1 June 1994 Volume 120 Number 11

Annals of Internal Medicine

Risk Factors for Intracranial Hemorrhage in Outpatients
Taking Warfarin
Elaine M. Hylek, MD, MPH, and Daniel E. Singer, MD

• Objective: To explore the rational use of anticoagu- Intracranial hemorrhage is the most feared complica-
lants, especially among the elderly, balancing an- tion of anticoagulant therapy. Outcomes are frequently
tithrombotic efficacy and risk for hemorrhage. Previous catastrophic, often resulting in death or severe neuro-
prospective studies have not provided powerful as- logic disability. The effect of intracranial hemorrhage is
sessments of risk factors for intracranial hemorrhage, as great as that of the thromboembolic events warfarin
the dominant complication in reversing the anticoagu- is used to prevent. As a result, the risk for intracranial
lant decision. hemorrhage is a critical feature of the decision to use
• Design: Case-control analysis. anticoagulation (1). The indications for use of anticoag-
• Setting: A large general hospital and its anticoagu- ulants are expanding, particularly among older patients.
lant therapy unit. For example, anticoagulant therapy is now recom-
• Patients: 121 consecutive adult patients taking war- mended to prevent stroke in patients with atrial fibril-
farin who were hospitalized with intracranial hemor- lation (2-6), and it has been found to significantly im-
rhage were each matched to three contemporaneous prove outcome in patients after myocardial infarction
controls randomly selected from among outpatients (7). Clinical trials of warfarin as primary preventive
managed by our hospital anticoagulant therapy unit. therapy for ischemic heart disease are ongoing (8). With
• Results: 77 patients had intracerebral hemorrhage this increase in the use of long-term warfarin therapy,
(46% fatal) and 44 had subdural hemorrhage (20% there is a pressing need to identify clinical features that
fatal). The prothrombin time ratio (PTR) was the domi- raise the risk of its most severe bleeding complications.
nant risk factor for intracranial hemorrhage. For each No previous study has identified independent risk fac-
0.5 increase in PTR over the entire range, the risk for tors for intracranial hemorrhage among patients taking
intracerebral hemorrhage doubled (odds ratio, 2.1; 95% warfarin. Previous prospective analytic studies that fo-
CI, 1.4 to 2.9). For subdural hemorrhage, the risk was cused on the entire spectrum of major bleeding compli-
unchanged over the PTR range from 1.0 to 2.0 but rose cations have contained few cases of intracranial hemor-
dramatically above a PTR of 2.0 (approximate interna- rhage (9-11).
tional normalized ratio, 4.0). Age was the only other We designed a case-control study focused exclusively
significant independent risk factor for subdural hemor- on intracranial hemorrhage occurring among outpatients
rhage (odds ratio, 2.0 per decade; CI, 1.3 to 3.1). For taking warfarin. We drew on an 11-year experience of
intracerebral hemorrhage, age was of borderline signif- one general hospital to provide case-patients and used
icance (odds ratio, 1.3 per decade; CI, 1.0 to 1.6) after the same hospital's large anticoagulant therapy unit to
controlling for PTR and the two other independent risk provide contemporaneous controls who were also tak-
factors: history of cerebrovascular disease (odds ratio, ing warfarin. This design was chosen to increase statis-
3.1; CI, 1.7 to 5.6) and presence of a prosthetic heart tical power for detecting risk factors for intracranial
valve (odds ratio, 2.8; CI, 1.3 to 5.8). hemorrhage while reducing possible bias.
• Conclusions: The results emphasize the importance
of maintaining the prothrombin time ratios under 2.0
and the need for especially careful use of warfarin in the Methods
elderly. Case-Patient Identification and Eligibility
Using a discharge log of consecutive admissions to the Mas-
sachusetts General Hospital during the period from 1 January
1981 through 31 December 1991, we identified 1881 patients
with a principal diagnosis of intracranial hemorrhage (ICD-9
codes 430, 431, 432.0, 432.1, and 432.9). Forty-one (2%) med-
ical records for these patients could not be located. In the
remaining 1840 case-patients, warfarin use was determined
from review of the neurologist, neurosurgery resident, and
attending staff physician admission notes. Intracranial hemor-
rhage was verified by computed tomographic (CT) scanning,
Ann Intern Med. 1994;120:897-902. lumbar puncture, or postmortem examination in all but one
case-patient. In this latter case-patient, the diagnosis was
From Massachusetts General Hospital, Boston, Massachusetts. based on clinical grounds because the patient died before di-
For current author addresses, see end of text. agnostic studies were done and no autopsy was performed.

