|Home | About | Journals | Submit | Contact Us | Français|
It is unknown how in-hospital cardiac arrest (IHCA) rates vary across hospitals and predictors of variability.
Measure variability in IHCA across hospitals and determine if hospital-level factors predict differences in case-mix adjusted event rates.
Get with the Guidelines Resuscitation (GWTG-R) (n=433 hospitals) was used to identify IHCA events between 2003-2007. The American Hospital Association survey, Medicare, and US Census were used to obtain detailed information about GWTG-R hospitals.
adult patients with IHCA
Case-mix adjusted predicted IHCA rates were calculated for each hospital and variability across hospitals was compared. A regression model was used to predict case-mix adjusted event rates using hospital measures of volume, nurse-to-bed ratio, percent ICU beds, palliative care services, urban designation, volume of black patients, income, trauma designation, academic designation, cardiac surgery capability and a patient risk score.
We evaluated 103,117 adult IHCAs at 433 US hospitals. The case mix adjusted IHCA event rate was highly variable across hospitals, median 1/1000 bed days (interquartile range: 0.7-1.3 events/1000 bed-days). In a multivariable regression model, case-mix adjusted IHCA event rates were highest in urban hospitals (rate ratio [RR] 1.1, 95% confidence interval [CI] 1.0-1.3, p=0.03) and hospitals with higher proportions of black patients (RR 1.2, 95% CI 1.0-1.3, p=0.01) and lower in larger hospitals (RR 0.54, 95% CI 0.45-0.66, p<.0001).
Case-mix adjusted IHCA event rates varied considerably across hospitals. Several hospital factors associated with higher IHCA event rates were consistent with factors often linked with lower hospital quality of care.
Prevention of cardiac arrest in hospitalized patients represents an important opportunity to improve hospital quality of care (1) as prior studies suggest many of these events are predictable and avoidable.(2-4) Hospitals have been increasingly focused on reducing IHCA and inpatient mortality.(5, 6) Substantial hospital resources are often devoted to early detection and prevention of respiratory and hemodynamic compromise in hospitalized patients.(6, 7) Accreditation standards mandate that hospitals have a written plan for in-hospital cardiac emergencies(8) and certification in Basic Life Support is required for most hospital-based healthcare providers in the United States (US).
Despite this focus on IHCA prevention, there has not been a national comparison of IHCA rates across hospitals and there are no specific guidelines to direct hospitals in how to best reduce arrest rates. As a result, IHCA event rates likely vary across hospitals. The lack of benchmarking data precludes hospitals from addressing their relative IHCA rates and their individual needs for quality improvement.
Some of the IHCA rate variability could be explained by differences in patient case-mix, as hospitals which provide care for sicker patients are likely to have more arrests. It is unknown however whether there is substantial variation in case-mix adjusted IHCA rates across hospitals, and whether specific hospital level factors (e.g. hospital volume, structural, demographic, and organizational features) predict these rates. This is important to understand, as clinicians and policymakers are increasingly focusing on understanding and identifying indicators of hospital care and quality which impact outcomes and care of hospitalized patients.(9, 10)
To address these issues, our primary aim was to determine how case-mix adjusted IHCA events varied across hospitals participating in the Get With the Guidelines-Resuscitation (GWTG-R), a large ongoing registry of in-hospital resuscitation events in the US. We also sought to determine if hospital level factors explained variability in rates of IHCA.
We used data from the GWTG-R (formerly, National Registry of Cardiopulmonary Resuscitation) to identify in-hospital cardiac arrest events. The GWTG-R is sponsored by the American Heart Association and it is the only prospective database of IHCA in the US.(10) The database was developed primarily for quality improvement, and it is structured to allow hospitals to systematically track treatments and outcomes of patients experiencing IHCA. Data are collected from participating acute care hospitals according to standardized Utstein definitions for in-hospital arrest.(11) Hospitals are encouraged to submit data on all consecutive cardiac arrest events in patients without do-not-resuscitate (DNR) orders. Resuscitation events are included in the database if the event elicits an emergency resuscitation response by hospital personnel and a code record is generated. Cardiopulmonary arrests are specifically defined as events with cessation of cardiac activity determined by the absence of a palpable pulse, apnea and unresponsiveness.
