Author + information
- Received February 3, 2003
- Revision received August 5, 2003
- Accepted September 17, 2003
- Published online February 4, 2004.
- ↵*Reprint requests and correspondence:
Dr. Alan S. Go, Division of Research, Kaiser Permanente of Northern California, 2000 Broadway St., 3rd Floor, Oakland, California 94612, USA.
Objectives This study was designed to determine the association between race and atrial fibrillation (AF) among patients with heart failure (HF).
Background Atrial fibrillation is known to complicate HF, but whether its prevalence varies by race, and the reasons why, are not well understood.
Methods We identified adults hospitalized with confirmed HF within a large integrated healthcare delivery system. We obtained information on demographics, comorbidity, vital signs, medications, and left ventricular systolic function status. “Atrial fibrillation” was defined as AF or atrial flutter documented by electrocardiogram or prior physician-assigned diagnoses. We evaluated the independent relationship between race and AF using multivariable logistic regression.
Results Among 1,373 HF patients (223 African Americans, 1,150 Caucasians), the prevalence of AF was 36.9% (95% confidence interval [CI] 34.3% to 39.5%). Compared with Caucasians, African Americans were younger (mean age 67 vs. 74 years, p < 0.001) and more likely to have hypertension (86.6% vs. 77.7%, p < 0.01) and prior diagnosed HF (79.4% vs. 70.7%, p < 0.01). African Americans had less prior diagnosed coronary disease, revascularization, hypothyroidism, or valve replacement. Atrial fibrillation was much less prevalent in African Americans (19.7%) than Caucasians (38.3%, p < 0.001). After adjustment for risk factors for AF and other potential confounders, African Americans had 49% lower odds of AF (adjusted odds ratio 0.51, 95% CI 0.35 to 0.76).
Conclusions In a contemporary HF cohort, AF was significantly less common among African Americans than among Caucasians. This variation was not explained by differences in traditional risk factors for AF, HF etiology and severity, and treatment.
More than five million Americans currently have chronic heart failure (HF), which remains the leading cause of hospitalization in patients 65 years and older (1,2). Because the incidence of HF markedly increases with advancing age, and in light of the aging U.S. population, the number of older patients with HF will increase substantially in the coming decades.
The risk of developing AF during long-term follow-up appears to be 5 to 10 times higher in patients with chronic HF than in persons without known HF (3–6). Approximately 15% to 30% of patients with HF will develop this arrhythmia at some time during their clinical course (3,7–9). Some studies have shown that the onset of AF in patients with HF is associated with clinical and hemodynamic deterioration due to loss of atrial contractility, tachycardia, and lack of atrioventricular synchrony, as well as a poorer long-term prognosis (10–12). Although the association between HF and AF is well documented, risk factors for developing AF in this setting are not well understood.
It has been postulated that among patients with HF, African Americans may have a lower prevalence of AF than Caucasians (13,14), as has been seen in the general adult population without HF (15–17). However, African Americans with HF have a higher morbidity and mortality rate than Caucasians with HF (10–12). Because AF is associated with higher morbidity and mortality rates in HF, it is unclear whether the racial difference in prevalence of AF may be diminished in the setting of HF. Previous studies comparing the prevalence of AF in African Americans with the prevalence of AF in Caucasians having HF have provided only crude estimates that were not adjusted for possible explanatory factors such as differences in HF etiology, left ventricular (LV) systolic function, comorbid illnesses, and variation in therapies received.
On the basis of existing published data, the prevalence of AF has not been clearly defined in different racial groups, along with potential factors that may explain any observed racial differences in HF. Establishing race as an independent risk factor may potentially help to better risk-stratify patients with HF. In addition, learning about different frequencies of complications of HF such as AF in various racial groups may help guide clinicians in the monitoring, evaluation, and management of these patients. Furthermore, learning more about which types of patients with HF develop AF may also yield important insights into the pathogenesis of AF in this condition.
To address these issues, we examined the association between race and prevalent AF in patients with HF receiving care within a large integrated healthcare delivery system.
The EPOCH (Epidemiology, Practice, Outcomes, and Cost of Heart Failure) Study is a cohort study with retrospective and prospective components. This cohort was assembled by taking a random sample of 1,700 from a total of 4,927 health plan members hospitalized with a primary discharge diagnosis of HF between July 1, 1999, and June 30, 2000, at any one of 16 Kaiser Permanente of Northern California hospitals. A HF hospitalization was defined as a primary discharge diagnosis of HF based on relevant ICD-9-CM codes (402.01, 402.11, 402.91, 425.0, 425.1, 425.2, 425.3, 425.4, 425.7, 425.8, 425.9, 428.0, 428.1, or 428.9) found in hospitalization databases.
