Author + information
- Received November 7, 2006
- Revision received January 8, 2007
- Accepted January 17, 2007
- Published online July 24, 2007.
- Albert Yuh-Jer Shen, MS, MD, FACC⁎,⁎ (, )
- Janis F. Yao, MS†,
- Somjot S. Brar, MD⁎,
- Michael B. Jorgensen, MD, FACC⁎ and
- Wansu Chen, MS†
- ↵⁎Reprint requests and correspondence:
Dr. Albert Yuh-Jer Shen, Department of Cardiology, 1526 Edgemont Street, 2nd Floor, Kaiser Foundation Hospitals, Los Angeles, California 90027.
Objectives This study was designed to study racial/ethnic differences in the risk for intracranial hemorrhage (ICH) and the effect of warfarin on ICH risk among patients with atrial fibrillation (AF).
Background Nonwhites are at greater risk for ICH than whites in the general population. Whether this applies to patients with AF and whether warfarin therapy is associated with comparable risk of ICH in nonwhites are unknown.
Methods We retrospectively identified a multiethnic stroke-free cohort hospitalized with nonrheumatic AF. Warfarin use and anticoagulation intensity were assessed by searching pharmacy and laboratory records. Crude ICH event rates were calculated by Poisson regression. Cox proportional hazard models were constructed to assess the independent effect of race/ethnicity on ICH after adjusting for age, gender, hypertension, diabetes, heart failure, and warfarin exposure.
Results Between 1995 and 2000, we identified 18,867 qualifying AF hospitalizations (78.5% white, 8% black, 9.5% Hispanic, and 3.9% Asian) and 173 qualifying ICH events over 3.3 years follow-up. Achieved anticoagulation intensity was lower among blacks but not different between the other groups. Warfarin was associated with increased ICH risk in all races, but the magnitude of risk was greater among nonwhites. There were no gender differences. The hazard ratio for ICH with whites as referent was 4.06 for Asians (95% confidence interval [CI] 2.47 to 6.65), 2.06 for Hispanics (95% CI 1.31 to 3.24), and 2.04 (95% CI 1.25 to 3.35) for blacks.
Conclusions Nonwhites with AF were at greater risk for warfarin-related ICH. Blacks, Hispanics, and Asians were at successively greater ICH risk than whites.
Atrial fibrillation (AF) is highly prevalent, is associated with a 5-fold increased risk of stroke (1), and may account for 10% to 15% of all strokes (2). In a pooled analysis of 5 primary stroke prevention trials, adjusted-dose warfarin reduced all strokes by 68% (4.5% vs. 1.4% per year) (2). This benefit was offset by a significant 0.3% increase in the annual risk of major bleeding (placebo 1.0%, warfarin 1.3%), the most severe of which was intracranial hemorrhage (ICH). The increase in ICH incidence is reported to be 0.2% per year (3,4). Go et al. (5) studied more than 11,000 nonvalvular AF patients in the Kaiser Permanente Northern California health plan and reported a 51% reduction in thromboembolic events with warfarin therapy but a doubling of ICH (0.46% vs. 0.23% per year).
Although most subjects in these studies were white, there are sparse data on whether the risks of anticoagulation among nonwhite patients parallel those of whites. Furthermore, there are demonstrated racial differences in the background incidence and types of stroke. The stroke incidences of Hispanics and blacks are more than twice as high as those of whites (6,7). Whereas ICH (intracerebral or subarachnoid hemorrhage) accounts for 15% to 20% of strokes in whites (8), it is substantially higher (20% to 30%) among blacks (9,10), Hispanics (6), and Asians (11,12). Death from ICH is also much more common among nonwhites (13).
Because of the higher baseline risk of ICH among nonwhites, it is important to assess whether there are racial/ethnic differences in AF-related stroke, particularly ICH, and whether nonwhite AF patients are at greater risk for ICH when treated with warfarin. This study’s objective was to evaluate the risk of ICH among white, black, Hispanic, and Asian patients with a history of AF hospitalization.
