Hylek EM, Heiman H, Skates SJ, Sheehan MA, Singer DE. Acetaminophen and Other Risk Factors for Excessive Warfarin Anticoagulation. JAMA. 1998;279(9):657–662. doi:10.1001/jama.279.9.657
From the General Medicine Division, Clinical Epidemiology Unit, Department of Medicine, Massachusetts General Hospital and Harvard Medical School, Boston.
Context.— Warfarin is highly effective in preventing thromboembolism, but increases
the risk of hemorrhage, particularly at an international normalized ratio
(INR) greater than 4.0. Identifying causes of excessive anticoagulation in
clinical practice could help target patients at risk for elevated INRs.
Objective.— To determine causes of INRs greater than 6.0 in a clinical practice
Design.— Case-control study.
Setting.— Outpatient anticoagulant therapy unit.
Patients.— Outpatients followed up prospectively from April 1995 to March 1996
who had been taking warfarin for more than 1 month, had a target INR of 2.0
to 3.0, and were able to be interviewed within 24 hours of their reported
INR. Case patients had INRs greater than 6.0; controls were randomly selected
from patients having INRs between 1.7 and 3.3.
Main Outcome Measures.— Factors associated with INRs greater than 6.0, including medication
use, recent diet, illness, alcohol consumption, and actual warfarin use.
Results.— A total of 93 cases and 196 controls were interviewed; they did not
differ in age, indication for warfarin, length of therapy, warfarin dose,
number of prescription medications, or previous INR or long-term INR variability.
Acetaminophen ingestion was independently associated in a dose-dependent manner
with having an INR greater than 6.0 (P for trend
<.001). For the highest-dose category of acetaminophen intake, 9100 mg/wk
or more, the odds of having an INR greater than 6.0 were increased 10-fold
(95% confidence interval [CI], 2.6-37.9). Other factors independently associated
with an INR greater than 6.0 were new medication known to potentiate warfarin
(odds ratio [OR], 8.5; 95% CI, 2.9-24.7), advanced malignancy (OR, 16.4; 95%
CI, 2.4-111.0), recent diarrheal illness (OR, 3.5; 95% CI,1.4-8.6), decreased
oral intake (OR, 3.6; 95% CI, 1.3-9.7), and taking more warfarin than prescribed
(OR, 8.1; 95% CI, 2.2-30.0). Higher vitamin K intake (OR, 0.7; 95% CI, 0.5-0.9)
and habitual alcohol consumption of from 1 drink every other day to 2 drinks
a day (OR, 0.2; 95% CI, 0.1-0.7) were associated with decreased risk.
Conclusions.— These data suggest that acetaminophen is an underrecognized cause of
overanticoagulation in the outpatient setting. Several other clinically important
risk factors were identified. Increased monitoring of INR values when such
risk factors are present or modification of the risk factors themselves should
reduce the frequency of dangerously high levels of anticoagulation.
WARFARIN is highly effective in preventing thromboembolism in a variety
of conditions.1 Its major complication is hemorrhage,
which can cause severe morbidity and death. In addition, fear of major hemorrhage
frequently dissuades patients and physicians from use of anticoagulants in
conditions for which such agents are beneficial. Major hemorrhage in patients
receiving warfarin is strongly associated with the intensity of anticoagulation.2- 5 In particular,
the risk of intracranial hemorrhage, the most feared complication of anticoagulation,
increases dramatically at international normalized ratio (INR) levels greater
Despite this dramatic relationship between excessive anticoagulation
and risk of hemorrhage, information on causes of very high INR levels in usual
clinical practice is scarce. We conducted a prospective case-control study
to identify factors associated with an INR greater than 6.0 among outpatients
taking warfarin whose target INR was 2.0 to 3.0. In this report, we quantify
the effects of several commonly encountered risk factors, focusing on the
impact of acetaminophen ingestion.
Study patients were identified from among the outpatients managed by
our anticoagulant therapy unit from April 24, 1995, through March 1, 1996.
To be eligible for the study, patients had to be taking warfarin for more
than 1 month, have a target INR of 2.0 to 3.0, and be able to participate
in a telephone interview or designate an appropriate proxy. Study patients
had to be interviewed within 24 hours of their reported prothrombin time test
to enhance the accuracy of patient responses.
