Risk Factors for Complications of Chronic Anticoagulation: A Multicenter Study

Methods

Setting and Patients

The five clinics in the consortium were selected to provide a mix of geographic locations, practice settings, and patient populations (Table 1). All practitioners had 1 or more years of experience in managing patients taking anticoagulant therapy.

Medical records were abstracted for all patients who were then receiving anticoagulant therapy and for those who had discontinued anticoagulation within the previous 18 to 24 months. We included the latter patients to avoid oversampling long-term survivors and those receiving long-term anticoagulation. When patients had received multiple courses of warfarin (that is, patients had extended periods in which they did not take warfarin), data for all courses were abstracted. All patients were eligible for inclusion except those who had received warfarin for 6 weeks or less.

Data Collection

Trained abstractors, using standard forms extensively tested for reliability, reviewed the inpatient and outpatient records of all patients. In a subsample of 50 charts, intra- and inter-rater agreement exceeded 95% for all items. The study coordinator co-reviewed at least 10% of charts at all sites and cross-checked clinic enrollment logs to ensure that all eligible patients were identified. Seven patients whose charts could not be located were excluded.

Indications for anticoagulation were divided into 7 main categories and 29 subcategories. For patients with more than one indication, the more serious problem (the one requiring a higher intensity or a longer duration of therapy) was deemed primary.

We reviewed all records from inpatient admissions and from the anticoagulation clinic, other medical and surgical clinics (excluding the psychiatry clinic), emergency departments, and walk-in clinics. If the anticoagulation practitioner maintained personal records apart from the formal medical record, we reviewed these as well. At the three Veterans Affairs medical centers, data were obtained from the hospital computer system.

During the period of study, only one of the medical centers reported results using the international normalized ratio (INR). Because values for the international sensitivity index (ISI) of the thromboplastins used at some of the participating centers did not become available until 1988, we analyzed all results using the prothrombin time ratio (PTR). All laboratories used standard North American thromboplastins, and providers in the clinics typically adhered to therapeutic recommendations initially published by the American College of Chest Physicians in 1986 [10, 11]. In general, the providers set the PTR target range at 1.5 to 1.8 (high intensity) for patients with mechanical valves and at 1.3 to 1.5 for most others (low intensity).

Outpatient visits were reviewed to determine the reason for the appointment, the occurrence of any intercurrent illnesses, blood pressure, and all medications prescribed in addition to warfarin, including the date each medication was started and stopped [12]. Using the medical records, we coded up to eight active surgical or medical conditions other than the indication or indications for anticoagulation. We recorded practitioners' comments about patients' medication compliance, cigarette smoking, and pattern of alcohol consumption.

Classification of Outcomes

We used a detailed scheme to classify complications as minor (no associated costs or medical consequences), serious [requiring treatment or medical evaluation], life-threatening, or fatal. Minor complications required no additional testing, referrals, or outpatient visits but were remarkable enough to report to the provider. Examples of minor bleeding included mild nosebleeds, bruising, mild hemorrhoidal bleeding, and microscopic hematuria. Examples of serious bleeding included overt gastrointestinal bleeding, occult gastrointestinal bleeding if endoscopic or radiographic studies were done, gross hematuria that prompted cystoscopy or intravenous urography or lasted more than 2 days, and hemoptysis. If blood was transfused, two units or less were given. We defined life-threatening bleeding as that leading to cardiopulmonary arrest, to surgical or angiographic intervention, or to irreversible sequelae such as myocardial infarction, neurologic deficit consequent to intracerebral hemorrhage, or massive hemothorax. Bleeding was also considered to be life-threatening if it resulted in two of the following consequences: 1) loss of three or more units of blood; 2) systolic hypotension [<90 mm Hg]; or 3) critical anemia (hematocrit of 0.20 or less). Fatal bleeding was defined as that which led directly to the patient's death.

