Association Between Hospital Process Composite Performance and Patient Outcomes After In-Hospital Cardiac Arrest Care

Importance  Survival rates after in-hospital cardiac arrest (IHCA) vary significantly among US centers; whether this variation is owing to differences in IHCA care quality is unknown.

Objective  To evaluate hospital-level variation to determine whether hospital process composite performance measures of IHCA care quality are associated with patient outcomes.

Design, Setting, and Participants  Using data from the American Heart Association’s Get With the Guidelines–Resuscitation (GWTG-R) program, we analyzed 35 283 patients 18 years or older with IHCA treated at 261 US hospitals from January 1, 2010, through December 31, 2012. We calculated the hospital process composite performance score for IHCA using 5 guideline-recommended process measures. Opportunity-based scores were calculated for all patients, aggregated at the hospital level, divided into quartiles, and then associated with risk-standardized survival and neurologic status by a test for trend. The scores were then evaluated through hierarchical logistic regression and reported as odds ratios per 10% increment in hospital process composite performance.

Interventions  Acute care treatments for IHCA.

Main Outcomes and Measures  The primary outcome was survival to discharge measured as risk standard survival rates, and the secondary outcome was favorable neurologic status at hospital discharge.

Results  Of the 35 283 adults included in this study, the median age was 67 years (interquartile range [IQR] 56-78 years), and 57.9% were male. The median IHCA hospital process composite performance was 89.7% (interquartile range, 85.4%-93.1%) and varied among hospital quartiles from 82.6% (lowest) to 94.8% (highest). The IHCA hospital process composite performance was linearly associated with risk-standardized hospital survival to discharge rates: 21.1%, 21.4%, 22.8%, and 23.4% from lowest to highest performance quartiles, respectively (P < .001). After adjustment, each 10% increase in a hospital’s process composite performance was associated with a 22% higher odds of survival (adjusted odds ratio, 1.22; 95% CI, 1.08-1.37; P = .01). Hospital process composite quality performance was also associated with favorable neurologic status at discharge (P = .004).

Conclusions and Relevance  The quality of guideline-based care for IHCA varies significantly among US hospitals and is associated with patient survival and neurologic outcomes.

More than 200 000 patients are treated for in-hospital cardiac arrest (IHCA) annually in the United States.^1,2 In-hospital cardiac arrest is associated with poor survival, yet survival to discharge rates vary significantly among US hospitals.³ Some process-of-care measures, such as shorter time to defibrillation, are associated with better survival after IHCA.^4-6 The Joint Commission, National Quality Forum, and American Heart Association (AHA) have expressed significant interest in developing performance measures specific to IHCA in the hopes of facilitating benchmarking and ultimately improving patient outcomes.⁷ An AHA consensus statement identified several strategies for improving survival from cardiac arrest and pinpointed best practices related to structure, care pathways, and quality improvement care opportunities.¹

We conducted this study to examine (1) the variability in IHCA process quality of care across US hospitals and (2) whether there is an association between a hospital measure of IHCA process quality of care and patient outcomes. To our knowledge, no previous study has examined variation between IHCA quality of care and patient outcomes using a hospital process composite performance measure. Understanding the association between process and survival may clarify the utility of these process measures to inform hospital-level quality for IHCA care.

The AHA’s Get With the Guidelines–Resuscitation (GWTG-R) program is an ongoing, prospective, hospital-based clinical registry and quality improvement program for patients with IHCA. This registry was created to (1) collect cardiac resuscitation care and outcomes data from hospitals and (2) generate evidence-based guidelines.