Kaiser Permanente Southern California (KPSC) is a prepaid health maintenance organization that has served at least 2.7 million members, about 14% of the region’s population, for each of the past 10 years. Membership demographics and socioeconomic and racial/ethnic composition are representative of California. Racial/ethnic information is administratively assigned on all hospitalized patients. The study protocol was approved by our institutional review board.
Inclusion and exclusion criteria
By searching hospital discharge records, we identified all first-time hospitalizations for AF or atrial flutter between January 1, 1995, and December 31, 2000, among approximately 2.2 million adult KPSC members. Patients were identified by searching the first 3 discharge diagnoses for International Classification of Disease, 9th Revision, Clinical Modification (ICD-9) codes 427.31 and 427.32 for AF and atrial flutter, respectively. Any AF was included, whether permanent or intermittent, because the risk of stroke does not differ between these groups (14–16). Atrial flutter was included, as the risk for stroke even among “lone atrial flutter” parallels that of AF (17). By referencing ICD-9 and Current Procedural Terminology codes, patients with a history of stroke or transient ischemic attacks, rheumatic heart disease, mitral valvuloplasty, mitral valve replacement predating stroke/transient ischemic attack, brain tumor, heart transplantation, and congenital heart disease were excluded. Patients with prosthetic mitral valves were excluded because of their particularly high risk of thromboembolic events and higher target international normalized ratio (INR) levels (18).
Other risk factors for stroke listed by the National Registry of Atrial Fibrillation (19) were identified using appropriate ICD-9 codes, including heart failure (HF), hypertension, and diabetes mellitus (DM). Race/ethnicity was recorded in our administrative database for 95% of hospitalized patients. For this study, we grouped patients by race/ethnicity into 5 categories: white, black, Hispanic, Asian, or other. The 2.8% of patients identified as belonging to the “other” group were excluded from analysis.
Follow-up of each patient began on the day of AF admission and ended with an outcome event (fatal or nonfatal ICH) or censoring event (non-ICH stroke death, ischemic stroke, mitral valve replacement, carotid endarterectomy within 2 weeks of a stroke, disenrollment from health plan, and the end of study on December 31, 2003). Patients who underwent mitral valve replacement were censored, as they may have had a different stroke risk profile (18,20).
Stroke event identification
Ischemic strokes were identified by ICD-9 codes 433–436; ICHs were identified by codes 430–432. Data sources included hospital records and a computerized matching of Kaiser membership data with California vital statistics death tapes. Only the primary diagnosis from hospital files and the underlying cause of death from mortality files were used. All computer-identified ICHs underwent chart review to confirm that all patients underwent brain imaging and that ICH was primary and nontraumatic. Nonqualifying events were censored at the time of analysis.
Warfarin use and anticoagulation intensity
About 92% of Kaiser members have prescription drug benefits and obtain their drugs through a KPSC-owned pharmacy. Patients are typically dispensed a 100-day supply of medication, which may last longer because of splitting pills. All members have laboratory test benefits and typically have their prothrombin time tested at a Kaiser-owned laboratory with an INR target of 2 to 3 for uncomplicated AF.
We searched our pharmacy and laboratory databases to identify warfarin prescriptions and patients’ achieved anticoagulation intensity. Patients with a filled warfarin prescription were considered to be continuously taking warfarin for the 100-day period after receiving the drug. For between-prescription gaps of 100 days or more, patients were considered to be continuously taking warfarin if they had an INR of 2 or greater in the intervening days. This was to account for pill-splitting and prescriptions filled at a non-Kaiser pharmacy. A therapeutic INR value was considered to be valid for up to 60 days if there were no subsequent INR values or if the warfarin supply was expected to have exhausted and there were no refills. Using this method, an individual patient may contribute to the follow-up times of both warfarin-treated and non–warfarin-treated groups. The proportion of follow-up time that patients were receiving warfarin according to this schema was designated “percent-time receiving warfarin.”
Approximately 60% of patients received ≥1 warfarin prescription. We recorded all INR values that were performed while patients were taking warfarin according to the schema described. To exclude patients who were prescribed but may not have taken warfarin, we included for analysis only patients who had ≥3 INR values while taking warfarin. The INR tests performed during periods when the patients were not expected to have been taking warfarin according to their supply of warfarin were not included in our anticoagulation intensity analyses.
