A, Cardiac arrest incidence. B, In-hospital case survival.
A, Unadjusted rates (correlation, −0.16; P = .003). B, Rates adjusted for patient factors only (r, −0.15; P = .004). Incidence is adjusted for the Centers for Medicare and Medicaid Services case-mix index for hospital admissions, and case survival is adjusted for patient factors (ie, age, sex, race/ethnicity, and preexisting conditions), including event characteristics (ie, initial arrest rhythm, year of admission, night vs day, and weekend vs weekday).
Regression models were adjusted for both patient and hospital factors to determine whether certain hospital factors may, in part, mediate this relationship. Incidence is adjusted for the Centers for Medicare and Medicaid Services case-mix index for hospital admissions and hospital factors (ie, teaching status, ownership, geographical region, urban vs rural, full-time–equivalent nurse ratio, number of beds, and certification as a trauma center). Case survival is adjusted for case-mix index, patient factors, and hospital factors (r, −0.07; P = .18).
eFigure 1. Correlation between hospital rates of cardiac arrest incidence and case survival, adjusted for patient factors only, after excluding 3 outliers. Incidence is adjusted for the Centers for Medicare and Medicaid Services case-mix index for hospital admissions, and case survival is adjusted for patient factors (ie, age, sex, race/ethnicity, and preexisting conditions), including event characteristics (ie, initial arrest rhythm, year of admission, night vs day, or weekend vs weekday).
eFigure 2. Mediation analysis of hospital factors that may account for the correlation between hospital rates of cardiac arrest incidence and case survival, after excluding 3 outliers. Regression models were adjusted for both patient and hospital factors to determine whether certain hospital factors may, in part, mediate this relationship. Incidence is adjusted for the Centers for Medicare and Medicaid Services case-mix index for hospital admissions and hospital factors (ie, teaching status, ownership, geographical region, urban vs rural, full-time-equivalent nurse ratio, number of beds, and certification as a trauma center). Case survival is adjusted for case-mix index, patient factors, and hospital factors.
Chen LM, Nallamothu BK, Spertus JA, Li Y, Chan PS, . Association Between a Hospital’s Rate of Cardiac Arrest Incidence and Cardiac Arrest Survival. JAMA Intern Med. 2013;173(13):1186-1195. doi:10.1001/jamainternmed.2013.1026
Copyright 2013 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
National efforts to measure hospital performance in treating cardiac arrest have focused on case survival, with the hope of improving survival after cardiac arrest. However, it is plausible that hospitals with high case-survival rates do a poor job of preventing cardiac arrests in the first place.
To describe the association between inpatient cardiac arrest incidence and survival rates.
Within a large, national registry, we identified hospitals with at least 50 adult in-hospital cardiac arrest cases between January 1, 2000, and November 30, 2009. We used multivariable hierarchical regression to evaluate the correlation between a hospital’s cardiac arrest incidence rate and its case-survival rate after adjusting for patient and hospital characteristics.
Main Outcomes and Measures
The correlation between a hospital’s incidence rate and case-survival rate for cardiac arrest.
Of 102 153 cases at 358 hospitals, the median hospital cardiac arrest incidence rate was 4.02 per 1000 admissions (interquartile range, 2.95-5.65 per 1000 admissions), and the median hospital case-survival rate was 18.8% (interquartile range, 14.5%-22.6%). In crude analyses, hospitals with higher case-survival rates also had lower cardiac arrest incidence (r, −0.16; P = .003). This relationship persisted after adjusting for patient characteristics (r, −0.15; P = .004). After adjusting for potential mediators of this relationship (ie, hospital characteristics), the relationship between incidence and case survival was attenuated (r, −0.07; P = .18). The one modifiable hospital factor that most attenuated this relationship was a hospital’s nurse-to-bed ratio (r, −0.12; P = .03).
Conclusions and Relevance
Hospitals with exceptional rates of survival for in-hospital cardiac arrest are also better at preventing cardiac arrests, even after adjusting for patient case mix. This relationship is partially mediated by measured hospital attributes. Performance measures focused on case-survival rates seem an appropriate first step in quality measurement for in-hospital cardiac arrest.