Study subjects were identified from the daily log of INR tests. At any
given time during the study period, nearly 2000 patients (75%) in our anticoagulant
therapy unit had an INR target of 2.0 to 3.0. We planned to recruit 100 case
patients and 200 controls to provide more than 85% power to detect a true
odds ratio of 3.0 or greater for risk factors having a prevalence of at least
10%, using P=.05 as the threshold for statistical
Case patients were identified as those having an INR greater than 6.0,
reported within 24 hours of the blood draw, and whose INR target was 2.0 to
3.0. The INR results greater than 6.0 are routinely verified by a duplicate
test. We focused on INR values greater than 6.0, because such levels convey
markedly increased risk of major hemorrhage2- 7
and are unlikely to be the result of usual intraindividual fluctuations. Of
111 patients identified, 96 (87%) were eligible to participate. The others
were excluded for the following reasons: 3 did not speak English and had no
appropriate proxy, 3 were admitted to the hospital, and 9 could not be reached
by telephone within the designated 24-hour time period. Of the 96 patients
who were eligible, 93 (97%) were interviewed. The other 3 patients refused
Control patients were randomly selected from among all the anticoagulant
therapy unit patients with a target INR of 2.0 to 3.0 whose actual INR value
recorded in the daily log was between 1.7 and 3.3. This is the range of INR
values that in our anticoagulant unit did not trigger a dose change. We sought
to enroll twice as many controls as cases during each week of the study. Of
279 randomly selected control patients, 216 were eligible to participate.
The others were excluded for the following reasons: 10 did not speak English,
3 were too cognitively impaired to participate meaningfully in the interview,
1 was severely hearing impaired, 1 was admitted to the hospital, and 48 patients
could not be reached by telephone within the 24-hour window after 3 attempts.
Of the 216 patients who were eligible, 196 (91%) agreed to be interviewed.
Two trained interviewers contacted patients by telephone, explained
the purpose of the study, obtained informed consent, and conducted a structured
interview from a previously piloted instrument. The script introduced the
study as follows:
We're calling some of our patients within a day of their blood
test to ask some questions about things like medications and diet. This is
part of a quality check on our procedures in an effort to improve our patient
We're calling some of our patients within a day of their blood
test to ask some questions about things like medications and diet. This is
part of a quality check on our procedures in an effort to improve our patient
There was no specific mention of out-of-range INRs. The interview lasted
approximately 10 to 15 minutes. Interviews were periodically observed to ensure
uniformity in data collection. Because of concerns about the risk of bleeding,
uniform blinding of the patients to their INR prior to the study interview
was not ethically feasible. Forty-eight case patients (52%) had been notified
of their INR value by clinical staff, unrelated to the study, before the study
interview. Our findings were essentially the same for both the blinded and
After consenting to participate, patients were asked to list the medications
they were currently taking and to identify any that were newly prescribed
within the preceding 4 weeks and the date treatment with the new medication
was started. If patients were unsure, the interviewer instructed the patients
to read directly from their medication list or pill bottle labels. Patients
were also asked about changes in medications or in dose of their medications
and use of over-the-counter medications during the previous 4 weeks. All medications,
prescription and over-the-counter, were classified according to their reported
effect on warfarin's metabolism, ie, as potentiators, inhibitors, or having
no effect.9- 11
To increase the reliability of patient responses, 4 different questions regarding
use of prescription and nonprescription medications were asked at different
points of the interview. For 21 case patients (22%) and 33 controls (17%),
a caretaker was identified as responsible for the patient's medications. Such
caretakers were then interviewed.
Patients were specifically asked to quantify their use of acetaminophen,
aspirin, other nonsteroidal anti-inflammatory drugs, and any other pain medications
during the 7 days preceding their blood test. Analgesic drugs were referred
to by their common brand names. If patients were unsure of the strength or
the specific acetaminophen preparation, the interviewer instructed the patient
to read the label of the pill bottle. Text describing the duration of acetaminophen
use was recorded on the interview form when available. Four case patients
and 1 control were unable to retrieve the specific acetaminophen dose; for
these patients, the lowest strength of acetaminophen was assigned. Our calculation
of acetaminophen use included acetaminophen contained in over-the-counter
cold and sinus formulations and in combination analgesics that included narcotics.
