Paradoxically, many studies have found that elderly persons are at higher risk for hemorrhagic complications than are younger patients [8-17]. It has been suggested that older patients are more prone to bleeding because they metabolize warfarin more slowly [18-20]; they have an elevated risk for drug interactions because of polypharmacy [21]; and they often have chronic illnesses, such as renal insufficiency [22], heart failure [9], cancer [10, 22], and cerebrovascular disease [23], that increase the risk for bleeding during warfarin therapy. On the other hand, numerous studies have not found elderly persons to be at greater risk for bleeding [24-31]. These discrepant findings have not been adequately explained.

Because of the clinical importance of this problem, we analyzed data from six sites to determine whether elderly patients who were receiving warfarin had a higher incidence of complications than similar younger patients who were also receiving warfarin.

Methods

Top Methods Results Discussion Author & Article Info References

We collected our data in two phases. The first phase involved a retrospective review of medical records from 1980 to 1990; other results from this review have been reported previously [30]. In the second phase, we analyzed data collected prospectively between 1990 and 1993, including the results from a multicenter, randomized trial of a computerized scheduling system for patients receiving warfarin [32].

Study Setting and Patients

Our study was done in the six anticoagulation clinics that make up the National Consortium of Anticoagulation Clinics. These clinics represent a mix of geographic locations, practice settings, and patient populations. Clinic personnel were responsible for managing anticoagulation therapy for all patients throughout the study, although in two clinics, personnel also provided primary medical care [33]. The sites were the University of California at Davis, Sacramento, California; Jefferson Medical College, Philadelphia, Pennsylvania; the University of Virginia, Charlottesville, Virginia; the Veterans Affairs Medical Center, Buffalo, New York; the Veterans Affairs Medical Center, Palo Alto, California; and the Veterans Affairs Medical Center, Seattle, Washington. Because Jefferson Medical College did not join the Consortium until 1992, data from that site were used only in the prospective phase of the study, which followed completion of the randomized trial of computerized scheduling.

Retrospective Data Collection

During the retrospective phase of our study, we abstracted the medical records of all patients who were currently receiving anticoagulation therapy and all patients whose anticoagulation therapy had been discontinued within the previous 18 to 24 months. If patients had received multiple courses of warfarin (that is, if there were extended periods during which a patient did not receive warfarin), we abstracted data from all courses. All patients were eligible for inclusion unless they had received warfarin for 6 weeks or less.

We have described our data collection methods in detail elsewhere [30]. Trained abstractors reviewed inpatient and outpatient records using standard forms that have been extensively tested for reliability. All records from inpatient admissions and visits to the anticoagulation clinic; other medical, surgical, and urgent care clinics (excluding visits made for psychiatric reasons); and emergency departments were abstracted. We also reviewed any records that were maintained separately from the formal medical record by a practitioner responsible for anticoagulation therapy. At each of the three Veterans Affairs medical centers, data were also retrieved from the hospital's information system. We excluded seven patients whose charts were missing.

Indications for anticoagulation therapy were organized into seven main categories and 29 subcategories. For patients who had more than one indication, the most serious problem (the one that required the greatest intensity or longest duration of therapy) was deemed the primary indication.

During the retrospective phase of our study, only one of the five medical centers reported results using the international normalized ratio. Because the international sensitivity index values of the thromboplastins used at some of the participating centers were unavailable before 1988, we analyzed all results using the prothrombin time ratio. However, we contacted each laboratory to determine whether the international sensitivity index values of the reagents used were known; this information was available from several of the study sites after 1985 and from all sites after 1990. All laboratories used standard North American thromboplastins that had international sensitivity index values ranging from 2.0 to 2.4. Practitioners typically adhered to therapeutic recommendations published by the American College of Chest Physicians (ACCP) in 1986 [34, 35], which set the target range as a prothrombin time ratio of 1.5 to 1.8 (high intensity) for patients with mechanical valves and 1.3 to 1.5 (low intensity) for most other patients.