1 June 1994 • Annals of Internal Medicine • Volume 120 • Number 11 897

Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/19800/ by a Fudan University User on 02/12/2017

 To be eligible for the study, patients had to be at least 18 used commercial laboratory has used thromboplastins with ISI
years old and taking warfarin as an outpatient. Patients with an values ranging from 1.9 to 2.1. The PTR was missing for four
anatomic abnormality or underlying bleeding diathesis predis- case-patients; the data for case-patients and controls were oth-
posing to intracranial hemorrhage, regardless of anticoagulant erwise complete.
therapy, were not included. Hemorrhages sustained as a result
of major head trauma (with skull fracture and loss of con- Selected Relevant Definitions
sciousness) also were not eligible.
Hypertension was defined as probable if the patient had such
Of 131 patients with intracranial hemorrhage identified, 10 a diagnosis listed in the medical record and as definite if the
were excluded as ineligible: Four patients had subarachnoid patient was receiving antihypertensive medication. When hy-
hemorrhage resulting from angiographically identified intracra- pertension was diagnosed, it was classified as definite 97% of
nial aneurysms; 2 had hemorrhage into primary or metastatic the time for case-patients and 93% of the time for controls. All
tumors; 2 had acute subdural hemorrhage after major head diagnoses of hypertension are used in the analyses.
trauma; 1 bled after multiple craniotomies and radiation ther- Patients with documented carotid or vertebrobasilar disease
apy for a recurrent craniopharyngioma; and another patient and those with a history of previous stroke were defined as
had aplastic anemia. having cerebrovascular disease. Diagnoses of carotid or verte-
brobasilar disease or both were confirmed by angiography in
71% of case-patients and 93% of controls, by Doppler studies
Controls: Source and Matching in 17% of case-patients and 2% of controls, and solely by
assessment by a neurologist in 12% of case-patients and 5% of
Controls were selected from the registry of the Massachu-
controls.
setts General Hospital's anticoagulant therapy unit. During the
study period, this unit managed warfarin dosing for approxi- Statistical Analysis
mately 8000 patients referred from all hospital clinical services.
The most common indications for anticoagulation were atrial Case-patients and controls were compared using chi-square
fibrillation, previous stroke, presence of prosthetic heart tests and the Fisher exact test, where appropriate, for categor-
valves, and venous thromboembolism. Approximately 40% of ical variables and using the Student Mest for continuous vari-
the patients managed by the anticoagulant therapy unit have ables. Univariate odds ratios were calculated using unmatched
prothrombin time tests done at the Massachusetts General and matched techniques. The matched odds ratios were pro-
Hospital. An additional 20% have their prothrombin times vided by the Mantel-Haenszel (12) summary statistic across
measured at one large commercial laboratory (via home phle- matched sets. The unmatched and matched techniques pro-
botomy services). The remaining 40% have their tests done at vided very similar estimates of odds ratios. We report the
various local laboratories. unmatched results. Confidence intervals for odds ratios were
Each case-patient was matched to three randomly selected calculated using the Taylor series method (13). The test of
controls taking warfarin at the time of the case-patient's intra- trend was done using the Cochran-Mantel-Haenszel test (14).
cranial hemorrhage. Matching was accomplished by first iden- Logistic regression models assessed the independent effect of
tifying all patients managed by the anticoagulant therapy unit multiple clinical features and the significance of interaction
at the date of hospital admission for the given case-patient. terms. Conditional logistic models (15) accounting for matching
Each potential control was assigned a random number. The provided estimates similar to those of the unmatched logistic
three controls with the lowest random numbers were selected. regression analyses. The estimates from the unmatched analy-
No control was used more than once. Three controls subse- ses are reported. Data were recorded in R:BASE (Microrim,
quently became case-patients. Matching was done to control Bellevue, Washington) from the paper data forms. Statistical
for any change in background risk over the 11-year study analyses were done using SAS (SAS Institute Inc., Cary,
period (for example, from changes in prothrombin time targets North Carolina) and GLIM (Numerical Algorithms Group Lim-
or technique of measuring the prothrombin time). ited, Oxford, United Kingdom).