The weekly acetaminophen dose was categorized according to the following scheme:
1 to fewer than 7 regular-strength (325-mg) tablets per week, 7 to fewer than
14 tablets per week, 14 to fewer than 28 tablets per week, and 28 tablets
or more per week.
Vitamin K intake was qualitatively assessed with 2 screening questions.
The first question asked about gross changes in diet over the past week, ie,
eating more, less, or the same as usual. The second question asked patients
to contrast their usual consumption of 12 specific foods high in vitamin K
content12 (avocado, broccoli, brussels sprouts,
cabbage, chickpeas, green peas, green tea, kale, lettuce, liver, spinach,
and greens [eg, beet, collard, dandelion, mustard, turnip]) with the intake
of these same items during the week preceding their prothrombin time test.
A value of 1 was assigned for each food item consumed, with the sum of such
values providing aggregate scores for the number of items from this list consumed
routinely and the number consumed in the prior week. Further quantification
of each item was coded as more, less, or the same as usual. Alcohol intake
was categorized as none, 1 to 2 drinks per week, 1 drink every other day to
2 drinks per day, more than 2 drinks per day, or binge. Information on intake
of vitamin K–rich foods was missing for 6 cases and 3 controls. Information
on alcohol use was missing for 1 case patient and 2 controls.
Other questions asked about diarrheal or febrile illness during the
previous week and recent hospitalization. We categorized patients as having
an advanced malignancy if they listed chemotherapeutic agents among their
medications or if pain related to cancer was the indication for analgesic
use. For such patients, metastatic disease or extensive local invasion was
subsequently confirmed by pathology or operative reports.
Compliance with warfarin was assessed with a series of questions about
missed or extra doses of warfarin and by patient report of the warfarin doses
actually taken on each day of the week preceding the prothrombin time test.
The patient-reported doses were compared with the anticoagulant unit–prescribed
doses. Other questions concerned age, sex, and race.
Data extracted directly from the anticoagulant unit database included
length of warfarin therapy, indication for anticoagulation, assigned warfarin
dose, and INR value immediately preceding the 4-week study period. The INR
preceding the study period was measured an average of 35 days prior to the
study INR. This INR measurement was not necessarily the patient's last INR
test prior to the study INR.
The study protocol was approved by the institutional review board at
Massachusetts General Hospital.
For univariate comparisons between patients with an INR greater than
6.0 and the controls, we assessed statistical significance using χ2 tests and the Fisher exact test, where appropriate, for categorical
data and the Student t test for continuous variables.
Odds ratios with 95% confidence intervals (CIs) were calculated by standard
methods.13,14 The test of trend
was done using the Cochran-Mantel-Haenszel test.15
The Kolmogorov-Smirnov test was used to test the equality of the continuous
skew distributions for acetaminophen intake in the groups of patients who
either knew or did not know their INR value prior to the study interview.16
Long-term variability in anticoagulant control was assessed by the coefficient
of variation of INR level after adjustment for changes in warfarin dose. For
each patient, adjustment for dose was accomplished using a continuous autoregressive
time series model of order 1 for the logarithm of INR level.17
All available INR values dating back to August 1993 were entered into the
variability model for each case and control.
Logistic regression models were used to assess the independent effect
of multiple clinical characteristics on the risk of having an INR greater
than 6.0 and the significance of interaction terms. Statistical analyses were
performed with SAS (SAS Institute, Cary, NC), S-plus (Statistical Science,
MathSoft, Seattle, Wash), and StatXact (Cytel Software, Cambridge, Mass) software.
We interviewed 93 case patients with an INR of greater than 6.0 and
196 controls with an INR in the range of 1.7 to 3.3 within 24 hours of their
reported prothrombin time test. In both the case and control groups the mean
age was 70 years, about half were female, and nearly all were white. The conditions
prompting anticoagulation also did not differ between groups, with atrial
fibrillation as the leading indication. Seventy-eight cases (84%) and 162
controls (83%) had been taking warfarin for longer than 3 months (Table 1).
