had diabetes, 504 (23 percent) had cardiac disease, 390 (18
percent) had angina or prior myocardial infarction, and 629 (29
percent) had prior amaurosis fugax, transient ischemic attack, or
stroke. The end-point status at two years was established for 2173
(98.5 percent). An additional 33 (1.5 percent) withdrew consent or
were lost to follow-up for other reasons, at a mean of 10.2±7.5
months after randomization. Figure 1
illustrates follow-up and imputation of events according to
treatment.
|
|
Laboratory Testing
Quest Diagnostics determined 48,931 INR values. The mean interval
between the dates of blood sampling was 27.9±12.6 days. The
mean daily INR for patients taking warfarin was 2.1 (median, 1.9).
Overall, 70.7 percent of daily INR values determined 28 or more days
after randomization were within the target range (1.4 to 2.8), 13.0
percent were above the range, and 16.3 percent were below the range.
There were no significant differences in INR values among patients
with different types of prior stroke (cryptogenic; small-vessel or
lacunar; severe stenosis, or occlusion of a large artery; other,
determined cause; and conflicting mechanism [there was more than one
diagnostic possibility]) (P=0.24 by F test with log-transformed INR
values).
Outcomes
The overall rate of the primary end point of death or recurrent
ischemic stroke of 16.9 percent (372 of 2206 patients) slightly exceeded
the 16 percent rate assumed in the trial design. In the primary
intention-to-treat analysis, there were no significant differences
between the warfarin and aspirin groups in the time to the primary end
point (P=0.25 by two-tailed log-rank test; hazard ratio for warfarin
as compared with aspirin, 1.13; 95 percent confidence interval, 0.92
to 1.38; two-year probability of an event, 17.8 percent with
warfarin and 16.0 percent with aspirin) (Table 2 and Figure 2).
Censoring data from subjects whose data were incomplete at the time
of loss to follow-up did not materially affect the outcome of the
primary analysis, and incorporating the interruption of study
medication as a time-dependent covariate showed that the effects of
warfarin and aspirin therapy did not differ.
|
|
The rates of major hemorrhage were low, with no significant differences
between treatment groups; the annual rates were 2.22 per 100
patient-years for warfarin and 1.49 per 100 patient-years for
aspirin (rate ratio, 1.48; P=0.10). Patients in the warfarin group
had significantly more minor hemorrhages than did those in the
aspirin group (Table
3). There was no significant difference between groups in the
time to the first occurrence of major hemorrhage or the primary end
point (P=0.16; hazard ratio with warfarin as compared with aspirin,
1.15; 95 percent confidence interval, 0.95 to 1.39) (Table 2).
|
There were also no significant differences in the time to a primary
end point between patients of different sexes, of different racial
or ethnic groups, or with different types of prior stroke (Table 2). Figure 3 shows
INR-specific rates of primary events plotted by the method of
Rosendaal et al.,19
with use of the last INR value before the event. The rates decline
for INR values until the INR interval of 1.5 to less than 2.0, but
change little thereafter.
|
Discussion
We observed no significant difference between treatment with warfarin
and treatment with aspirin in the prevention of recurrent ischemic
stroke or death or in the occurrence of serious adverse events in
this large cohort of patients with inferred noncardioembolic ischemic
stroke. Not only did the use of warfarin not lead to a 30 percent
reduction in the risk of recurrent stroke — the reduction used to
estimate the sample size — but it was also associated with a
nonsignificant, 13 percent higher increase in risk over that with
aspirin. Treatment with warfarin did not result in excess event
rates during the first 30 days or in a significant increase in the
rates of hemorrhage; these potential outcomes affected the trial
design because of concern that either of these outcomes would offset
any benefit of warfarin.
Two observations suggest that the demographic characteristics
of the study population and outcomes compare favorably with those
of other trials of aspirin or warfarin. First, the event rate among
patients assigned to aspirin was similar to that in other trials of
aspirin for the prevention of recurrent ischemic stroke.10,11,12,14
Furthermore, the low rates of hemorrhage with warfarin were similar to
those in warfarin-treated patients with stroke associated with
atrial fibrillation whose INR values were similar to those of our
patients.5,7 Our
finding that the rate of recurrent stroke with warfarin was similar
to the rate with aspirin suggests that warfarin is an effective
therapy in patients with a prior ischemic stroke. However, in our
trial, warfarin was not superior to aspirin. If anything, the
reverse was true; warfarin did not decrease the rate of severe
recurrent stroke, as it does in patients with prior stroke
associated with atrial fibrillation.5,6,7
Moreover, warfarin costs more than aspirin and requires close
monitoring.