Approximately 200 000 adults have hospitalizations complicated by a cardiac arrest annually in the United States.1 Because less than 20% survive to hospital discharge,2 most prior studies3- 9 have focused on identifying factors associated with improved in-hospital survival. However, this emphasis may not accurately reflect a hospital’s overall performance in resuscitation outcomes because it ignores the hospital’s cardiac arrest incidence rate.10
It is possible that hospitals with high case-survival rates (high performers) do a poor job of preventing cardiac arrests and, therefore, only seem to have better resuscitation outcomes because more lower-risk patients have an arrest and are able to survive resuscitation efforts.10 If this was the case, it would raise questions about current efforts by organizations, such as The Joint Commission, to disseminate performance measures for in-hospital resuscitation that focus only on case-survival rates. Alternatively, high-performing hospitals may excel at both preventing and treating in-hospital cardiac arrest (ie, low incidence and high case-survival rates), or there may be no association between incidence and case-survival rates.
To assess whether case-survival rates are effective at identifying high-performing hospitals, we examined the relationship between a hospital’s cardiac arrest incidence rate and its case-survival rate, using data from a large, national in-hospital cardiac arrest registry. In addition, we examined hospital factors that may mediate any relationship found between incidence and case-survival rates for cardiac arrest.
We conducted our study within Get With the Guidelines (GWTG)–Resuscitation, formerly known as the National Registry of Cardiopulmonary Resuscitation. The GWTG-Resuscitation is a large, prospective, national registry of patients with in-hospital cardiac arrest, and its design has been previously described.11 To summarize briefly, the registry enrolls patients with a pulseless cardiac arrest, which is defined as the absence of a palpable central pulse, apnea, and unresponsiveness in patients without do-not-resuscitate (DNR) orders. Using standardized Utstein-style definitions,12,13 quality improvement personnel at participating hospitals collect information about clinical characteristics and outcomes for consecutive patients with in-hospital cardiac arrest. Data accuracy is enhanced through training of research staff; quality assurance activities, such as routine data reabstraction; and standard software that flags outlier values and incomplete data. The GWTG-Resuscitation also provides information on hospital characteristics and cardiac arrest incidence because the registry data have been linked with data from the American Hospital Association. The institutional review board at the Mid-America Heart Institute waived the requirement for informed consent because the study used deidentified data.
Between January 1, 2000, and November 30, 2009, we identified 111 634 patients 18 years or older at 566 hospitals with an index in-hospital cardiac arrest (Figure 1). Because we were interested in looking at the relationship between a hospital’s cardiac arrest incidence rate and case-survival rate, we excluded 4731 patients from 37 hospitals with missing information about cardiac arrest incidence and 133 patients with missing data about survival to discharge. We also excluded 2 hospitals with 1134 patients during the entire study period because these hospitals lacked data about hospital characteristics. To ensure that each hospital had sufficient case volume, we restricted our analyses to those hospitals with at least 50 cardiac arrest cases during the study period (3362 patients from 168 hospitals excluded). Finally, we excluded 121 patients from 1 hospital with an extreme outlier cardiac arrest incidence rate (17 times the standard deviation). Our final study sample was composed of 102 153 adults with an in-hospital cardiac arrest at 358 hospitals.
Our outcome of interest was the correlation between a hospital’s incidence rate and case-survival rate for cardiac arrest. The hospital’s case-survival rate was determined by dividing the number of patients with in-hospital cardiac arrest who survived to hospital discharge by the number of patients who experienced cardiac arrest. The incidence rate for cardiac arrest at each hospital was determined by dividing the number of cardiac arrest cases by the number of total hospital admissions (expressed as cases per 1000 admissions) during the hospital’s period of participation in the registry. This latter variable was obtained by linking hospitals within the GWTG-Resuscitation with data from the American Hospital Association, which provides the total number of admissions annually at each hospital. If a hospital participated for only part of the year (eg, enrollment year), the total number of admissions was prorated by the hospital’s length of participation within the GWTG-Resuscitation for that year.
