Age and the Risk of Warfarin-Associated Hemorrhage: The Anticoagulation and Risk Factors In Atrial Fibrillation Study

Abstract

OBJECTIVES

To assess whether older age is independently associated with hemorrhage risk in patients with atrial fibrillation, whether or not they are taking warfarin therapy.

DESIGN

Cohort study.

SETTING

Integrated healthcare delivery system.

PARTICIPANTS

Thirteen thousand five hundred fifty-nine adults with nonvalvular atrial fibrillation.

MEASUREMENTS

Patient data were collected from automated clinical and administrative databases using previously validated search algorithms. Medical charts were reviewed from patients hospitalized were for major hemorrhage (intracranial, fatal, requiring ≥2 units of transfused blood, or involving a critical anatomic site). Age was categorized into four categories (<60, 60–69, 70–79, and ≥80), and multivariable Poisson regression was used to assess whether major hemorrhage rates increased with age, stratified by warfarin use and adjusted for other clinical risk factors for hemorrhage.

RESULTS

A total of 170 major hemorrhages were identified during 15,300 person-years of warfarin therapy and 162 major hemorrhages during 15,530 person-years off warfarin therapy. Hemorrhage rates rose with older age, with an average increase in hemorrhage rate of 1.2 (95% confidence interval (CI) 1.0–1.4) per older age category in patients taking warfarin and 1.5 (95% CI=1.3–1.8) in those not taking warfarin. Intracranial hemorrhage rates were significantly higher in those aged 80 and older (adjusted rate ratio=1.8, 95% CI=1.1–3.1 for those taking warfarin, adjusted rate ratio=4.7, 95% CI=2.4–9.2 for those not taking warfarin) than in those younger than 80.

CONCLUSION

Older age increases the risk of major hemorrhage, particularly intracranial hemorrhage, in patients with atrial fibrillation, whether or not they are taking warfarin. Hemorrhage rates were generally comparable with those reported in previous randomized trials, indicating that carefully monitored warfarin therapy can be used with reasonable safety in older patients.

Anticoagulation therapy with warfarin effectively reduces the risk of ischemic stroke associated with atrial fibrillation but also increases the risk for major hemorrhage.^1–4 Although previous randomized trials of warfarin for atrial fibrillation reported low rates of hemorrhage, these studies included few patients aged 80 and older and selected their patients more rigorously than in actual clinical practice.^1,5 It is uncertain whether the low bleeding rates observed in trial settings apply to patients with atrial fibrillation in usual clinical care. Prior studies have also provided conflicting evidence as to whether older age is an independent risk factor for warfarin-associated hemorrhage.^6–8 As increasing numbers of elderly patients take warfarin for atrial fibrillation,⁹ more-precise data are needed on hemorrhage rates in the oldest patients.

Most observational studies of warfarin did not specifically address the risk of hemorrhage in patients with atrial fibrillation. Studies generally included patients taking warfarin for mechanical heart valves or venous thromboembolic diseases, conditions that may have different clinical characteristics and target anticoagulation intensities than atrial fibrillation. Previous studies also lacked sufficient outcome events to examine the association between age and different types of hemorrhage, an important concern because risk factors and consequences of hemorrhage differ between intracranial and extracranial hemorrhages.¹⁰ To address these concerns, data from the large AnTicoagulation and Risk Factors In Atrial Fibrillation (ATRIA) cohort were analyzed to assess whether older age is independently associated with risk of intracranial and extracranial hemorrhage in patients with nonvalvular atrial fibrillation.

METHODS

ATRIA is a cohort study of 13,559 adults aged 18 and older with diagnosed atrial fibrillation and who received care within Kaiser Permanente of Northern California, a large integrated healthcare-delivery system. Details of the cohort assembly have been described previously.¹¹ Cohort members were identified by searching electronic inpatient, outpatient, and electrocardiographic databases for physician-assigned International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9) diagnoses of atrial fibrillation (427.31) between July 1, 1996, and December 31, 1997. To focus on nontransient, nonvalvular atrial fibrillation, patients with ICD-9 diagnoses of mitral stenosis, valvular repair or replacement, transient postoperative atrial fibrillation, or concurrent hyperthyroidism were excluded.¹¹

Hemorrhage Outcomes Assessment

Hospitalization and billing databases were searched electronically through August 31, 1999, for primary and secondary discharge diagnoses of intracranial hemorrhage and primary discharge diagnoses of extracranial hemorrhage. Secondary discharge diagnoses of extracranial hemorrhage were not included, because initial validation studies revealed that few of these events represented true incident major hemorrhages.¹² Hemorrhages that occurred as a complication of a hospitalization for another problem were not included. Because Kaiser Permanente is an integrated healthcare system, hemorrhagic events that incurred medical bills were identified even if they occurred at non-Kaiser Permanente medical facilities.