It is unlikely that the range of INR values chosen was too low
to show the superiority of warfarin. Treatment targeted to the same
range of values was successful for the prevention of first strokes
in patients with atrial fibrillation. Published graphs showing the
effect of the INR on the risk of stroke showed curves similar in
shape to those in our results, flattening for INR values of 1.5 to
2.0 and remaining relatively stable for higher values up to 3.0.
However, the event rates in relation to the same range of INR values
(1.5 to 3.0) among patients with atrial fibrillation were well below
that in our study.5,6,7
We considered using higher INR values than those used in studies
of patients with atrial fibrillation, but observations from other
studies published during the course of our study supported our
concern about safety.20,21,22,23,24
Higher rates of major hemorrhage could have stopped the trial before
efficacy could be validly tested, as happened for the Stroke
Prevention in Reversible Ischemia Trial, an open-label comparison of
warfarin with lower-dose aspirin after transient ischemic attacks
and stroke that used an INR range of 3.0 to 4.5 (mean, 3.5).25
Higher INR ranges than those we used in other, nonstroke settings
have had mixed results with respect to safety as compared with
studies of warfarin alone26
or in combination with aspirin.27
The overall percentages of patients with INR values in, above,
or below the target range in our study also compare favorably with
the percentages in other trials. These findings argue against the
possibility that warfarin's lack of superiority to aspirin was due
to high percentages of patients with low INR values. Because reports
of studies showing the success of warfarin in patients with atrial
fibrillation did not present data on the time course of INR values
during the trials in graphic form, no direct time-based comparisons
with our data are possible.
As a direct test of warfarin versus aspirin for the prevention
of recurrent ischemic stroke in a broad clinical setting (excluding patients
with stroke due to embolism), our study necessarily included
patients with a variety of types of prior ischemic stroke. Because
it is not always easy to separate different types of stroke, regardless
of the classification scheme used,28,29,30
some patients with cardiogenic embolism may have been included. If
so, they did not favorably affect the findings with regard to the
effect of warfarin. The recurrence rates in patients with different
types of prior ischemic stroke are similar to those found in the
Stroke Data Bank of the National Institute of Neurological Disorders
and Stroke28
and the Northern Manhattan Stroke Study31
but differ somewhat from those in other studies.32
Like the studies of tissue plasminogen activator for acute stroke,33
our study did not find significant differences in the effects of
treatment among patients with different clinically identifiable types
of prior ischemic stroke. Despite our study's lack of sufficient
power to show such differences, our data nonetheless suggest some
possible selective treatment effects. Aspirin was slightly, but not
significantly, superior to warfarin in patients with large-vessel
and lacunar infarcts. Patients with large-vessel strokes are
currently under study.34
If aspirin is superior to warfarin in lacunar stroke, that finding
will support the idea that there is a mechanistic link between
lacunar disease and large-intracranial-artery atheroma.35,36
Cryptogenic stroke, in which the prevalence of superficial brain
convexity infarcts and lack of evidence of large-artery disease have
made clinically occult embolism15
or coagulopathy37
the leading presumed causes, was the only clinically identified
stroke type for which a possible benefit of warfarin was suggested
by our data; but the reduction in risk was small (8 percent) and not
statistically significant.
Warfarin offered no additional benefit over aspirin in preventing
recurrent ischemic stroke in the population we studied. Patients with
other, established reasons for warfarin use may take comfort in the
evidence of safety and lack of significant difference overall, as
compared with aspirin. However, aspirin, either alone or in
combination with some other antiplatelet agents,38
appears to be a well-justified choice for the prevention of recurrent
ischemic stroke.
Supported by a grant (RO1-NS-28371) from the National Institute
of Neurological Disorders and Stroke. Medications and placebos were
supplied by Dupont Pharmaceuticals and Bayer.
* Participants in the study group are
listed in the Appendix.
Source Information
From the Neurological Institute (J.P.M., R.M.L., R.L.S.) and the
Department of Biostatistics (J.L.P.T., B.L.), Columbia Presbyterian Medical
Center, New York; Massachusetts General Hospital, Boston (K.L.F., J.P.K.);
Stanford University Medical Center, Palo Alto, Calif. (G.W.A.); the University
of Kentucky Medical Center, Louisville (L.C.P.); University of Iowa Health
Care, Iowa City (H.P.A.); the University of California at San Diego, San Diego
(C.M.J.); and the State University of New York at Buffalo, Buffalo (P.P.).
Address reprint requests to Dr. Mohr at the Neurological
Institute, 710 W. 168th St., New York, NY 10032, or at [log in to unmask].
References
Edward E.
Rylander, M.D.
Diplomat American
Board of Family Practice.
Diplomat American
Board of Palliative Medicine.