Several patient- and hospital-level factors were collected for this study. Patient factors within the GWTG-Resuscitation included age, race/ethnicity (white, black, or other), sex, initial cardiac arrest rhythm (asystole, pulseless electrical activity, ventricular fibrillation, or pulseless ventricular tachycardia), year of admission, location (intensive care unit, monitored unit, nonmonitored unit, emergency department, or other) in which the cardiac arrest occurred, and time of cardiac arrest (night vs day or weekend vs weekday).8 In addition, the registry collected information on the presence or absence of the following conditions within 24 hours of the cardiac arrest: heart failure; myocardial infarction or ischemia; arrhythmia; hypotension; renal, hepatic, or respiratory insufficiency; diabetes mellitus; metabolic or electrolyte abnormality; acute central nervous system event (stroke or other); pneumonia; septicemia; major trauma; and malignant tumor.
Information about several hospital factors were also obtained from the American Hospital Association. These factors included a hospital’s teaching status (major, minor, or nonteaching), ownership (nonprofit, public, or private), location (urban vs rural), geographic region (North and Mid-Atlantic, South and Atlantic, North Central, South Central, or Mountain/Pacific), certification status as a trauma center, number of beds (<200, 200-349, 350-499, or ≥500), and the ratio of full-time-equivalent registered nurses to beds at the hospital. Finally, we also obtained information about a hospital’s case-mix index—a composite measure of patients’ illness severity for all Medicare discharges (not simply those with an inpatient cardiac arrest)—from the Centers for Medicare and Medicaid Services.14
To describe patient and hospital characteristics, we categorized the study sample by hospital quintiles of cardiac arrest incidence. Trends across hospital quintiles were examined for continuous variables using the linear trend test and for categorical variables using the Mantel-Haenszel trend test.
We then examined the correlation between a hospital’s incidence rate and case-survival rate for cardiac arrest with sequential multivariable models. First, the correlation between a hospital’s unadjusted incidence rate and its unadjusted case-survival rate was examined using ordinary linear models.
Second, we examined the correlation between a hospital’s risk-adjusted incidence rate and risk-adjusted case-survival rate, after adjusting for patient factors only. This analysis is the main analysis for this study. To complete this analysis, both hospital rates were risk adjusted for differences in patient case mix across hospitals. To determine risk-adjusted incidence rates, we first estimated predicted incidence rates using a random-effects logistic model for incidence rates. In this model, we adjusted for a hospital’s Centers for Medicare and Medicaid Services case-mix index, allowing us to control for differences in illness severity for all admitted patients among hospitals. For each hospital, we then calculated risk-adjusted incidence rates as follows: ([predicted incidence rate at the hospital/expected incidence rate for the same hospital] × mean unadjusted incidence rate for our entire hospital sample). This approach is the same as that taken to calculate risk-adjusted case-survival rates, leading to shrinkage of incidence rate estimates toward the global mean for hospitals with fewer patients.
Risk-adjusted cardiac arrest case-survival rates at hospitals were estimated using multivariable hierarchical logistic regression models, with survival to discharge as the dependent variable. We used 2-level hierarchical models to adjust for clustering of outcomes within hospitals, with individual hospitals modeled as random effects and patient characteristics (collected by the GWTG-Resuscitation and described in the Study Variables section) modeled as fixed effects within each hospital.15 From the multivariable model, we used hospital-specific random slope coefficients to calculate risk-adjusted case-survival rates as follows: ([predicted case-survival rate at a hospital/expected case-survival rate at the same hospital] × mean case-survival rate at all hospitals in our sample). This model is similar to one that has been used by others to calculate risk-standardized mortality rates for other conditions.16
If a significant correlation was found between a hospital’s cardiac arrest incidence rate and case-survival rate even after adjusting for patient case mix, we determined a priori to conduct a mediation analysis to assess whether certain hospital factors accounted for this relationship. This was evaluated by examining the correlation between a hospital’s incidence rate and case-survival rate after additionally adjusting for hospital factors (described in the Study Variables section) in our regression models for hospital incidence and case-survival rates. We then systematically adjusted for all patient factors and each hospital factor individually in our final models to assess whether any single hospital factor accounted for a substantial proportion of the correlation between these 2 hospital rates (ie, led to a significant attenuation of any observed correlations).