Data Collected Results

Clinical features of case-patients and controls were extracted Clinical Course of Case-Patients
primarily from hospital records, with supplementation in a few
instances from physician office records. Variables were entered During the 11-year study period, 121 patients with
on a predesigned data form and included indication for antico- intracranial hemorrhage were eligible; 77 hemorrhages
agulant therapy; race; sex; age; prothrombin time ratio (PTR);
duration of warfarin therapy; history of diagnosed hyperten- were intracerebral and 44 were subdural (Table 1).
sion, stroke, transient ischemic attack, diabetes mellitus, myo- Three of the patients with subdural hemorrhage had a
cardial infarction, atrial fibrillation, and congestive heart fail- history of trivial head trauma; the others had no known
ure; and medications. Clinical features of the intracranial antecedent head trauma. For the patients with intrace-
hemorrhages were also recorded. Clinical data for the controls
rebral hemorrhages, headache was the most common
were current at the admission date of the matched case-patient
with intracranial hemorrhage. presenting feature (53%), followed by nausea and vom-
The prothrombin time ratio was expressed as the ratio of the iting (40%) and unresponsiveness (36%). Seventy-eight
patient's value divided by the simultaneously reported control percent of the case-patients presented to the emergency
value. For case-patients, we used the PTR on admission, or, if department within 24 hours of the onset of symptoms
available, the PTR closest to the reported onset of symptoms.
and 87% within 48 hours. In contrast, only 36% of the
For controls, the PTR closest to that of the case-patient ad-
mission date was recorded from the anticoagulant therapy unit patients with subdural bleeding came to medical atten-
database. Because values for the international sensitivity index tion within this same period. Forty-six percent of pa-
(ISI) for thromboplastins were not universally reported before tients with intracerebral bleeding died, and 17% sur-
1988, we analyzed our data using the PTR rather than the vived with major neurologic deficits that prevented
international normalized ratio (INR). Since 1988, Massachu-
setts General Hospital's hematology laboratory has used Sim- subsequent independent living.
plastin Automated (Organon Teknika Corporation, Durham, The most common presenting complaint in those pa-
North Carolina), with values for the ISI ranging from 1.9 to tients with a subdural hemorrhage w a s also headache
2.0. During the study period, this company (previously General (68%), followed by confusion (39%) and ataxia (32%).
Diagnostics) supplied the thromboplastin and assisted our lab-
oratory in selecting lots of comparable sensitivity. Studies Twenty percent of patients with subdural bleeding died
were routinely done to minimize year-to-year variation in before discharge, and 9% had major neurologic se-
thromboplastin sensitivity. Since 1988, the most frequently quelae.

898 1 June 1994 • Annals of Internal Medicine • Volume 120 • Number 11

Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/19800/ by a Fudan University User on 02/12/2017

 Table 1. Clinical Course of Intracranial Hemorrhages above 2.0. The relation between the PTR and the rela-
tive odds of intracranial hemorrhage is shown in Figure
Variable Intracerebral Subdural
1. Age also achieved significance as a risk factor for
(n = 77) Jn = 44) both intracerebral and subdural hemorrhage in the
n(%) univariate analysis, although it was more important as a
Presenting features (not mutually risk factor for subdural hemorrhage (Table 2). The mean
exclusive) age of patients with intracerebral hemorrhage was 69
Headache 41 (53) 30 (68) years (range, 28 to 92 years) compared with a mean age
Nausea and vomiting 31 (40) 12 (27) of 64 years (range, 18 to 92 years) for controls (P =
Unresponsiveness 28 (36) 11 (25)
Hemiparesis 26 (34) 13 (30)
0.003). For patients with subdural hemorrhage, the
Aphasia 15 (20) 9(20) mean age was 73 years (range, 52 to 88 years) compared
Ataxia 13 (17) 14 (32) with 64 years (range, 20 to 88 years) for controls (P =
Confusion 7(9) 17 (39) 0.001). Additional risk factors for intracerebral bleeding
Estimated time from onset of
symptoms to presentation
included a history of cerebrovascular disease (odds ra-
<24h 60 (78) 12 (27) tio, 2.9; 95% CI, 1.7 to 4.9) and the presence of a
24 to 48 h 7(9) 4(9) prosthetic heart valve (odds ratio, 2.0; CI, 1.1 to 3.8).
48 to 72 h 3(4) 7(16) For subdural hemorrhage, other variables achieving
3to7d 5(6) 12 (27) significance in the univariate analyses included a history
>7d 2(3) 9(21)
Outcome by discharge
of atrial fibrillation (odds ratio, 2.9; CI, 1.4 to 5.8), a
Death 35 (46) 9(20) history of myocardial infarction (odds ratio, 0.3; CI, 0.1
Major deficit 13 (17) 4(9) to 0.7), and duration of warfarin therapy more than 5
Minor deficit 22 (28) 22 (50) years (odds ratio, 2.8; CI, 1.3 to 6.2).
No sequelae 7(9) 9(20)

Multivariate Analyses
Comparison of Case-Patients and Controls
In multiple logistic models the PTR remained the
The case-patients and controls did not differ signifi- dominant independent risk factor for both intracerebral
cantly in sex; race; or past medical history of diagnosed and subdural hemorrhage (Table 3). For subdural hem-
hypertension, diabetes mellitus, or congestive heart fail- orrhage, a term representing the square of the PTR was
ure (Table 2). The PTR was the dominant risk factor for significant. As a result, the estimated odds of subdural
both intracerebral and subdural hemorrhage, with the hemorrhage increased 7.6-fold as the PTR increased
risk for both increasing dramatically for PTR values from 2.0 to 2.5. For intracerebral hemorrhage, the risk

