Lowering the Intensity of Oral Anticoagulant Therapy: Effects on the Risk of Hemorrhage and Thromboembolism

Background  Oral anticoagulation is effective in the prevention of arterial thromboembolism. The major drawback of coumarin therapy is the increased risk of hemorrhage. Implementing the optimal intensity of oral anticoagulation, ie, the level at which thromboembolic events are prevented without introducing an excessive bleeding risk, is an important step to improve the safety of oral anticoagulant therapy.

Methods  We observed all patients of the Leiden Anticoagulation Clinic who were treated because of a mechanical heart valve or atrial fibrillation between January 1, 1995, and January 1, 1998, or because of cerebral ischemia between January 1, 1994, and January 1, 1998. In 1996, at the halfway point of follow-up, the target intensity for patients with a mechanical heart valve was lowered from 4.0 (range, 3.6-4.8) to 3.5 (range, 3.0-4.0) international normalized ratio, and for atrial fibrillation and cerebral ischemia, from 3.5 (range, 3.0-4.5) to 3.0 (range, 2.5-3.5) international normalized ratio. We compared in-cidence rates of hemorrhage and thromboembolism before and after the introduction of lower target intensities.

Results  Higher target treatments were given to 2341 patients (2863 patient-years) and lower target treatments to 2256 patients (2260 patient-years). After introduction of the lower target ranges, the overall incidence rate of major untoward events declined from 5.7 (95% confidence interval [CI], 4.9-6.7) to 3.6 (95% CI, 2.8-4.4) per 100 patient-years. The incidence of major bleeding fell from 3.6 (95% CI, 2.9-4.4) to 2.7 (95% CI, 2.1-3.5) and the incidence of major thromboembolism from 2.0 (95% CI, 1.5-2.5) to 0.8 (95% CI, 0.5-1.3) per 100 patient-years.

Conclusion  Implementation of lower target intensities for coumarin therapy decreased the complication risk.

Over the last decades, the efficacy of oral anticoagulant treatment with coumarin derivatives for the prevention of arterial thromboembolism has been demonstrated for several indications.¹

The major adverse effect of coumarin therapy is hemorrhage, which is associated with patient characteristics and the intensity of anticoagulation.² The incidence of bleeding events increases with higher intensities of anticoagulation, whereas the risk of thromboembolism rises at lower intensities.³ Determining the optimal intensity of anticoagulation to balance the thromboembolic and bleeding risk is therefore crucial to improve the overall effectiveness of treatment with coumarins.

In attempts to find this optimum, randomized trials have been conducted in which 2 intensities of oral anticoagulation were compared.^4,5 In 1993, an alternative method was presented to determine the optimal intensity level in observational studies.⁶ This method subsequently has been applied to various indications,^3,7,8 including determination of which international normalized ratio (INR) was associated with the lowest incidence of events. The implicit reasoning was that a target INR at this intensity would minimize the number of untoward events. Although this reasoning is plausible, it has not been validated, which is the objective of the present analysis.

In 1996, the Dutch target ranges for the prevention of arterial thrombosis were adjusted according to the findings in 2 studies on the optimal level of oral anticoagulation in patients with a mechanical heart valve³ or atrial fibrillation and recent cerebral ischemia.⁷ This offered an opportunity to directly assess the effects of introducing a better target intensity in coumarin therapy.

The change of target INRs occurred during a prospective follow-up study⁹ on thromboembolic and hemorrhagic events in patients anticoagulated for artificial heart valves, atrial fibrillation, and postmyocardial infarction. This offered the opportunity to compare the incidence rates of hemorrhage and thromboembolism under 2 anticoagulant regimens in an ongoing prospective study based within routine practice.