Specially trained nurse or research coordinators at each facility routinely abstract data from medical records and code review sheets regarding: patient demographics, pre-event data, event data, and patient outcomes. The American Heart Association provides rigorous quality control and oversight for all GWTG-R data collection, analysis, reporting and research studies. Additional details regarding the GWTG-R study design are described in detail elsewhere.(10)
Our study population included patients hospitalized at facilities participating in GWTG-R with documented IHCA. We included in our analysis cardiac arrests in patients 18 years or older, admitted between January 1, 2003 and December 31, 2007 at hospitals primarily providing care for adult patients. Subsequent arrests in a single patient were included in our analysis if designated as discrete events according to standard GWTG-R definitions. Pediatric hospitals were excluded from the analysis.
To calculate each hospital’s crude event rate, we calculated the number of IHCA divided by bed days. Hospital bed-days were selected as the incidence “denominator” measurement of the population at risk, analogous to person-time. (12, 13) Hospital bed-days were estimated from data reported to the annual American Hospital Association survey and linked with the closest year of data in GWTG-R.
Because IHCA presumably occurs more frequently in populations of severely ill patients, we included several variables in our model to adjust for potential differences in the IHCA event rate that may be due to variation in the acuity level of patients for which a hospital provides care.(14) These variables included trauma hospital designation, cardiac surgery capability, academic designation, and a patient risk score. We specifically assumed that hospitals designated as trauma centers (level 1 and 2), hospitals with cardiac surgery capabilities, and academic hospitals would be more likely to provide care for overall sicker patients, with a consequent higher risk of cardiac arrest.
Trauma designation (level 1, 2 versus other) and academic designation (indicated by membership in the American Association of Medical College’s Council of Teaching Hospitals and Health Systems), were obtained from the American Hospital Association annual survey. The capability of a hospital to perform cardiac surgery procedures was determined from Medicare based on billing codes International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 361x or Diagnosis Related Group (DRG) codes (106, 107, 109, 547, 548, 549, 550).
We then used inpatient Medicare claims from 2001-2002 (i.e., not during the years of the current study) at all U.S. acute care hospitals to estimate an inpatient mortality logistic regression model. Covariates associated with patient risk included age, gender, race, Elixhauser comorbidity predictors (Appendix A), and DRGs with >500 cases. The c-statistic for this model was 0.85. (15) This model was then applied to claims from all Medicare patients hospitalized at each GWTG-R hospital from 2003-2007 and a mortality risk probability was assigned to each patient. As an estimate of the fraction of patients at each hospital at risk for in-patient death, we then determined the fraction of Medicare patients at each GWTG-R hospital whose risk score exceeded the 75th percentile nationally for risk of inpatient death. We later tested the sensitivity of our main results by varying this risk score cut point from the 66th to 90th percentile nationally. The mean risk score at each hospital was included as a continuous variable in the model.
Prior investigations have suggested a relationship between hospital volume and resuscitation outcomes (i.e. larger hospitals have better IHCA outcomes).(16) Bed size (number of beds/hospital) was included in our model as a measure of hospital volume. Other organizational factors included categorical measures of percentage of intensive care unit (ICU) beds, and nurse-to-bed ratio in our model using data from the American Hospital Association annual survey.
As one aspect of palliative care services involves helping critically ill patients and their families make decisions about end-of-life care and resuscitation preferences (e.g. DNR determination), availability of palliative care services (based on data from the American Hospital Association survey) was included in our model as a potentially relevant hospital organizational factor.