Table 2. Clinical Features of Case-Patients with Intracerebral Hemorrhage and of Subdural Hemorrhage Compared
with Their Controls
Variable Intracerebral Control Odds Ratio Subdural Control Odds Ratio
Hemorrhage (n = 231) (95% CI) Hemorrhage (n = 132) (95% CI)
(n = 77) (n = 44)
% %
Male sex 49 58 0.7 (0.4 to 1.2) 66 61 1.2 (0.6 to 2.5)
Race
White 91 88 98 98
Other 9 12 0.8 (0.3 to 1.8) 2 2 1.0 (0.1 to 9.9)
Medical history
Hypertension 51 47 1.2 (0.7 to 2.0) 48 52 0.9 (0.4 to 1.7)
Atrial fibrillation 49 38 1.6 (0.9 to 2.7) 61 36 2.9 (1.4 to 5.8)
Myocardial infarction 25 29 0.8 (0.5 to 1.5) 14 36 0.3 (0.1 to 0.7)
Congestive failure 22 16 1.5 (0.8 to 2.8) 27 21 1.4 (0.6 to 3.1)
Diabetes mellitus 22 15 1.6 (0.9 to 3.2) 20 14 1.6 (0.7 to 4.0)
Cerebrovascular disease 60 34 2.9 (1.7 to 4.9) 39 31 1.4 (0.7 to 2.8)
Prosthetic valve 22 13 2.0(1.1 to 3.8) 25 18 1.5 (0.7 to 3.4)
Prothrombin time ratio
<1.5 32 42 1.0 31 49 1.0
1.5 to 2.0 41 48 1.2 (0.6 to 2.1) 26 42 1.0 (0.4 to 2.3)
>2.0 27 10 3.6 (1.7 to 7.5) 43 9 7.4 (2.9 to 18.9)
Age
<65y 32 47 1.0 16 42 1.0
65 to 79 y 44 41 1.6 (0.9 to 2.8) 54 45 3.2 (1.3 to 8.0)
>80y 23 12 2.8 (1.3 to 5.8)* 30 14 5.7 (2.0 to 16.4)t
Duration of warfarin therapy
<3 mo 14 14 0.9 (0.4 to 2.0) 16 11 1.5 (0.6 to 3.9)
3 mo to 5 y 63 72 0.7 (0.4 to 1.2) 52 74 0.5 (0.2 to 1.3)
>5y 23 14 1.9 (1.0 to 3.6) 32 14 2.8 (1.2 to 6.2)

* P = 0.006.
t P = 0.001.

1 June 1994 • Annals of Internal Medicine • Volume 120 • Number 11 899

Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/19800/ by a Fudan University User on 02/12/2017

