General Emergency Medicine · CU Anschutz Medical Campus
Methods
We conducted a retrospective cohort study using CARES data collected from January 1, 2013 to December 31, 2019 (n=457 621). Since location (public and private) may impact biases and willingness to perform B‐CPR, we examined differences in B‐CPR rates and outcomes by sex and stratified by location (public and private). To test our hypothesis that receipt of B‐CPR differed by sex within predominantly Black neighborhoods, we stratified our model by neighborhood composition. We examined adult, nontraumatic cardiac arrest events that occurred in the out‐of‐hospital setting. Because the data were deidentified, the Duke University Institutional Review Board determined this study to be exempt from requirements for informed consent.CARES is a US‐based prospective registry of OHCA with a catchment area of ≈175 million people in 47 states across the United States. The registry includes all nontraumatic OHCAs where resuscitation was attempted by a 9‐1‐1 responder, including cardiopulmonary resuscitation (CPR), defibrillation, or both. Data were collected via EMS run sheets and hospital records. Each CARES record is reviewed for accuracy through a computer audit algorithm and routinely checked by EMS and hospital users as well as a CARES administrator.18 These data have been used in multiple publications in high‐impact journals.19, 20, 21To account for neighborhood characteristics, we used census tract estimates from the American Community Survey conducted by the US Census Bureau.22 Variables collected include census tract‐level median age, proportion race, proportion ethnicity, median education, and median income.The data that support the findings of this study are available from the corresponding author on reasonable request.Consistent with our prior work, we defined B‐CPR as delivery from a bystander, excluding CPR from law enforcement or EMS first responders. We excluded pediatric patients (aged <18 years) and those who experienced OHCA from traumatic injury. We also excluded arrest events that occurred in a residential institution (eg, skilled nursing facility) or health care center and arrest events that were witnessed by EMS. We defined AED application as an AED applied by either a bystander or a first responder before EMS arrival. For the purposes of this variable, a bystander was defined as any layperson, whereas a first responder (non‐EMS) was defined as personnel who responded in an official capacity as part of the organized medical team but was not the designated transporter of the patient to the hospital. Race and ethnicity, which is determined by the EMS provider, were modeled categorically (ie, White, Black, Hispanic, and other). The other category included American Indian or Alaska Native, Asian, and Native Hawaiian or Pacific Islander. Age was modeled as a continuous variable. Location of cardiac arrest was defined as private (ie, home) or public (ie, street/highway, public building, place of recreation, or other public location). Event time of day was grouped on the basis of assumed daily activity (6:00 am–8:59 am, 9:00 am–3:59 pm, 4:00 pm–6:59 pm, 7:00 pm–10:59 pm, and 11:00 pm–5:59 am). Response time was categorized as >6 or <6 minutes. Sex was modeled as a binary variable (men/women).
Census tract characteristics were included from the American Community Survey and added into the CARES data set merged by census tracts. Each OHCA event location was geocoded and then mapped to the corresponding census tract. In some instances, records were missing census tract observations when the census tracts did not meet the population threshold to provide statistically reliable estimates of the population. In addition, given the focus on neighborhood census tract data and sex, we excluded cases where the contribution to the census tract data was considered sparse (<5 cases). There was no bias between the distributions of sparse cases that occurred in public or private locations. We used the continuous variable of proportion race or proportion ethnicity to create a categorical variable classifying neighborhood census tracts by >30% individuals residing in a census tract. The 30% cutoff was based on the definition of White flight, where White residents leave an area when sizes of minority populations increase.23, 24 A follow‐up survey also demonstrated that White individuals perceive mixed race communities as less desirable.25 This was confirmed through subsequent work.26 Given this, we used a 30% cutoff. We considered examining higher proportions for race and ethnicity, but after consultation, a biostatistician determined we did not have adequate sample size. We focused our analysis on racial and ethnic groups that have historically experienced prejudice or inequity in the United States, specifically Black and Hispanic populations. We also included a category of >30% White individuals. Similar to the approach taken in prior publications, if the proportion of White, Black, or Hispanic residents was not >30%, we classified the neighborhood as integrated.19 Sex was recorded as clinician‐identified sex (man or woman). Final categories included >30% Black, >30% Hispanic, >30% White, and integrated residents.