For all patients hospitalized with an ischemic stroke or ICH event, we reviewed their medical records to confirm whether or not they were taking warfarin, as well as the INR values at the time of admission. Comparison of patients’ warfarin use status at the time of stroke defined by our computerized algorithm and that of our chart review showed a 90% consistency rate (kappa = 0.78). Because warfarin-related stroke rates are associated with the achieved anticoagulation intensity, we analyzed the INR values over the follow-up period and at the time of ICH by race/ethnicity. The association between race/ethnicity and INR value at the time of stroke was studied using 3 different INR cut points (3.0, 3.5, and 4.0).
Continuous variables are reported as mean ± SD or median and interquartile range as appropriate. Continuous and categorical variables were compared among racial/ethnic groups with the Kruskal-Wallis test or chi-square test, respectively. Overall and risk factor-stratified crude event rates were calculated using log-linear (Poisson) regression with a generalized estimating equations approach and are reported as per 100 person-years of follow-up. Cox proportional hazard models were used to calculate the effect of race/ethnicity on the risk of stroke after adjusting for age, gender, hypertension, DM, HF, and percentage time receiving warfarin. Rate ratios and hazard ratios are reported with 95% confidence intervals (CI). Adjusted survival curves were constructed from the Cox model by setting the covariates at their means. All analyses were conducted with SAS software (version 9.1 for Windows, SAS Institute, Cary, North Carolina).
In the 6-year period that ended December 2000, we identified 22,196 first-time AF hospitalizations, of which 18,867 qualified for analysis (Table 1).The cohort median age was 74 years; nonwhites were about 6 years younger than whites. More than half of whites were older than 75 years, whereas only about a third of nonwhites were. A lesser proportion of Asians and Hispanics were women than among whites. Hypertension and HF were more prevalent among blacks. Whites were least likely to be diabetic. Hispanics and Asians had more stroke risk factors than the other 2 groups. About 40% to 45% of all patients were not prescribed warfarin at any time, whereas about 25% of patients took warfarin >80% of follow-up. There were no between-group differences.
Number of ICH events and follow-up person-years
Over the 9-year follow-up period, 185 ICH cases were identified. Ten were hemorrhagic conversions of index ischemic strokes, and 2 followed head trauma. By matching our AF cohort to California state death records for the entire study period, we found no cases of death due to ICH that were not captured into our database. There were thus 173 qualifying ICH events that occurred over 63,204 person-years of follow-up (0.27 ICH events/100 person-years). The numbers of ICH (total person-year follow-up) for each group are as follows: white 112 (49,324), black 19 (5,004), Hispanic 23 (6,206), and Asian 19 (2,669).
ICH event rate ratios while taking or not taking warfarin
As shown in Table 2,the crude rate of ICH for the entire cohort was 0.47 per 100 patient-years while taking warfarin and 0.15 while not taking warfarin, a rate ratio of 3.12. Although the risk of ICH increased with warfarin therapy in all groups, there were striking differences in the magnitude of increased risk. There was a 15-fold increased risk of ICH among Asians taking warfarin, whereas the rate ratios for Hispanics and blacks were 4.8 and 5.0, respectively. Warfarin-related ICH risk was lowest among whites at 2.3. As expected, the magnitude of risk was greater among older persons, but there were no gender differences in warfarin-related ICH in this cohort. Patients with hypertension had a 3.3-fold risk of ICH while taking warfarin. Those with 1 or more risk factors had a 3-fold risk for ICH, whereas there was only a single ICH event in each group among patients without risk factors. The ICH risk increased with warfarin whether or not DM or HF was present.
Adjusted hazard ratios
After adjusting for age, gender, hypertension, DM, HF, and percent time treated with warfarin, Hispanics and blacks had twice the risk for ICH as whites, whereas Asians were at 4 times the risk (Table 3,Fig. 1).There was a 4% increased risk of ICH with each advancing year in age. Patients with >41% time receiving warfarin had a 1.8- to 2.1-fold risk of ICH compared with those who were not treated with warfarin. Hypertension, DM, and HF were not independently associated with ICH in this cohort.