 anticoagulant therapy. In particular, anticoagulants
largely reverse the increased risk for stroke among pa-
tients with atrial fibrillation, a group that constitutes 4%
of persons older than 60 years and 10% of persons older
than 80 years (16). The decision to use anticoagulants
hinges on the balance between the decreased risk for
thromboembolism and the increased risk for hemor-
rhage. Intracranial hemorrhage assumes great impor-
tance in these considerations because it is the only
complication that is comparable in severity to the
thromboembolic events that anticoagulants prevent. The
anticoagulant decision would clearly benefit from pow-
erful studies identifying risk factors for intracranial
hemorrhage. Low absolute event rates have hindered
previous prospective studies. Estimates of the absolute
rate of anticoagulant-related intracranial hemorrhage
Figure 1. Odds ratio of intracranial hemorrhage compared with range from 0.3% per year observed in the first five trials
prothrombin time ratio. The unadjusted odds ratio for intracra- in atrial fibrillation (2-6) to approximately 2% per year
nial hemorrhage is shown for different levels of the prothrom- reported from an observational study (9) as well as from
bin time ratio. The prothrombin time ratios presented are the a randomized trial studying patients older than 75 years
medians of the following intervals: 1.0 to 1.5, 1.6 to 1.7, 1.8 to
1.9, 2.0 to 2.1, 2.2 to 2.3, and 2.4 to 3.5. (17). Rates of 0.5% to 0.6% per year were found in two
large observational studies from anticoagulant therapy
units (10, 11). Unfortunately, rates of intracranial hem-
rose fourfold for each unit increase in the PTR. (The orrhage as low as 1% to 2% per year can reverse the
corresponding odds ratio for a 0.5 increase in the PTR decision to use anticoagulants (1).
was 2.1.) Age was the only other independent risk fac- We chose a case-control design to overcome the
tor for subdural hemorrhage (odds ratio, 2.0 per decade; problem of the low absolute event rate of intracranial
CI, 1.3 to 3.1). hemorrhage. Our study's 121 case-patients represent
For intracerebral hemorrhage age was of borderline roughly the equivalent of 20 000 person-years of pro-
statistical significance after controlling for the PTR and spective follow-up. Large population-based studies have
the other independent risk factors: history of cerebro- estimated that anticoagulants raise the risk for intracra-
vascular disease (odds ratio, 3.1; CI, 1.7 to 5.6) and nial hemorrhage tenfold (18-20), implying that nearly all
presence of a prosthetic heart valve (odds ratio, 2.8; CI, this risk is attributable to anticoagulant therapy itself
1.3 to 5.8). The independent variables had no significant and dictating that our controls should also have been
correlations and no significant two-way interactions. taking anticoagulants.
Combined Analysis Our large number of case-patients allowed for sepa-
When all 121 case-patients of both types were com- rate risk assessment for intracerebral and subdural hem-
bined, independent risk factors for intracranial hemor-
rhage were PTR (odds ratio, 4.5 per unit increase; CI, orrhage. The PTR was the dominant independent risk
2.6 to 8.0), history of cerebrovascular disease (odds factor for both. The relation between intensity of anti-
ratio, 2.3; CI, 1.4 to 3.7), presence of a prosthetic heart Table 3. Independent Risk Factors for Intracranial Hem-
valve (odds ratio, 2.1; CI, 1.2 to 3.8) and age (odds
orrhage in Outpatients on Warfarin Therapy
ratio, 1.4 per decade; CI, 1.2 to 1.7).
Risk Factor Odds Ratio P Value
Analysis of Case-Patients and Controls Managed by (95% CI)
the Anticoagulant Therapy Unit
Fifty-five case-patients with intracranial hemorrhage Subdural hemorrhage
had had their warfarin therapy managed by the antico- Prothrombin time ratio 7.6 (2.7 to 21.7)* <0.001
agulant therapy unit. Analyses restricted to these case- Age 2.0(1.3to3.1)t 0.002
Intracerebral hemorrhage
patients and their respective controls represent a case- Prothrombin time ratio 4.1 (1.9 to 8.7)$ <0.001
control study nested within the defined anticoagulant Cerebrovascular disease 3.1 (1.7 to 5.6) <0.001
therapy unit population. The effects of previously iden- Prosthetic heart valve 2.8 (1.3 to 5.8) 0.009
tified independent risk factors for intracranial hemor- Age 1.3(1.0tol.6)t 0.05
Intracranial hemorrhage
rhage were similar in this subset of our study popula- Prothrombin time ratio 4.5 (2.6 to 8.0)* <0.001
tion: prothrombin time ratio (odds ratio, 8.6 per unit Cerebrovascular disease 2.3 (1.4 to 3.7) <0.001
increase; CI, 3.3 to 22.6), cerebrovascular disease (odds Prosthetic heart valve 2.1 (1.2 to 3.8) 0.01
ratio, 2.0; CI, 1.0 to 4.1), prosthetic heart valve (odds Age 1.4(1.2tol.7)t <0.001
ratio, 1.6; CI, 0.6 to 3.9), and age (odds ratio, 1.5 per * This odds ratio applies to a comparison of prothrombin time ratios,
decade; CI, 1.1 to 2.0). 2.5 compared with 2.0. The model generating this odds ratio includes a
significant (prothrombin time ratio)2 term. Note that odds ratios for
prothrombin time ratio in the model for intracerebral hemorrhage and
Discussion for all intracranial hemorrhage apply to a full unit change in prothrom-
bin time ratio.
Recent randomized trials have greatly expanded the t Per decade.
number of persons with strong indications for long-term t Per unit.

900 1 June 1994 • Annals of Internal Medicine • Volume 120 • Number 11

Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/19800/ by a Fudan University User on 02/12/2017