We examined the odds of receipt of B‐CPR by sex stratified by location considering neighborhood‐level race or ethnicity. We hypothesized that women would be less likely to receive B‐CPR in predominantly Black neighborhoods compared with men. As secondary analyses, we examined AED application (including bystanders and first responders) and survival to hospital discharge.
We analyzed the data using STATA 17.1 (StataCorp, College Station, TX). Using logistic regression modeling, we analyzed whether there were differences in B‐CPR rates by patient sex. Given that bystanders may respond to arrests differently in public compared with private locations, we stratified the analysis by location to account for potential biases in bystander response. We included year in the model to account for potential changes in OHCA trends from 2013 to 2019. We constructed models for the odds of overall B‐CPR delivery by patient sex. We constructed the models accounting for neighborhood (census tract level) and data reporting from the EMS agency (EMS agency) as random effects. Individual‐ and neighborhood‐level variables with P<0.15 for association in a univariate analysis were entered into a multivariable model as fixed effects, stratified by location. Variables were also included on the basis of clinical criteria. To account for the nonlinearity of age and median household income in all models, the polynomials of each variable were added in the equation. We tested the interaction of sex and neighborhood race and ethnicity. The final multivariable regression model included B‐CPR (primary outcome), sex (exposure), neighborhood race and ethnicity, EMS agency, time of arrest, year, demographics (age, race, and ethnicity), arrest witness status, and neighborhood‐level socioeconomic status defined by median household income. To answer our primary hypothesis of whether there is difference in receipt of B‐CPR by sex within public locations and predominantly Black neighborhoods, we stratified our model by neighborhood composition. As a secondary analysis, we examined the univariate and multivariable models for AED application and survival to hospital discharge stratified by location. The final multivariable model for survival to hospital discharge included survival to hospital discharge (outcome, yes/no), sex (exposure), B‐CPR, AED application, neighborhood race and ethnicity, time of event, year, demographics (age, race, and ethnicity), witness status, response time, and neighborhood‐level socioeconomic status defined by median household income. Given the large amount of missingness for response time, the survival to hospital discharge model was displayed using multiple imputation.
We conducted a sensitivity analysis imputing the missing patient race categories assuming that the data are missing at random. Variables were imputed and then rounded, as appropriate, before examining final estimations. We also conducted a sensitivity analysis including neighborhood‐level education in all models.
Results
From 2013 to 2019, the CARES contained 457 621 cardiac arrest events. Given the study question surrounding B‐CPR and AED application by neighborhood, any case that was pediatric (<18 years of age), occurred in a health care facility or nursing home, was EMS witnessed, had missing census tract data, or had sparse census tracts was excluded (Figure). The figure details the missing data, which comprised 4% of the population. Given the low percentage of missing data, methods to address missing data were not needed for the primary outcome of interest (B‐CPR). Of these, 309 662 events were included in the final analysis.
EMS indicates emergency medical services.
Within this cohort, the mean±SD age was 62±17 years, 111 758 of 309 622 (36%) were women, and 64 372 of 309 622 (21%) were Black patients. In addition, 121 047 of 309 622 (39%) received B‐CPR, 21 931 of 56 461 (39%) received AED application in public locations, and 30 718 of 309 622 (10%) survived to hospital discharge. Overall, 56 461 of 309 622 (18%) arrests occurred in public locations, 63 649 of 309 622 (20%) arrests occurred in predominantly Black neighborhoods, and 58 218 of 309 622 (19%) arrests occurred in predominantly Hispanic neighborhoods, whereas 178 865 of 309 822 (58%) occurred in predominantly White neighborhoods (Table 1).