The mean INR for the cases was 8.3 (range, 6.1-29.8) and 2.4 (range,
1.7-3.3, which was an eligibility criterion) for the controls. For most cases,
their excessively high INR value represented a recent deterioration in control
of anticoagulation. The mean INR value for the prothrombin time test obtained
immediately prior to the 4-week study period was 2.5, with 73% of these values
in the acceptable range of INR of 1.7 to 3.3. These measures were very similar
to the controls' values (mean INR immediately prior to the 4-week study period
was 2.4 with 74% of values between 1.7 and 3.3) (Figure 1). The weekly warfarin dose did not differ for cases (mean,
28.9 mg) vs controls (mean, 26.7 mg) (P=.22). Three
cases (3%) and 9 controls (5%) had had their warfarin dose increased within
the 2 weeks prior to the study INR. The coefficient of variation used to assess
long-term INR variability did not differ significantly between cases and controls.
Twenty cases (22%) reported recently starting to take a medication known
to potentiate warfarin's anticoagulant response compared with 7 controls (4%).
These potentiating medications were antibiotics in 12 of the 20 cases and
in 2 of the 7 controls (Table 1).
Case patients were taking a mean of 5.1 prescription medications other than
warfarin vs 4.3 for controls (P=.01).
Fifty-two cases (56%) and 70 controls (36%) reported taking acetaminophen
in the week preceding the study INR (Table
1) . These included 11 cases (12%) and 6 controls (3%) who reported
taking an acetaminophen-narcotic combination preparation, mostly codeine and
oxycodone formulations. Fever was the reason for taking acetaminophen in only
7 case patients and 4 controls. More controls than cases reported taking aspirin,
but the percentage was small in both groups. Use of other nonsteroidal anti-inflammatory
drugs was uncommon and did not differ between case and control groups.
The number of vitamin K–rich foods consumed in the past week was
inversely correlated with an INR greater than 6.0 (P
for trend, .001) (Table 1). Sixty-one
cases (66%) compared with 79 controls (40%) reported routinely consuming only
1 or none of the 12 listed foods high in vitamin K content. Usual consumption
of these items did not differ significantly from that during the week preceding
the prothrombin time test (for cases, 1.6 vs 1.4, respectively; for controls,
2.2 vs 2.0, respectively). An overall decrease in oral intake during the previous
7 days was also strongly associated with having an INR greater than 6.0.
Patients were asked to quantify their usual alcohol intake and their
intake over the preceding 7 days. Seventy-five cases (81%) reported never
or rarely drinking alcohol compared with 121 controls (62%). Alcohol consumption
was inversely correlated with the risk of having an INR greater than 6.0 (P for trend, .002) (Table 1). For most cases and controls, the amount of alcohol consumed
during the week prior to the prothrombin time test was the same as usual.
Too few patients reported drinking more than 2 drinks per day to meaningfully
assess the effect of this amount of alcohol intake.
Twenty cases (22%) compared with 15 controls (8%) reported having had
diarrhea during the week preceding their prothrombin time test. Few patients
had had a recent fever, but fever was more commonly reported by case patients.
Recent hospitalization was unusual but was more common among the cases, 7
cases (8%) vs 1 control (1%). Nine cases (10%) and 2 controls (1%) had an
advanced malignancy (Table 1).
Eleven cases (12%) and 5 controls (3%) reported taking more warfarin
than the anticoagulant unit had prescribed (Table 1). For 6 of these 11 case patients, the higher dose of warfarin
they were taking had been prescribed by a treating physician who had not coordinated
care with the anticoagulant unit.
Case patients were more likely than controls to ingest acetaminophen,
and they ingested greater amounts than controls. Among those patients reporting
taking acetaminophen, the mean weekly dose was 6756 mg for the cases (approximately
21 tablets per week of the regular-strength dose of 325 mg) compared with
2938 mg for the controls (approximately 9 tablets per week) (P<.001). Beginning at the equivalent of 7 regular-strength tablets
per week, compared with no intake of acetaminophen, there was a dose-dependent
monotonic increase in the adjusted relative odds for having an INR greater
than 6.0, increasing from 3.5 (95% CI, 1.2-10.0) for 2275-mg to 4549-mg intake
of acetaminophen per week, to 6.9 (95% CI, 2.2-21.9) for 4550-mg to 9099-mg
intake per week, to 10.0 (95% CI, 2.6-37.9) for 9100-mg intake or more per
week (Table 2). These large effects
were seen at standard doses of acetaminophen. Indeed, only 5 cases reported
a cumulative weekly dose exceeding 18.2 g (or 56 regular-strength tablets)
per week corresponding to the maximum recommended daily dose of 2.6 g.10 Acetaminophen intake of the 48 cases who knew their
INR prior to the study interview (median weekly intake, 4175 mg; mean, 7304
mg) did not differ from that of the 45 cases who did not know their INR at
the time of the interview (median weekly intake, 3125 mg; mean, 5109 mg) (P=.37).