 coagulant therapy and the risk for major bleeding is well hemorrhage in patients with a history of past or current
known (21) but has not been well quantified for intra- stroke.
cranial hemorrhage. We found that for each 0.5 increase It is unclear why patients with prosthetic heart valves
in PTR, the risk for intracerebral bleeding doubled. For should be at increased risk for intracerebral hemor-
subdural hemorrhage, the risk was unchanged between rhage. The effect of prosthetic heart valves appears to
PTRs ranging from 1.0 to 2.0 but rose dramatically for be independent of higher intensity of anticoagulation. It
those above 2.0. Our risk estimates relate to the PTRs also seems unlikely that these intracerebral hemor-
that actually occurred in the study patients, regardless rhages were initially embolic infarctions because such
of their target PTR. It is likely that the intensity of hemorrhagic transformations are unusual and occur
anticoagulation is an even greater risk factor for intra- mainly with large ischemic strokes (31). Several recent
cranial hemorrhage than our data indicate because the randomized controlled trials in patients with prosthetic
PTR is not well standardized (22, 23). One can antici- heart valves have shown decreased rates of major
pate that the association between INR and intracranial bleeding with less intense anticoagulation without loss
hemorrhage would be stronger and might reveal a of antithrombotic efficacy (32, 33). Our data further
monotonic relation with risk even at the low end of support the use of lower intensities of anticoagulation in
"therapeutic" intensity of anticoagulation (for example, patients with prosthetic valves.
INR, 2.0 to 3.0). Even in our data set, most of the Only 4% of the case-patients with intracranial hem-
intracranial hemorrhages occurred at PTRs below 2.0. orrhage had a history of head trauma (including the two
Since 1988, the two laboratories serving most of our case-patients with major head trauma who were exclud-
study patients have recorded ISI values for their throm- ed). The rarity of antecedent trauma in anticoagulant-
boplastins. This ISI range has been 1.9 to 2.1. As a related cases of chronic subdural hemorrhages has been
result, our PTR threshold of 2.0 corresponds to an INR previously documented (34, 35). Acute nontraumatic
range of 3.7 to 4.3. subdural hemorrhage seems almost exclusively depen-
The high mortality rate associated with anticoagulant- dent on anticoagulant therapy (34).
related intracranial hemorrhage has been attributed to We did not find any independent association between
the increased volume of the hematoma, reported to be duration of anticoagulant therapy and intracranial hem-
approximately double that occurring with spontaneous orrhage, nor did we find any risk associated with sex or
intracerebral hemorrhage (17, 24). Radberg and col- treated hypertension. Because patients on warfarin ther-
leagues (24) also reported an increased hematoma vol- apy are highly selected and closely followed, a cohort
ume and mortality rate with higher INR values. of truly untreated hypertensive patients who are receiv-
Previous studies have reported conflicting conclusions ing warfarin therapy may not exist.
about the role of age (9, 10, 25). For example, Fihn and Potential methodologic limitations of this study merit
colleagues (10) showed a relative risk of 0.8 for persons discussion. In a case-control study, controls should be a
older than 65 years with an upper 95% CI limit estimate representative sample of the population that gives rise
of 1.0, whereas Landefeld and Goldman (9) found a to the case-patients (36). Ideally, controls for this study
highly significant (P < 0.001) relative risk of 3.2. The would have been a random selection of patients taking
report of the Second Stroke Prevention in Atrial Fibril- warfarin who would have come to our hospital because
lation (SPAF-II) Trial has noted an intracranial hemor- of a bleeding complication. Such a population is difficult
rhage rate of 1.8% per year for patients receiving anti- to define unambiguously, and no registry of such pa-
coagulation who were older than 75 years compared tients exists. However, the patients receiving care from
with 0.5% per year for patients receiving anticoagula- our anticoagulant therapy unit do constitute a compara-
tion who were 75 years or younger (17). Age was a ble cohort. Our substudy of case-patients arising from
strong independent risk factor for subdural hemorrhage the anticoagulant therapy unit and their controls consti-
in our study, nearly doubling the risk with each 10-year tutes an unbiased case-control study nested within the
increase, perhaps reflecting the increased fragility of anticoagulant therapy unit cohort. The fact that our
bridging veins and the effect of cerebral atrophy (26). results were largely the same in this subgroup argues
For intracerebral hemorrhage, age was of borderline strongly for the lack of substantial selection bias over-
statistical significance in the multiple regression model. all.
The incidence rates for intracerebral hemorrhage in pa- Our case-control design does not directly provide the
tients not taking warfarin clearly rise with age (27), an absolute risk for intracranial hemorrhage associated
effect attributed to age-related pathologic changes (28, with each risk factor. However, the average absolute
29). It is not so clear that age increases the risk for risk observed in recent prospective studies can be cou-
intracerebral hemorrhage among patients taking antico- pled with our findings (including the distribution of risk
agulants, where other features, in particular, the inten- factors in our control group) to provide estimates of the
sity of anticoagulation, assume greater importance. absolute risk in any risk factor category. For example,
Cerebrovascular disease and prosthetic heart valves using the risk estimate of van der Meer and colleagues
were the other identified independent risk factors for (11)—0.6% per year—one can estimate that a PTR
intracerebral bleeding. Levine and Hirsh (30) have pre- greater than 2.0 conveys an important absolute risk for
viously noted that major bleeding in patients taking intracranial hemorrhage of 2% per year. Finally, we
warfarin was especially likely in patients with ischemic note that although case-control studies can add effi-
cerebrovascular disease and was frequently intracere- ciency and power to the study of infrequent events,
bral in origin, and Landefeld and Goldman (9) have they may fail to identify important but uncommon risk
reported a univariate relative risk of 6.6 for intracranial factors.