Records from outpatient visits were reviewed to ascertain the reason for the appointment, the occurrence of any intercurrent illnesses, all medications prescribed in addition to warfarin, and the dates on which therapy with each medication was started and stopped [36].

Prospective Study

The prospective data collection was done over a 3-year period (1990 to 1993). During the first year of this period, we did a randomized trial of a computerized scheduling intervention [32]. Because the scheduling system produced no statistically significant differences in control of anticoagulation or frequency of complications, we have included data collected during that 1-year period in our present analysis. During the 2 years after completion of the trial, we continued to collect identical information on patients who had participated in the trial and on all eligible patients subsequently enrolled into participating clinics for the management of warfarin therapy. Jefferson Medical College did not participate in the trial of computerized scheduling.

Patients who were actively enrolled in one of the six participating clinics or who were newly referred to one of the clinics were eligible for this portion of the study if the total planned duration of their anticoagulation therapy was 6 weeks or longer. Before the start of prospective data collection, we abstracted the medical records of all active eligible patients in each clinic. These data were collected according to the same protocol used in the retrospective study. At the three clinics located in university medical centers, our trial was exempted from requirements for verbal or written informed consent by the local institutional review boards, and all eligible patients were enrolled. At the three Veterans Affairs clinics, the local review boards required informed consent, and we invited the participation of all eligible patients in person or by mail; 5.2% of patients in these clinics declined to participate.

At each visit to each anticoagulation clinic during the prospective phase of the study, all data, including those on the prothrombin time ratio (or international normalized ratio) results, warfarin dosage, complications, and follow-up plans, were entered directly into a notebook computer.

Classification of Outcomes

In both the retrospective and prospective phases of data collection, we used the same detailed scheme to classify bleeding 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, 2 units or fewer were given. Life-threatening bleeding was defined as that leading to cardiopulmonary arrest, surgical or angiographic intervention, or 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 led to at least two of the following consequences: loss of 3 or more units of blood; systolic hypotension (systolic blood pressure less than 90 mm Hg); or critical anemia (hematocrit less than equals 0.20). Fatal bleeding was defined as that leading directly to the death of the patient.

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

Deviation in the Prothrombin Time Ratio

We previously described [37] a method that can be used to characterize the degree to which a patient's prothrombin time ratios deviate from his or her target prothrombin time ratio over time, and we have shown that the level of variability in the prothrombin time ratio is statistically significantly related to the risk for complications [37]. For this study, we calculated the deviation in the prothrombin time ratio over time () using a minor modification of the method we previously reported: Equation 1 In this formula, n is the number of measurements of prothrombin time before a terminating event (such as complication, cessation of warfarin therapy, or end of study) and is the time since the previous measurement of prothrombin time in weeks. If more than one prothrombin time ratio had been determined during a 1-week period, we used the last one.

(1)

Analysis

Data from the retrospective and prospective phases of the study were analyzed both separately and jointly. Only the first course of warfarin for each patient was included in the analysis. A total of 443 patients contributed follow-up time to both the retrospective and prospective phases of the study, and, to avoid counting them twice, we assigned these patients to the prospective cohort. No substantive differences in our results were seen when these 443 patients were combined with the retrospective cohort. The baseline characteristics of patients within cohorts were compared across age groups using one-way analysis of variance. For all analyses, we counted only time accumulated during the first course of warfarin therapy for each patient. In addition, except for the Poisson regression (see below), we counted only the first complication that each patient had in the severity classification of interest.

According to the method of Gurwitz and coworkers [20], we calculated a mean dose-adjusted prothrombin time ratio by dividing the mean prothrombin time ratio for the course of therapy by the mean daily warfarin dose. We compared the mean dose-adjusted prothrombin time ratio for five age strata (< 50 years, 50 to 59 years, 60 to 69 years, 70 to 79 years, and more than equals 80 years). We adapted the method of Gurwitz and coworkers by computing a value for each year of therapy for each patient and assigning it to the appropriate age group.