The presence of clinical HF was confirmed using the Framingham Heart Study criteria for HF based on information from inpatient medical records for the index hospitalization (18). For the purpose of our analyses, we excluded 327 patients who were neither Caucasian nor African American based on our a priori hypothesis. Hispanic/Latino patients were not considered Caucasian.
We performed a cross-sectional analysis using data from 1,150 Caucasian and 223 African American patients enrolled in the EPOCH study.
Definition of prevalent AF
“Prevalent AF” was defined as the presence of AF or atrial flutter on electrocardiogram (ECG) during the index hospitalization and/or as indicated by a diagnosis found in medical records, hospitalization databases, or ambulatory visit databases. “Electrocardiographic AF” was defined as the presence of an irregularly irregular rhythm with fibrillatory waves and no defined P-waves (19). “Electrocardiographic atrial flutter” was defined as the presence of typical or atypical flutter waves with an atrial rate ranging from 250 to 350 per minute (19). Diagnoses were based on physician-assigned diagnoses in the medical records and/or the presence of corresponding ICD-9-CM codes for AF (427.31) or atrial flutter (427.32) in hospital discharge or ambulatory visit clinical databases during the five years before the index hospitalization. We have previously demonstrated the utility of using these clinical databases for identifying diagnosed AF (15). We also used the same time frame for ascertainment of a history of AF for both African American and Caucasian subjects from clinical databases and medical records.
Data collected on demographic characteristics, co-morbid conditions, vital signs at presentation, and medications were obtained from medical records, supplemented by information from automated hospitalization, ambulatory visit, pharmacy, and laboratory health plan databases during the five years before the index hospitalization. Demographic information included age, gender, and self-reported race from automated databases. For patients without self-reported race information, we obtained race data from admission medical records. Information about health-related behaviors, including past and current use of tobacco, alcohol, heroin, amphetamine, or cocaine, was based on medical records. Medical history obtained from medical records as well as inpatient and outpatient clinical databases included diagnosed hypertension, prior angina or coronary disease, prior known HF, known cardiomyopathy or LV systolic dysfunction, valvular disease (mitral and/or aortic stenosis or insufficiency) or valve repair, diabetes mellitus, chronic lung disease, hyperthyroidism, hypothyroidism, prior ischemic stroke or transient ischemic attack, known peripheral arterial disease, and prior coronary revascularization (i.e., percutaneous coronary intervention or coronary artery bypass surgery).
Information on medications at presentation included the use of aspirin, beta-adrenergic antagonists, calcium channel antagonists, angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers (ARBs), diuretics, or spironolactone based on medical records and filled prescriptions from the comprehensive outpatient pharmacy database during the 120 days before admission. Physical exam findings at presentation, including the presence or absence of tachycardia (≥120 beats/min), inspiratory rales on pulmonary exam, and an S3gallop rhythm, were obtained by chart review. Dictated reports of chest X-rays performed at presentation for the index hospitalization were reviewed for diagnoses of cardiomegaly, pleural effusions, and interstitial/alveolar edema.
Duration of health plan membership and utilization of outpatient services during the five years before the index hospitalization were obtained from administrative files.
Presumed etiology of HF
“Ischemic HF” was defined as being present when a patient had a documented history of coronary heart disease, which included myocardial infarction, angina, coronary artery disease, or coronary revascularization. All other cases of HF were considered “nonischemic.”
LV systolic function status
To assign systematically LV systolic function status, we used information on the quantitative estimate of LV ejection fraction (LVEF), with a preference for radionuclide scintigraphy over echocardiography (transthoracic or transesophageal) followed by left ventriculograms performed during cardiac catheterization. This was based on the expected accuracy of these tests for quantitatively measuring LVEF. Whenever possible, we used the intrahospitalization or most recent post-hospitalization LVEF assessment rather than pre-index hospitalization results, as this more likely reflected the clinical status associated with the index hospitalization. If quantitative measurements were not available, we used qualitative descriptions of LV systolic function that were grouped into categories of severity (normal, mildly reduced, moderately reduced, or severely reduced). In this case, reduced systolic function was defined as LVEF <40% or a qualitative description of moderately or severely reduced systolic function.
