Racial and Ethnic Differences in Bystander CPR for Witnessed Cardiac Arrest

Data Source

The Cardiac Arrest Registry to Enhance Survival (CARES) is a prospective, multicenter registry of persons who have had an out-of-hospital cardiac arrest in the United States, with a current catchment area that includes approximately 167 million residents, which represents 51% of the U.S. population. The registry was established by the Centers for Disease Control and Prevention and Emory University, and has been previously described^14,15 (details are provided in the Supplementary Appendix, available with the full text of this article at NEJM.org). The registry includes all persons with a nontraumatic (i.e., not caused by a trauma) out-of-hospital cardiac arrest for whom resuscitation was attempted and who were identified by emergency medical service (EMS) agencies. Standardized international Utstein definitions for reporting clinical variables and outcomes associated with cardiac arrest were used to ensure the uniformity of the data included in the registry.¹⁶ The study was approved by the institutional review board of Saint Luke’s Hospital, which waived the requirement for informed consent because the study involved deidentified data. The fifth, sixth, and last authors vouch for the accuracy and completeness of the data in this report; author contributions are listed in the Supplementary Appendix.

Study Population

Between January 1, 2013, and December 31, 2019, we identified 460,827 persons with a nontraumatic out-of-hospital cardiac arrest. We were interested in adults with witnessed out-of-hospital cardiac arrest, so we excluded 222,795 unwitnessed arrests and 12,739 pediatric arrests (Fig. S1 in the Supplementary Appendix). Also excluded were 56,272 persons whose arrests were witnessed by EMS personnel (i.e., there was no opportunity for a layperson bystander to provide CPR) and 22,899 persons with arrests that occurred at a nursing home or health care facility (since these locations had on-site health care professionals). In addition, we excluded 30,559 cardiac arrests in persons with unknown or missing information on race or ethnic group and 4590 arrests that occurred in persons of other races (4018 Asians and 572 Native Americans or Alaska Natives) in order to focus the comparison on the differences in out-of-hospital cardiac arrest between Black or Hispanic persons and White persons. We further excluded 47 arrests in which there was missing information on bystander CPR and 872 arrests that were not linked to census-tract data. Our final study cohort consisted of 110,054 witnessed out-of-hospital cardiac arrests that were reported by 1614 EMS agencies.

Study Outcomes

The primary outcome was the initiation of bystander CPR, defined as CPR initiated by any layperson (family member, medical provider, or other person) who was not a 911 responder (fire, police, or EMS employee). The independent variable was race or ethnic group (Black or Hispanic vs. non-Hispanic White). For cases included in CARES, race and ethnic group are reported by persons who had a cardiac arrest or their family members, whenever possible, or are reported by EMS personnel when the person dies during resuscitation and no family member or acquaintance is available to provide race or ethnic-group information.

We analyzed the incidence of bystander CPR according to the race or ethnic group of persons who had out-of-hospital cardiac arrests that occurred at home and in public locations. Analyses were further stratified according to the racial or ethnic makeup and the income composition of the neighborhood in which the arrest occurred. The address of each out-of-hospital cardiac arrest that was included in CARES was geocoded to a U.S. census tract. Census tracts were used as proxies for neighborhoods because they typically represent economically and socially homogeneous groups of approximately 1200 to 8000 residents.¹⁷ Neighborhood-level information on racial and ethnic makeup and income were linked to each geocoded address with data from the 2019 American Community Survey.¹⁸ Using previously gathered data regarding the distribution of the racial composition of census tracts included in CARES,¹⁹ we categorized neighborhoods as predominantly White (>80% White), majority Black or Hispanic (>50% Black or Hispanic), or integrated. Integrated neighborhoods were those that did not meet the criteria for a predominantly White or majority Black or Hispanic neighborhood. Neighborhoods were also classified as high-income (median annual household income, >$80,000), middle-income ($40,000 to $80,000), or low-income (<$40,000).

Statistical Analysis

Owing to the large sample size, characteristics of Black or Hispanic persons and White persons at baseline were compared with the use of standardized differences, in which a standardized absolute difference of more than 10 percentage points was considered clinically meaningful.²⁰

To assess for racial and ethnic differences in the incidence of bystander CPR, multivariable hierarchical logistic regression models were constructed separately for out-of-hospital cardiac arrests that occurred at home and those that occurred in public locations. Besides race and ethnic group, these models adjusted for the age and sex of the person who had a cardiac arrest, the calendar year of arrest, the cause of the arrest (presumed cardiac, respiratory, or other), and urbanicity (according to U.S. census urban–rural tract classification: urbanized [≥50,000 residents], urban cluster [nonurbanized areas, ≥2500 residents]; or rural [<2500 residents])²¹ as fixed effects. Because an EMS agency might have worked in more than one census tract, each combination of EMS agency and census tract was modeled as a unique random effect to account for clustering of patient outcomes within the site. In all models, the effect of race was categorized according to between-cluster and within-cluster effects, with the latter representing the association between the race or ethnic group of a person who had an arrest and the likelihood of bystander CPR within an individual neighborhood.

To examine whether racial and ethnic differences in bystander CPR were explained by neighborhood factors, we repeated the above analyses of out-of-hospital cardiac arrests that occurred at home and in public locations for each neighborhood racial or ethnic-group designation and each income strata. In addition, we examined the number of Black or Hispanic persons as compared with the number of White persons for survival to hospital discharge and for favorable neurologic survival (survival with a discharge Cerebral Performance Category score of 1 or 2 out of 5, in which 1 denotes no-to-mild neurological disability and 2 denotes moderate disability) after an out-of-hospital cardiac arrest. The analyses for survival to hospital discharge and favorable neurologic survival initially were adjusted for the same variables that were used for the outcome of bystander CPR. The analyses were further adjusted for the presence or absence of bystander CPR and the cardiac-arrest rhythm that was initially detected (since this variable may be influenced by receipt of bystander CPR). To account for potential bias owing to missing data regarding race or ethnic group, we used inverse probability weighting to generate all model estimates.

Finally, we evaluated whether racial or ethnic differences in bystander CPR were present in different public locations (i.e., workplace settings [commercial or industrial building], street or highway, recreational facility, public transportation center [e.g., airport or bus terminal], or other), since the number of potential bystanders and their familiarity with the person having an out-of-hospital cardiac arrest would differ according to the location. We constructed a hierarchical model for arrests in a public location and adjusted for the age and sex of the person who had the arrest, calendar year, the race or ethnic group of the person, the cause of the arrest (i.e., cardiac, respiratory, other), urbanicity, public location category, neighborhood racial and ethnic makeup, and neighborhood income.

Because we did not prespecify that there would be correction for multiplicity when conducting tests, results are reported as point estimates and 95% confidence intervals. The widths of the confidence intervals have not been adjusted for multiplicity, so the intervals should not be used to infer definitive associations. All analyses were performed with SAS software, version 9.4 (SAS Institute).