For each age category, we did survival analyses to compute the incidence of first-time bleeding [38, 39]. To calculate the unadjusted risk for bleeding according to age at the time of a bleeding event, we computed conditional maximum likelihood estimates using the Mantel-Haenszel procedure to compare incidence rates of first-time hemorrhagic events for the five age strata [40]. Data were censored after the first bleeding complication or after cessation of warfarin therapy. We computed exact 95% CIs. In all analyses, a two-sided P value of 0.05 was considered significant.

We then did multivariate Cox analyses (SPS, Inc., Chicago, Illinois) using age at the start of warfarin therapy to obtain conditional maximum likelihood estimates of the relative risk for bleeding after adjustment for the intensity of anticoagulation therapy and for deviation in the prothrombin time ratio, both of which we had previously found to be statistically significantly related to risk for bleeding [41]. We generated three series of Cox models in which time to first complication was the dependent variable. One set of models was constructed for each of three severity levels: minor, serious, and life-threatening and fatal combined. This approach was repeated for the retrospective, prospective, and combined cohorts. Age in years was initially forced into all models as an independent variable, along with intensity of anticoagulation therapy, presence of three or more comorbid conditions, duration of warfarin therapy, presence of hypertension, indication for anticoagulation therapy, and clinic site. Intensity of therapy was defined according to the indication for anticoagulation therapy according to ACCP recommendations [34, 35]. We have previously reported that, in addition to the intensity of anticoagulation therapy and the variability in prothrombin time ratio, the presence of three or more comorbid conditions is a statistically significant predictor of bleeding events [30]. We did not adjust for time-dependent covariates, such as use of drugs that can interfere with warfarin or the achieved intensity of anticoagulation therapy. The use of potentially interfering medications was not associated with the risk for bleeding in our previous studies [30]. The final Cox model consisted of age in years, the intensity of anticoagulation therapy, and the deviation in prothrombin time ratio. To determine whether there was an age threshold at which complications became more common, we constructed another set of Cox models with the same covariates, except that age was coded as a dummy variable in our five categories.

We also did Poisson regression analyses (STATA, Stata Corp., College Station, Texas) using age at the start of warfarin therapy to estimate the absolute risk for bleeding by means of the incident rate ratio, which is the exponential of the regression coefficient. The dependent variable was the total number of bleeding complications. Other investigators have used Poisson regression in these circumstances [13, 17, 42]. As in the Cox models, we adjusted for the intensity of anticoagulation therapy and the deviation in prothrombin time ratio. Time receiving therapy was treated as total exposure time. Unlike Cox models, Poisson models permit the analysis of multiple events occurring in the same person [43].

All aspects of the study were conducted without the involvement of the entities providing funding.

Results

Top Methods Results Discussion Author & Article Info References

Patient Characteristics and Anticoagulation Management

In the combined retrospective and prospective phases of our study, 2376 patients received 2631 courses of warfarin; 471 patients received warfarin solely during the retrospective phase (1980 to 1990). The mean age at the start of therapy was 58.3 years, and 73% of patients were men (Table 1). Compared with younger patients, elderly patients received anticoagulation therapy more often for atrial fibrillation and prevention of stroke and less often for deep venous thrombosis and pulmonary embolism. The average duration of follow-up was 81 weeks.

The mean warfarin dose required to maintain a given prothrombin time ratio decreased progressively with age. The mean warfarin dose was inversely related to age (P < 0.001), whereas the dose-adjusted prothrombin time ratio increased with age (P < 0.001; Table 1). To maintain a similar mean prothrombin time ratio, patients 80 years of age and older required a warfarin dose that was one third to one half the size of that given to patients younger than 50 years of age.

Although dosage requirements changed with age, warfarin therapy did not appear to be more difficult to control with increasing age in patients as old as age 80 years, as evidenced by minimal differences in the variability in prothrombin time ratio, the interval between visits to the anticoagulation clinic, or the frequency of changes in warfarin dosage (Table 1). The oldest patients, however ( 80 years of age), did show greater variability in prothrombin time ratio and had more frequent determinations of prothrombin time ratio and more frequent dosage adjustments.