We examined baseline differences by race (African Americans vs. Caucasians) and by AF status using the Student ttest for continuous variables and chi-square test for categorical variables. The Fisher exact test was used for categorical variables when the assumptions of the chi-square test were not met (5 or fewer observations in any cell). Using this criterion, the Fisher exact test was used in the univariate analyses by race for cocaine use, amphetamine use, heroin use, and history of mitral stenosis.
We used staged multivariable logistic regression models to assess the independent relationship between race and prevalent AF after sequentially controlling for different categories of potential confounders. Variables available for these models included race, age, gender, known risk factors for AF (such as hypertension, coronary disease, prior diagnosed HF, diagnosed cardiomyopathy, hyperthyroidism, mitral stenosis, chronic lung disease, and valvular repair), LV function status, medications at presentation, other comorbid conditions, physical examination findings, and chest X-ray findings. In the first multivariate model, we adjusted for only age and gender. In the next model, we added known risk factors for AF. For the final model, we added additional comorbid conditions, medications at presentation, LV systolic function status, physical examination findings, and chest X-ray findings. Of note, digoxin use was not included in the multivariable models given its high correlation with the presence of AF. In addition, we performed stratified analyses by HF etiology (ischemic vs. nonischemic), gender (men vs. women), and age (<60 vs. ≥60 years) and found no statistically significant interactions between these variables and race; therefore, results from the final multivariable model are presented. In the final multivariate model, we evaluated the goodness of fit using the Hosmer-Lemeshow test. Finally, to address whether the occurrence of transient HF due to primarily diastolic dysfunction and AF may explain any racial difference in prevalent AF, we excluded patients who presented with tachycardia and who were identified with AF by admission ECG alone and repeated the final multivariable model. All analyses were performed using SAS version 8.0 statistical software (SAS Institute, Cary, North Carolina).
This study was approved by the Kaiser Foundation Research Institute's institutional review board.
Baseline characteristics and prevalence of AF
Among 1,373 patients hospitalized for HF, there were 223 African American and 1,150 Caucasian subjects, with mean age of 73 years and 48% being women (Table 1). Seventy-two percent of all subjects had a history of diagnosed HF, and 31% had known cardiomyopathy. Nearly 80% of patients had a history of diagnosed hypertension. A history of coronary disease was also common (65%), with approximately one-third of patients having had a prior myocardial infarction and 27% having had a prior coronary revascularization. Diagnosed valvular heart disease was relatively uncommon, with the most prevalent diagnoses being mitral regurgitation (9%) and aortic stenosis (7%). Among our cohort, 46% had chronic lung disease. Approximately 90% of patients were also taking a diuretic on hospital admission, with more than 65% taking an angiotensin-converting enzyme inhibitor and/or an ARB. More than 85% of patients with confirmed HF had inspiratory rales on lung examination, with 56% having interstitial edema on admission chest X-ray. Cardiomegaly on chest radiography was present in the majority (80%) of patients, and 49% of patients had evidence of pleural effusion. Heart failure with preserved LV systolic function (defined as LVEF ≥40% or qualitative assessment of normal or mildly reduced systolic function) was present in 48% of subjects.
A total of 506 persons had evidence of AF, for an overall prevalence in the cohort of 36.9% (95% CI 34.3% to 39.5%). Among the patients with evidence of AF, 241 (47.6%) were identified by ECG plus prior diagnoses found in medical records or clinical databases, 170 (33.6%) by prior diagnoses alone, and 95 (18.8%) by ECG alone during the index hospitalization.
Race and AF
Compared with Caucasians, African Americans were younger but equally likely to be women (Table 1). Hypertension, prior diagnosed HF, prior cardiomyopathy, and radiographic cardiomegaly were more common in African Americans than Caucasians. African Americans were less likely to have a history of angina, hypothyroidism, coronary revascularization, or valvular repair/replacement. There were no significant racial differences in current or former tobacco or alcohol use. Overall, illicit drug use was low in both groups, with slightly higher current or former cocaine and heroin use among African Americans. African Americans were less likely than Caucasians to be taking a beta-adrenergic antagonist on admission, but there were no significant differences in the use of angiotensin-converting enzyme inhibitors, ARBs, spironolactone, diuretics, aspirin, or calcium channel antagonists. There were no significant racial differences in selected measures of HF severity at presentation, including the presence of tachycardia, an S3gallop rhythm, or interstitial edema on admission chest X-ray. In addition, there were also no significant racial differences in the distribution of LV systolic function status (LVEF ≥40%, <40%, or unknown). Of note, duration of membership during the five years before the index admission was similar between African American and Caucasian patients (mean ± SD: 4.4 ± 1.3 vs. 4.5 ± 1.3 years, respectively), and outpatient utilization was not significantly different during the 12 months before hospitalization in both groups (data not shown).