The design of the GWTG-R has been previously described in detail.⁸ Briefly, all patients with a confirmed IHCA (defined as apnea, unresponsiveness, and lack of a palpable central pulse) who received cardiopulmonary resuscitation are identified and enrolled by staff at participating hospitals. Cases are recognized by centralized collection of cardiac arrest flow sheets, review of hospital paging-system logs, routine checks of code carts (carts stocked with emergency medications and equipment), pharmacy tracer drug records, and hospital billing charges for code-cart charges. The GWTG-R uses standardized Utstein Style definitions for clinical variables and outcomes. Utstein Style refers to consensus reporting guidelines for cardiac resuscitation. Utstein Style originated from an international multidisciplinary meeting in 1990 and has been updated several times.^9,10 Data completeness and accuracy are ensured by rigorous training, certification of hospital staff, and the use of standardized software with internal data checks.^8,11 A prior report⁸ shows an error rate in data abstraction of 2.4. This study was approved by the Duke University Institutional Review Board. The requirement for informed consent was waived.

From January 1, 2010, through December 31, 2012, we identified 48 189 adults aged 18 years or older with IHCA across 351 US GWTG-R hospitals. To avoid inflation in variance owing to small numbers, patients were excluded if they were enrolled at sites with fewer than 20 admissions overall, had a mean of fewer than 5 cardiac arrests per year, or had participated in the GWTG-R for less than a year (eTable 1 in the Supplement). In addition, we excluded hospitals whose hospital characteristics were missing (n = 16). Only the index cardiac arrest for each patient was included in this analysis. Within the GWTG-R hospitals, we excluded cardiac arrests that occurred in operating rooms, procedural suites, and the emergency department because these arrests are known to be different and to have a survival advantage compared with those that occur in wards and intensive care units (eTable 1 in the Supplement). After these exclusions, our final study population included 35 283 patients from 261 GWTG-R hospitals.

Five American College of Cardiology/AHA guideline–recommended acute care resuscitation process measures were evaluated among individuals who were eligible to receive them. The guideline-recommended process-of-care measures chosen a priori by consensus of our research team were as follows: (1) device confirmation of correct endotracheal tube placement,⁵ (2) a monitored or witnessed cardiac arrest event,⁶ (3) time to first chest compression less than or equal to 1 minute, (4) time to first defibrillation delivered at less than or equal to 2 minutes for ventricular tachycardia (VT) or ventricular fibrillation (VF),⁴ and (5) administration of epinephrine or vasopressin for pulseless events (pulseless VT or VF or pulseless electrical activity or asystole) within 5 minutes.¹² These 5 measures were chosen based on standard guideline recommendations, evidence of the association of individual measure with outcome,^4-6 and completeness of data.¹² Inclusion and exclusion criteria for each process measure were defined according to standard eligibility definitions for IHCA put forth by the AHA for the GWTG-R (eTable 2 in the Supplement). Another measure, time to second defibrillation greater than 2 minutes, was considered for inclusion in this study but was removed at the request of the AHA GWTG-R task force after an observational analysis called into question the practice of withholding shocks for at least 2 minutes to allow for cardiopulmonary resuscitation. In addition, 3 measures were not selected for inclusion given suspected universal adherence: time to assisted ventilation, chest compressions provided, and defibrillation shock provided.

Hospital process composite performance scores were calculated using opportunity-based scoring, which is defined as the sum of correct care divided by total care opportunities.¹³ Each patient at a GWTG-R hospital contributed care opportunities to the relevant hospital’s overall composite performance score.¹⁴ Each patient, depending on the initial rhythm, could contribute a maximum of 4 (nonshockable) or 5 (shockable) opportunities to the model. For example, if a patient arrested and had a shockable rhythm during rhythm analysis, then he or she would potentially be eligible to contribute all 5 care opportunities to the hospital’s performance score. If correct care only occurred for the monitored or witnessed event and time to defibrillation was less than 2 minutes, then only 2 of the 5 opportunities were counted as received. In contrast, a patient with pulseless electrical activity or asystole as an initial rhythm would only contribute a maximum of 4 opportunities to the model. Opportunity scoring implicitly weights each measure in proportion to the percentage of eligible patients at each hospital.¹³ Hospitals were divided into equal quartiles after rank-ordering of hospital process composite performance scores.

