*Includes percutaneous coronary intervention (PCI), fibrinolytic, or
coronary artery bypass graft reperfusion strategies and spontaneous reperfusion.
AMI indicates acute myocardial infarction.
Bradley EH, Herrin J, Wang Y, McNamara RL, Webster TR, Magid DJ, Blaney M, Peterson ED, Canto JG, Pollack, CV, Krumholz HM. Racial and Ethnic Differences in Time to Acute Reperfusion Therapy for Patients Hospitalized With Myocardial Infarction. JAMA. 2004;292(13):1563-1572. doi:10.1001/jama.292.13.1563
Author Affiliations: Section of Health Policy
and Administration, Department of Epidemiology and Public Health (Drs Bradley
and Krumholz and Ms Webster), Section of Cardiovascular Medicine (Drs McNamara
and Krumholz and Mr Wang), and Robert Wood Johnson Clinical Scholars Program
(Dr Krumholz), Department of Medicine at the Yale University School of Medicine,
New Haven, Conn; Yale-New Haven Hospital Center for Outcomes Research and
Evaluation, New Haven (Dr Krumholz); Flying Buttress Associates, Charlottesville,
Va (Dr Herrin); Kaiser Permanente Clinical Research Unit, Denver, Colo (Dr
Magid); Department of Preventive Medicine and Biometrics and the Division
of Emergency Medicine at the University of Colorado Health Sciences Center,
Denver (Dr Magid); Genentech Inc, South San Francisco, Calif (Dr Blaney);
Division of Cardiology, Department of Medicine, Duke University, Durham, NC
(Dr Peterson); Center for Cardiovascular Prevention, Research, and Education,
Watson Clinic, Lakeland, Fla (Dr Canto); and Department of Emergency Medicine,
Pennsylvania Hospital, Philadelphia (Dr Pollack).
Context Nonwhite patients experience significantly longer times to fibrinolytic
therapy (door-to-drug times) and percutaneous coronary intervention (door-to-balloon
times) than white patients, raising concerns of health care disparities, but
the reasons for these patterns are poorly understood.
Objectives To estimate race/ethnicity differences in door-to-drug and door-to-balloon
times for patients receiving primary reperfusion for ST-segment elevation
myocardial infarction; to examine how sociodemographic factors, insurance
status, clinical characteristics, and hospital features mediate racial/ethnic
Design, Setting, and Patients Retrospective, observational study using admission and treatment data
from the National Registry of Myocardial Infarction (NRMI) for a US cohort
of patients with ST-segment elevation myocardial infarction or left bundle-branch
block and receiving reperfusion therapy. Patients (73 032 receiving
fibrinolytic therapy; 37 143 receiving primary percutaneous coronary
intervention) were admitted from January 1, 1999, through December 31, 2002,
to hospitals participating in NRMI 3 and 4.
Main Outcome Measure Minutes between hospital arrival and acute reperfusion therapy.
Results Door-to-drug times were significantly longer for patients identified
as African American/black (41.1 minutes), Hispanic (36.1 minutes), and Asian/Pacific
Islander (37.4 minutes), compared with patients identified as white (33.8
minutes) (P<.01 for all). Door-to-balloon times
for patients identified as African American/black (122.3 minutes) or Hispanic
(114.8 minutes) were significantly longer than for patients identified as
white (103.4 minutes) (P<.001 for both). Racial/ethnic
differences were still significant but were substantially reduced after accounting
for differences in mean times to treatment for the hospitals in which patients
were treated; significant racial/ethnic differences persisted after further
adjustment for sociodemographic characteristics, insurance status, and clinical
and hospital characteristics (P<.01 for all).
Conclusion A substantial portion of the racial/ethnic disparity in time to treatment
was accounted for by the specific hospital to which patients were admitted,
in contrast to differential treatment by race/ethnicity inside the hospital.
Many studies have demonstrated different patterns of cardiovascular
care by racial and ethnic groups,1- 4 but
few have investigated the relative contributions of sociodemographic, economic,
clinical, or health system factors to these differences. Understanding the
sources of racial and ethnic differences in cardiovascular care is paramount
to designing effective interventions to eliminate disparities, which has been
designated as a national priority.5- 7 Therefore,
studies are needed to examine the broad range of factors that might explain
racial and ethnic differences in care.