African Americans had a 50% lower prevalence of AF (19.7%, 95% CI 14.7% to 25.6%) than Caucasians (38.3%, 95% CI 35.4% to 41.1%, p < 0.001) (Fig. 1). There was no significant difference in the prevalence of atrial flutter between African Americans and Caucasians (4.5% vs. 6.5%, respectively, p = 0.25). After adjusting for differences in age and gender, African American race was associated with 45% decreased odds (95% CI 22% to 61%) of prevalent AF (Table 2). In the final model that adjusted for differences in age, gender, known risk factors for AF (hypertension, coronary disease, prior diagnosed HF, mitral stenosis, valvular repair, chronic lung disease, and hyperthyroidism), differences in HF therapy before admission, LV systolic function status (LVEF ≥40%, <40%, or unknown), and other co-morbid conditions, African Americans still had 49% (95% CI 24% to 65%) decreased odds of having AF compared with Caucasians (Table 2). In the final model, age, prior HF, prior known mitral insufficiency, prior known valvular disease, diabetes, selected medications at admission (aspirin, calcium channel blockers, and diuretics), LV systolic function status, tachycardia at presentation, presence of S3on physical exam, and cardiomegaly on admission chest X-ray were also significant correlates of AF. The final multivariable model had acceptable goodness of fit (p = 0.70 by Hosmer-Lemeshow test). To address whether racial differences in prevalent AF may be explained by the differential occurrence of transient HF due to rapid AF and diastolic dysfunction, we excluded the 34 patients who presented with tachycardia and whose AF was identified only by the admission ECG, but we found that the results were similar (adjusted odds ratio 0.54, 95% CI 0.36 to 0.79).
Atrial fibrillation is highly prevalent among patients with HF and can lead to adverse consequences, including tachycardia-related cardiomyopathy, reduction in LV preload attributable to disorganized atrial contractions, increased risk of systemic embolus, and overall poorer long-term outcome (10–12). However, whether AF occurs at the same frequency among different race/ethnic groups has not been well understood. In a large cohort of patients hospitalized with confirmed HF, we found that AF affected more than one-third of patients. Furthermore, we observed that the prevalence of AF was nearly 50% less among African American than Caucasian patients and that this difference only minimally decreased after adjustment for known risk factors for AF, relevant patient characteristics, severity of HF presentation, LV systolic dysfunction status, presumed HF etiology, and medical management before hospital admission.
Our finding of a prevalence of AF of 36.9% in HF patients was slightly higher than the National Heart Failure Project sample of Medicare beneficiaries hospitalized for HF (29.5%) (20), although the exact definition of AF used in the National Heart Failure Project was not clear. However, our prevalence estimate was similar to that reported by Aronow et al. (35.2%) in a smaller sample of elderly, long-term nursing home residents diagnosed with HF (11). These relatively minor differences are likely explained by differences in populations and methods used to ascertain AF.
We observed that AF is approximately 50% less prevalent in African Americans than Caucasians with HF, which is consistent with estimates from prior studies (13,14). In a cohort study of 398 patients age 50 years or older hospitalized with HF, Vaccarino et al. (14)reported a prevalence of AF of 28.0% in Caucasians and 14.6% in African Americans. Among 163 consecutive patients admitted for HF at an urban hospital, Afzal et al. (13)reported a prevalence of AF of 42% in Caucasians and 21% in African Americans. However, these studies included relatively modest numbers of subjects, and more importantly, did not adjust for potential confounders to examine the independent relationship between race and AF. In our study, African Americans generally had a more adverse risk profile for AF (3), including a higher frequency of diagnosed hypertension, prior diagnosed HF or cardiomyopathy, and diabetes, although they were also younger and had a lower prevalence of diagnosed hypothyroidism. The lower prevalence of AF in African Americans with HF in this cohort persisted even after adjustment for these known risk factors for AF, as well as presumed HF etiology (ischemic vs. nonischemic), differences in cardiac medication use at presentation, other relevant co-morbid illnesses, selected measures of HF severity, and LV systolic function status.