Outcomes from cardiac arrest have been related to socioeconomic factors in prior studies.(17, 18) We used fee-for-service Medicare hospital claims to determine each hospital’s inpatient racial composition, reflecting the proportion of claims from black patients (considered a valid measure of race in Medicare data) (19) treated annually for cardiovascular conditions (i.e. diagnosis-related group in major diagnosis category five). We linked zip codes of each hospital’s fee-for-service Medicare patients to 2000 U.S. Census data on median household income to approximate the socioeconomic status of the hospital’s inpatient population. Urban versus non-urban designation was ascertained from Medicare’s Hospital Cost Report Information System. As hospital geographic region (obtained from the US census) could confound hospital level demographic effects we also included region as a covariate in our regression model.
To determine the event rate of IHCA events at GWTG-R hospitals we divided the number of arrest events at each hospital by each hospital’s annual bed-days. The numerator was determined from events recorded in GWTG-R and the denominator from the annual reported bed-days per facility. This calculation is reported as events per 1000 bed-days.
Crude event rates were adjusted for case-mix by using a generalized estimating equation (GEE) regression model (20) with the following covariates: bed size, nurse-to-bed ratio, percent ICU beds, palliative care services, urban designation, volume of black patients, median per capita income, trauma designation, academic designation, cardiac surgery capability, and patient risk score. We used GEEs to control for multiple observations of the same hospital across years. Statistical significance across groups was determined using the Kruskal-Wallis test.
To estimate how measures of overall hospital quality and outcomes independently predicted case-mix adjusted IHCA event rates, we estimated a multivariable negative binomial regression model with the following covariates: organizational structure (bed size, nurse-to-bed ratio, percentage of ICU beds, palliative care service availability), demographic characteristics (urban, hospital racial composition, income), and other (trauma designation, academic designation, cardiac surgery capability, patient risk score). We estimated a negative binomial regression model after determining that event rates were excessively dispersed to satisfy the Poisson model assumptions.(21, 22) Results are presented as rate ratios (RR). A p value <.05 was considered statistically significant for all calculations.
Statistical analysis were performed using SAS version 9.1 (SAS Institute Inc, Cary, North Carolina) and Stata version 11, College Station, Texas. This study was approved by the Institutional Review Board at the University of Pennsylvania.
We evaluated a total of 103,117 in-hospital cardiac arrest events at 433 hospitals contributing data to GWTG-R from 2003-2007. In this sample of hospitals, most were medium or large in size (i.e. > 100 beds) (90%), with more than 5% ICU beds (86%). Many were in urban locations (41%) and/or designated as level 1 or 2 trauma centers (42%). Additional characteristics of these hospitals are listed in Table 1.
The case mixed adjusted IHCA event rate was highly variable across hospitals, median 1/1000 bed days (interquartile range [IQR]: 0.7 to 1.3 events/1000). The distribution of these event rates is illustrated in Figure 1. Notably, 8% of hospitals (n=35) had case mix adjusted event rates that were 50% below the median and 3% of hospitals (n=13) had case mix adjusted event rates that were twice the median.
Even when grouping hospitals of similar characteristics, there was significant variation in the mean rate of excess IHCA across hospital size, cardiac surgery capability, and hospital racial composition (Table 2).
In a multivariable regression model accounting for measures of hospital volume, structural, demographic, and organizational features (Table 3), we found that hospitals in urban location (RR 1.1, 95% CI 1.0-1.3, p=0.03) and hospitals with higher proportions of black patients (RR 1.2, 95% CI 1.0-1.3, p=0.01) were both independent predictors of higher IHCA rates. Larger hospitals (RR 0.54, 95% CI 0.45-0.66, p<.0001) had lower IHCA rates.