Demographic, cardiac arrest event, and hospital characteristics were compared by hospital performance quartiles. Individual process measures (eligibility and received) were also compared across performance quartiles. Pearson χ² tests were used to compare categorical variables across hospital performance quartiles; Kruskal-Wallis tests were used to compare continuous variables across hospital performance quartiles. Categorical variables were presented as percentages, and continuous variables were presented as medians and interquartile ranges (IQRs). Cochran-Armitage trend tests were used to compare differences across quartiles for individual performance measures. The primary outcome of interest was survival to discharge measured as risk-standardized survival rates (RSSRs), and the secondary outcome was favorable neurologic status at hospital discharge.

The RSSRs were calculated based on a previously validated model, which was developed to facilitate comparisons across hospitals.¹⁵ Validation and derivation characteristics for this model have been previously described.¹⁵ According to this method, the RSSR was calculated for each hospital by dividing a hospital’s predicted survival by the expected survival, multiplied by the cohort’s unadjusted survival. Survival to discharge was modeled by hierarchical logistic regression, including patient risk factors as fixed effects and a random intercept for hospitals. Hospital-level predicted survival was calculated as the mean of model-based predictions across patients at a given hospital, including an empirical Bayes prediction of the hospital effect.^16,17 Hospital-level expected survival was calculated in the same way, using only the fixed-effects portion of the model, with the hospital effect set equal to 0 (representing a typical hospital). This same process was implemented to obtain the RSSR with patients grouped by hospital quartile. The full model included the following risk factors: age, sex, event location, initial cardiac arrest rhythm, myocardial infarction present on admission, prior heart failure, renal insufficiency, hepatic insufficiency, hypotension, septicemia, acute stroke, diabetes mellitus, metabolic or electrolyte abnormality, metastatic or hematologic malignant tumor, major trauma, mechanical ventilation, dialysis, and intravenous vasopressor use. Race was not included in the published RSSR model because it is associated with the quality of the treating hospital.¹⁵ We performed a sensitivity analysis, adding race to the RSSR model, to determine whether its addition attenuated the association between quality of care and outcomes (eTable 3 in the Supplement). Patient race was self-identified and abstracted from the medical records by the GWTG-R staff. Differences in hospital RSSR and risk-adjusted favorable neurologic status across hospital process composite performance quartiles were assessed using linear regression weighted by the number of patients within a hospital to address nonconstant variance.

Differences between unadjusted outcomes across hospital process composite performance quartiles were assessed by linear regression weighted by the number of patients at each site. The association between hospital process composite performance and risk-adjusted survival was also evaluated directly through hierarchical logistic regression. Specifically, the hospital process composite performance score was added as a continuous covariate to the hierarchical model described above. Variables included in this model were age, race, sex, event location, initial cardiac arrest rhythm, whether myocardial infarction was present on admission, prior heart failure, renal insufficiency, hepatic insufficiency, hypotension, septicemia, acute stroke, diabetes, metabolic or electrolyte abnormality, metastatic or hematologic malignant tumor, major trauma, mechanical ventilation, dialysis, and intravenous vasopressor use. The linear association with survival was tested by a χ² test, and odds ratios (ORs) were reported per 10% increment in hospital process composite performance score. This approach differs from the previous assessment of the RSSRs in that the linear association is evaluated on the log-odds scale rather than the absolute scale.

To determine the number of lives that could be saved if all hospitals operated at the level of the best-performing hospital, we first identified the best-performing hospital as the one with the highest risk-adjusted survival. From the full covariate-adjusted hierarchical model, we estimated the effect of being treated at this hospital. For all patients in the data set, we used the hierarchical model to predict their survival probability given their fixed covariates and the best hospital effect. The predictions were summed over all patients in the sample to estimate the overall predicted survival if all patients were treated at the best hospital.

Favorable neurologic status at discharge was assessed as a secondary outcome for each patient with cardiac arrest. Cerebral performance categories (CPCs) are defined as follows: 1, good cerebral performance; 2, moderate disability; 3, severe disability; 4, coma or vegetative state; and 5, brain death. Favorable neurologic status was defined according to Utstein Style criteria as having a CPC score of 1 or 2 at hospital discharge and is defined according to Utstein Style criteria.¹⁰ We also report clinically significant favorable neurologic status with a CPC score of 1 (eTable 4 in the Supplement).