Acute reperfusion for patients presenting with ST-segment elevation
myocardial infarction (STEMI) provides a good opportunity to examine the interplay
of patient and health system features and racial and ethnic differences in
cardiovascular care. Prompt reperfusion markedly improves survival for patients
with STEMI8- 11 and
is an indicator of quality of care used by the Centers for Medicare &
Medicaid Services and the Joint Commission on Accreditation of Healthcare
Organizations. Recent reports indicate that patients identified as African
American/black or as nonwhite minority experience significantly longer times
to fibrinolytic therapy (door-to-drug times) and percutaneous coronary intervention
(PCI) (door-to-balloon times) than patients identified as white,12- 15 raising
concerns of health care disparities. However, existing studies have not comprehensively
investigated the factors that explain or mediate the observed racial and ethnic
differences in time to reperfusion therapy.
Accordingly, we used data from the National Registry of Myocardial Infarction
(NRMI) 3 and 416 and the American Hospital
Association Survey of Hospitals17 to examine
time to reperfusion therapy for patients with STEMI. We sought to characterize
racial and ethnic differences in time between hospital arrival and receipt
of reperfusion therapy, and we examined the role of sociodemographic factors,
insurance status, clinical characteristics, and health system factors in explaining
these differences. This information is essential for understanding possible
reasons for differences in this key quality indicator and for formulating
interventions that would support national efforts to eliminate racial and
ethnic disparities in cardiovascular care.
We conducted a retrospective, observational study using admission and
treatment data from a national cohort of patients with STEMI. The sample for
this analysis comprised patients enrolled in the NRMI, an observational registry
sponsored by Genentech Inc,18,19 who
had STEMI or left bundle-branch block and who were hospitalized from January
1, 1999, through December 31, 2002, to receive acute reperfusion therapy with
either fibrinolytic therapy or primary PCI. The Figure shows the sample selection process and specific exclusion
criteria. We began with the 830 473 hospitalized patients who were enrolled
in NRMI 3 or NRMI 4 and who met any of the following criteria for acute myocardial
infarction (AMI): level of total creatine kinase or creatine kinase MB that
was 2 or more times the upper limit of the reference range or elevations in
alternative cardiac markers; electrocardiographic evidence of AMI; nuclear
medicine testing, echocardiography, or autopsy evidence of AMI; or a diagnosis
of AMI according to the International Classification of
Diseases, Ninth Revision, Clinical Modification (code 410.X1). The
diagnosis code 410.X1 was required for NRMI 4.
We then excluded patients who had neither ST-segment elevation (≥2
leads) nor left bundle-branch block on the first electrocardiogram to identify
an ideal group for primary reperfusion therapy. From this group, we excluded
patients who were transferred in from other hospitals as well as patients
whose AMI symptom onset was after admission or who were missing data on symptom
onset time and who had no chest pain present at admission. The percentage
of patients missing data on symptom onset time did not vary by race/ethnicity.
To focus on primary reperfusion, we then eliminated patients who either did
not receive primary reperfusion with fibrinolytic therapy or PCI or who received
reperfusion therapy more than 6 hours after hospital arrival.
We developed 2 cohorts based on whether the patient received fibrinolytic
therapy or PCI. If NRMI data indicated that the patient received both fibrinolytic
therapy and PCI, we considered only the first one received. In each cohort,
we excluded patients with unknown door-to-treatment time and those admitted
to a hospital outside the United States. The absence of door-to-treatment
time did not vary significantly by race/ethnicity. To avoid including hospitals
that administered reperfusion therapy uncommonly, we also excluded patients
who were admitted to hospitals that reported to the NRMI fewer than an average
of 5 cases per year of fibrinolytic therapy (for the fibrinolytic therapy
cohort) or of primary PCI (for the PCI cohort) over the study period. Finally,
we excluded patients for whom race/ethnicity information was missing. The
institutional review board at the Yale University School of Medicine determined
that this protocol was exempt from review because we used existing data that
had no patient identifiers.