The Leiden Anticoagulation Clinic is a regional clinic in the Netherlands. All patients of the clinic who were treated with coumarins for a mechanical heart valve, atrial fibrillation, or cerebral ischemia between January 1, 1995, and January 1, 1998, were eligible for the study. For patients with cerebral ischemia, inclusion started 1 year earlier, on January 1, 1994. Patients who were already taking coumarin therapy at the start of follow-up and patients starting treatment were enrolled. There were no exclusion criteria.

In the Netherlands, the target intensities of oral anticoagulant therapy for the prevention of arterial thrombosis were adjusted in 1996. The target intensity for patients with a mechanical heart valve was lowered from 4.0 (range, 3.6-4.8) to 3.5 (range, 3.0-4.0) INR, whereas the target for atrial fibrillation and cerebral ischemia was shifted from 3.5 (range, 3.0-4.5) to 3.0 (range, 2.5-3.5) INR. This resulted in patients being assigned to 2 target intensities over time, which were the subject of comparison in this study. Because the implementation of a new target range takes some time to assess the effect, we chose a washout period of 1 month after adjustment of the target intensities to minimize carryover effects; events and patient-years of this period were excluded.

Patients visited the anticoagulation clinic regularly or were visited at home, at mean intervals of 3 weeks. During each visit, a trained nurse filled out a short medical questionnaire addressing intercurrent diseases, changes in medication, the occurrence of bleeding and thromboembolic events, and hospital admissions. An antecubital blood sample was taken to determine the prothrombin time (expressed in INR).

The following information was collected: date of birth, sex, indication and duration of treatment, dates and results of routine INR measurements, hospital admissions, and death. In addition, we collected discharge letters for all registered hospital admissions, with results of x-rays, computed tomographic scans, laboratory tests, and autopsy reports, when relevant.

Outcome events were major hemorrhage and thromboembolism, whichever event occurred first under each anticoagulant intensity regimen. Diagnostic criteria for major events were adopted from the Leiden Artificial Valve and Anticoagulation study³ and have been applied previously.¹⁰

Major bleeding events were extracranial or intracranial. Extracranial hemorrhage was defined as acute blood loss, internal or external, leading to death or to hospital admission for treatment or observation. Admissions for diagnostic purposes only or hemorrhages that occurred while the patient was admitted for another indication were not considered outcome events. Intracranial and spinal hemorrhage were defined as acute or subacute events of neurologic impairment, with hemorrhage proven by computed tomographic scan, surgery, or autopsy.

Thromboembolism consisted of cerebral infarction, myocardial infarction, and peripheral arterial embolism. Cerebral infarction was defined as an acute neurological deficit, proven by computed tomographic scan or autopsy. The diagnosis of myocardial infarction required 2 or more of the following: history of chest discomfort, typical rise of specific cardiac enzymes, or the development of new Q waves on the electrocardiogram. Peripheral arterial embolism was defined as sudden peripheral ischemia, proven by duplex scanning, angiography, surgery, or autopsy.

For all possible outcome events, discharge letters were reviewed by an expert panel. The panel members were at all times blinded for the intensity of oral anticoagulation at the time of the event. Moreover, they were unaware that the applied target intensity was a study item.

The principal aim of the trial was to compare the incidences of outcome events between the different anticoagulant intensity strategies on an intention-to-treat basis. The incidence rates of events within each patient group and their 95% confidence intervals (CIs) were derived by standard calculations, assuming a Poisson distribution of the number of events. We compared the occurrence of outcome events in the 2 treatment groups in terms of hazard ratios, which were obtained from Cox proportional hazards models.

The study group consisted of 483 patients with mechanical heart valves, comprising 1116 patient-years of follow-up. During the study, 457 hospital admissions occurred (41 per 100 patients per year). Full clinical information was available in more than 99% of these admissions. Between January 1, 1995, and May 19, 1996, 414 patients (523 patient-years) were treated with a target intensity of 4.0 (range, 3.6-4.8) INR. After the washout period, 440 patients (593 patient-years) received coumarin therapy with a target of 3.5 (range, 3.0-4.0) INR until the end of follow-up on January 1, 1998. Characteristics of both groups are presented in Table 1. The distribution of sex and age was similar for both target intensities.