Overall, missing data rates were low. At least one covariate was missing for 8.1% of the patients, and 2 covariates were missing for 0.2% of patients. The most common variables with missing data were race/ethnicity (6.9%) and nighttime arrest (1.2%). These data were assumed to be missing at random and were imputed using IVEware software.17 Five imputed data sets were generated and pooled for the imputed results. Results with and without imputation were not meaningfully different, so only the former are presented. Finally, we repeated the correlation analyses using restricted cubic splines and found no evidence for a nonlinear relationship in our models (P = .55 in the model adjusted for patient factors and .90 in the model adjusted for patient and hospital factors).
All statistical analyses were conducted using SAS statistical software, version 9.1.3 (SAS Institute, Inc), IVEware (University of Michigan), and R version 2.6.0 (Free Software Foundation). All tests for statistical significance were 2-tailed and were evaluated at a significance level of .05.
Outcome Sciences Inc serves as the data collection (through their Patient Management Tool) and coordination center for the GWTG-Resuscitation. The University of Pennsylvania serves as the data analytic center and prepared the data for research purposes. All analyses for this study were performed at the Mid-America Heart Institute.
Among 102 153 patients with an in-hospital cardiac arrest, the overall case-survival rate was 19.2%. At the patient level, the overall cardiac arrest incidence rate was 4.54 per 1000 admissions.
Across the 358 hospitals, the median cardiac arrest incidence rate across the hospitals was 4.02 per 1000 admissions (interquartile range [IQR], 2.95-5.65 per 1000 admissions; range, 0.90-18.4 per 1000 admissions) (Figure 2A). Patient and event characteristics differed across hospital quintiles of incidence rates (Table 1 and Table 2). Compared with patients at hospitals with the highest cardiac arrest incidence rates, patients at hospitals with lower incidence rates were more likely to be white and have an initial rhythm of ventricular fibrillation or pulseless ventricular tachycardia. Moreover, patients at hospitals with lower incidence rates were less likely to have certain conditions, such as renal insufficiency, hepatic insufficiency, and respiratory insufficiency. Table 3 lists the characteristics of hospitals stratified by hospital quintile of incidence rates. Compared with hospitals in the highest quintile of cardiac arrest incidence, hospitals in the lowest quintile were more likely to be nonprofit, have higher nurse-to-bed ratios, and possess a larger number of beds.
The median cardiac arrest case-survival rate was 18.8% (IQR, 14.5%- 22.6%; range, 2.5%-46.9%) (Figure 2B). In unadjusted analysis, a significant negative correlation was found between a hospital’s case-survival rate and incidence rate (r, −0.16, P = .003) (Figure 3A). This finding suggests that hospitals with high case-survival rates for in-hospital cardiac arrest also had lower cardiac arrest incidence rates, and hospitals with low case-survival rates had higher cardiac arrest incidence rates. Notably, the negative correlation between these 2 rates was minimally changed after adjusting for differences in patient case mix (r, −0.15; P = .004) (Figure 3B). In sensitivity analyses excluding 3 outliers with a crude incidence rate of greater than 15 per 1000 admissions, our results were similar (r, −0.15; P = .005) (eFigure 1 in the Supplement).
Finally, we examined whether hospital factors mediated the relationship between a hospital’s cardiac arrest incidence and case-survival rate. Adjustment for hospital and patient characteristics led to a significant attenuation of the relationship between these 2 rates (r, −0.07; P = .18) (Figure 4). This finding suggests that the correlation between cardiac arrest incidence and case survival was partly attributable to the hospital factors in the model. In sensitivity analyses excluding 3 outliers with a crude incidence rate of greater than 15 per 1000 admissions, our results were similar (r, −0.08; P = .13) (eFigure 2 in the Supplement). In further exploratory analyses, each hospital factor led to some attenuation of the relationship between a hospital’s cardiac arrest case-survival and incidence rates (Table 4). The largest attenuation (ie, the greatest shift in the correlation estimate from a negative to a positive direction) occurred after adjusting solely for geographic region or a hospital’s nurse-to-bed ratio.