The reason for the difference in prevalence of AF by race remains unclear. One possibility is that the difference could be a result of unmeasured confounders such as differential access to care related to socioeconomic status that could have led to lower rates of prior diagnoses of AF (ascertainment bias). However, our sample included only insured patients who belonged to an integrated healthcare delivery system and who had equal access to hospitalization based on their basic membership coverage, which decreases the likelihood of differential decision-making regarding hospitalization for HF. Both African American and Caucasian subjects had similar mean duration of membership and outpatient utilization rates before the index hospitalization, suggesting equal opportunity to identify previously diagnosed AF. In addition, our observation that there were similar rates of the presence of an S3gallop, pulmonary edema on chest X-ray, and similar distributions of ejection fraction status between Caucasians and African Americans suggests that confounding by severity of HF is less likely in our sample. Another possible explanation for the lower prevalence of AF in African Americans than in Caucasians with HF may be intrinsic racial differences in atrial membrane stability, atrial conduction pathways, or genetic polymorphisms leading to different susceptibility to the development of AF. For example, polymorphisms have been found to be associated with racial differences in risk for HF (21)and response to treatment for HF (22). In addition, recent data from the Cardiovascular Health Study have demonstrated a smaller average left atrial size among elderly African Americans than among Caucasians, which may help to partially explain the lower prevalence of AF in African Americans (23).
Our study had several strengths. This was a large, contemporary cohort of patients hospitalized with HF that included excellent representation of African Americans. The cohort was assembled from a well-characterized managed care population that is diverse and highly representative of the California state population (24). Prevalent AF was determined by reviewing admission ECGs using standardized criteria in addition to the systematic review both of medical records and of comprehensive automated clinical databases for outpatient and inpatient diagnoses of AF. The validity of our automated databases for the diagnosis of AF is high (15). Information on traditional risk factors for AF, targeted co-morbid conditions, and cardiac medications were obtained from medical records in combination with comprehensive automated outpatient, hospital discharge, pharmacy, and laboratory databases (15,25).
Our study also had several limitations. Because of the cross-sectional design of this study, we cannot accurately determine whether AF developed before or after the initial diagnosis of HF. Atrial fibrillation that was differentially distributed between African American and Caucasian patients before the onset of HF could have led to an overestimate of racial differences in prevalence of AF. However, because the prevalence of AF in the general population is only slightly lower among African Americans than among Caucasians (15), this is an unlikely explanation for the magnitude of our and other studies' findings. Another limitation is that we could not accurately measure the duration of HF or other co-morbid conditions. However, the time frame for ascertainment of the various co-morbid conditions as well as prior diagnosed outpatient or inpatient HF (from medical records and clinical databases) was the same for both groups. Our cohort was comprised of only hospitalized patients with HF who likely represent a more severe spectrum of HF, which may lead to a higher estimate of the overall prevalence of AF but not the observed racial differences. Finally, we analyzed data from an insured population in northern California, so results may not be completely generalizable to uninsured patients or other geographic areas.
In sum, we found that African American race was independently associated with a lower prevalence of AF when compared with Caucasians in the setting of HF. The reasons for a lower prevalence of AF among African Americans with HF remain unclear, but further research into the potential racial differences in clinical, physiological, and genetic factors that affect the development of AF is needed.
We thank Wendy Lee and Galina Zlotnikov for their technical assistance and helpful support in this project.
☆ Supported by a research grant from Pharmacia Corporation, Inc.
- atrial fibrillation
- angiotensin II receptor blocker
- electrocardiogram/electrocardiographic/ electrocardiography
- heart failure
- left ventricle/ventricular
- Received February 3, 2003.
- Revision received August 5, 2003.
- Accepted September 17, 2003.
- American College of Cardiology Foundation
- ↵National Center for Health Statistics GE. 1993 Summary: National Hospital Discharge Survey. Hyattsville, MD: 1995:1–63
- American Heart Association. 2002 Heart and Stroke Facts Statistical Update. Dallas, TX: American Heart Association, 2001:19
- Onundarson P.T.,
- Thorgeirsson G.,
- Jonmundsson E.,
- Sigfusson N.,
- Hardarson T.
- Pedersen O.D.,
- Bagger H.,
- Kober L.,
- Torp-Pedersen C.
- Middlekauff H.R.,
- Stevenson W.G.,
- Stevenson L.W.
- Dries D.L.,
- Exner D.V.,
- Gersh B.J.,
- Domanski M.J.,
- Waclawiw M.A.,
- Stevenson L.W.
- Pozzoli M.,
- Cioffi G.,
- Traversi E.,
- Pinna G.D.,
- Cobelli F.,
- Tavazzi L.
- Hajat C.,
- Dundas R.,
- Stewart J.A.,
- et al.
- Goldberger A.L. Goldberger E.