Treatment and Outcome Variation in Out-of-hospital Cardiac Arrest Among Four Urban Hospitals in Detroit

Abstract

Aims

To determine whether out-of-hospital cardiac arrest (OHCA) post-resuscitation management and outcomes differ between four Detroit hospitals.

Introduction

Significant variation exists in treatment/outcomes from OHCA. Disparities between hospitals serving a similar population is not well known.

Methods

Retrospective OHCA data was collected from the Detroit-Cardiac Arrest Registry (DCAR) between January 2014 to December 2019. Four hospitals were compared on two treatments (angiography, do not resuscitate (DNR)) and two outcomes (cerebral performance category (CPC) ≤ 2, in-hospital death). Models for death and CPC were tested with and without coronary angiography and DNR status.

Results

999 patients at hospitals A - D differed (p<0.05) before multivariable adjustment by age, race, witnessed arrest, dispatch-emergency department (ED) time, TTM, coronary angiography, DNR order, and in-hospital death. Rates of death and CPC ≤ 2 were worse in Hospital A (82.8%, 10%, respectively) compared to others (69.1%, 14.1%). After multivariable adjustment, Hospital A performed angiography less compared to B (OR=0.17) and was more likely to initiate new DNR status than B (OR=2.9), C (OR=16.1), or D (OR=3.6). CPC ≤ 2 were worse in Hospital A compared to B (OR=0.27) and D (OR=0.35). After sensitivity analysis, CPC ≤ 2 odds did not differ for A versus B (OR=0.58, adjusted for angiography) or D (OR=0.65, adjusted for DNR). Odds of death, despite angiography and DNR differences, were worse in Hospital A compared to B (OR=1.87) and D (OR=1.81).

Conclusion

Differing rates of DNR and coronary angiography was associated with observed disparities in favorable neurologic outcome, but not death, between four Detroit hospitals.

Background

Out-of-hospital cardiac arrests (OHCA) occur in over 350,000 individuals in the United States annually ¹. Survival from OHCA has remained consistently low in the United States for years, at 7% ^{2, 3}. Prior studies have shown marked variation in intervention and outcomes in OHCA when comparing large geographic divisions such as cities, counties, and states ⁴. Nearly twenty years ago, Detroit, MI had a survival rate of 0.2% whereas Seattle, WA had a survival rate of 16.0% ^{5, 6}. Variation in OHCA survival rates can be attributed to patient characteristics, event-related circumstances, and a host of factors involved in pre-hospital or post-resuscitation care ^6–8. Prior studies have evaluated survival differences as a function of “early” chain of survival factors (e.g. early bystander cardiopulmonary resuscitation (CPR), automatic external defibrillator (AED) accessibility, emergency medical services (EMS) response time), with a lesser focus on “late” chain of survival interventions (coronary artery angiography (CAG), withdrawal of care decisions, etc.) ⁹

Since the 2002 report of a 99.8% chance of fatality from a cardiac arrest in Detroit, MI, the city has made significant strides to improve OHCA care and outcomes ^{5, 10}. Some of the efforts that increased OHCA survival rates include community bystander CPR training in targeted areas, dispatch-assisted CPR, and the implementation of non-transporting fire units in a medical capacity. SaveMIHeart, a program to improve sudden cardiac arrest survival in Michigan communities, and the Cardiac Arrest Registry to Enhance Survival (CARES) are two pivotal examples of recent efforts to improve OHCA care and outcomes in recent time ¹¹. The data from the Detroit Cardiac Arrest Registry (D-CAR) and CARES have been utilized to monitor survival trends between since 2014, the effect of the COVID-19 pandemic on OHCA in Detroit, and other trends in monthly arrest rates, bystander CPR, hospital interventions, and annual survival outcomes ^{10, 12}.

Unlike OHCA outcome and treatment variation observed between large geographic areas (e.g. city, county, state, country), disparities within a single city tertiary hospital systems remain unstudied ^{8, 13–17}. We hypothesized that OHCA outcomes would differ significantly between tertiary care hospitals with overlapping patient populations, and that these disparities would, in turn, be at least partially explained by differences in “late” phase treatment decisions.