Thromboembolic complications were also classified as minor, serious, life-threatening, or fatal. For example, mild superficial thrombophlebitis was considered a minor event. Serious thromboembolic events included transient ischemic attacks or suspected stroke, recurrent deep venous thrombosis, and pulmonary embolism without respiratory or hemodynamic compromise. Life-threatening events included massive pulmonary embolism, stroke with residual neurologic deficit, and systemic embolism.

All serious, life-threatening, and fatal complications were independently reviewed by a physician-investigator at the local site and three investigators at the coordinating center. Using standardized criteria, we determined whether deaths were related to warfarin-associated bleeding or to a thromboembolic complication. Disagreements were reconciled by discussion.

Analysis

Using conditional maximum likelihood estimates of relative risk, we evaluated potential risk factors by comparing the incidence of first-time complications between groups with and without each factor, stratifying as appropriate with the Mantel-Haenszel procedure [13]. We carried out survival analyses to compute the incidence of first-time bleeding among patients with and without each variable that was significant in the first analysis [14, 15]. Unless otherwise noted, data were censored after the first bleeding complication or after the cessation of warfarin therapy.

All variables with P values ≤ 0.1 by the log-rank test were entered in a stepwise fashion into a set of multivariate Cox regression models that controlled for clinic site [16]. A separate regression model was constructed using the same technique but with the addition of calendar month as a covariate. A two-sided P value of 0.05 was considered significant.

To evaluate the risk for recurrent complications, we carried out a similar set of survival and Cox regression analyses that were restricted to patients who had an initial bleeding complication.

We calculated the incidence of first-time bleeding and thromboembolic complications at four different PTR ranges (<1.3, 1.3 to 1.49, 1.5 to 2.0, and >2.0) by dividing the number of events occurring in patients with PTRs in each range by the total number of patient-weeks accumulated for that range. The time that elapsed between two PTR determinations was always attributed to the later value.

To analyze the effect of variability in a patient's PTR on the incidence of complications, we computed a variance growth rate reflecting the degree to which a patient's PTR deviated from his or her target PTR over a prolonged time interval. The mathematical derivation of this coefficient is based on the assumptions that the patient's PTR varies continuously over time in response to many independent perturbations and that the variance of the change in PTR increases as the interval between measurements increases [17]. The variance growth rate is a measure of the time-weighted variance of the PTR about the patient's target PTR over the interval of treatment and is calculated by the formula: Equation 1

where σ² is the variance growth rate, n is the number of all PTR measurements before the event (occurrence of a complication, cessation of warfarin, or end of the study), and τ is the duration (in weeks) since the last PTR determination (had to be 1 week or longer). Only patients who received anticoagulation for 12 weeks or more were included in univariate analyses of the variance growth rate, because this was the minimum duration of therapy required to obtain a reliable estimate of this variable. Because the target PTR was not always noted in the medical records before the publication of the American College of Chest Physicians standards in 1986 [10], we computed an approximation of the desired target for each patient. This target value was based on a discounted average of a patient's serial PTR values using an initial estimate derived from the average PTR for all patients receiving anticoagulation for the same indication at the same site.

Results

The study sample included 928 patients who received 1103 distinct courses of therapy and made 33 962 clinic visits (see Table 1). We excluded 115 courses of therapy that were administered at nonparticipating institutions. The mean duration of follow-up was 1.9 years, ranging from 6 months to 2.8 years among the five clinics. The shortest course of therapy was 1 week and the longest was 22 years. Fifty-four percent of the courses of therapy exceeded 6 months in duration (Figure 1), and 66% were initiated after 1985. On average, patients had PTR determinations every 3.3 weeks, ranging from 2.5 to 4.8 weeks among the five clinics.

The 928 patients included 750 men and 178 women. The mean age (±SD) at the start of therapy was 57.3 ± 13.7 years. The primary indication for anticoagulation was deep venous thrombosis or pulmonary embolism in 274 patients (30%), valvular heart disease or cardiac prosthesis in 239 patients (26%), atrial fibrillation in 130 patients (14%), cerebral or systemic embolism in 196 patients (21%), other circulatory conditions in 64 patients (7%), and other causes in 25 patients (3%).