Study Demographic Characteristics Overall and Stratified by Location
Data are given as number (percentage) unless otherwise indicated. Missing data: B‐CPR, 13 (0%); *AED application by a bystander or first responder before arrival of emergency medical services, 6 (0%); survival to hospital discharge, 823 (0%); location type, 5 (0%); time of arrest, 114 396 (37%); response time, 167 773 (54%); age, 177 (0%); sex, 18 (0%); race, 133 (0%); ROSC, 92 (0%); first arrest rhythm, 31 (0%); and CPC score, 279 427 (90%). AED indicates automatic external defibrillator; B‐CPR, bystander cardiopulmonary resuscitation; CPC, cerebral performance category; and ROSC, return of spontaneous circulation.
Forty‐three percent of women who had an arrest in public received B‐CPR, whereas 47% of men who had an arrest in public received B‐CPR (P<0.01; Table 1). Receipt of B‐CPR varied by neighborhood composition in public locations (P<0.01). Among those who had an arrest in public, 36% of women had an AED applied, whereas 40% of men had an AED applied (P<0.01; Table 1).
Before examining our data in an adjusted model, we examined frequencies of our distributions of receipt of B‐CPR by sex and neighborhood categories. In public, women were less likely to receive B‐CPR compared with men in predominantly Black neighborhoods (32% versus 35%), predominantly Hispanic neighborhoods (41% versus 44%), and predominantly White neighborhoods (47% versus 51%) (Table 2).
Differences in Receipt of B‐CPR by Men and Women in Diverse Neighborhoods
Data are given as number (percentage). B‐CPR indicates bystander cardiopulmonary resuscitation.
P<0.05.
AED application varied by neighborhood composition (P<0.01). In public, women were less likely to receive AED application compared with men in predominantly Hispanic neighborhoods (28% versus 32%) and predominantly White neighborhoods (37% versus 42%) (Table S1). In public, women were less likely to survive to hospital discharge compared with men in predominantly White neighborhoods (22% versus 23%; Table S2).
In the home, women were more likely to receive B‐CPR in predominantly White neighborhoods compared with men (adjusted odds ratio [aOR], 1.05 [95% CI, 1.02–1.08]; Table 3). Women were more likely to receive B‐CPR in predominantly Hispanic neighborhoods compared with men (aOR, 1.06 [95% CI, 1.01–1.12]). Women were less likely than men to receive AED application in the home within predominantly Black neighborhoods (aOR, 0.93 [95% CI, 0.88–0.98]), predominantly Hispanic neighborhoods (aOR, 0.91 [95% CI, 0.85–0.98]), and predominantly White neighborhoods (aOR, 0.90 [95% CI, 0.87–0.93]) (Table 3). Women were more likely to survive to hospital discharge compared with men in predominantly White and predominantly Black neighborhoods (Table 3).
Association of B‐CPR, AED Application, and Survival to Hospital Discharge With Sex Stratified by Location and Neighborhood
n=118 429
1.05 (1.02–1.08)
n=27 674
0.80 (0.75–0.85)
n=35 857
1.03 (0.98–1.08)
NS
n=7217
0.87 (0.76–0.98)
0.03
n=27 075
1.06 (1.01–1.12)
0.02
n=6182
0.83 (0.73–0.96)
0.01
n=2132
1.04 (0.84–1.29)
NS
n=526
0.87 (0.51–1.48)
NS
n=118 431
0.90 (0.87–0.93)
n=27 677
0.78 (0.73–0.84)
n=35 857
0.93 (0.88–0.98)
n=7217
0.88 (0.78–0.99)
n=27 075
0.91 (0.85–0.98)
n=6182
0.74 (0.64–0.87)
n=2132
0.99 (0.74–1.35)
n=526
0.98 (0.52–1.86)
n=118 198
1.05 (1.01–1.10)
n=27 509
0.93 (0.87–0.99)
n=35 797
1.13 (1.03–1.23)
n=7175
1.03 (0.89–1.19)
n=27 022
1.02 (0.93–1.12)
n=6141
0.97 (0.83–1.14)
n=2128
1.04 (0.74–1.45)
n=520
1.23 (0.68–2.23)
B‐CPR and AED model is controlling for age, race, arrest witnessed status, year of arrest, median household income, and time of arrest. Survival to hospital discharge model is controlling for B‐CPR, AED application, age, race, time of event, response time, witness status, and median household income. AED indicates automatic external defibrillator; aOR, adjusted odds ratio; B‐CPR, bystander cardiopulmonary resuscitation; and NS, not significant.