Methods

The study was reviewed by the Wayne State University Institutional Review Board (IRB) as exempt. Detroit Cardiac Arrest Registry (D-CAR) encompasses elements of the Cardiac Arrest Registry to Enhance Survival (CARES). Patient encounters were analyzed from the American Heart Association’s CARES, which encompasses over 2,500 hospitals and 2,300 EMS agencies from 31 participating states in the United States ^{18, 19}.

Inclusion Criteria

Patients in CARES from January 1st, 2014 to December 31st, 2019 presenting to one of four tertiary care hospitals in the city of Detroit were included. This time period was chosen because the CARES data in these years are complete and prior published work has shown that EMS characteristics underwent temporary changes in 2020 due to the COVID-19 pandemic ¹². Patients were excluded if they were <18 years old, the OHCA cause was trauma, resuscitation was not initiated, they had a pre-existing DNR order, they were transferred to another facility, or they died prior to hospital admission.

Registry Data Management

OHCA with a resuscitation attempt are uploaded from the Detroit EMS database into CARES. The Detroit CARES coordinator conducts a quality check on each record to ensure the record is accurately transferred from the EMS database into CARES. If any values are not properly transferred from the EMS database into CARES, the coordinator will check the EMS database for the correct value and manually enter it into CARES. The coordinator manually enters any missing records or missing values. Data are reviewed by the site CARES coordinator and by the physicians leading the project and performing the analysis.

City and Pre-Hospital Characteristics

Detroit, Michigan is a major city in the United States. 78% of the Detroit population is African-American ²⁰. 1/3 of Detroit residents have 3+ co-morbidities (which include, but are not limited to, heart disease, cancer, diabetes, and stroke) and 36% of the residents live in poverty ^21–23. Detroit falls below the national average on multiple social determinants of health, but these effects are distributed in a fairly homogenous pattern throughout the city its EMS service area ²⁴. Detroit EMS is the main service providing medical care and transporting patients to hospitals within the city. During the study period, the Detroit Fire Department utilized 6 non-transporting squads and 27 fire engines. These units have been licensed at the medical first-responder level since 2014. There are 27 Basic Life Support (BLS) ambulances and 9 Advanced Life Support (ALS) ambulances operated by the fire department as well. In addition, there are 8 ambulances from 4 private companies stipulated by the city to distribute coverage during peak hours. The typical EMS response for a cardiac arrest includes a medical first responder engine and the closest available ambulance. The first arriving unit could be either depending on their availability. ALS units are not preferentially dispatched due to the limited number available and the large size and low population density of the city. Detroit contains 142.9 square miles, equivalent to >3 times the size of San Francisco despite a ~25% smaller population. The closest available unit is sent to every call.

Hospital Characteristics

There are four tertiary care hospitals that reside within Detroit city lines under the umbrella of 3 hospital systems, separated by a median 10.3 miles. The names of each hospital were de-identified for this report and labeled hospital A-D in no particular order. All four are teaching hospitals are affiliated with an academic institution, have robust residency and fellowship programs and each hospital’s emergency department sees over 70,000 patient visits annually. Three hospitals are level 1 trauma centers and one hospital is a level 2 trauma center. All four hospitals are designated as ST-elevation myocardial infarction (STEMI) and stroke centers by the local medical control authority. Each hospital is also equipped with the ability to perform targeted temperature management and has 24-hour angiographic capabilities. Angiography and coronary artery bypass grafting (CABG) are combined as the composite of revascularization procedures for all analyses. All references to angiography or CABG in the manuscript refer to both/either (i.e. revascularization). The hospitals in the study have similar resources – more information regarding hospital strain, staffing or finances are unable to be disclosed due to potential unblinding of a hospital’s identity.