A clinical outcomes committee reviewed the hospitalization medical records of all potential outcome events using a formal study protocol. Hemorrhagic events were categorized as occurring on or off warfarin based on chart review. The international normalized ratio (INR) at presentation before reversal of excessive anticoagulation was obtained from review of the medical record and the inpatient health plan laboratory database. Intracranial hemorrhages associated with major head trauma (e.g., neurosurgical procedure, high-velocity motor vehicle accident, and skull fracture) were excluded, because interest was primarily in the risk of spontaneous hemorrhages. Major extracranial hemorrhages were defined as fatal, requiring transfusion of two or more units of packed blood cells, or hemorrhage into a critical anatomic site (other than intracranial).

Patient Characteristics

Data on patient age and sex were obtained from electronic databases. Diagnoses of specific medical conditions were obtained through automated searches of electronic clinical inpatient and ambulatory databases for relevant ICD-9 codes during the 5 years before study enrollment. This search strategy was validated by reviewing a subset of medical records, as has been described previously.¹³ Clinical risk factors in the Outpatient Bleeding Index (prior ischemic stroke, prior gastrointestinal hemorrhage, anemia (hemoglobin <13 g/dL in men and <12 g/dL in women), serum creatinine >1.5 mg/dL, coronary artery disease, and diabetes mellitus)¹⁴ and other potential risk factors for hemorrhage (prior hematuria, prior other hemorrhage, hepatic cirrhosis, and mechanical fall complicating a previous hospitalization) were specifically searched for. Although nonsteroidal antiinflammatory drug (NSAID) and aspirin use increase hemorrhage risk,^6,15 their use was not captured comprehensively in the pharmacy database because of their availability without prescription. Attempts were made to ascertain whether aspirin was a potential confounder by reviewing the outpatient medical charts of 232 subjects not taking warfarin, as part of a previously described case-control study.¹² Information on aspirin use by outpatient warfarin users was lacking; instead, aspirin use is provided from chart review of 467 inpatients taking warfarin who were hospitalized for a thromboembolic or hemorrhagic event. The rationale was that, if aspirin use was not associated with older age, it would not confound the association between age and hemorrhage.

Warfarin Exposure

Warfarin exposure was ascertained from automated clinical, pharmacy, and laboratory databases using an algorithm previously validated through chart review.¹² Patients who were newly started on warfarin were defined as those without a previous filled prescription for warfarin in the 12 months before cohort enrollment and with no more than one outpatient INR measurement in the year before enrollment. The start date of warfarin was considered the date the prescription was dispensed. Length of warfarin exposure was calculated based on the number of days supplied per prescription. If there were gaps between prescriptions, patients were considered to be continually taking warfarin if the gaps were less than 60 days or if there were INR measurements at least every 42 days.

For patients taking warfarin, information on anticoagulation intensity and INR variability were obtained, because these factors have been associated with hemorrhage risk.³ An adapted linear interpolation method was used to determine the person-time each patient spent at different INR intervals.¹⁶ If a person was taking warfarin according to pharmacy records but the interval between INR measurements was greater than 8 weeks, the INR values for this extended period were not interpolated and instead were categorized as “not available”; 18% of the INR person-time of warfarin users met this criterion. INR variability was calculated by using the modified formula reported previously.³

Statistical Analyses

Patients who were taking warfarin were analyzed separately from those who were not. Because individual patients could have multiple periods on or off warfarin, a generalized estimating equation approach was used to account for correlation of observations obtained from the same patient.¹⁷ Each patient was followed prospectively until the occurrence of a hemorrhagic event, the end of the study period, disenrollment from the health plan, or death. Patients suffering an intracranial hemorrhage remained at risk for an extracranial hemorrhage and vice versa. Event rates for each age category were calculated by dividing the number of hemorrhagic events by the total person-years of follow-up in the specified warfarin category.