In the 1950 patient-years of follow-up, 510 first-time bleeding events occurred (Table 2). The incidence of first bleeding episodes according to our severity classification was as follows: minor bleeding, 17.3 events/100 patient-years (337 first events); serious bleeding, 7.5 events/100 patient-years (147 events); life-threatening bleeding, 1.1 events/100 patient-years (22 events); and fatal bleeding, 0.2 events/100 patient-years (4 cerebral hemorrhages). When all 1332 bleeding events (including multiple events in the same patient) were considered, rates of minor and serious bleeding were substantially higher (Table 2).

The cumulative incidence of fatal bleeding among all patients was 1% at 1 year and 2% at 3 years, after which no additional fatalities occurred. The cumulative incidences of first episodes of life-threatening and serious bleeding at 1, 2, 4, and 8 years of follow-up were 1%, 2%, 5%, and 9% and 12%, 20%, 28%, and 40%, respectively (Figure 2). More than one half of the serious and life-threatening bleeding episodes originated in the gastrointestinal tract (Table 3).

Of the 156 patients who had one bleeding event that was serious or life-threatening, 50 (32%) had a second event; in 31 of these 50 patients (62%), the second event occurred within 1 year of the initial event (Figure 3).

Risk Factors for Bleeding

Of the many variables examined, four demonstrated an independent relation to the risk for a first episode of bleeding. These four variables were a mean PTR of 2.0 or greater, a shorter duration of anticoagulation, a greater variability in the PTR, and the presence of three or more comorbid conditions.

The cumulative incidence of first-time serious bleeding was significantly lower among patients with a mean PTR of less than 2.0 when compared with patients who had a mean PTR of 2.0 or more (P = 0.003) (Figure 4). The incidence of serious bleeding when the mean PTR was 2.0 or more was 26 events per 100 patient-years as compared with 11 events per 100 patient-years when the PTR was in the range of 1.3 to 1.5 (relative risk, 3.0 [95% CI, 1.9 to 4.7]; P < 0.001) (Table 4). A similar trend was observed for fatal and life-threatening bleeding combined (relative risk, 3.0 [CI, 1.4 to 6.2]) (P = 0.002). The cumulative incidence of serious bleeding at a PTR of 1.3 to 1.5 was identical to that at a PTR of 1.5 to 2.0.

An initial hemorrhagic complication was far more likely during the early phases of treatment. During the first 3 months of therapy, serious and life-threatening bleeding occurred at rates of 21 and 1.8 episodes per 100 patient-years, respectively. After adjustments were made for intensity of therapy, the relative risks for serious bleeding during the first 3 months of treatment compared with the rest of the first year, the second year, and anytime thereafter were 1.9 (CI, 1.3 to 3.0), 3.0 (CI, 1.8 to 4.8), and 5.9 (CI, 3.8 to 9.3), respectively. Relative risks for life-threatening bleeding, computed in a similar fashion, were 2.6, 3.4, and 1.7, respectively. Because of the relatively small number of events (22 total), 95% CIs for these latter estimates of relative risk all overlapped 1.0.

When we divided patients into tertiles with low (sigma < 0.12), moderate (sigma = 0.12 to 0.20), and high (sigma > 0.2) variability, the incidence of serious bleeding was significantly related to the PTR variance growth rate (Figure 5). The relative risk for serious bleeding among patients in the highest tertile compared with those in the lowest was 1.6 (CI, 1.2 to 2.7; P = 0.003).

We did not detect a higher rate of initial bleeding complications among patients with any one underlying illness such as hypertension, diabetes, or peptic ulcer disease. However, patients with three or more chronic conditions, other than the indication or indications for anticoagulation, were significantly more likely to have a bleeding complication than those with fewer than three conditions (relative risk, 1.4 [CI, 1.1 to 2.5]; P = 0.0008). This finding could not be attributed to concomitant use of drugs known to interfere with warfarin metabolism because the use of such drugs was infrequent.