1 June 1994 • Annals of Internal Medicine • Volume 120 • Number 11 901

Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/19800/ by a Fudan University User on 02/12/2017

 Recent randomized trials have established the efficacy ing complications in oral anticoagulant therapy. Arch Intern Med.
1993;153:1557-62.
of long-term warfarin therapy in preventing thromboem-
12. Fleiss JL. Statistical Methods for Rates and Proportions. New York:
bolic complications of several common chronic condi- John Wiley and Sons; 1981:123-33.
tions. Nonetheless, warfarin remains a demanding and 13. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic Re-
search Principles and Quantitative Methods. Belmont, California:
risky therapy and should be used only in those groups Lifetime Learning Publications, 1982:299.
of patients where the expected benefit clearly exceeds 14. Snedecor GW, Cochran WG. Statistical Methods. Seventh edition.
the risk. The randomized trials coupled with observa- Ames, Iowa: Iowa State University Press; 1980:206-8.
15. Hosmer DW, Lemeshow S. Logistic Regression for Matched Case-
tional epidemiologic studies have helped to identify Control Studies. Applied Logistic Regression. New York: John
groups at high risk for embolic events in certain condi- Wiley & Sons; 1989:187-215.
tions (37). Much less information is available on risk 16. Wolf PA, Abbott RD, Kannel WB. Atrial fibrillation: a major con-
tributor to stroke in the elderly. Arch Intern Med. 1987;147:1561-4.
factors for intracranial hemorrhage. Our study shows 17. Stroke Prevention in Atrial Fibrillation Investigators. Warfarin versus
the importance of the effect of increasing age on risk for aspirin for prevention of thromboembolism in atrial fibrillation:
subdural hemorrhage with warfarin and of previous ce- Stroke Prevention in Atrial Fibrillation II Study. Lancet. 1994;343:
687-91.
rebrovascular disease and of prosthetic heart valves on 18. Franke CL, de Jonge J, van Swieten JC, Op de Coul AA, van Gyn
the risk for intracerebral hemorrhage. Most importantly, J. Intracerebral hematomas during anticoagulant treatment. Stroke.
1990;21:726-30.
our study shows the dramatic increase in risk for intra- 19. Wintzen AR, de Jonge H, Loeliger EA, Bots GT. The risk of intra-
cranial hemorrhage with increasing PTR, especially cerebral hemorrhage during oral anticoagulant treatment: a popula-
PTRs greater than 2.0, underlining the need for careful tion study. Ann Neurol. 1984;16:553-8.
20. Whisnant JP, Cartlidge NE, Elveback LR. Carotid and vertebral-
control of anticoagulation at the lowest effective inten- basilar transient ischaemic attacks: effect of anticoagulants, hyper-
sity, particularly among the many older patients for tension, and cardiac disorders on survival and stroke occurrence—a
whom warfarin is now indicated. population study. Ann Neurol. 1978;3:107-15.
21. Levine MN, Hirsh J, Landefeld S, Raskob G. Hemorrhagic compli-
Acknowledgments: The authors thank E. Francis Cook, ScD, and Terry cations of anticoagulant treatment. Chest. 1992;102(Suppl.):352S-
S. Field, ScD, for advice on study design and analysis; Robert A. 363S.
Hughes, MD, for facilitating use of the database; and Lynn Oertel, MS, 22. Bussey HI, Force RW, Bianco TM, Leonard AD. Reliance on pro-
RN-C, for help with data collection. thrombin time ratios causes significant errors in anticoagulation
therapy. Arch Intern Med. 1992;152:278-82.
Requests for Reprints: Elaine M. Hylek, MD, MPH, General Internal 23. Hirsh J. Substandard monitoring of warfarin in North America: time
Medicine Unit, Massachusetts General Hospital, Bulfinch 1, Boston, for change [Editorial]. Arch Intern Med. 1992;152:257-8.
MA 02114. 24. Radberg JA, Olsson JE, Radberg CT. Prognostic parameters in
spontaneous intracerebral hematomas with special reference to an-
Current Author Addresses: Drs. Hylek and Singer: General Internal ticoagulant treatment. Stroke. 1991;22:571-6.
Medicine Unit, Massachusetts General Hospital, Bulfinch 1, Boston, 25. Sixty-Plus Reinfarction Study Research Group. Risks of long-term