Multivariable Poisson regression models were used to assess whether age was associated with higher hemorrhage risk, grouping age into four ordinal categories (<60, 60–69, 70–79, and ≥80) and testing for an increasing trend in hemorrhage risk per incremental age category while adjusting for other potential risk factors for hemorrhage. Only those covariates with P<.1 according to univariate analysis were included in the final multivariable model for each outcome to avoid overfitting of the models. For patients receiving warfarin therapy, potential covariates also included INR intensity (dichotomized as ≥4.0 and <4.0) and INR variability (dichotomized as >90th and ≤90th percentile of variability).^3,18 Finally, separate multivariable models were developed to assess the effect of age on the risk of intracranial and major extracranial hemorrhages.

All analyses were performed using SAS software, version 9.1 (SAS Institute, Inc., Cary, NC). The institutional review boards of the collaborating institutions approved the study.

RESULTS

The cohort was followed for a median of 2.4 years (inter-quartile range 1.8–2.8 years). The mean age ± standard deviation of the cohort was 71 ± 15, and 28% of the total person-years were from patients aged 80 and older. Older patients were more likely to have risk factors for ischemic stroke, such as hypertension and coronary disease, and risk factors for hemorrhage, such as prior gastrointestinal hemorrhage, anemia, and renal insufficiency (Table 1). INRs of 4.0 and higher and high INR variability did not vary significantly by age (P=.90 and .08, respectively, Table 1).

Table 1.

Clinical Characteristics of 13,559 Patients with Atrial Fibrillation, Stratified by Age and Warfarin Status

	On Warfarin					Off Warfarin
	Age (number of person-years)
Patient Characteristics	Mean Age ± SD	<60 (1,453)	60–69 (3,269)	70–79 (6,767)	≥80 (3,818)	Mean Age ± SD	<60 (2,493)	60–69 (2,946)	70–79 (5,253)	≥80 (4,934)
		%					%
Women	74 ± 11	22.9	34.9	44.9	53.7	74 ± 14	26.8	39.2	45.7	54.8
Prior ischemic stroke	74 ± 11	7.7	9.2	12.5	15.6	77 ± 9	1.4	3.8	6.5	8.2
Diagnosed hypertension	72 ± 11	39.6	52.0	55.7	54.7	73 ± 12	28.2	51.1	53.6	52.3
Known coronary disease	73 ± 10	15.9	28.7	33.5	33.8	75 ± 10	9.3	25.7	30.9	31.8
Diabetes mellitus	71 ± 10	15.7	23.1	19.4	13.9	72 ± 11	9.4	17.1	18.6	10.8
Prior fall during hospitalization	76 ± 10	1.1	2.1	2.5	5.0	79 ± 12	1.0	2.4	4.0	8.7
Prior gastrointestinal bleed	73 ± 10	2.1	1.9	3.4	3.1	76 ± 12	2.0	3.4	5.8	8.3
Prior hematuria	73 ± 10	0.9	2.2	1.8	1.8	75 ± 11	0.7	1.5	1.9	2.1
Prior other bleed	73 ± 11	0.6	0.7	1.0	1.3	72 ± 11	0.5	1.3	0.9	1.0
Hepatic cirrhosis	69 ± 9	0.6	0.8	0.5	0.3	71 ± 12	0.9	1.2	1.4	0.7
Anemia*	74 ± 10	6.8	8.6	12.6	18.4	76 ± 12	6.8	12.7	17.6	25.3
Serum creatinine >1.5mg/dL	74 ± 10	3.8	7.6	10.1	13.6	76 ± 11	4.6	7.4	8.7	12.5
INR ≥4.0	72 ± 2	1.7	1.7	1.7	1.8	—	—	—	—	—
INR variability >90th percentile	71 ± 3	3.8	3.8	3.4	3.7	—	—	—	—	—

^*Anemia defined as hemoglobin < 13 g/dL in men and < 12 g/dL in women.

SD = standard deviation: INR = international normalized ratio.

Hemorrhagic Events

A total of 170 major hemorrhages were identified during 15,300 person-years of follow-up in patients taking warfarin, and 162 hemorrhages were identified during 15,530 person-years in patients not taking warfarin. There were 72 intracranial and 98 major extracranial hemorrhages that occurred in patients taking warfarin (annualized rate 0.47%, 95% confidence interval (CI)=0.37–0.59 and 0.64%, 95% CI=0.53–0.78, respectively) and 46 intracranial and 116 major extracranial hemorrhages that occurred in patients not taking warfarin (annualized rate 0.29%, 95% CI=0.22–0.39 and 0.75%, 95% CI=0.62–0.90, respectively). Eighty-nine percent of the 214 major extracranial hemorrhages involved the gastrointestinal tract.