As the country works to increase its ability to measure and distinguish hospitals on the basis of quality, cardiac arrest survival has emerged as a potential performance measure. However, it is important to ensure that this measure does not “reward” hospitals that do a poorer job of preventing cardiac arrests, thus artificially making them appear to have higher survival rates. In a large national registry of in-hospital cardiac arrest, we found substantial variation in rates of cardiac arrest incidence and case survival across hospitals. There was a sizable negative correlation between a hospital’s incidence and case-survival rates for cardiac arrest, suggesting that hospitals that have higher survival rates after cardiac arrest also have lower cardiac arrest incidence rates. This correlation persisted after adjustment for patient factors. These findings suggest that cardiac arrest case-survival rates are not confounded by incidence rates because a positive correlation between the 2 was not observed; therefore, a focus on comparing case-survival rates across similar hospitals would be a useful hospital quality performance measure. Furthermore, we found that additional adjustment for certain hospital factors significantly attenuated the negative correlation between incidence and case survival, thereby suggesting that these hospital factors may, in part, mediate (and explain) why some hospitals are better at both preventing and treating in-hospital cardiac arrest.
Most prior studies of in-hospital cardiac arrest have examined case-survival and incidence rates independently of one another. For instance, prior work has found that case-survival rates differ by time of arrest (ie, nights or weekends),8 arrest location, initial cardiac arrest rhythm,9,18 time to defibrillation,4 and processes of care during acute resuscitation (eg, delays in medication administration).7,19 Others have reported that hospitals with urban locations, a higher proportion of black patients, and fewer hospital beds have higher rates of cardiac arrest incidence.20
Together, the hospital factors we examined seemed, in part, to mediate the relationship between a hospital’s incidence and case-survival rate for cardiac arrest. For example, the correlation between incidence and case-survival decreased by one-quarter when comparing hospitals with similar nurse staffing. This finding suggests that care in certain types of hospitals (eg, hospitals with higher nurse staffing) may be associated with both decreased cardiac arrest incidence and improved survival to discharge. In fact, prior studies21 have demonstrated an association between registered nurse staffing and lower rates of mortality and adverse events. Additional studies are needed to understand whether a causal relationship exists between nurse staffing and cardiac arrest outcomes and to identify other modifiable factors that may mediate the relationship between hospital rates of case-survival and cardiac arrest incidence.
It is important to acknowledge that the overarching goal of many clinicians and policymakers is to reduce the mortality rate from in-hospital cardiac arrest for all hospital admissions—a goal that can be achieved by lowering the incidence rate, increasing case survival, or both. However, given limited resources for measurement development and collection, as well as inherent difficulties in capturing the true case mix for all hospital admissions (the denominator for cardiac arrest incidence), we believe our results support the pragmatic approach already taken by others to focus on case survival as an initial measure of hospital performance for in-hospital cardiac arrest. Future efforts to identify factors that distinguish high-performing hospitals from low-performing hospitals would benefit from examining the relationship of hospital factors to both cardiac arrest incidence and case survival.
Our study had several potential limitations. First, although we were able to adjust for a number of patient and hospital factors, the possibility of unmeasured confounding in our use of registry data exists. Second, our sample was restricted to hospitals that participated in the GWTG-Resuscitation; therefore, our findings may not be generalizable to all hospitals. Third, we were able to examine only a handful of hospital factors, and we did not have specific information on nurse and physician staff experience and training or on quality improvement resuscitation programs implemented by a hospital (eg, mock codes, debriefing after codes, rapid response teams, increased use of telemetry, and remote intensive care unit monitoring). Third, our data lacked information about variation in hospital practices concerning end-of-life care. In particular, some hospitals may be more aggressive about obtaining DNR orders for patients with a poor prognosis. These hospitals may have lower incidence rates and higher case-survival rates than their peers. Furthermore, although we excluded patients with DNR orders from the numerator for cardiac arrest incidence rate, we did not have data to exclude all patients with DNR orders from the denominator for our incidence rate. Thus, we may have overestimated the correlation between unadjusted incidence and case survival.