Achieved anticoagulation intensity
The median monthly number of INR values ranged from 2.51 in Asians to 2.67 in Hispanics (p = 0.05). The median duration of patient time that INR was ≥2 was approximately 70% in the entire cohort (Table 4).The patient time in subtherapeutic range (INR <2) was greater in blacks than in the others. Overall, 54.5% of patient-time was in the therapeutic range (INR 2 to 3), with blacks having a lower percentage of patient time in this range. Hispanics and Asians were in the therapeutic range for a similar percentage of follow-up time as whites. For approximately 9% of patient time, the INR values were >3, with no differences seen between groups.
INR at the time of ICH
The INR threshold value above which ICH and other major hemorrhagic risk increases varied between studies (21–23); therefore, INR values at the time of ICH hospitalization were analyzed using 3.0, 3.5, and 4.0 as cut points (Table 5).There was no statistically significant difference between groups in the proportion of ICH patients with INR >3.0 or ≥4.0, but minorities were more likely than whites to have an INR ≥3.5 at the time of ICH.
The U.S. populace is becoming increasingly racially diverse. One in 4 residents is nonwhite, according to the 2000 Census. In some states (e.g., California), non-Hispanic whites constitute but half the population. However, literature on the epidemiology, prognosis, and treatment of nonwhites with AF are remarkably sparse. The Cardiovascular Health Study, which involved about 5,000 subjects, suggested that the incidence of AF in blacks was less than half that of whites, but only 5% of the total cohort was black (24). Data on nonblack minorities are even scarcer. In the Atrial Fibrillation Investigators pooled analysis (2), 90% were white. Even in the recent AFFIRM (Atrial Fibrillation Follow-up Investigation of Sinus Rhythm Management) study (15), which included 4,060 patients, only 11% were nonwhite.
Racial differences in ICH
There are important racial/ethnic differences in the incidence and mortality of ICH. In whites, ICH accounts for 15% to 20% of all strokes (6), whereas in Chinese (25), Japanese (26), and blacks (9), that percentage may be 1.5- to 2.0-fold higher. In the Northern Manhattan Stroke Study (6), with whites as referent, the age-adjusted relative risk of ICH was 2 to 3 for both Hispanics and blacks of either gender. Klatsky et al. (12) found that Northern California Asians had a 1.6-fold risk of ICH compared with whites. Furthermore, the Centers for Disease Control and Prevention reported a higher ICH mortality among Asians compared with whites (27). Among patients who died of stroke in the U.S. between 1995 and 1998, ICH accounted for 38% in Asians, 35% in Hispanics, 24% in blacks, and 18% among whites (13). These reports consistently indicate a higher baseline risk of ICH among minorities.
Risk factors for ICH
Advanced age, hypertension, and anticoagulation intensity are associated with a greater risk for warfarin-related bleeding, especially ICH (21,28). In our cohort, nonwhites were 6 to 7 years younger than whites, yet the adjusted hazard ratio for ICH was 2- to 4-fold than that of whites. There was a slightly higher proportion of hypertensives among Asians and blacks, but this does not appear sufficient to explain the excessive risk observed.
Achieved anticoagulation intensity
In this cohort, INR values were therapeutic more than half the time. Anticoagulation intensity was subtherapeutic more often among blacks than the other 3 groups, among which there were no differences. The observed excessive risk of ICH among nonwhites can therefore not be explained by over-anticoagulation.
Polymorphisms of the P450 cytochrome CYP2C9, the enzyme responsible for metabolizing warfarin, partially determine the dose required. Several studies have suggested racial differences in population frequencies of certain polymorphisms (29) that result in a higher risk of warfarin-related serious bleed (30). However, other studies have suggested that CYP2C9 polymorphisms do not fully account for the population differences in warfarin dose requirement (29). Variants in the gene for vitamin K epoxide reductase complex 1 (VKORC1), the target enzyme of warfarin, also affect patients’ response to warfarin (31,32). Recent studies (33,34) have found that the frequency of the haplotypes predictive of a low warfarin dose was significantly higher in the Asian-American population (∼89%) and lower in the African-American population (10% to 14%) than in the European-American population (37% to 42%). Whether this influenced our observation of higher ICH risks among nonwhites is not known.