Twenty-nine percent of the cohort, with a mean age of 70 ± 11, was considered to be newly started on warfarin. Only three intracranial and four major extracranial hemorrhagic events were observed during the first month after starting warfarin. The annualized rate of intracranial hemorrhage in the first month of taking warfarin was 0.92%, compared with 0.46% after the first month (crude relative rate=2.0, 95% CI=0.6–6.7). The rate of major extracranial hemorrhage was 1.2% in the first month of taking warfarin, compared with 0.61% afterward (crude relative rate=2.0, 95% CI=0.7–5.8). There were too few events to assess whether hemorrhage rates differed by age in the early phase of warfarin therapy.

Age and Risk for Hemorrhage

Upon univariate analysis, the rate of all major hemorrhages rose with older age, in patients taking and not taking warfarin. In patients who were taking warfarin, the average relative increase in the rate of hemorrhage from one age category to the next older age category was 1.3 (95% CI=1.1–1.6). In patients who were not taking warfarin, the increase was 1.7 (95% CI=1.5–2.1) per age category. Adjusting for potential risk factors for hemorrhage somewhat attenuated these results; the relative increase in hemorrhage rate in warfarin users was 1.2 (95% CI=1.0–1.4) per age category once prior ischemic stroke, history of gastrointestinal hemorrhage, anemia, renal insufficiency, INR at presentation, and INR variability were adjusted for. For patients not taking warfarin, the relative increase in hemorrhage rate was 1.5 (95% CI=1.3–1.8) per age category (adjusting for prior stroke, gastrointestinal hemorrhage, anemia, renal insufficiency, and diabetes mellitus).

Upon univariate analyses, the unadjusted rate of intracranial hemorrhage remained relatively flat from age 60 to 80 and then increased sharply at 80. This increase was observed in patients taking warfarin (adjusted rate ratio (RR)=1.8, 95% CI=1.1–3.1, comparing patients aged ≥80 with those <80) and in patients not taking warfarin (adjusted RR=4.7, 95% CI=2.4–9.2, comparing patients aged ≥80 with those <80) (Figure 1).

Figure 1.

Unadjusted age-specific rates of intracranial hemorrhage of 13,559 patients with nonvalvular atrial fibrillation taking and not taking warfarin. CI = confidence interval.

The relationship between extracranial hemorrhage and age was less consistent. In patients taking warfarin, rates of extracranial hemorrhage were significantly higher at age 70 to 79 than in those younger than 60, with an adjusted RR of 2.0 (95% CI=1.1–8.2). At age 80 and older, the risk leveled off, with an adjusted RR of 2.1 (95% CI=0.7–6.4) (Figure 2). In contrast, the increase in extracranial hemorrhage rates in patients who were not taking warfarin rose in a linear fashion (Figure 2), with each older age category associated with an increase in hemorrhage rate of 1.3 (95% CI=1.1–1.7) after multivariable adjustment.

Figure 2.

Unadjusted age-specific rates of major extracranial hemorrhage of 13,559 patients with nonvalvular atrial fibrillation taking and not taking warfarin. CI = confidence interval.

Aspirin was unlikely to be a significant confounder of the relationship between age and hemorrhage. In a sample of 232 outpatients not taking warfarin described in the Methods section, the proportion of patients taking aspirin did not vary significantly across age categories (39.4% documented aspirin use in patients aged <60, 57.1% in patients aged 60–69, 44.0% in patients aged 70–79, and 53.1% in patients aged ≥80, chi-square P=.33). Of 467 inpatients taking warfarin and hospitalized with an outcome event, aspirin use was uncommon, with only 8.4% of patients taking both medications, and no differences in aspirin use by age were observed (chi-square P=.78).

DISCUSSION

The risk of major hemorrhage rose with older age whether subjects were taking warfarin or not in this large cohort of adults with nonvalvular atrial fibrillation. A notable finding was that the rates of intracranial hemorrhage—the most devastating form of hemorrhage—rose strikingly at the age of 80 and older.

Reported rates of warfarin-associated hemorrhage in patients with atrial fibrillation vary considerably, probably representing at least some differences in the baseline bleeding risk of the source populations. Rates ranged from 0.76% per year¹⁹ to approximately 10% per year in a study of patients aged 75 and older.²⁰ Prior studies have also provided conflicting data on whether older age is a risk factor for warfarin-associated hemorrhage.^{3,6–8,21,22} Advantages of the current study include a considerably higher number of outcome events, more person-years of observation accumulated in the oldest patients, and separate analyses between intracranial and extracranial hemorrhages.