Sensitivity analyses varied the cut point for the risk score, without qualitative changes in the results of the model. As the risk score in our model was based on Medicare patients, we also conducted a subsequent analysis and restricted GWTG-R data to patients >=65 years of age. We then estimated the correlation coefficient of IHCA rates at each hospital for elderly vs. non-elderly patients. We then re-estimated the negative binomial regression model using the event rate of elderly patients as the dependent variable. The resulting rate ratios for the model did not significantly change except for academic hospitals (RR 0.74, 95% CI 0.63-0.88, p<.0004).
There is significant and important variation in the rate of cardiac arrest in hospitalized patients that is not explained by case mix. We also identified hospital factors that were independently predictive of IHCA event rates. These findings have meaningful policy implications—as IHCA rates are easily measurable and case-mix adjustment is straightforward, IHCA event rates can potentially be used for tracking hospital performance.
Prior studies have identified variability and potential causes for variability in out-of-hospital cardiac arrest incidence, but less is known about variability and the causes of variability in IHCA incidence.(23, 24) As cardiac arrest in hospitalized patients is both a function of a patients’ underlying illness severity and a hospitals’ process of care for treating acutely and chronically ill patients, valid comparisons of IHCA event rates must account for risk adjustment of both patient and hospital level factors. This has been evident in prior work of coronary artery bypass graft procedural outcomes illustrating that if risk adjustment is inadequate, outcomes will appear worse in hospitals which provide care for patients with more severe illnesses.(25) We included in our model several measures of patient and hospital case-mix previously studied and readily available in the American Hospital Association survey. Using these variables, we demonstrated that case-mix adjustment of IHCA rates is feasible-and non-GWTG-R hospitals can calculate (Appendix A) their own case-mix adjusted event rate and understand how they compare with other hospitals.
Beyond case-mix, we sought to identify other useful factors that may explain variability in hospital event rates. While the causal pathway for IHCA is complex and multi-factorial, it is important for individual facilities to know how their adjusted IHCA rates compare to other facilities, particularly peer institutions. This knowledge can be a key motivator for quality improvement; Berwick et al refers to this as “move your dot”—an approach for hospitals looking to compare their organizational performance with in-hospital mortality across hospitals and then improve this rate relative to peer institutions.(26) Clearly, hospitals cannot change their fixed characteristics (size, demographics etc) and it is unlikely that these factors inherently represent the direct mechanisms impacting event rate (i.e. the physical size of the hospital is not protective against IHCA). It is likely however, that these hospital measures provide clues to potentially modifiable factors that impact rates of arrest in hospitalized patients (e.g., large hospitals may be more likely to have rapid response teams).
Other hospital factors predictive of event rate in our model were worrisome and raise equity concerns (e.g., race). The high “excess” IHCA rates at hospitals with large numbers of black patients are troubling, as they provide further evidence that these hospitals may be providing lower quality inpatient care.(27) Prior work shows that black patients with cardiovascular disease are more likely to receive care at hospitals with fewer evidence-based therapies, longer delays in reperfusion treatment, and overall worse risk-adjusted outcomes from cardiac surgery.(28-30) Others have shown that black patients with in-hospital cardiac arrest had longer times to defibrillation than other patients.(18) Clearly a solution to reducing IHCA event rate would not involve altering the demographics of a hospitals’ patient population but rather examining overall process and quality measures in place at hospitals with high rates of arrest in hospitalized patients. Additional work is needed to determine how to best reduce unexpected cardiac emergencies for all hospitalized patients regardless of race.
Our findings related to hospital volume (i.e., smaller hospitals had a higher IHCA event rate and larger hospitals had a lower IHCA event rate) could reflect differences in hospital resources relative to size. Large hospitals may have more resources in place to recognize early indications of clinical deterioration, more advanced technology, and greater overall intensity of services. Although a higher nurse-to-bed ratio was not associated with lower IHCA event rate, there may be other markers of staffing volume (e.g. housestaff coverage, in-house overnight attending coverage, critical care nurse coverage) that we did not measure which may contribute to the observed lower IHCA event rate at large hospitals.