Because dosing studies of warfarin have involved mostly whites, physicians likely dose warfarin for patients of all races either empirically or based on nomograms derived from predominantly white patients (35). However, several studies have shown that to achieve the same INR range, Asians require lower warfarin doses, whites require intermediate doses, and blacks require higher doses (36,37). It is unknown whether nonwhites in this cohort had greater fluctuation in INR levels and whether this influenced the different rates of ICH observed.
An important consideration is whether all races/ethnicities require the same anticoagulation intensity. You et al. (38) suggested that a target INR of 1.8 to 2.4 is associated with less thromboembolic and hemorrhagic events than INRs of 2 to 3 for Chinese patients. There are few studies addressing racial differences in the optimal INR range. It is important for future studies to assess whether anticoagulation intensity should be modified based on race.
This cohort consists of patients who had been hospitalized with AF as one of the primary diagnoses. One study suggested that as few as 30% of AF patients are ever hospitalized (39). Our results may not apply to patients who were never hospitalized. ICD-9 codes were used for all disease identification in this study. A recent study (40) showed that ICD-9 coding for AF and other stroke risk factors was >80% sensitive and specific. Although ICD-9 codes may not capture all morbidities, they have excellent negative predictive value (0.98) and moderate positive predictive value for ICH (0.54 to 0.77) (41). This limitation was addressed by manual confirmation of all ICH.
In this study, hypertension was considered a dichotomous risk factor (present/absent). We could not adjust for the stage of hypertension or the adequacy of blood pressure control. Nor could we control for other factors known to affect the risk of ICH (e.g., alcohol or illicit drug use) or drugs that interact with warfarin (e.g., acetaminophen or herbs), which use may be heterogeneous among racial groups. Despite our use of both pharmacy and laboratory values, we cannot ascertain the day-to-day fluctuation of anticoagulation intensity and whether this influenced the rates of ICH. Finally, racial/ethnic classification was administratively assigned rather than self-reported. Our administrative database did not differentiate between East and South Asians, but regional demographics indicate that East Asians predominate.
In this multiethnic cohort of patients hospitalized with AF, all races experienced increased risk of ICH while taking warfarin. However, nonwhite patients, especially Asians, experienced substantially higher risk for warfarin-related ICH compared with whites. Furthermore, the adjusted hazard ratio for nonwhites was 2- to 4-fold that of whites.
- Abbreviations and Acronyms
- atrial fibrillation
- confidence interval
- diabetes mellitus
- heart failure
- International Classification of Disease, 9th Revision, Clinical Modification
- intracranial hemorrhage
- international normalized ratio
- Received November 7, 2006.
- Revision received January 8, 2007.
- Accepted January 17, 2007.
- American College of Cardiology Foundation
- Wolf P.A.,
- Abbott R.D.,
- Kannel W.B.
- Sacco R.L.,
- Boden-Albala B.,
- Gan R.,
- et al.
- Thrift A.G.,
- Dewey H.M.,
- Macdonell R.A.,
- McNeil J.J.,
- Donnan G.A.
- Hu H.H.,
- Sheng W.Y.,
- Chu F.L.,
- Lan C.F.,
- Chiang B.N.
- Ayala C.,
- Croft J.B.,
- Greenlund K.J.,
- et al.
- Hart R.G.,
- Pearce L.A.,
- Rothbart R.M.,
- McAnulty J.H.,
- Asinger R.W.,
- Halperin J.L.
- Fuster V.,
- Ryden L.E.,
- Cannom D.S.,
- et al.
- Psaty B.M.,
- Manolio T.A.,
- Kuller L.H.,
- et al.
- Tanaka H.,
- Ueda Y.,
- Hayashi M.,
- et al.
- Ayala C.,
- Greenlund K.J.,
- Croft J.B.,
- et al.
- Dang M.-T.N.,
- Hambleton J.,
- Kayser S.R.
- Yu H.C.,
- Chan T.Y.,
- Critchley J.A.,
- Woo K.S.
- Lip G.Y.,
- Golding D.J.,
- Nazir M.,
- Beevers D.G.,
- Child D.L.,
- Fletcher R.I.
- Kokotailo R.A.,
- Hill M.D.