Statistical Analysis

The Kruskal-Wallis and Fisher’s exact tests were utilized for descriptive comparison of hospitals A – D by EMS, patient, and clinical characteristics. All data were analyzed using the R statistical programming language (Rstudio V.1.2.5)²⁵.

Logistic regression was used to compare the hospitals by four variables of interest (response variables, i.e. 4 total regression models). These included two clinical outcomes (Cerebral Performance Category [CPC] ≤ 2; in-hospital death) and two “late” phase treatment decisions (coronary angiography, new initiation of DNR). The primary measure for comparison (i.e. independent variable) was Hospital A-D (as a nominal categorical variable). Hospital A-D was adjusted in each model for covariates chosen a priori for their prior published associations with OHCA clinical outcomes ¹². These included patient characteristics (age, race black/white/other, sex) and “early” phase/pre-hospital/EMS treatment characteristics (bystander CPR, arrest witnessed by bystander/witnessed by EMS/unwitnessed, initial rhythm shockable/unshockable, time from EMS dispatch to ED arrival, initiation of targeted temperature management). Adjusted odds ratios (aOR) with 95% confidence intervals (95% CI) were then calculated from the models.

Sensitivity Analysis

We performed a pre-planned sensitivity analysis on the models for death and CPC ≤ 2, under the assumption that disparities in the two “late” phase treatment decisions we studied could account for disparities in clinical outcomes. For instance, if one hospital had a higher propensity for initiating a new DNR status, that hospital’s death rate would be expected to increase (i.e. since the propensity for new DNR is expected to be related to the propensity for death). Thus, models for death and CPC outcome were tested with vs. without the addition of angiography and DNR initiation as additional covariates. Outcome disparities between hospitals were compared with vs. without these adjustments for treatment propensity, to evaluate the degree to which these disparities may or may not be related to the treatment differences. Requests for access to the study’s analytical code and data can be directed to the corresponding author.

Sample Size Considerations

The entire available sample from the registry meeting date and inclusion criteria was used. To accommodate ≥10 events per variable for logistic regression (to limit the chance of model overfitting), 110 events per outcome were needed (9 base model covariates, and 11 for the sensitivity analysis).

Results

A total of 4185 patients were transported by Detroit EMS as a non-traumatic OHCA to one of the four Detroit hospitals. Inclusion/exclusion characteristics are presented in Figure 1. A total of 999 patients were admitted into one of the four hospitals in Detroit between January 1^st, 2014 – December 31^st, 2019. 3186 patients were excluded from the analysis because they died prior to hospital admission – either pre-hospital or in the emergency department. Hospitals A – D were significantly different (p<0.05), before multivariable adjustment by age, race, witnessed/unwitnessed arrest, dispatch-ED time, TTM, angiography utilization, DNR initiation, and in-hospital death, but not CPC ≤ 2. Rates of death and CPC ≤ 2 (Figure 2) were least favorable in Hospital A (82.8%, 10%, respectively) compared to all others (69.1%, 14.1%). Table 1 displays the characteristics of admitted non-traumatic OHCA patients from the 4 Detroit hospitals.

Figure 1:

Flow diagram of study subjects

Figure 2:

Unadjusted OHCA outcomes per hospital

Table 1:

Demographic and Clinical Characteristics of Study Patients

Characteristic	Hospital A % (count) or median (IQR) [n=209]	Hospital B % (count) or median (IQR) [n=270]	Hospital C % (count) or median (IQR) [n=204]	Hospital D % (count) or median (IQR) [n=316]	p-value
Age	57 (45.5-67)	60 (51-68)	61 (50-69)	61 (49-71)	p=0.017
African American	71.3% (149)	72.2% (195)	73.5% (150)	85.1% (269)	p<0.001
White	13.4% (28)	18.1% (49)	10.8% (22)	6% (19)	p<0.001
Unknown Race	14.4% (30)	8.1% (22)	10.8% (22)	8.5% (27)	p=0.1
Other Race	1% (2)	1.5% (4)	4.9% (10)	0.3% (1)	p<0.001
Unwitnessed	43.5% (91)	38.9% (105)	33.3% (68	31% (98)	p=0.018
Witnessed by Bystander	38.8% (81)	42.2% (114)	39.2% (80)	46.5% (147)	p=0.243
Witnessed by EMS	17.7% (37)	18.9% (51)	27.5% (56)	22.5% (71)	p=0.062
Bystander CPR	34% (71)	30.4% (82)	25% (51)	30.4% (96)	p=0.258
Dispatch ED Time	28 (21.1-35.53)	29.35 (22.47-36.84)	26.41 (20-33)	26.08 (20-33.95)	p=0.005
CAG	7.7% (16)	33% (89)	10.8% (22)	13.6% (43)	p<0.001
TTM	22% (46) Yes	77% (208) Yes	29.9% (61) Yes	62.7% (198) Yes	p<0.001
DNR	52.6% (110)	33% (89)	7.4% (15)	27.5% (87)	p<0.001
CPC 1	7.2% (15)	10.7% (29)	9.3% (19)	11.1% (35)	p=0.47
CPC 2	2.9% (6)	3.3% (9)	2.5% (5	3.5% (11)	p=0.914
CPC 3	5.3% (11)	6.3% (17)	4.9% (10)	5.7% (18)	p=0.922
CPC 4	1.9% (4)	11.1% (30)	8.8% (18)	13.6% (43)	p<0.001
Death (CPC 5)	82.8% (173)	68.5% (185)	74.5% (152)	66.1% (209)	p<0.001
CPC < 3	10% (21	14.1% (38)	11.8% (24)	14.6% (46)	p=0.416

A. New DNR Initiation After Hospital Admission

Figure 3A presents the results of the adjusted analysis for between-hospital differences in new DNR initiation after hospital admission. Patients made DNR during hospitalization were older (OR = 1.34; 95% CI: 1.1–1.6). Patients were less likely to be made DNR at Hospital B (OR = 0.36; 95% CI: 0.23–0.56), C (OR = 0.06; 95% CI: 0.03–0.12), and D (OR = 0.28; 95% CI: 0.18–0.42) compared to hospital A and less likely to receive coronary angiography/CABG (OR = 0.63; 95% CI: 0.40–0.99) and TTM (OR = 0.62; 95% CI: 0.44–0.87).

Figure 3 – Multivariable-adjusted analyses of OHCA cohort for specified outcome measures.

Figure 3A: Multivariable-adjusted odds of initiating new DNR (Post-admission)

B. Coronary Angiography/CABG Performed

Figure 3B presents the results of the adjusted analysis for between-hospital differences in coronary angiography/CABG performed. Patients receiving coronary angiography/CABG were less likely to be female (OR = 0.65; 95% CI: 0.45–0.96) or have been made DNR (OR = 0.62; 95% CI: 0.39–0.98). Patients receiving coronary angiography/CABG were more likely to have had shockable rhythms (OR = 7.34; 95% CI: 4.80–11.2) and be at hospital B compared to hospital A (OR = 6.4; 95% CI: 3.3–12.5).

Figure 3B:

Multivariable-adjusted odds of utilizing Coronary Artery Angiography (CAG)

C. In-hospital Death

Figure 3C presents the results of the adjusted analysis for in-hospital death. Patients were less likely to die at hospital B (aOR = 0.53; 95% CI: 0.31–0.91) and D (aOR = 0.55; 95% CI: 0.33–0.91) in comparison to hospital A. Patients who died were more likely to have an EMS-witnessed arrest (aOR = 2.22; 95% CI: 1.13–4.37), less likely to have a shockable rhythm (OR = 0.53; 95% CI: 0.36–0.79), and more likely to be made DNR (aOR = 10.1; 95% CI: 5.9–17.1). After further adjusting for the rates of angiography and DNR (sensitivity analysis, Supplemental Figure 3C), the disparities in death remained similar for hospitals B (aOR = 0.53; 0.31–0.91) and D (aOR=0.55,0.31–0.91) compared to hospital A.