Our study was limited in that our estimates of event rate were based on registry data, and while extensive checks were in place to insure a high likelihood of capturing most events at an institution, we cannot verify that all IHCA resuscitations were identified. Our denominator (hospital bed days) may not be recorded accurately at all facilities and may not account for all the potential person-time that could contribute to a hospital’s population at risk of cardiac arrest, including patients seen in clinics, hospital visitors, etc. Nevertheless, bed-days are a commonly measured indicator of hospital size that likely reflects the general dimensions of the population at risk.(12)
We were also limited in our findings and generalizations about event rates at small hospitals (<100 beds) as these are underrepresented in GWTG-R. It is unlikely however that these event rates can be estimated from other existing administrative datasets. Notably, our results also did not change when we excluded small hospitals from the analysis.
It is possible that hospitals participating in GWTG-R may have been motivated to join the registry because they have lower or higher rates of arrest than non-GWTG-R hospitals. This may be important for contextualizing our reported event rates, which therefore might be higher or lower than the general population of US hospitals.
Finally, we identified differences in event rate according to broad hospital categories, but we were unable to fully explain the mechanism that accounts for differences in case-mix adjusted IHCA. Specifically, there may be differences in end-of-life policies, use of rapid response teams, quality improvement initiatives, or other organizational or demographic factors (e.g. insurance status)(31) that could further explain the variability in event rate across facilities that were not included in our model.(32, 33) In addition, validated quality measures for other cardiovascular diseases (e.g. AMI, heart failure) were not included in our analysis but may correlate with IHCA rates as better surrogates for quality and require further study.
There is significant variability in case-mix adjusted IHCA event rates across hospitals and hospital v measures of volume and demographic features explained variation in case-mix adjusted IHCA event rates.
We wish to thank Amy Praestgaard for her assistance with data analysis. No compensation was provided to this consultant.
Funding Source: This research was supported by funding from the Robert Wood Johnson Foundation Clinical Scholars program at the University of Pennsylvania (Merchant). This research was also supported by an unrestricted grant from the Institute for Health Technology Studies (Washington, DC), as well as grant 1-R01-HL086919 from the National Heart, Lung, and Blood Institute. This project was also funded, in part, under a grant from the Pennsylvania Department of Health, which specifically disclaims responsibility for any analyses, interpretations, or conclusions. Dr. Groeneveld was additionally supported by a Research Career Development Award from the Department of Veterans Affairs’ Health Services Research and Development Service. None of the above funding sources were involved in the design or conduct of the study, collection, management, analysis, or interpretation of the data, or preparation, review, or approval of the manuscript.
Conflict of Interest Merchant, Yang, Bradley: no conflicts of interest to disclose
Becker: Speaker honoraria/consultant fees: Philips Healthcare, Seattle, WA. Institutional grant/research support: Philips Healthcare, Seattle, WA; Laerdal Medical, Stavanger, Norway; NIH, Bethesda, MD; Cardiac Science, Bothell, Washington
Berg: Institutional grant/research support: NIH, Bethesda, MD
Nadkarni: Institutional grant/research support: Laerdal Foundation for Acute Care Medicine, Stavanger, Norway; NIH, Bethesda, MD; AHRQ , Bethesda, MD.