When entered into a multivariate Cox regression model that controlled for clinic site, all four variables found to be significant in the univariate model (intensity of anticoagulation, time on therapy, the PTR variance growth rate, and presence of three or more comorbid conditions) remained significant (Table 5). When added to the model, calendar time was highly significant, indicating that changes over time in treatment and patient factors had important effects; however, all four of the other risk factors remained significant as well.

We did not find a higher incidence of initial bleeding among patients with a record of alcohol abuse. Among 140 patients with a history of alcoholic binge drinking, however, the relative risk for serious bleeding, after adjustment for intensity of treatment, was 1.3 when they were compared with patients who had no record of alcohol abuse (CI, 0.8 to 1.9; P = 0.15). Compared with all other patients, binge drinkers older than 65 years showed a nonsignificant trend toward more frequent serious bleeding (relative risk, 1.7; CI, 0.8 to 3.9).

The relative risk for a first time serious bleeding event was 1.9 (CI, 1.3 to 3.0) times greater in women than in men after adjustment was made for intensity of treatment. Further investigation revealed that the excess in bleeding events among women was almost exclusively due to vaginal or uterine bleeding and occurred mainly at one of the two (non-Veterans Affairs) clinics with sizable female populations.

In multiple analyses of the relation between age and the incidence of an initial hemorrhage, age was not a significant independent risk factor. Other factors that were not significantly related to the risk for bleeding were race, indication for anticoagulation, history of stroke, history of atrial fibrillation, elevated blood pressure, any particular type of intercurrent illness, or use of potentially interfering medications. In addition, no increase in bleeding was observed among patients taking aspirin or aspirin-containing medications, but these were seldom used.

Among 156 patients who had either an initial serious or life-threatening bleeding episode, the cumulative proportion with recurrent bleeding at 3 years was 60% in those whose first event had a gastrointestinal source and 33% in those whose initial bleeding originated at another site (P = 0.05). No other variable helped to distinguish those at risk for recurrent bleeding.

Risk Factors for Thromboembolism

There were 147 thromboembolic events, one of which was fatal. Twenty of these 147 events were categorized as life-threatening and 115 as serious, yielding rates of 1.0 and 5.9 per 100 patient-years, respectively. Of the variables analyzed, only intensity of therapy was significantly related to the risk for an initial thromboembolic event; the rate of thromboembolism was 16 per 100 patient-years at a PTR of less than 1.3 and 4.7 per 100 patient-years at a PTR of 1.3 to 1.5 (relative risk, 3.6; CI, 2.1 to 6.4) (see Table 4). When the interval between courses of anticoagulant therapy [that is, those periods when patients were not receiving warfarin] was included in the analysis, the incidence of thromboembolism at a PTR of less than 1.3 was 71.2 events per 100 patient-years (relative risk, 11.5; CI, 7.5 to 17.7).

Discussion

Our study provided estimates of the incidence of bleeding and thromboembolic complications among unselected patients taking warfarin who were followed in anticoagulation clinics. Because our study involved a large and heterogeneous patient sample, our results can be generalized to clinical practice settings more readily than those from many clinical trials.

Risk for Bleeding

The overall rate of fatal bleeding in our study was 0.2 events per 100 patient-years, which is similar to the rates (0.06 to 2.6) observed in studies of anticoagulation for valvular heart disease or after myocardial infarction [6, 18-27]. Our rates of 1.1 and 1.6 per 100 patient-years for first-time bleeding and any life-threatening bleeding, respectively, are lower than the rates of 1.7 to 3.7 observed in recent studies of anticoagulation in patients with atrial fibrillation [1-5]. On the other hand, these rates are higher than those of 0.4 to 0.9 found in several trials involving patients who had valvular heart disease or had had a myocardial infarction [6, 18-21, 23]. These comparison studies may not, however, provide appropriate benchmarks because all dealt with patients who received anticoagulant therapy for a specific indication, and several excluded many patients who would have been treated with warfarin in clinical practice.