In our analysis, the missing CPC score rate was 17.7%, which aligns with previously published estimates^4,6 of missing CPC data in the GWTG-R. After a histogram review, we discovered that several hospitals were missing more than 50% of CPC data (eFigure 1 in the Supplement). As a result, we excluded hospitals with less than 75% of CPC scores available for their patients (n = 88) for this secondary analysis. The subsequent missing discharge CPC score rate among 173 hospitals was 2.3%.

To avoid survivor bias and to facilitate adequate hospital-level comparisons for neurologic status, all patients were included in this analysis. Patients who died during the hospitalization were assigned a CPC score of 5 (brain death).

C indexes were calculated to determine model diagnostics. The C index for calculation of predicted survival for the RSSR rates is 0.694 and 0.704 for risk-standardized favorable neurologic status. After race is added to the aforementioned models for the sensitivity analysis, C indexes were 0.697 for predicted survival in the RSSR and 0.708 for favorable neurologic status. For our continuous risk-adjusted models, the C indexes are 0.716 and 0.733 for survival and favorable neurologic status, respectively. All P values were 2-tailed and significant at P < .05.

Our analysis included 35 283 patients with IHCA from 261 GWTG-R hospitals from January 1, 2010, through December 31, 2012. The overall population eligible for each measure ranged from 14.5% for time to defibrillation of 2 minutes or less to 97.0% for time to first compressions of 1 minute or less (Table 1). The median hospital process composite performance score was 89.7% (IQR, 85.4%-93.1%). The hospital process composite performance varied significantly among the GWTG-R hospitals and ranged from 47.6% to 94.2% (Figure 1). Hospitals in quartiles 1, 2, 3, and 4 had median hospital process composite performance scores of 82.6% (IQR, 78.9%-84.3%), 88.0% (IQR, 86.7%-88.9%), 91.5% (IQR, 90.4%-92.3%), and 94.8% (IQR, 93.9%-95.9%), respectively (eFigure 2 in the Supplement).

Table 1 reveals the variation in use of individual guideline-recommended IHCA process quality-of-care measures among patients in our cohort. Hospitals in the highest quartile for the hospital process composite performance had significantly higher adherence to all individual guideline measures for IHCA compared with hospitals in other quartiles. There was significant variability in adherence to guideline-recommended IHCA process quality-of-care measures. The greatest increase across quartiles for hospital performance of individual measures was seen for confirmation of endotracheal tube placement (70.8% in quartile 1 to 94.3% in quartile 4, P = .01) and first defibrillation shock at 2 minutes or less for VT or VF (49.4% in quartile 1 to 66.5% in quartile 4, P = .01). The measures with the greatest overall adherence and the lowest degree of variance (although significant) were (1) monitored or witnessed cardiac events (P for increase across quartiles = .004) and (2) time to first compressions of 1 minute or less (P for increase across quartiles = .01) (Table 1).

Table 2 and Table 3 list the baseline characteristics of patients and hospitals within each quartile of hospital process composite performance. Compared with patients treated at hospitals in the lowest hospital process composite performance quartile, patients at the hospitals in the highest hospital process composite performance quartile were younger, slightly less likely to be male, more likely to be black, and less likely to have VT or VF at the time of cardiac arrest. Patients at the hospitals in the highest hospital process composite performance quartile were more likely to be in the intensive care unit at the time of arrest, less likely to have cardiac arrests at night, and more likely to undergo interventions, such as mechanical ventilation, hemodialysis, vasopressors, arterial catheters, and vascular access.

Compared with hospitals in the lowest hospital process composite performance quartile, hospitals in the highest hospital process composite performance quartile were more likely to have cardiac surgery capabilities and more likely to be teaching hospitals. Best-performing hospitals were also more likely to have a mean medium and large number of hospital beds (Table 3).