Several statistically significant differences were seen between patients in the retrospective cohort and those in the prospective cohort, and most of these differences reflected changes in anticoagulation practice over the course of the study (Table 1). Compared with patients in the retrospective cohort, those in the prospective cohort were more likely to be given warfarin for atrial fibrillation and less likely to be given warfarin for valvular disease. The mean prothrombin time ratio was significantly higher across all age groups for patients in the retrospective phase who were treated for conditions requiring high- or low-intensity anticoagulation therapy than for patients in the prospective phase.

Incidence of Bleeding Complications

The unadjusted incidence rate of minor and serious bleeding complications did not increase significantly with increasing age at the time of the event (Table 2). For patients younger than 50 years of age, the rate for a first episode of minor bleeding was 18.0 per 100 patient-years, and the rate for a first episode of serious bleeding was 9.3 per 100 patient-years. For patients 80 years of age and older, the rates were 16.3 and 4.4, respectively. However, life-threatening or fatal bleeding complications occurred more often among the oldest patients. The incidence of these events was 0.75 per 100 patient-years in patients younger than 50 years of age and 3.38 in patients 80 years of age and older (relative risk, 4.50 [95% CI, 1.3 to 15.6]) (Table 2). Thus, although the overall risk for all types of bleeding episodes combined did not appear to increase with age, a life-threatening or fatal event was more than four times more likely to occur in a very elderly patient than in a patient younger than 50 years of age. This finding must be interpreted cautiously, however, given the small number of events that were seen (five) and the wide CIs.

To adjust for duration of therapy and the effects of factors other than age that may also be related to the incidence of bleeding complications, we did extensive multivariate analyses. In a series of Cox models, age at the start of warfarin therapy was entered as a continuous variable. In all of these analyses, comorbidity was not statistically significant after age had been forced into the models, nor was duration of warfarin therapy, presence of hypertension, reason for anticoagulation therapy, or clinic site. Thus, our final Cox models included age, intensity of anticoagulation therapy, and deviation of prothrombin time ratio Table 3 and Table 4.

The deviation in prothrombin time ratio () was the strongest risk factor for bleeding complications, with an adjusted relative risk ranging from 2.2 to 28.8 in the six final models. Intensity of therapy was statistically significantly related to the incidence of serious bleeding complications and of life-threatening and fatal bleeding complications combined but not to the incidence of minor bleeding complications. After we adjusted for the deviation in prothrombin time ratio and the intensity of anticoagulation therapy, age was unrelated to the risk for any type of bleeding complication in the retrospective data set Table 3 and Table 4. In the prospective and combined data sets, increasing age was statistically significantly associated with a slightly increased risk for minor bleeding complications (adjusted relative risk, 1.01 per year of age). Increasing age was not associated with the risk for serious bleeding complications in the combined data set (relative risk, 0.99 [95% CI, 0.98 to 1.00]). The risk for life-threatening and fatal events combined was not associated with increasing age in either data set. When age was analyzed as a categorical variable using the combined data set, no statistically significant relation with risk for bleeding was seen, with the exception of life-threatening or fatal bleeding among patients 80 years of age and older (Figure 1).

View larger version (26K): [in this window] [in a new window]   Figure 1. Adjusted relative risks for bleeding complications and 95% Cls according to age category. Relative risks were obtained by Cox regression using combined retrospective and prospective cohorts with adjustment for intensity of warfarin therapy and deviation in prothrombin time ratio.

To analyze the risk for multiple bleeding events and confirm the results of the Cox models, we created a series of Poisson regression models using the same independent variables used in the Cox models Table 3 and Table 4. As in the Cox analysis, a slight, albeit statistically significant, relation was seen between advancing age and risk for minor bleeding complications. A slight negative association was found between age and the risk for serious bleeding (incident rate ratio, 0.99 per year of age), whereas no important association was seen for life-threatening and fatal bleeding combined (incident rate ratio, 1.00 to 1.03).