Our rates of fatal and serious bleeding were substantially lower than those seen in studies in which patients received anticoagulation routinely for various indications [28-33] (see Table 6). There are at least two probable reasons for these discrepancies. First, all patients in our study were followed by experienced practitioners in clinics dedicated to anticoagulation. The organization of care delivery was similar to ours in only one previous study [29]. Second, most of our patients were treated after 1986, when lower-intensity regimens of warfarin were popularized. Moreover, estimates from our study may be more accurate because only one of these comparison studies had a larger sample size [32] and only two involved more than one institution [28, 32].

Our findings are strongly supported by a recent prospective study in which the incidence of bleeding was similar to that found in our retrospective study [34]. That prospective trial, conducted in the same clinics during 1990 to 1991, found incidences of first-time life-threatening, serious, and minor bleeding of 0.4, 5.7, and 17 events per 100 patient-years, respectively; the corresponding rates in this retrospective study were 1.1, 7.5 and 17.5.

Even though we applied a much more liberal definition of “major” bleeding in our study, our rates for first-time serious bleeding were similar to those reported by other investigators. We deliberately categorized many events as serious that would have been considered minor using the classification schemes of other investigators. For example, we classified a hemorrhage in a patient who received a two-unit transfusion as serious, but this would have been categorized as “minor” or even ignored in other studies [30, 35]. We adopted a patient-oriented perspective, in which a blood transfusion would probably not be perceived as trivial, and considered any bleeding that required diagnostic or therapeutic intervention to be serious. Although epistaxis, occult gastrointestinal bleeding, or hematuria are not life-threatening events, having to endure the expense, inconvenience, discomfort, and possible risk of a posterior nasal packing, barium enema, or excretory urogram might well be viewed as a serious matter by the patient. Our approach is supported by the study of Lancaster and associates [36], who found that bleeding complications, even those relatively minor in nature, significantly impair patients' perceptions of their own health and well-being [36].

Risk Factors for Bleeding

We examined several factors suspected of raising the risk for warfarin-associated bleeding. Like many previous investigators, we found that a mean PTR of more than 2.0 was associated with a significantly greater risk for hemorrhage [37]. During the earliest years of this retrospective study, the laboratories at several of the participating clinics did not use thromboplastins with a known ISI. They did, however, use standard North American thromboplastins with an estimated ISI value of 2.2 to 2.4 [11]. Using these ISI values, a PTR of 2.0 would be equivalent to an INR of 4.6 to 5.3, which is above the upper bounds of what is presently considered more intense anticoagulation.

The second major factor influencing the risk for initial bleeding was how recently warfarin therapy had been started. The risk was highest during the first 3 months of therapy and gradually diminished in the subsequent 2 or 3 years, a trend that has been noted in other studies [28, 32, 33]. At the start of therapy, dosages are adjusted and PTR values are often erratic. Moreover, patients with an unrecognized predisposition toward bleeding are likely to manifest it early during therapy. The decline in the rate of bleeding over time probably reflects the withdrawal of high-risk patients who had bleeding and the improved management of others as they achieved a steady state.

A third factor, not studied by others, that was significantly related to bleeding was variability in the PTR. Reasons for an unstable PTR include variations in drug metabolism, medication compliance, diet, use of other drugs, or fluctuations in a coexisting illness such as heart failure. This variability may also reflect actions taken by the health care provider (for example, frequent dosage changes that interfere with attainment of a steady state and exaggerate underlying random biologic variability). We found that patients who had more than four dosage adjustments per year bled 25% more often than patients who had fewer adjustments (P = 0.004).