Unadjusted survival to discharge was 22.4% overall, ranging from 20.7% in the lowest quartile to 23.6% in the highest quartile (P < .001) (Table 4). After patient and event characteristics were adjusted for, the RSSRs were 21.1%, 21.4%, 22.8%, and 23.4% from lowest to highest quartile, respectively (P < .001 for trend) (Figure 2). Each 10% increase in hospital process composite performance was associated with a 22% higher adjusted odds of survival (adjusted OR, 1.22; 95% CI, 1.08-1.37; P = .01) (Table 4).

Significant differences were also found in favorable neurologic status at discharge based on hospital process quality of care, with hospitals in the highest hospital process composite performance quartile having the highest percentage of patients with favorable neurologic status. After adjustment, favorable neurologic status was 17.7%, 17.0%, 17.5%, and 19.9% from the lowest to highest quartile, respectively (P < .001 for trend; Table 4 and eTable 4 in the Supplement). A sensitivity analysis of patients with a CPC score of 1 (clinically significant favorable neurologic status) also revealed improved neurologic status by hospital process composite performance quartiles (eTable 4 in the Supplement).

Successful treatment of IHCA requires rapid implementation of several processes of care within a short and defined period.^1,12 We found significant variation in process quality-of-care achievement for patients with IHCA treated at US hospitals. Furthermore, we found that patients treated at hospitals with greater adherence to IHCA guideline–recommended therapies had higher survival rates. The association between process quality-of-care measures and outcomes was evident after adjusting for patient and hospital characteristics. We estimate that an additional 22 990 to 24 200 lives could be saved if all hospitals had similar IHCA quality to that of the best-performing hospitals.

Time to defibrillation has been established as an important measure for IHCA care. Patients with VT or VF with timely defibrillation (ie, within 2 minutes) were 50% more likely to survive compared with patients who had delays in defibrillation.⁴ Previous work³ has also demonstrated a link between variation in hospital performance with time to defibrillation and survival. Hospitals with the best performance for timely defibrillation had 41% higher adjusted survival compared with the worst-performing hospitals (quartile comparisons).³ We found significant variation not only in time to defibrillation but also in other processes of guideline-recommended acute care for IHCA.

Our cross-sectional analysis supports an association between greater adherence to process measures for IHCA and higher survival rates. Every 10% increase in hospital process composite performance measures among hospitals in our analysis was associated with a 1.22 higher adjusted odds of survival. An association between hospital process composite performance and outcomes has been demonstrated for other cardiovascular conditions, such as acute stroke, heart failure, and myocardial infarction care.^14,18-20 However, a previous analysis²¹ revealed that hospitals with better outcomes for heart failure, acute myocardial infarction, and pneumonia did not have the best survival for IHCA. In light of this information, the authors of the previous study concluded that public reporting of IHCA measures could provide new information on hospital quality. We found that process measures for IHCA care vary appreciably and are significantly associated with survival and neurologic outcome. The Joint Commission and National Quality Forum have proposed performance measures for IHCA care, including survival to discharge, time to defibrillation, and endotracheal tube confirmation. Our work supports the importance of addressing the process quality of IHCA care. We found that several of these process measures contribute important information related to clinical outcomes. As a result, a hospital process composite performance score used with other medical conditions (eg, congestive heart failure, coronary artery disease)^14,22,23 may be a more appropriate measure of quality of care for patients with IHCA.

In our analysis, we estimated that predicted survival to discharge of the best-performing hospital was 34.5% (compared with an observed survival of 22.4%). On the basis of an estimated 190 000 to 200 000 IHCAs per year in the United States, we estimate an additional 22 990 to 24 200 lives would be saved per year if all hospitals operated at the level of the highest-performing hospital. Although this is an estimate only, it helps to shed light on the effect of ensuring timely and high-quality care for IHCA.