The principal outcome was the time between hospital arrival and delivery
of reperfusion therapy. Time to fibrinolytic therapy (ie, door-to-drug time)
and time to PCI (ie, door-to-balloon time) were analyzed as separate outcomes.
We modeled each as a continuous variable, given recent evidence supporting
the continuous relationship between time to reperfusion and 1-year mortality.20 Because the distributions of the outcome variables
were skewed, we log transformed21,22 the
outcome measures for performing parametric analysis. To improve the clinical
interpretation of the results, we converted the transformed values from the
model back to their original units, ie, minutes, using geometric means21,22 and simulation with 10 000 reiterations.23 Compared with the arithmetic mean, the geometric
mean gives less weight to outlying values and thus better reflects the median
The primary independent variable was patient race/ethnicity, coded as
a set of dummy variables indicating patients’ racial/ethnic group, which
was abstracted from the medical records using the following categories: white,
African American/black, Hispanic, Asian/Pacific Islander, American Indian
or Alaska native, and other or unknown race/ethnicity. Admission or triage
staff recorded race/ethnicity as the patient was registered, using hospital-defined
race/ethnicity options; in NRMI, patients were assigned to only 1 race/ethnicity
Patient-level variables other than race/ethnicity included sex, age
(<65, 65-80, >80 years), insurance status, and clinical characteristics.
Clinical characteristics consisted of medical history (current smoker, chronic
renal insufficiency, previous AMI, hypertension, family history of coronary
artery disease, hypercholesterolemia, heart failure, previous percutaneous
transluminal coronary angioplasty, previous coronary artery bypass graft surgery,
chronic obstructive pulmonary disease, stroke, angina, diabetes); presentation
characteristics (chest pain, systolic blood pressure, pulse rate, heart failure);
and the results of the first electrocardiogram obtained after hospital arrival
(number of leads with ST-segment elevation, left bundle-branch block, AMI
location, ST-segment depression, nonspecific ST-segment or T-wave changes,
Q wave). We included the reported time between symptom onset and hospital
arrival, arrival time and day of week, and whether a prearrival electrocardiogram
had been performed. We also included calendar time, calculated as the number
of days between January 1, 1999, and the hospital admission date, as an independent
variable to account for any secular trends as well as for differing reporting
periods by hospitals, although it was nonsignificant in all models.
Hospital characteristics were obtained from the American Hospital Association
Annual Survey of Hospitals17 and the SMG Marketing
Group data set24 and included Census region,
urban/rural location (urban defined as location in a county with
a population ≥50 000) and teaching status (ie, participation in a
residency or fellowship training program accredited by the Liaison Committee
on Medical Education), hospital ownership type (government, not-for-profit,
for-profit), and cardiac facilities (presence of cardiac surgery capability,
catheterization laboratory only, or neither for the fibrinolytic therapy sample).
We combined hospital urban/rural location and teaching status to include the
resulting 4-level variable. For the door-to-drug analysis, we estimated the
hospital’s annual fibrinolytic therapy volume (0-14, 15-30, >30 cases)
and the percentage of all primary reperfusion cases that were performed with
fibrinolytic therapy rather than PCI (<20%, 20%-90%, >90%) from the NRMI
database. For the door-to-balloon analysis, we estimated the hospital’s
annual PCI volume (<20, 20-40, >40 cases) and percentage of all primary
reperfusion cases performed with PCI rather than drug therapy (<20%, 20%-90%,
>90%) from the NRMI database.
We performed separate analyses for the fibrinolytic therapy and PCI
cohorts. We first examined overall geometric means for door-to-treatment times
for each racial/ethnic group, which we termed “crude” means, by
comparing log time to treatment between groups using t tests
adjusted for clustering of patients within hospitals.25 Then,
to explore how these crude differences might be mediated by patient-level
and hospital-level factors, we estimated a sequence of hierarchical models26 for each cohort. We used hierarchical models because
patients were clustered within hospitals (the unit of enrollment in the NRMI).
In all models the intercept and calendar time were modeled as random effects
to account for hospital-specific effects, ie, differences in mean time to
treatment and in improvement in time to treatment at the hospital level.