Patients who had experienced one serious or life-threatening hemorrhage were at a remarkably high risk for recurrence. As has been noted by others, those whose first bleed was from a gastrointestinal source were particularly at risk for a second event [28].

An important aspect of our study is that several commonly cited risk factors for bleeding, such as advanced age, were not found to be significant [28, 31, 38]. Older patients did have significantly higher prothrombin times for a given dose of warfarin than younger patients, as recently reported by Gurwitz and colleagues [39], but they had no excess bleeding complications [39]. The findings of our study are consistent with those of investigators who did not detect a significant relation between age and the incidence of bleeding during warfarin therapy [30, 32, 33, 37]. This absence of an association with age is also consistent with the extremely low rates of bleeding seen in the mostly elderly participants of recent controlled trials of warfarin in chronic atrial fibrillation [1-3, 5]. The discrepancy among studies regarding age may simply reflect the fact that most of the data showing an association with bleeding during warfarin therapy were collected before modern monitoring standards were available.

Hypertension is commonly cited as a risk factor for intracranial bleeding [28, 31, 32, 37]. We did not observe this association in our cohort. In part, the absence of such an association may have been due to the fact that we were able to obtain blood pressure readings for only 10 686 of the 33 962 outpatient visits reviewed, leading us to overlook patients who were intermittently hypertensive. Alternatively, practitioners were aware of the risks and either avoided prescribing anticoagulant therapy for poorly controlled hypertensive patients or were careful to ensure good blood pressure control.

Some investigators have found that patients with cerebrovascular disease who receive anticoagulation to prevent stroke have a higher incidence of intracranial bleeding [28, 37], but others have not observed this relation [30, 33]. Controlling for intensity and duration of treatment, we found no increase in bleeding complications related to any specific indication for anticoagulation, including cerebrovascular disease. This finding is not consistent with studies that have suggested that patients with deep venous thrombosis or pulmonary embolism [37] or atrial fibrillation [28] are at a higher risk for bleeding. In the former case, the increased risk for bleeding may simply reflect the fact that most patients with deep venous thrombosis or pulmonary embolism are treated for 3 to 6 months (that is, during the period of highest risk for hemorrhage).

Although we did not find that a history of alcohol abuse or poor compliance conferred a higher risk for treatment complications, data on these two variables may have been recorded unreliably in the medical record. The trend we saw toward a higher rate of bleeding among elderly binge drinkers is certainly plausible.

Contrary to our initial hypothesis, the use of medications that interfere with the metabolism of warfarin, such as certain antibiotics or antiepileptic agents, did not lead to a higher incidence of bleeding. Clinicians in participating clinics used such drugs infrequently and carefully followed the patients requiring them. Surprisingly, patients with a history of peptic ulcer disease did not bleed more often than patients without ulcer disease as reported by others [32]. We surmise that this was a result of careful patient selection and recent widespread use of potent histamine-2 blockers.

We also found no clinically significant seasonal variation in PTR values or in the incidence of complications, which suggests that increased consumption of cruciferous vegetables, rich in vitamin K, during the spring and summer, had little effect on control of the PTR.

Risk for Thromboembolism

We observed a remarkably high risk for thromboembolic events when a patient's mean PTR was less than 1.3. This was the case for both venous and arterial events. Unexpectedly, we found no difference between the risk for thromboembolism at a PTR in the range of 1.3 to 1.5 and that at a PTR between 1.5 and 2.0. These two ranges roughly correspond to those currently designated as “low” and “high” intensity in guidelines from the American College of Chest Physicians [10]. We wonder whether practitioners strove to maintain the PTR in the high end of the low-intensity range and in the low end of the high-intensity range, minimizing the differences in both rates of bleeding and thromboembolism that might have occurred.