Our study has several limitations. First, our data are observational; therefore, we cannot prove causation between process-of-care measures and outcomes. In addition, we cannot be sure that unmeasured confounders may have contributed to the association between composite quality of care and outcomes. Second, hospitals participating in the GWTG-R may be more interested in improving quality of care compared with nonparticipating hospitals. Although our sample may not be generalizable to all US hospitals, the degree of variability in actual process quality may be even greater among all US hospitals because the GWTG-R hospitals are self-selected with a presumed greater interest in improving outcomes from IHCA. Third, our secondary outcome of neurologic function was only analyzed among a subset of hospitals that did not have large amounts of missing data (173 of 261 hospitals), which may further limit generalizability and interpretation to hospitals that routinely collect CPC data on most of their patients with cardiac arrest. Fourth, we included 5 guideline-recommended process measures to create our hospital process composite performance score. Although we carefully chose measures based on the 2010 AHA guideline recommendations and the availability and completeness of data, we acknowledge that other processes or structural measures may be more or less associated with survival and neurologic outcomes. Fifth, we restricted our analysis to data collected from January 1, 2010, through December 31, 2012, to account for secular trends in survival to discharge (survival known to be lower in earlier years). Sixth, each measure was weighted equally in our analysis, yet some measures may be more associated with outcomes than others. Nonetheless, hospital process composite performance measures developed for other measures have treated individual measures similarly. Seventh, hospital variability in obtaining do-not-attempt-resuscitation (DNAR) orders before cardiac arrest likely influences a hospital’s survival rate such that hospitals with the best survival rates may be more aggressive in obtaining DNAR orders. The GWTG-R does not include patients in its registry if they have DNAR orders before their IHCA; however, we did not find that better-performing hospitals were more likely than their counterparts to establish DNAR orders during the hospitalization (eTable 5 in the Supplement). Eighth, we used standard definitions provided by the AHA’s GWTG-R to determine the population inclusion and exclusion criteria for each guideline measure. We acknowledge that the population inclusion for these measures may change over time as a result of expert consensus or new evidence, yet our analysis closely mimics feedback provided to each GWTG-R hospital for each measure and reflects consensus of the measure at the time of our analysis. Ninth, our data represent a cross-sectional association between adherence to process quality and outcomes. A longitudinal study accounting for change in hospital process performance is needed to establish an association between quality and outcomes.

Significant opportunities remain for improving adherence to guideline-recommended care overall and with individual process-of-care measures. Of importance, enhancing process quality of care may improve outcomes for the many patients with IHCA.

Accepted for Publication: November 20, 2015.

Corresponding Author: Monique L. Anderson, MD, MHS, Duke Clinical Research Institute, Duke University Medical Center, 7022 N Pavilion, PO Box 17969, Durham, NC 27715 (monique.anderson@duke.edu).

Published Online: February 24, 2016. doi:10.1001/jamacardio.2015.0275.

Author Contributions: Dr Anderson had full access to all 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: Anderson, Nichol, Berg, Bradley.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Anderson, Nichol.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Anderson, Dai, Thomas.

Obtained funding: Anderson.

Study supervision: Nichol, Berg, Peterson.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Chan reported receiving funding from grant 1R01HL123980 from the National Heart, Lung, and Blood Institute. Dr Peterson reported receiving industry funding from Janssen and consulting funding from Boehringer Ingelheim, Janssen, Sanofi, Bayer, Merck, and AstraZeneca. No other disclosures were reported.

Funding/Support: This study was supported by the Duke Clinical Research Institute (Durham, North Carolina). Dr Anderson is personally funded by the following career development award: National Institutes of Health Common Fund research supplements to promote diversity in health-related research under award 3U54AT007748-02S1.

Role of the Funder/Sponsor: The Duke Clinical Research Institute is responsible for the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Group Information: American Heart Association’s Get With the Guidelines–Resuscitation Investigators are listed in the eAppendix in the Supplement.

Additional Contributions: We thank Barbara Lytle, MS, for her project leadership and Erin Hanley, MS, for her editorial contributions to the manuscript. Neither Ms Lytle nor Ms Hanley received compensation for their contributions apart from their employment at the institution where this study was conducted. Ms Lytle and Ms Hanley are employees of the Duke Clinical Research Institute, Duke University Medical Center, Durham, North Carolina.