We first estimated a hierarchical model including only race/ethnicity
to assess the extent to which crude racial/ethnic differences in time to treatment
were attenuated by hospital-specific effects. Then, in successive hierarchical
models, we adjusted for covariates in the following sequence: (1) patient
age, sex, and insurance status; (2) clinical characteristics; (3) time between
symptom onset and hospital arrival, arrival time and day of week, and whether
a prehospital electrocardiogram was performed; and (4) hospital characteristics.
Covariates considered in each model have been shown in previous literature
to be associated with β-blocker use or were empirically associated with β-blocker
use in this data set. We validated the normality assumptions of all models
by inspecting the quartile-quartile plots of the patient-level residuals and
the Mahalanobis distance of the hospital-level residuals.27 The
proportion of variance in the outcome explained was indicated by the R2 for each model. Statistical analyses were
performed using SAS versions 6.12 and 8.2 (SAS Institute Inc, Cary, NC), HLM
version 5.04 (SSI, Lincolnwood, Ill), and Stata version 8.0 (Stata Corp, College
The cohorts included 73 032 patients receiving fibrinolytic therapy
in 1052 hospitals and 37 143 patients receiving primary PCI in 434 hospitals. Table 1, Table 2, Table 3, and Table 4 describe the patient and hospital characteristics for both
the fibrinolytic therapy and the PCI patient cohorts separately. In both cohorts,
about 5% were identified as African American/black, about 4% as Hispanic,
about 2% as Asian/Pacific Islander, less than 1% as American Indian or Alaska
Native, and about 4% as “other” race or ethnicity. The geometric
mean for door-to-drug time for all racial and ethnic groups combined was 34.3
minutes (95% confidence interval [CI], 34.1-34.4 minutes) (Table 1). The geometric mean door-to-balloon time was 104.7 minutes
(95% CI, 104.2-105.1 minutes) (Table 2).
Many patient and hospital covariates differed significantly by racial
and ethnic groups (Tables 1-4), highlighting the importance of adjusting
for these factors in analysis of racial and ethnic differences in the outcome.
White patients tended to be older than patients of other racial and ethnic
groups. African American/black patients were more likely to be female compared
with white patients. Insurance status differed significantly, with nonwhite
patients more likely to have Medicaid only or no insurance compared with white
patients. Differences in clinical characteristics were pronounced, and patients
in the nonwhite groups were generally more likely than white patients to be
current smokers and to have chronic renal insufficiency, diabetes, or hypertension.
Racial and ethnic group differences are also apparent in the geographic region
of the hospital to which patients were admitted, urban/rural location, teaching
status, ownership type of the hospital, availability of cardiac facilities,
and volume of reperfusion cases (fibrinolytic therapy and PCI) performed annually
at the hospital.
As shown in Table 5, the crude
geometric mean of door-to-drug time was significantly longer for patients
identified as African American/black (41.1 minutes), Hispanic (36.1 minutes),
and Asian/Pacific Islander (37.4 minutes) compared with patients identified
as white (33.8 minutes; P<.01 for all). These
crude differences by race/ethnicity were attenuated to varying degrees, depending
on the racial/ethnic group, after accounting for the overall mean door-to-drug
time at the hospital in which the patient was treated (Table 5, model 0).
The racial and ethnic group differences in door-to-drug time remained
significant, although further attenuated, after adjustment for age, sex, and
insurance status (Table 5, model 1);
clinical characteristics (Table 5, model
2); time of arrival, time since symptom onset, and having a prehospital electrocardiogram
performed (Table 5, model 3); and hospital
characteristics (Table 5, model 4).
After full adjustment for patient-level and hospital-level factors (Table 5, model 4), the differences by racial/ethnic
group in door-to-drug times were 5.1 (95% CI, 4.2-5.9) minutes longer for
patients identified as African American/black compared with white patients
(P<.001), 1.3 (95% CI, 0.4-2.3) minutes longer
for Hispanic compared with white patients (P = .006),
and 1.7 (95% CI, 0.4-3.0) minutes longer for Asian/Pacific Islander compared
with white patients (P = .01).