To help define the temporal relationship between acetaminophen and its
potentiation of warfarin's effect, we examined the duration of acetaminophen
use among those patients taking the doses that conferred the greatest risk.
Twenty-three case patients and 12 controls reported taking more than 4550
mg of acetaminophen during the 7-day period preceding their prothrombin time
test. Information on duration of acetaminophen use was recorded for 19 of
the 23 case patients and for 8 of the 12 controls. All of the cases and controls
had started taking acetaminophen since their previous (prestudy period) INR.
The predominant indication for acetaminophen was analgesia for a variety of
conditions causing acute pain. Fifteen of these 19 case patients compared
with 4 of the 8 controls reported taking acetaminophen for longer than 1 week.
Of these 4 controls, 1 had had an intervening INR of 5.9 prompting a decrease
in warfarin dose.
Independently significant determinants of INR values greater than 6.0
included the effect of advanced malignancy, decreased oral intake, diarrheal
illness, newly started treatment with potentiating medication, and taking
more warfarin than the dose prescribed by the anticoagulant unit (Table 3). Subjects with higher vitamin
K consumption and habitual moderate consumption of alcohol, 1 drink every
other day to 2 drinks per day, were less likely to have an INR greater than
6.0. Recent hospitalization, fever, and number of prescription medications
were not independent risk factors in the multivariate analyses. The independent
variables had no significant 2-way interactions.
We focused on the determinants of INR values greater than 6.0 because
such values represent a marked deviation from control and are high enough
to clearly increase the risk of major hemorrhage.2- 7
The INR values greater than 6.0 are unlikely to result from intraindividual
fluctuation in warfarin response. We limited our study to patients whose target
INR was 2.0 to 3.0, since including patients with different INR targets would
confound our results.
Our study design enabled us to quantify the real-world use of acetaminophen,
a widely available over-the-counter medication, among a predominantly older
population taking warfarin. Because the risk of life-threatening hemorrhage
increases with age,4,6,7
intervention studies of acetaminophen's effect on warfarin in this older population
would not be feasible.
The highly significant dose-response relationship between acetaminophen
and warfarin's effect was independent of other factors thought to alter response
to anticoagulation. For patients who reported taking the equivalent of at
least 4 regular-strength (325-mg) tablets per day for longer than 1 week,
the odds of having an INR greater than 6.0 were increased 10-fold above those
taking no acetaminophen. Risk decreased with lower intakes of acetaminophen,
reaching the background level of risk at an intake of 6 or fewer 325-mg tablets
Several case reports attest to the potentiation of warfarin by acetaminophen.18,19 Older clinical experiments also support
such an effect.20,21 In a double-blind
crossover study, Rubin et al22 assigned 15
healthy male volunteers on a stable dose of warfarin to 2 weeks each of acetaminophen,
4 g/d and placebo. This dose of acetaminophen increased the prothrombin time
more than 1.75 times the control in 7 of 15 subjects compared with 1 in the
placebo group, and produced an increase to more than 2.0 times the control
in 5 of the 15 subjects compared with none in the placebo group. The potentiating
effect was initially detected after 7 days of acetaminophen intake and peaked
at 12.5 days.22 While both warfarin and acetaminophen
are eliminated primarily through hepatic metabolism, the actual mechanism
by which acetaminophen potentiates warfarin is not well described.
We also assessed the impact of other clinical features on the risk of
excessive anticoagulation. We found a powerful and highly significant relationship
between having an INR greater than 6.0 and advanced malignancy. Possible mechanisms
include a direct effect on coagulation proteins, treatment-related hepatic
effects, or a potentiating interaction with warfarin. Case patients were also
more likely to have recently started taking a medication known to potentiate
warfarin's anticoagulant response. The underlying mechanisms of interaction
with warfarin have been previously described.9,11,23
Of the 14 case patients who had recently started taking an antibiotic, only
2 patients were prescribed an antibiotic that did not have a reported potentiating
effect on warfarin. This observation highlights the need to closely monitor
patients taking medications that interfere with warfarin or to choose alternative
therapies, if possible.