AED application by a bystander or first responder before arrival of emergency medical services. Loss of sample size in models because of missing data in time of arrest and response time variables.
We reported the multiple imputation analysis of the survival to hospital discharge model because of loss of sample size.
In public locations, women who arrested in predominantly White neighborhoods had a 0.80 (95% CI, 0.75–0.85) decreased adjusted odds of receiving B‐CPR compared with men in predominantly White neighborhoods, while controlling for confounding variables. Women in predominantly Black neighborhoods had a 0.87 (95% CI, 0.76–0.98) decreased adjusted odds of receiving B‐CPR compared with men (Table 3). Similarly, women in predominantly Hispanic neighborhoods had a 0.83 (95% CI, 0.73–0.96) adjusted odds of receiving B‐CPR compared with men. These differences persisted when examining AED application in public locations, with women being less likely than men to receive AED application in predominantly White neighborhoods (aOR, 0.78 [95% CI, 0.73–0.84]) and predominantly Hispanic neighborhoods (aOR, 0.74 [95% CI, 0.64–0.87]) (Table 3). While controlling for process variables, including B‐CPR among arrests that occurred in public locations, there was no significant difference in sex and neighborhood location on survival to hospital discharge (Table 3). These differences did not change when using multiple imputation to address missing data in race and ethnicity (Table S3) or adding neighborhood‐level income (Table S4).
An interaction of sex and neighborhood race and ethnicity was not significant when included in the model for B‐CPR, AED application, or survival to hospital discharge.
Discussion
In this study, we found that women with OHCA in public locations have a decreased odds of receiving B‐CPR and AED application; these findings did not differ significantly by neighborhood composition. As reflected in previous studies, the findings from this study further confirm that women are less likely than men to receive B‐CPR in public locations.11, 27 This sex disparity has been observed in the United States as well as in other international countries.28, 29 Building on previous findings, we added the component of considering underrepresented minority racial and ethnic neighborhoods and found that sex inequity persists, but there is no effect modification by neighborhood composition. This may be partially attributable to the use of the concept of White flight without incorporating a temporally sensitive component to the neighborhood variable. Regardless, we saw poorer receipt of B‐CPR for women compared with men in predominantly Black neighborhoods, which highlights the importance of this issue. These findings have implications for resuscitation intervention implementation and education efforts, especially considering methods to address health disparities.
There was a lack of receipt of B‐CPR among women in public, and it was seen within all neighborhood compositions. This is important to consider from a CPR training perspective. Studies have demonstrated the ability to geospatially map areas with higher proportions of Black and Hispanic patients to target high‐risk areas with CPR training.30, 31, 32 The TAKE10 initiative, conducted in Texas, demonstrated the feasibility of this approach.33 Additional considerations may be made to consider how to integrate targeted training around performing CPR on women within these geographic efforts. Another approach may be to quantitatively consider creating an overall metric to assess community‐level inequity and effects of training interventions on B‐CPR and survival to hospital discharge from OHCA. When considering interventions, it is important to also consider the effect on both B‐CPR and survival to hospital discharge, as a recent publication from North Carolina found that community OHCA interventions improved B‐CPR in both Black and White populations, but survival to hospital discharge only increased significantly in the White population.34 Future work may consider testing these intervention strategies to increase CPR training for these high‐risk communities while considering training efforts to address B‐CPR on women across neighborhood compositions.
Furthermore, we acknowledge that there may be intersectionality with patient sex and race, ethnicity, and neighborhood characteristics. This is an area that warrants future investigation with a larger sample size.