Limitations

We took every precaution to minimize bias by extensively testing all data-collection tools and carefully checking all data for reliability and validity. We only carried out statistical tests that were selected before the data were analyzed to reduce further the potential for bias and problems of repeated testing. Even so, our study had several limitations that deserve comment. First, the investigation was retrospective, dependent on medical records that were occasionally incomplete or missing documentation of important items. Data on PTR or INR values or warfarin dosages were rarely incomplete, but missing data frequently posed a problem for assessments of patient compliance or alcohol intake. A second related problem was that we classified the severity of complications using data from medical records that could not always be verified for accuracy and that may have been missing useful clinical details. For example, computed tomography was not always done immediately after the development of neurologic symptoms.

A third potential liability was that four of the participating clinics did not report INR values during the first several years studied. Because the ISI values for the thromboplastin reagents used at these sites often were unknown, especially during the earlier years of study, PTR values from the five clinics may not always have been comparable (Table 6). Although all these centers currently report INR values, using PTR results only was common practice during this period. To address this issue directly, we contacted each laboratory to determine whether the ISI values of the reagents used may have been known even though results were being reported in terms of the PTR. In several instances, this information was available, and ISI values ranged between 2.0 and 2.4. We presumed that the thromboplastins used at other periods had similar ISI values. This same assumption was made by the Committee on Antithrombotic Therapy of the American College of Chest Physicians in setting forth guidelines for the management of anticoagulated patients based on the PTR. Unfortunately, this assumption may not have always been a safe one. In a survey of 190 hospital laboratories, Bussey and colleagues [40] found that 33% used thromboplastins with ISI values of 2.61 to 2.80 and another 19% used unlabeled reagents.

Finally, despite our relatively large sample size, the power of subgroup analyses was limited, and such results should be applied cautiously in patient management.

Clinical Implications

Based on our findings and those of others, it should be possible to reduce the risk for complications by attending to modifiable risk factors, such as intensity of treatment and PTR variability, and by considering nonmodifiable risk factors, such as comorbidity, when deciding to initiate anticoagulation. We found that the incidence of bleeding and thromboembolic complications remained approximately constant throughout a PTR range of 1.3 to 2.0 but increased sharply when the PTR was greater or less than these limits. Because variability in the PTR was an important risk factor, a patient with a highly erratic PTR should be maintained well within his or her assigned target range to avoid unexpected periods of inadequate or excessive anticoagulation.

No specific diagnostic or age group of patients seems to be at an inordinately high risk for bleeding. Older persons may require lower doses of warfarin but do not necessarily need more frequent monitoring. Patients should be monitored carefully during the first few months of therapy when the risk for bleeding is highest. As many as 50% of patients on long-term treatment may be expected to experience a serious hemorrhagic complication, but the risk for a fatal or life-threatening event is far lower. Patients who develop a hemorrhagic complication and continue to take warfarin are at high risk for a recurrence, especially if the site of initial bleeding is the gastrointestinal tract. Patients should be advised that minor bleeding is common.

Appendix

Members of the Warfarin Anticoagulation Follow-up Study Group at the time of this study included Catherine Callahan, MS; Stephan D. Fihn, MD, MPH; Jorja Henikoff, MS; Daniel L. Kent, MD; Esther Kohler, CRT; Mary McDonell, MS; Donald Martin, PhD; James Reiss; Nancy Roben, CRNP; and Domokos Vermes, PhD: Seattle Veterans Affairs Medical Center (Coordinating Center), Seattle, Washington; John Banas, MD; Mary Pasko, PharmD; and Marc Stern, MD: Buffalo Veterans Affairs Medical Center, Buffalo, New York; Rose Hutchinson, RN, and Richard H. White, MD: University of California, Davis, Medical Center, Sacramento, California; Robert W. Coleman, RPh, and Frederick Yee, RPh: Palo Alto Veterans Affairs Medical Center, Palo Alto, California; Daniel M. Becker, MD; Pam Buncher, RNP; Jane Spencer Bopp, RNP; Linda Krongaard-DeMong, RNP; and Frederick Walker III, MD: University of Virginia Medical Center, Charlottesville, Virginia.