We confirmed a highly significant inverse relationship between stable
consumption of vitamin K–rich foods and having an INR greater than 6.0.
The interaction between dietary vitamin K and warfarin is well established.24- 27 Although
our assessment of vitamin K ingestion was crude, the significance of this
finding supports the concept28 that individuals
with diets rich in vitamin K are less sensitive to fluctuations in warfarin's
anticoagulant effect. Decreased oral intake in general was also associated
with excessive anticoagulation. The mechanism for this is not clear. It may
involve an aggregate decrease in other foods containing lesser amounts of
vitamin K or may be secondary to other mechanisms such as decreased binding
of warfarin to plasma proteins in the fasting state.29,30
We also observed a nearly 4-fold increase in risk of having an INR greater
than 6.0 following a diarrheal illness. In a recent case report, an acute
diarrheal illness was thought to cause excessive anticoagulation through malabsorption
of vitamin K.31
Long-term alcohol use has been reported to increase the metabolism of
warfarin, likely due to ethanol-induced stimulation of hepatic microsomal
enzymes.32 Such an effect was not seen in an
experiment involving healthy young male volunteers.33
Our data suggest that habitual light to moderate alcohol intake in our predominantly
older patient population does inhibit warfarin's anticoagulant response.
Case-control studies must address concerns about bias. The primary concern
of selection bias is minimized by nesting the case-control study within a
known population.34 To this end we identified
all potential case and control subjects during the study period and enrolled
the vast majority of all of the cases and a random sample of controls. Although
participation rates of the cases and controls were slightly different, it
is unlikely that this small difference could explain the dose response or
the 10-fold relative odds conferred by the highest acetaminophen doses. It
is also unlikely that patients at higher baseline risk for poor anticoagulant
control comprised the case population using acetaminophen because there was
no significant difference between cases and controls in long-term INR variability
or in their previous INR measurement. Our results are necessarily based on
subjects' self-report. We sought to increase the accuracy of this information
by interviewing patients within 24 hours of their prothrombin time test and
by restricting the retrospective periods of interest to either 7 or 30 days.
Although 52% of case subjects had been notified of their INR before the study
interview, raising the possibility of a recall bias, our findings were the
same when analyses were restricted to case patients who did not know their
INR level prior to the study interview. Error in patients' reports could not
explain our findings since such misclassification tends to obscure significant
associations and bias toward observing no effect. We minimized interviewer
bias by using a scripted interview and by observing interviews periodically
to ensure consistency in questioning subjects. A strong potentiating effect
of acetaminophen was not anticipated prior to the study, further reducing
the likelihood of interviewer bias.
The association we observed between acetaminophen and excessive INR
values has several important features of a causal relationship.35
It is strong, dose-dependent, and biologically plausible, follows an appropriate
temporal sequence, and is supported by previous case reports and small-scale
clinical experiments. In addition, the relationship persisted after control
for a variety of potential confounders.
Our data suggest that acetaminophen is an underrecognized cause of anticoagulant
instability. Of note, several pharmacology references either do not mention
an interaction between acetaminophen and warfarin or diminish its clinical
Acetaminophen is generally preferred over nonsteroidal anti-inflammatory agents
as an analgesic and antipyretic agent because it does not inhibit platelet
function or raise the risk of injury to the gastric mucosa.40- 42
The experience of our control subjects indicates that 36% of all patients
taking warfarin in our anticoagulant therapy unit also take acetaminophen
at any given time. Thirty percent of all INR values greater than 6.0 in our
anticoagulation unit population are attributable to intake of acetaminophen
equivalent to 7 or more tablets per week.43
Acetaminophen remains a valuable therapy for patients taking warfarin.
However, our findings should encourage clinicians to query patients about
their use of acetaminophen and to clarify its indication. Those patients taking
warfarin who also require sustained high doses of acetaminophen need close
monitoring of their INR levels.
The current study identifies powerful risk factors for excessive anticoagulation
in real-world practice. Knowledge of such risk factors should result in changes
in the management of anticoagulant therapy that will reduce hemorrhagic complications.
In turn, reduced risk of hemorrhage should encourage more widespread use of
warfarin among the many patients who would benefit from long-term anticoagulation.44