It is important to consider whether additional language should be incorporated into the B‐CPR training curriculum to remove potential biases or fears of performing B‐CPR in public on women. A recent study asked members of the public reasons why individuals may fear performing B‐CPR in public on women, with reasons cited including fear of inappropriately touching the woman or harming her.35 Tailoring CPR training curriculum to address some of these issues may help reduce known inequities. Furthermore, the American Heart Association is looking specifically at these sex‐based issues and has included additional materials in courses to address these barriers. For example, in the 2020 CPR Guidelines section, the American Heart Association is trying to resolve some of the known concerns (eg, taking bra on or off before providing compressions, defibrillation, or both).36, 37 It is both the science and the training that must support changing this curriculum and seeing what type of impact that has on different audiences.
Given the known inequities in OHCA by race and ethnicity and differences in receipt of B‐CPR by sex, it is also worth considering whether additional accommodations should be made to ensure that the curriculum tailored to addressing barriers to B‐CPR on women is linguistically appropriate and culturally competent. To support this notion, studies have demonstrated a need for better access to CPR training curriculums in a variety of languages.38, 39 The High Arrest Neighborhoods to Decrease Disparities in Survival (HANDDS) program, conducted in Denver, CO, developed a curriculum to understand barriers to learning and performing CPR and implementing train‐the‐trainer programs.40
Another critical link in the chain of survival is the provision of CPR by 9‐1‐1 dispatchers to a bystander over the telephone, referred to as dispatch‐assisted CPR or telephone CPR. There may be opportunity to consider methods to address known inequities through educating the dispatchers. A recent publication from Korea noted that telephone CPR attenuated the sex disparity in provision of B‐CPR.41 Future work may consider simulation studies, which could be used to evaluate different methods of coaching as well as protocol modification to consider some of the barriers, such as women being less likely to be identified as having a cardiac arrest and more likely to have passed out or fainted.35 The role of the dispatcher represents a critical area for further investigation.
There are limitations with registry data, specifically selection bias, and possible missing data. We acknowledge these limitations but believe that the sensitivity analyses strengthen these findings. Given the coverage of the data, these findings are generalizable to the US population. The definition of White flight was used to motivate the neighborhood categorical assignments, but the current variable does not fully encapsulate the temporal and spatial dynamics of White flight. Further analysis is warranted to fully assess the impact of this demographic phenomenon on outcomes and processes within the cardiac resuscitation literature. Another limitation is our inability to discern the underlying barriers and factors associated with these disparities, and ability to measure implicit bias. Additional research, including qualitative work, is needed to further understand these limitations.
Conclusions
Women with an arrest in public locations have a decreased likelihood of receiving B‐CPR and AED application. These findings were consistent within predominantly White, Black, and Hispanic neighborhoods, with no evidence of effect modification by neighborhood composition. These differences were not seen when examining survival to hospital discharge. This information has implications for considering targeted education strategies and future public health community programming to reduce disparities around bystander response to OHCA.
Sources of Funding
Dr Blewer is funded as a Building Interdisciplinary Research Careers in Women's Health scholar through a National Institutes of Health National Institute of Arthritis and Musculoskeletal and Skin Diseases grant (K12AR084231).
Disclosures
Dr Blewer is funded through a National Institutes of Health (NIH) National Institute of Arthritis and Musculoskeletal and Skin Diseases scholar award (K12AR084231), NIH National Heart, Lung, and Blood Institute, an American Heart Association research grant, and a Laerdal Foundation grant. Dr Ong reports grants from the Laerdal Foundation, Laerdal Medical, and Ramsey Social Justice Foundation for funding of the Pan‐Asian Resuscitation Outcomes Study. Dr Sasson is employed by the American Heart Association. Dr. McNally and Ms. Al‐Araji declare funding through a Centers for Disease Control and Prevention CARES Expansion and Modernization Grant. The remaining authors have no disclosures to report.
Supporting information
Tables S1–S4
CS
Dr. Comilla Sasson
Emergency Medicine · General Emergency MedicineUniversity of Colorado Anschutzcomilla