In summary, we found that a significant negative correlation exists between a hospital’s incidence and case-survival rates for in-hospital cardiac arrest. Hospitals that excelled at preventing cardiac arrests also had higher survival rates for cardiac arrest cases, and this correlation persisted after adjustment for patient case mix. We found evidence that certain hospital factors, in part, mediated this relationship, but only one of the factors we examined—a hospital’s nurse-to-bed ratio—is potentially quickly modifiable. Future studies are needed to determine which hospital factors are surrogate markers of a hospital’s resuscitation performance and which may lead to improvements in cardiac arrest incidence and case survival.
Accepted for Publication: February 3, 2013.
Corresponding Author: Lena M. Chen, MD, MS, Division of General Medicine, University of Michigan, North Campus Research Complex, 2800 Plymouth Rd, Bldg 16, Room 407E, Ann Arbor, MI 48109 (firstname.lastname@example.org).
Published Online: May 20, 2013. doi:10.1001/jamainternmed.2013.1026.
Author Contributions: Dr Chen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Chen and Chan.
Acquisition of data: Chen and Chan.
Drafting of the manuscript: Chen and Chan.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Chen, Li, and Chan.
Study supervision: Chan.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study is supported by Career Development Grant Award K23HL102224 from the National Heart Lung and Blood Institute (Dr Chan) and Career Development Grant Award K08HS020671 from the Agency for Healthcare Research and Quality (Dr Chen).
Role of the Sponsors: The GWTG-Resuscitation is sponsored by the American Heart Association, which had no role in the study design, data analysis, or manuscript preparation and revision. This material is the result of work supported with resources of the Veterans Affairs Health Services Research and Development Center for Clinical Management Research, Veterans Affairs Ann Arbor Healthcare System.
Group Members: Besides the author, Paul S. Chan, MD, MSc, members of the Get With the Guidelines (GWTG)–Resuscitation Clinical Work Group/National Registry of Cardiopulmonary Resuscitation Science Advisory Board and Adult Task Force include the following: Robert A. Berg, MD, Children’s Hospital of Philadelphia; Paul S. Chan, MD, Mid-America Heart Institute and the University of Missouri; Mary E. Mancini, RN, PhD, University of Texas at Arlington; Emilie Allen, MSN, RN, Parkland Health & Hospital System; Scott Braithwaite, Yale University School of Medicine; Michael W. Donnino, MD, Beth Israel Deaconess Medical Center; Kathy Duncan, RN, Institute for Healthcare Improvement; Dana P. Edelson, MD, University of Chicago; Brian Eigel, PhD and Lana Gent, PhD, American Heart Association; Robert T. Faillace, MD, St Joseph’s Regional Medical Center; Romergryko G. Geocadin, MD, The Johns Hopkins University School of Medicine; Elizabeth Hunt, MD, MPH, PhD, Johns Hopkins Medicine Simulation Center; Karl B. Kern, University of Arizona Medical Center; Lynda Knight, RN, Lucile Packard Children’s Hospital at Stanford; Kenneth LaBresh, MD, RTI International; Timothy J. Mader, MD, Tufts University; Raina Merchant, MD, University of Pennsylvania School of Medicine; Vincent N. Mosesso Jr, MD, University of Pittsburgh School of Medicine; Vinay M. Nadkarni, MD, University of Pennsylvania School of Medicine; Graham Nichol, MD, MPH and Samuel Warren, MD, University of Washington; Joseph P. Ornato, MD, and Mary Ann Peberdy, MD, Virginia Commonwealth University Health System; Comilla Sasson, MD, MS, University of Colorado; and Mindy Smyth, MSN, RN.
Additional Contributions: Mary Jane Giesey, BA, provided research assistance and was compensated for her work.