MA 02114. anticoagulant therapy in elderly patients after myocardial infarction.
Lancet. 1982;1:64-8.
26. Yamashima T, Friede RL. Why do bridging veins rupture into the
References virtual subdural space? J Neurol Neurosurg Psychiatry. 1984;47:
1. Gustafsson C, Asplund K, Britton M, Norrving B, Olsson B, Marke 121-7.
L. Cost effectiveness of primary stroke prevention in atrial fibrilla- 27. Broderick J, Brott T, Tomsick T, Leach A. Lobar hemorrhage in the
tion: Swedish national perspective. BMJ. 1992;305:1457-60. elderly. The undiminishing importance of hypertension. Stroke.
2. The Boston Area Anticoagulation Trial for Atrial Fibrillation Inves- 1993;24:49-51.
tigators. The effect of low-dose warfarin on the risk of stroke in 28. Masuda J, Tanaka K, Ueda K, Omae T. Autopsy study of incidence
patients with nonrheumatic atrial fibrillation. N Engl J Med. 1990; and distribution of cerebral amyloid angiopathy in Hisayama, Japan.
323:1505-11. Stroke. 1988;19:205-10.
3. Connolly SJ, Laupacis A, Gent M, Roberts RS, Cairns J A, Joyner C. 29. Vonsattel JP, Myers RH, Hedley-Whyte ET, Ropper AH, Bird ED,
Canadian Atrial Fibrillation Anticoagulation Study. J Am Coll Car- Richardson EP Jr. Cerebral amyloid angiopathy without and with
diol. 1991;18:349-55. cerebral hemorrhages: a comparative histological study. Ann Neu-
4. Ezekowitz MD, Bridgers SL, James KE, Carliner NH, Colling CL, rol. 1991;30:637-49.
Gornick CC, et al. Warfarin in the prevention of stroke associated 30. Levine M, Hirsh J. Hemorrhagic complications of long-term antico-
with nonrheumatic atrial fibrillation. N Engl J Med. 1992;327:1406- agulant therapy for ischemic cerebral vascular disease. Stroke.
12. 1986;17:111-6.
5. Petersen P, Boysen G, Godtfredsen J, Andersen ED, Andersen B. 31. Hart RG, Lockwood KI, Hakim AM. Immediate anticoagulation of
Placebo-controlled, randomised trial of warfarin and aspirin for pre- embolic stroke: brain hemorrhage and management options. Stroke.
vention of thromboembolic complications in chronic atrial fibrilla- 1984;15:779-89.
tion. The Copenhagen AFASAK study. Lancet. 1989;1:175-9. 32. Turpie AG, Gunstensen J, Hirsh J, Nelson H, Gent M. Randomised
6. Preliminary report of the Stroke Prevention in Atrial Fibrillation comparison of two intensities of oral anticoagulant therapy after
Study. N Engl J Med. 1990;322:863-8. tissue heart valve replacement. Lancet. 1988;1:1242-5.
7. Smith P, Arnesen H, Holme I. The effect of warfarin on mortality 33. Saour JN, Sieck JO, Mamo LA, Gallus AS. Trial of different inten-
and reinfarction after myocardial infarction. N Engl J Med. 1990; sities of anticoagulation in patients with prosthetic heart valves.
323:147-52. N Engl J Med. 1990;322:428-32.
8. Meade TW. Low-dose warfarin and low-dose aspirin in the primary 34. Wintzen AR, Tyssen JG. Subdural hematoma and oral anticoagulant
prevention of ischemic heart disease. Am J Cardiol. 1990;65:7C- therapy. Arch Neurol. 1982;39:69-72.
11C. 35. Mattle H, Kohler S, Huber P, Rohner M, Steinsiepe KF. Anticoag-
9. Landefeld CS, Goldman L. Major bleeding in outpatients treated ulation-related intracranial extracerebral haemorrhage. J Neurol
with warfarin: incidence and prediction by factors known at the Neurosurg Psychiatry. 1989;52:829-37.
start of outpatient therapy. Am J Med. 1989;87:144-52. 36. Rothman KJ. Modern Epidemiology. Boston: Little, Brown; 1986:
10. Fihn SD, McDonell M, Martin D, Henikofif J, Vermes D, Kent D, et 63.
al. Risk factors for complications of chronic anticoagulation. Ann 37. Atrial Fibrillation Investigators. Risk factors for stroke and efficacy
Intern Med. 1993;118:511-20. of anti-thrombotic therapy in atrial fibrillation. Arch Intern Med.
11. van der Meer FJ, Rosendaal FR, Vandenbroucke JP, Briet E. Bleed- 1994: [In press.]

902 1 June 1994 • Annals of Internal Medicine • Volume 120 • Number 11

Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/19800/ by a Fudan University User on 02/12/2017