During follow-up, 2111 patients who were treated for atrial fibrillation at the Leiden Anticoagulation Clinic were enrolled. Clinical information could be obtained in 98% of 1650 hospital admissions. Between the start of follow-up on January 1, 1995, and August 31, 1996, the target intensity was 3.5 (range, 3.0-4.5) INR (1631 patients and 1899 patient-years). Afterward, when a target of 3.0 (range, 2.5-3.5) INR was applied, 1586 patients were studied for 1479 patient-years. The 2 groups did not differ on relevant characteristics (Table 1).

Among the 356 patients with anticoagulant treatment for cerebral ischemia, 214 hospital admissions were reported. Clinical information could be retrieved in 99%. Between January 1, 1994, and October 27, 1996, when a target of 3.5 (range, 3.0-4.5) INR was applied, 296 patients (441 patient-years) were treated. Afterward, until the end of follow-up on January 1, 1998, 230 patients (188 patient-years) were treated with a target of 3.0 (range, 2.5-3.5) INR. The main characteristics of the 2 patient groups were again similar (Table 1).

Lowering of the target levels led to the intended lowering of the achieved levels of oral anticoagulation in daily practice, as illustrated in Figure 1. The mean achieved intensity of anticoagulation dropped from 3.9 to 3.5 INR in patients with mechanical heart valves, from 3.4 to 3.1 INR in those with atrial fibrillation, and from 3.5 to 3.2 INR in patients treated because of cerebral ischemia.

Under the more intense anticoagulation regimens, the follow-up time spent within the target range was 71%, with 21% of time spent below the target range vs 8% above. After introduction of the lower target intensities, the time in range fell to 58%. Fourteen percent of the follow-up time was spent below the target range. Twenty-eight percent was spent above the target range, mainly within 0.5 INR from the upper INR limit (17%).

The change led to a reduction of high-intensity INRs, with nearly 60% less time spent above 5.0 INR (116 vs 48 patient-years), while the volume of low INRs (<2.0) remained about the same (43 vs 53 patient-years).

Table 2 presents the incidence rates of major bleeding and thromboembolism for the 3 patient groups during the 2 anticoagulant intensity regimens.

The overall incidence rates of major untoward events were 5.7 (95% CI, 4.9-6.7) per 100 patient-years in the higher target group and 3.6 (95% CI, 2.8-4.4) per 100 patient-years in the lower target group, corresponding with a hazard ratio of 0.6 (95% CI, 0.5-0.9).

The incidence of major bleeding events fell from 3.6 (95% CI, 2.9-4.4) to 2.7 (95% CI, 2.1-3.5) per 100 patient-years under the milder anticoagulation intensity, yielding a hazard ratio of 0.7 (95% CI, 0.5-1.0). A reduction of intracerebral hemorrhage, of which the incidence decreased from 0.9 (95% CI, 0.6-1.3) to 0.5 (95% CI, 0.2-0.8) per 100 patient-years, contributed most to the lowering of major hemorrhage.

The incidence of major thromboembolism also declined after adjustment of the target INR levels, from 2.0 (95% CI, 1.5-2.5) to 0.8 (95% CI, 0.5-1.3) per 100 patient-years (hazard rate, 0.5; 95% CI, 0.3-0.8). When we limited the thromboembolic end points to cerebral ischemia and peripheral thrombosis, like in the major atrial fibrillation trials, the hazard rate became 0.6 (95% CI, 0.3-1.4).