As shown in Table 2 and Table 6, the geometric mean of door-to-balloon
times for patients identified as African American/black (122.3 minutes) or
Hispanic (114.8 minutes) were significantly longer than for patients identified
as white (103.4 minutes) (P<.001 for all). A marked
proportion of these differences were attenuated for all racial/ethnic groups
when we accounted for differences in the geometric mean time to treatment
of patients treated at the hospital at which a given patient was treated (Table 6, model 0).
Consistent with the analysis of door-to-drug time, racial/ethnic group
differences in door-to-balloon time remained significant, although attenuated
in most cases, after adjusting for age, sex, and insurance status (Table 6, model 1) and then further adjusting
for clinical characteristics such as covariates (Table 6, model 2); time of arrival, time since symptom onset, and
whether a prehospital electrocardiogram was performed (Table 6, model 3); and hospital characteristics (Table 6, model 4). In the fully adjusted model (Table 6, model 4), door-to-balloon time remained 8.7 (95% CI, 6.7-10.8)
minutes longer for African American/black patients compared with white patients
(P <.001) and 3.7 (95% CI; 1.3-6.1) minutes longer
for Hispanic patients compared with white patients (P = .002).
We found marked differences in time to reperfusion by race/ethnicity,
with patients identified as African American/black having on average about
20% longer door-to-drug and door-to-balloon times than patients identified
as white. This difference equates to times that are 7 minutes longer for door-to-drug
time and 19 minutes longer for door-to-balloon time. We also found significant
differences in times to acute reperfusion therapy for patients identified
as Hispanic and for those identified as Asian/Pacific Islander, although the
magnitude of these differences was more modest.
While differences in door-to-treatment times have been shown in previous
studies,12- 15 our
study advances that research in several ways. We examine multiple racial/ethnic
categories, whereas previous studies of differences in time to acute reperfusion
focused only on African American/black vs white comparisons12,15 or
on nonwhite vs white comparisons,14 which might
oversimplify our understanding of difference by racial/ethnic groups. Due
to the relatively small numbers of patients identified as Asian and American
Indian, however, we had little statistical power to detect differences among
these groups, as reflected in the wide CIs for the estimates for these groups.
In addition, unlike previous studies, we used hierarchical models to examine
both hospital-level and patient-level factors that simultaneously and adequately
account for the clustering of patients within hospitals. Without such modeling,
it is difficult to explicitly examine the effect of the hospitals on patients’
time to reperfusion therapy; estimates of all effects may be incorrect, and
standard errors are typically underestimated. Also, unlike previous studies
of racial/ethnic differences in time to treatment, we use sequential models
to investigate the extent to which hospital-level and patient-level factors
mediate, and thus potentially explain, these disparities. Understanding the
relative importance of these factors is critical for targeting effective interventions
to eliminate such disparities.
Our findings reveal that a substantial portion of the racial and ethnic
disparity in time to treatment is accounted for by the hospital to which a
patient is admitted, in contrast to differential treatment by race and ethnicity
inside the hospital. For instance, the crude difference in door-to-balloon
time between African American/black and white patients was reduced by 33%
after accounting for differences between the hospitals in which patients were
treated. More striking, the crude difference in door-to-balloon times between
Hispanic patients and white patients was reduced by nearly 75% after accounting
for differences between the hospitals in which they were treated.
However, the racial/ethnic group differences were not attenuated substantially
by the addition of structural hospital characteristics to the models. This
suggests that the crude race/ethnicity-related differences in time to treatment
were not largely attributable to traditional proxies for hospital quality
such as volume, teaching status, or urban/rural location. Rather, the crude
differences by race/ethnicity were accounted for by other unmeasured differences
in hospitals in which certain patients received care. The result is important
for designing interventions that address such disparities and highlights the
importance of improving the quality of care in hospitals in which minority
groups are more likely to be treated.
Nonetheless, holding the hospital in which care was received constant,
there remained racial and ethnic disparities in door-to-drug and door-to-balloon
times. These within-hospital differences by race/ethnicity were independent
of differences in patients’ clinical characteristics, sociodemographic
factors, insurance status, or structural hospital characteristics. Although
the magnitude of this independent race/ethnicity effect was modest, recent
research indicates that small delays can importantly influence 1-year mortality.20 The effect is notable given the comprehensive set
of covariates for which we adjusted. Furthermore, from the perspective of
health care equity, systematically longer times to treatment for certain racial/ethnic
groups that are not explained by clinical characteristics raise concerns.