Nichol: Institutional grant/research support: Resuscitation Outcomes Consortium (NIH U01 HL077863-05) 2004-2015; Co-PI, Evaluation of Video Self-Instruction in Compressions-Only CPR (Asmund S. Laerdal Foundation for Acute Medicine) 2007-2010; PI, Randomized Trial of Hemofiltration After Resuscitation from Cardiac Arrest (NHLBI R21 HL093641-01A1) 2009-2011; PI, Randomized Field Trial of Cold Saline IV After Resuscitation from Cardiac Arrest (NHLBI R01 HL089554-03) 2007-2012; Co-I, Resynchronization/Defibrillation for Advanced Heart Failure Trial (RAFT) (200211UCT-110607) 2003-2010; Co-I, Novel Methods of Measuring Health Disparities (1RC2HL101759-01) 2009-2011; Co-I, Cascade Cardiac Resuscitation System (Medtronic Foundation) 2010-2015; PI, Research Collaborator: Gambro Renal Inc., Lakewood, CO, Sotera Wireless, San Diego, CA, Lifebridge Medizintechnik AG, Ampfing, Germany, Other: Chair, AHA Executive Database Steering Committee; Chair, Mission:Lifeline EMS Task Force, Co-Chair, AHA Resuscitation Science Symposium Planning Committee; Member, AHA Advanced Cardiac Life Support Subcommittee; Member, AHA Epidemiology and Statistics Committee; Member, Pacific Mountain Affiliate Board of Directors, American Heart Association, Received travel reimbursement, AHA
Carr: Institutional grant/research support: NIH, AHRQ
Abella: Speaker honoraria/consultant fees: Philips Healthcare, Seattle, WA; Medivance Corporation, Louisville, CO. Institutional grant/research support: Philips Healthcare, Andover, MA; Doris Duke Foundation, New York City, NY; American Heart Association, Dallas, TX; NIH, Bethesda, MD. In-kind research support: Laerdal Medical Corp, Stavanger, Norway.
Groeneveld: Federal employee
Complete author information
Raina M. Merchant MD MS University of Pennsylvania, 423Guardian Street 1022 Blockley Hall, Philadelphia, PA 19104 raina.merchant/at/uphs.upenn.edu office: (215) 746-7990 fax: (215) 662-3953
Lin Yang MS University of Pennsylvania 423Guardian Street Blockley Hall, 12th floor, Philadelphia, PA 19104 yanglin/at/mail.med.upenn.edu, office (215) 898-2569 fax (215) 573-8778
Lance B. Becker MD University of Pennsylvania 125 South 31st Street, Suite 1200, Philadelphia, PA 19104 lance.becker/at/uphs.upenn.edu, office (215) 746-3625, fax (215) 746-1224
Robert A. Berg MD Children’s Hospital of Philadelphia, 3615 Civic Center Boulevard, Philadelphia, PA, 19104 bergra/at/email.chop.edu, office (215) 590-7430 fax (267) 426-5480
Vinay Nadkarni MD Children’s Hospital of Philadelphia, 3615 Civic Center Boulevard, Philadelphia, PA, 19104 215 nadkarni/at/email.chop.edu, office (215) 590-7430 fax (267) 426-5480
Graham Nichol MD MPH University of Washington, 325 9th Ave, Seattle, WA 98104 Nichol/at/u.washington.edu, office (206) 521-1728, fax (206) 521-1784
Brendan G. Carr, MD MS University of Pennsylvania 929 Blockley Hall, Philadelphia, PA, 19104. carrb/at/upenn.edu, office (215) 573-3976 fax (215) 573-2265
Nandita Mitra PhD University of Pennsylvania, 212 Blockley Hall, Philadelphia, PA 19104 nanditam/at/mail.med.upenn.edu, office (215) 573-4467 fax (215) 573-2265
Steven M. Bradley MD MPH University of Washington, 325 9th Ave, Seattle, WA 98104 sbradley/at/medicine.washington.edu office, (206) 521-1728, fax (206) 521-1784
Benjamin S. Abella MD MPhil 3400 Spruce Street, Ground Ravdin, University of Pennsylvania, Philadelphia, PA, 19104. Benjamin.abella/at/uphs.upenn.edu, office (215) 615-0785 fax (215) 662-3953
Peter W. Groeneveld MD MS University of Pennsylvania, 423Guardian Street 1229 floor Philadelphia, PA 19104 petergro/at/mail.med.upenn.edu, office (215) 898-2569 fax (215) 573-8778
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.