We analyzed separately those patients who started treatment during our study, which included 687 patients (426 patient-years), when the higher target ranges were applied and 580 patients (266 patient-years) when the lower target ranges were applied. For both intensity regimens, the incidence of hemorrhage and thromboembolism was considerably higher in this group of patients who newly started treatment than among patients who already received treatment before the start of the study. The effect of a lower target range was also more pronounced in these patients. In the higher intensity group, the overall incidence of major untoward events was 9.9 (95% CI, 7.0-13.1) compared with 4.9 (95% CI, 2.6-8.0) per 100 patient-years in the lower intensity group. The incidence of major bleeding events fell from 6.1 (95% CI, 3.9-8.7) to 3.0 (95% CI, 1.3-5.5) per 100 patient-years, and the incidence of major thromboembolism declined from 3.8 (95% CI, 2.1-5.9) to 1.9 (95% CI, 0.6-3.9).

To assess how the effect of changing the intensity ranges evolved over time, we drew Kaplan-Meier curves for the cumulative hazard (Figure 2). At the lower target intensity, the hazard rate was at all times lower than for the higher target intensity, and both incidence rates were approximately constant over time.

The introduction of lower target intensities for treatment with oral anticoagulants for 3 arterial indications provided the opportunity to study the effect of the intensity of oral anticoagulation on the incidence of untoward events in a routine clinical setting. After an observational study⁹ aimed at finding the optimal intensity of anticoagulation and subsequent lowering of the target intensity, we compared new target INRs for patients with mechanical heart valves (INR, 3.0-4.0)³ and atrial fibrillation and cerebral ischemia (INR, 2.5-3.5)⁷ with the formerly applied targets (INR, 3.6-4.8 and 3.0-4.5, respectively) on an intention-to-treat basis. For all 3 indications, the incidence rates of bleeding and thromboembolic events declined after the introduction of the optimal target intensity. The overall 40% decrease of severe complications was impressive.

Some limitations in the study design arose from the unintentional character of the study. We did not assign the target intensities to the participants, and the different intensity levels were not applied simultaneously. Moreover, some of the patients experienced the change in target intensity and, hence, participated in both study groups. On the other hand, the change affected all patients eligible for treatment, irrespective of any patient characteristics, and the indications for oral anticoagulation did not change. Nevertheless, because the groups were not studied concurrently, we cannot exclude the possibility of changes in medical behavior, although it seems unlikely that these would account for the large effects we observed in a short period.

After the change in target ranges, the mean achieved intensity of anticoagulation decreased within all 3 study groups by 0.3 to 0.4 INR. Before the lower target intensities were applied, 71% of the follow-up time was spent within the target range, compared with 58% thereafter. The explanation for this apparently declined achievement is that the new target ranges were not only lowered but also narrowed, presuming a 0.5-INR lowered but equally wide target range yielded more than 70% of follow-up time spent within the target range. However, the observed success of 58% under the narrowed targets is in agreement with results of other studies.^3,11

As predicted on theoretical grounds, the incidence rate of major hemorrhage declined after implementation of the lower target intensities in each of the studied patient groups. The overall achieved result, a relative risk of 0.7 (95% CI, 0.5-1.0), ie, a risk reduction of 30%, was even better than the expected 11% bleeding reduction based on a bleeding risk index.¹²

In the overall analysis, this beneficial effect was mainly due to the prevention of intracranial hemorrhages, whereas the incidence of extracranial hemorrhage decreased only slightly. The lowered intracranial bleeding rate was only observed in patients treated because of atrial fibrillation or previous cerebral ischemia, whose main difference from those treated because of a mechanical heart valve was older age. Older persons are known to have an increased bleeding risk,^13,14 probably due to age-dependent comorbidity, and may therefore benefit most from a less intensive anticoagulant treatment. Overall, the reduction in bleeding risk was substantial and concerned the most severe type of hemorrhage.

Surprisingly, a decline in thromboembolic risk was also observed. When we excluded myocardial infarction as a thromboembolic end point, based on the argument that patients with atrial fibrillation or a mechanical heart valve do not receive oral anticoagulants to prevent myocardial infarction, this finding persisted. Because the downward trend was shown for all 3 indications and for the distinct thromboembolic events, it seems unlikely that this finding could be explained by chance. We investigated several possible explanations for this finding.