These remaining differences may result from unmeasured aspects of patients’
clinical characteristics or other issues such as differences in patient preferences,
communication patterns between clinicians and patients, or clinician/institutional
bias, which may influence patterns of care.
There are several study design issues to consider in interpreting these
findings. We used a national sample of hospitals and thus had diverse groups
of patients, enabling us to assess differences across multiple racial and
ethnic groups. However, racial/ethnic groups are heterogeneous, and the assignment
of patients to these groups is inherently imperfect.28 We
used race/ethnicity codes recorded by admissions or triage staff as the patient
was registered, and these may or may not reflect the patients’ self-identified
race/ethnicity.29- 33 Nevertheless,
to the degree that these reflect how clinicians view a patient’s race
and ethnicity, this approach is reasonable for understanding potential differential
treatment based on perceived racial and ethnic groups. In addition, we were
unable to obtain information on other patient socioeconomic factors that might
vary by race and ethnicity, such as education, occupation, and income. We
did, however, adjust for insurance status, an important aspect of socioeconomic
status in this context. Furthermore, we did not assess differences in patient
preferences or clinician bias, which others have suggested might account for
racial and ethnic differences in treatment.3,34- 36 Finally,
our sample was drawn from hospitals that participated in the NRMI, which tend
to have greater AMI volume, are more likely to be nonprofit, and are larger
than hospitals that do not participate in the NRMI; furthermore, time-to-treatment
data are self-reported by hospitals. However, the NRMI provides the only national,
longitudinal data on the clinical characteristics and time to treatment for
patients with AMI, and we believe these issues are unlikely to have substantially
affected our findings.
Our study has important implications for efforts to eliminate disparities
in time to acute reperfusion. Although efforts to increase awareness of racial/ethnic
disparities inside the hospital are important, our findings suggest the need
for parallel efforts directed toward improving the care in hospitals that
are lagging in their quality and in which minority patients may be more likely
to receive their care. Experts have identified possible explanations for the
racial and ethnic disparities in cardiovascular care,1,3,37- 39 including
differences in clinical presentation, access to care, and clinician bias.
We conclude that the genesis of racial/ethnic disparities in time to
treatment is complex, involving differences in the hospitals accessed by minority
groups as well as differential treatment inside the hospital. Interventions
to eliminate racial/ethnic disparities are likely to fall short of their goals
unless they are accompanied by systemic changes that can ensure all patients
have access to high-quality hospitals.
Corresponding Author: Harlan M. Krumholz,
MD, SM, Yale University School of Medicine, 333 Cedar St, PO Box 208088, New
Haven, CT 06520-8088 (email@example.com).
Author Contributions: The authors had full
access to all of the data in the study and take responsibility for the integrity
of the data and the accuracy of the data analyses.
Study concept and design: Bradley, Herrin,
Wang,McNamara, Webster, Blaney, Krumholz.
Acquisition of data: Wang, Blaney, Peterson.
Analysis and interpretation of data: Bradley,
Herrin, Wang, Magid, Canto, Pollack.
Drafting of the manuscript: Bradley, Herrin.
Critical revision of the manuscript for important
intellectual content: Bradley, Wang, McNamara,Webster, Magid, Blaney,
Peterson, Canto, Pollack,Krumholz.
Statistical analysis: Bradley, Herrin, Wang,Krumholz.
Administrative, technical, or material support:Webster,
Blaney, Canto, Pollack.
Funding/Support: This study was supported by
National Heart, Lung, and Blood Institute (NHLBI) grant R01HS10407-01. Dr
Bradley is supported by the Patrick & Catherine Weldon Donaghue Medical
Research Foundation (grant 02-102) and by a grant from the Claude D. Pepper
Older Americans Independence Center at Yale University (P30AG21342).
Role of the Sponsors: Neither the NHLBI nor
the Patrick & Catherine Weldon Donaghue Medical Research Foundation had
any involvement in the design or conduct of the study; data management or
analysis; or manuscript preparation, review, or authorization for submission.
Genentech Inc, South San Francisco, Calif, provided access to the data without