The most common sites of serious bleeding were the genitourinary tract (accounting for 24% of all bleeding episodes), the gastrointestinal tract (accounting for 22%), and the skin and soft tissues (accounting for 19%). The gastrointestinal tract was by far the most common site for life-threatening or fatal bleeding; it was involved in 64% of all events. Intracranial bleeding was responsible for 20% of all life-threatening or fatal bleeding complications. Of the four fatal hemorrhages, all were intracranial and all occurred during the retrospective phase of the study (before 1990). No statistically significant differences were seen in the site of bleeding complications, including intracranial hemorrhages, among different age groups, although the relatively small number of events limited the statistical power of the analysis.

Discussion

Top Methods Results Discussion Author & Article Info References

In this large, long-term, multicenter study, we found that the overall incidence of any type of hemorrhagic complication was no higher in elderly patients receiving warfarin than in younger patients receiving warfarin. In addition, we saw no consistent trends across age groups or types of bleeding complications in bivariate or multivariate analyses.

However, when examined separately, patients 80 years of age and older appeared to have a risk for life-threatening or fatal bleeding that was three to four times higher than that of patients younger than 50 years of age. Although this result was seen in both univariate and multivariate analyses, it must be interpreted with great caution because it is based on only five events occurring in 93 patients. Moreover, the risk for minor or serious bleeding events and the overall risk for any bleeding event was no higher in patients 80 years of age and older than in younger patients. On the other hand, this finding is clinically plausible, given that older persons have a higher risk for intracranial bleeding [17, 23] and a limited physiologic reserve for responding to acute stress (such as gastrointestinal hemorrhage, which accounted for nearly two thirds of life-threatening or fatal bleeding episodes). Moreover, the incidence rate (3.4 fatal or life-threatening bleeding events per 100 patient-years) that we saw in persons 80 years of age and older is similar in magnitude to that found in other studies. For example, an incidence rate of 4.2 per 100 patient-years was reported by the Stroke Prevention in Atrial Fibrillation investigators for patients 75 years of age and older [14].

The lack of a strong and consistent association between age and warfarin-related bleeding is consistent with the results of eight other studies published since 1970 [24-31]. On the other hand, this finding is at variance with those of 10 other studies, which have reported a positive association between age and warfarin-related bleeding [8-17]. These discrepant findings remain unexplained, although it has been suggested they represent differences in patient populations [44]. The varied nature of the clinics participating in our study suggest that our findings are not restricted to a single type of patient, clinical setting, or style of practice. Furthermore, our mainly negative findings cannot readily be attributed to a lack of statistical power, because we included 2376 patients who contributed a total of 3702 years of follow-up.

Taken together with the results of other studies, our findings reinforce the view that the risk for warfarin-related bleeding complications is, on average, no higher in elderly patients than in younger patients receiving warfarin at the same intensity. Recent surveys [45-47] suggest that many patients in this age group who might benefit from warfarin are not receiving this drug. In many instances, the decision to not prescribe warfarin appears to be based on a physician's belief that the risk for complications in elderly persons is inordinately high [48]. Our results suggest that, when clinically warranted, warfarin therapy should not be withheld on the basis of age alone. Warfarin should only be prescribed when the expected benefit, in terms of decreased risk for thromboembolic events, clearly exceeds the risk for hemorrhagic complications. This is especially true for patients 80 years of age and older.

Like many investigators, we found that the warfarin dose needed to maintain a given prothrombin time ratio or international normalized ratio diminishes with age [18-20, 49]. Some investigators have interpreted such findings to mean that elderly patients are at greater risk for bleeding complications and require more intensive monitoring [20]. Our findings suggest that this is not uniformly true. We observed a remarkable consistency of the mean prothrombin time ratio in patients as old as 80 years of age. Moreover, variability in the prothrombin time ratio did not consistently increase with age. As we and others have previously reported [17, 26], we found that greater deviations in prothrombin time ratio were associated with a statistically significantly higher risk for bleeding complications. Moreover, up to the age of 80 years, elderly patients did not appear to be more difficult to manage; patients younger than 80 years of age did not require more frequent monitoring or dosage adjustments than patients who were younger still.