The target intensities of the Dutch anticoagulation clinics were altered according to the optimal intensities found in 2 retrospective cohort studies.^3,7 For patients with atrial fibrillation and for those treated after cerebral ischemia, the new targets were based on a population with atrial fibrillation and cerebral ischemia. Although the optimal intensity for patients with cerebral ischemia was confirmed later,¹⁰ the optimum for patients with atrial fibrillation has yet to be determined.

As patients are treated with coumarins for longer periods, those susceptible to thromboembolism and bleeding may be responsible for a higher incidence of untoward events in the earlier phase of treatment and thereby for the lower incidence thereafter. Because many patients in our trial were participants in both treatment periods, it might be argued that attrition of susceptible patients led to the decreased incidence of outcome events in the latter period. Because we included all patients treated in the 2 periods, this is an unlikely explanation. Nevertheless, to completely exclude such an effect, we did an analysis on patients who were new to coumarin therapy under each anticoagulant regimen. Again, we found a considerable decline of the thromboembolic and bleeding risk after introduction of the lower target ranges, indicating that our results could not be explained by a selection of less susceptible patients. We confirmed previous findings of a higher risk of complications in patients starting coumarin treatment.¹⁰ Furthermore, we calculated hazard rates over time for the 2 anticoagulant regimens to exclude decreasing hazard with time. The incidence rates were approximately constant and always lower for the lower intensity, indicative of a change in risk immediately after the change in the target intensities.

The decreased incidence of thromboembolism could also be explained if some bleeding events had consequently been misclassified as thromboembolic, leading to an apparent reduction of "thromboembolism" after lowering of the target levels. However, this possibility is unlikely given the strict definitions of untoward events and the simultaneous decrease of cerebral ischemia and myocardial infarction.

Use of a lower target intensity might be more comfortable to prescribing physicians, as patients are less likely to be overdosed to a dangerous extent. Hence, drastic downward changes in anticoagulant treatment resulting in nonprotective intensities and a high thromboembolic risk would less frequently occur. However, while the lower target range led to a decreased amount of INRs greater than 5.0, underanticoagulation (INR, <2.0) did not materially change. A paradoxical explanation is that the decreased incidence of thromboembolism resulted from the decline of higher intensities. No plausible physiological mechanism can explain this phenomenon, although the observation has been reported before.⁷ It might be that patients with subtherapeutic anticoagulant intensities were treated more aggressively under the more intense anticoagulant regimens, leading to an overanticoagulated state at the time of the thromboembolic event, which was actually caused by the preceding low intensity. However, we did not find any evidence for this concept in our data set.

Implementation of the optimal intensity of oral anticoagulation as the target for coumarin therapy decreases the complication risk. The decline of bleeding events was expected, but no satisfying explanation for the decreased thromboembolic risk was found. The outcome of our study nevertheless supports the superiority of the newly introduced lower levels of oral anticoagulant treatment and serves as a clinical validation of our method to determine the optimal intensity of anticoagulation. Further research to determine better target intensities for all current indications of oral anticoagulation is therefore worthwhile and recommended.

Corresponding author and reprints: Frits R. Rosendaal, MD, PhD, Department of Clinical Epidemiology, Leiden University Medical Center, C9-P, PO Box 9600, 2300 RC Leiden, the Netherlands (e-mail: F.R.Rosendaal@lumc.nl).

Accepted for publication June 20, 2003.

This study was supported by grants 94.001 from the Dutch Thrombosis Foundation, 28-2542 from The Prevention Fund, and 96.114 from the Netherlands Heart Foundation, all located in Den Haag.

We thank the members of the expert panel: Ward L. E. M. Bollen, MD, PhD, Ernst E. van der Wall, MD, PhD, and Bea C. Tanis, MD, PhD. Furthermore, we thank Janneke Reehuis-Doornbos for her secretarial and data management work.