The relationship between age and hemorrhage risk was found to be somewhat attenuated after adjustment for other medical conditions, suggesting that clinical factors related to aging mediate some of the higher risk. That an independent relationship still remained after multivariable adjustment indicates that other age-related features contribute to hemorrhage risk and were not identified in this study. It is probable that age-related risk factors such as cerebral amyloid angiopathy and leukoaraiosis explain some of the association between age and risk of intracranial hemorrhage.^23,24 It is also likely that the greater prevalence of gastrointestinal pathology in older adults contributes to the higher risk of extracranial hemorrhages.^25,26 Although there is concern that older patients are more difficult to manage on warfarin, there were no significant differences in INR control across age categories, making it unlikely that INR instability mediated the observed age effects.

The hemorrhage rates observed in patients who were taking warfarin were similar to the hemorrhage rates in patients who were not taking warfarin. Because prior studies have demonstrated that warfarin clearly increases hemorrhage risk, this finding suggests that clinicians may selectively prescribe warfarin for patients who have a lower intrinsic risk for hemorrhage. This probable selection effect was more apparent in extracranial hemorrhages, which clinicians may more easily anticipate than intracranial hemorrhages. Intracranial hemorrhages are difficult to predict, and several of the known risk factors—namely, prior stroke and hypertension—are also indications for prescribing warfarin for atrial fibrillation.

Hemorrhage rates in this observational study of actual clinical care were similar to the low rates reported in randomized trial settings, in which there was a large net benefit observed with warfarin.^1,27 The majority of patients in this cohort were prevalent users of warfarin, which may have contributed to the low rates of hemorrhage. Patients newly started on warfarin may be more likely to suffer hemorrhagic complications. Indeed, higher rates of hemorrhage were observed in patients during their first month on warfarin therapy, but the small number of events makes such an estimate imprecise. It would be valuable for future assessments of warfarin safety to focus specifically on the initial phase of drug therapy. It is also possible that low hemorrhage rates were observed, because dedicated anticoagulation clinics, which have been shown to improve anticoagulation control, cared for most patients.²⁸ Anticoagulation in the appropriate INR range is integral to maximizing stroke prevention while minimizing hemorrhage risk.^10,29 This study demonstrates that warfarin can be used with reasonable safety in clinical settings, even in elderly patients, if monitored carefully.

This study had several limitations. As an observational study of actual clinical practice, it was subject to nonstandardized data collection, resulting in periods of missing warfarin exposure and unavailable INR data. Because warfarin treatment was not randomly assigned, confounding by contraindication may occur in that physicians are less likely to anticoagulate patients at higher risk for hemorrhage. Such residual confounding is less likely to affect the estimates of intracranial hemorrhage, because there are few validated clinical risk factors for intracranial hemorrhage. Identifying only hospitalizations with a primary diagnosis of extracranial hemorrhage may have missed some extra-cranial events, although previous validation studies of the cohort indicate that secondary discharge diagnoses of extracranial hemorrhages rarely represented valid incident events and that including such events in the search algorithm would have added negligibly to the total number of major extracranial bleeds. Finally, comprehensive data on nonprescription aspirin and NSAID use were not available, because nonprescription medications were not routinely recorded in the health plan's pharmacy database. If older patients were more likely to take NSAIDs, then use of NSAIDs may confound some of the age effect. There was no significant association between aspirin use and age in samples of outpatients and inpatients, making it less likely that aspirin use, at least, confounded the relationship between age and hemorrhage.

This study provides clear evidence that major hemorrhage rates rise with older age in patients with atrial fibrillation, whether taking warfarin or not. In particular, the rates of intracranial hemorrhage rose sharply at the age of 80 and older. Because patients with atrial fibrillation are generally older and have multiple comorbid conditions, care must be taken to minimize modifiable risk factors for hemorrhage when using warfarin, particularly avoiding excessively high anticoagulant intensities. These results also support vigilance when initially starting warfarin therapy, although the absolute rates of major hemorrhage in this study were reassuringly low even in the oldest patients and comparable with rates reported in randomized trials. These findings indicate that well-managed warfarin therapy can be used safely in clinical practice to achieve substantial benefit in reducing the risk of atrial fibrillation-associated ischemic stroke.