Figure 3C:

Multivariable-adjusted odds of In-Hospital Death, After Adjusting for Treatment Propensity

D. CPC ≤ 2 – Good (1) or Moderate (2) Neurological Outcome

Figure 3D presents the results of the adjusted analysis for favorable neurologic outcome (CPC ≤ 2). Patients with CPC ≤ 2 were younger (OR = 0.56; 95% CI: 0.42–0.74) and had a shorter time from dispatch to ED (OR = 0.73; 95% CI: 0.53–0.99). Patients with CPC ≤ 2 were less likely to have an unwitnessed arrest (OR = 0.45; 95% CI: 0.26–0.81) or have an EMS-witnessed arrest (OR = 0.14; 95% CI: 0.05–0.41) compared to bystander witnessed. Patients with CPC ≤ 2 were more likely to have undergone coronary angiography/CABG (OR = 1.94; 95% CI: 1.12–3.37) and received TTM (OR = 4.53; 95% CI: 2.79–7.35). Multivariable-adjusted odds of CPC ≤ 2 were worse in hospital A compared to hospital B (OR = 0.27, 0.14–0.53) and D (OR = 0.35, 0.19–0.66). After additional adjustment (sensitivity analysis, Supplemental Figure 3D) for the different treatment propensities observed (angiography use and DNR initiation), CPC ≤ 2 odds no longer differed for A versus B (0.58, 0.28–1.2) or D (0.65, 0.32–1.27).

Figure 3D:

Multivariable-adjusted odds of CPC ≤ 2, After Adjusting for Treatment Propensity

Discussion

The current study presents three novel findings. First, differences were found in “later” phase OHCA care at tertiary hospitals from different health systems, in initiation of DNR status and use of cardiac angiography. Second, despite the 4 hospitals having similar resources and a shared geography, significant differences existed in clinically important outcomes including death and neurologic function (CPC). As discussed in the introduction, it is well-described that differences in these outcomes exist between larger geographic areas. There has been some effort on the extent of defining differences in smaller geographies. One study examined OHCA mortality differences specifically between tertiary and non-tertiary hospitals in Copenhagen, Denmark ²⁶. Another study specifically examined TTM initiation amongst OHCA patients presenting to hospitals within the province of Southern Ontario²⁷. To our knowledge this is the first report to show outcome differences of tertiary-care hospitals within a city located in the U.S. served by the same EMS medical control authority. Third, the inconsistencies in initiation of DNR status and cardiac angiography at least partially accounts for the disparities between hospitals for neurologically intact survival (CPC≤2). Additionally, our results demonstrate that comparative disparities in OHCA care between large areas like cities and states may miss important differences at the hospital level.

More reports in OHCA focus on the “early” chain of survival – pre-hospital and ED care – than the “late” chain of survival. Girotra et. al. explained that 41% of county-level variation in outcome could be attributed to demographics, arrest characteristics, and bystander CPR ⁸. Based on the same report from Girotra et. al, the remaining 59% of variation is left unexplained. The variation is likely attributed to differences in post-resuscitation care between hospitals. In 2013, Richardson et. al stated early DNR placement is associated with a decrease in interventions, survival to discharge, and variability in treatments between hospitals ²⁸. Similarly, Swor et. al recently indicated that variation between hospitals could be attributed to DNR practice patterns, and early DNR was inversely associated with survival to discharge ^{29, 30}. More recently, Huebinger et. al noted that post-arrest care varied significantly between hospitals throughout Texas, with higher rates of TTM, LHC, and PCI associated with a higher rate of survival to discharge ³¹. Even after adjusting for observed correlates of DNR status, our study showed markedly different rates of DNR initiation after hospital admission.