Our study had several methodologic strengths. Our patient sample was large and diverse, including patients of both sexes who represented a broad spectrum of ages and clinical characteristics. To minimize error, we extensively tested all data collection tools. We separately analyzed retrospectively and prospectively collected data to test for consistency of findings. Even so, however, our study had several limitations that deserve comment. First, during the retrospective phase, data collection relied on abstraction of medical records that occasionally were incomplete or lacked documentation. Data were rarely missing for prothrombin time ratios (or international normalized ratios) or warfarin dosages.

A second, related, problem was that we classified the severity of complications using the data taken from medical records, which could not always be verified for accuracy and may have lacked useful clinical details. For example, computed tomography was not always done immediately after neurologic symptoms occurred.

A third potential liability was our reliance on the prothrombin time ratio rather than the international normalized ratio; we used the former because our study spanned a 15-year period. Because the international sensitivity index values for the thromboplastin reagents used at these sites were sometimes unknown, especially during the earlier years of the study, prothrombin time ratio values from the six clinics may not always have been comparable. We contacted each laboratory and learned that international sensitivity indexes of reagents used were uniformly between 2.0 and 2.4.

Fourth, in both the retrospective and prospective studies, selection bias may have been caused by referral patterns or clinical decisions that kept high-risk elderly patients from being enrolled at the participating anticoagulation clinics. We think that this is unlikely, because the number of patients in all clinics increased substantially during the years studied, and most of the new referrals were elderly. Moreover, if this were the case, a similar bias must have existed at all participating clinics, because we detected no statistical differences in outcomes among elderly persons at any individual site.

Fifth, adjustments for potentially confounding factors, including the use of concomitant drugs such as aspirin and other nonsteroidal anti-inflammatory drugs, were limited because therapy with these drugs was started and stopped so often. Furthermore, we did not consider the dosages of these drugs in our analyses. The use of these agents may have played a role in increasing the risk for bleeding, a role that we did not detect. It is also possible that our data on comorbidity or other clinical factors were insufficient to allow us to fully adjust for potentially confounded effects.

In summary, we show that, when properly managed, elderly patients receiving warfarin appear to have no greater risk for hemorrhagic complications than do younger patients receiving warfarin to achieve standard levels of intensity. Moreover, despite their greater sensitivity to warfarin, elderly patients do not necessarily require closer follow-up or monitoring than younger patients. However, the risk for life-threatening or fatal bleeding, although rare, may be slightly greater as patients enter the eighth and ninth decades of life.

Appendix

The following are members of the National Consortium of Anticoagulation Clinics and other persons who contributed to this study:

Seattle Veterans Affairs Medical Center, Seattle, Washington (Coordinating Center): Catherine M. Callahan, MS Hyg; Stephan D. Fihn, MD, MPH; Jorja G. Henikoff, MS; Daniel L. Kent, MD; Esther Kohler, CRT; Nancy Roben, CRNP; Mary B. McDonell, MS; Donald C. Martin, PhD; James Reiss, MD; and Domokos Vermes, PhD.

Buffalo Veterans Affairs Medical Center (Buffalo, New York): John Banas, MD; Mary Pasko, PharmD; and Marc Stern, MD.

University of California at Davis Medical Center (Sacramento, California): Rose Hutchinson, RN, and Richard H. White, MD.

Palo Alto Veterans Affairs Medical Center (Palo Alto, California): Robert W. Coleman, RPh, and Frederick Yee, RPh.

University of Virginia Medical Center (Charlottesville, Virginia): Daniel M. Becker, MD; Pam Buncher, RNP; Jane Spencer-Bopp, RNP; Linda Krongaard-DeMong, RN; and Frederick Walker III, MD.

Jefferson Medical College (Philadelphia, Pennsylvania): Gino Merli, MD, and Deborah Fritz, RN.

Dr. White: Division of General Internal Medicine, Primary Care Center, Room 3120, 2221 Stockton Boulevard, Sacramento, CA 95817.