The Girotra report also noted the impact of coronary angiography on survival ⁸. Evidence from Kim et. al reports post-arrest coronary angiography is associated with better survival in OHCA ³². Khera et. al submits early coronary angiography is associated with increased hospital discharge and improved neurological outcome in OHCA ³³. A recent report by Lemkes et. al indicates that immediate coronary angiography for patients with a shockable rhythm and no signs of STEMI was not found to be better than delayed angiography with respect to overall survival at 90 days ³⁴. Additionally, a recent meta-analysis indicated the positive influence of CAG on survival to hospital discharge. The results of the aforementioned studies demonstrate the variability in outcome of CAG post-OHCA. Like rates of DNR-initiation, there were marked variances in angiography use between hospitals even after accounting for differences in “early-chain” confounders like initial rhythm.

A single hospital in the current investigation had significantly less interventions and worse outcomes compared to the other three hospitals. Hospital A had the least favorable rates of death and good neurological outcome when compared to the other three. Hospital A was more likely to initiate new DNR status and less likely to perform a coronary angiography. We have recently noted that most cardiac catheterizations from these hospitals occur later in the hospital stay; therefore, the initiation of DNR and rate of cardiac catheterization may be related. Initiation of DNR order had an odds ratio of 10.09 for predicting death after adjustment for treatment propensity, (FIGURE 3C). Despite multivariate and propensity analysis, we were not able to fully account for the increased death rate at Hospital A. In this data set, we are not able to track other markers of intensity of care or variation in other treatment protocols between hospitals.

In many of the figures below, arrest witnessed by EMS appeared to portend a poorer outcome when compared to bystander witnessed. However the rate of shockable rhythm was much higher in the bystander witnessed group 26% versus 12% in the EMS witnessed and most EMS witnessed arrest occur in patients with deteriorating medical conditions and are often not primary cardiac arrests.

These findings suggest differences in hospitals serving a similar population at a hospital level and emphasis is needed on more standardized care among hospitals to improve survival. Observed disparities in the clinically important outcome of neurologically intact survival were found to be associated with the differences in how patients were treated, which are modifiable factors.

Limitations

Our study has several important limitations. The current study has the equivalent limitations as all retrospective analyses. Second, as in any observational study, there may be unmeasured confounders which account for the results we observed. Since this is, to our knowledge, the first report to examine tertiary-care hospital disparities within a city located in the U.S., our results may not be representative of hospitals in other regions. Our covariate selection was limited by the data available in CARES, which most notably does not collect reliably coded data on comorbidities and socioeconomic factors. One of the major limitations of the study is that DNR and angiography may be surrogate markers for the aggressiveness and quality of care between hospitals. It is valid to assume that Hospital A has some challenges with post-arrest care that are specific and unmeasured. Another limitation of the study is the confounder of immortal time bias. Angiography may be a surrogate marker for length of in hospital survival. We were unable to analyze the relationship between length of stay and angiography. Lastly, physician opinion/judgement and patient or family preferences may also factor into late phase decisions. These decisions are inherently complex and multifactorial, with multiple stakeholders (e.g. different physician specialties, patients, and families) driving the final decision. While the current dataset lacks granularity to address these nuances, future prospective studies could better characterize the effect on between-hospital treatment differences that occur due to factors such as caregiver and patient attitudes, family/legally authorized representative decisions, social contexts, physician risk-stratification of prognosis, and physician bias.

Conclusion

This is the first investigation showing significant differences in treatment and survival between close-proximity tertiary hospitals serving a similar patient population within a U.S. city. Differing rates of DNR and coronary angiography was associated with observed disparities in favorable neurologic outcome, but not death, between four Detroit hospitals. Multivariate analysis failed to adjust for the increased death rate at one hospital. Addressing variation in all aspects of care of OHCA patients is an important tool to attempt to improve patient outcomes.

Supplementary Material

1

Figure 3CX: Multivariable-adjusted odds of In-Hospital Death, Before Adjusting for Treatment Propensity

Figure 3DX: Multivariable-adjusted odds of CPC ≤ 2, Before Adjusting for Treatment Propensity