Variation Across Hospitals in In-Hospital Cardiac Arrest Incidence Among Medicare Beneficiaries

IMPORTANCE Although survival for in-hospital cardiac arrest (IHCA) has improved substantially over the last 2 decades, survival rates have plateaued in recent years. A better understanding of hospital differences in IHCA incidence may provide important insights regarding best practices for prevention of IHCA. OBJECTIVE To determine the incidence of IHCA among Medicare beneficiaries, and evaluate hospital variation in incidence of IHCA. DESIGN, SETTING, AND PARTICIPANTS This observational cohort study analyzes 2014 to 2017 data from 170 hospitals participating in the Get With The Guidelines-Resuscitation registry, linked to Medicare files. Participants were adults aged 65 years and older. Statistical analysis was performed from January to December 2021


Introduction
In-hospital cardiac arrest (IHCA) affects nearly 290 000 hospitalized patients each year in the United States 1 and is associated with poor survival and a high risk of neurological disability among survivors. 2,3Over the past decade, hospitals have devoted considerable effort toward improving IHCA survival by improving care delivery during the acute resuscitation phase [4][5][6] and the postresuscitation phase. 7Despite these efforts, IHCA survival rates have plateaued in recent years, with mean survival of approximately 25%. 2,8,9Given the high incidence and low survival for IHCA, efforts targeted toward prevention of IHCA could be impactful.Cardiac arrest in hospitalized patients is rarely sudden, and may be a result of inappropriate triage, delays in diagnosis, or treatment in deteriorating patients.1][12] It is possible that some hospitals have developed innovations or systems of care (eg, rapid response teams) that identify and treat unstable patients before they progress to IHCA.Understanding the extent to which the incidence of IHCA varies across hospitals in the US and the factors associated with this variation is a critical first step toward identifying best practices for IHCA prevention.
To address this gap in knowledge, we linked Get With the Guidelines-Resuscitation (GWTG-R) database with Medicare inpatient claims summarized by hospital to calculate hospital rates of IHCA incidence among Medicare beneficiaries and identify hospital factors associated with IHCA incidence.
An improved understanding of variation in IHCA incidence and the associated factors is important for developing strategies for reducing the burden of IHCA in the United States.

Methods
The University of Iowa institutional review board deemed this cohort study exempt and granted a waiver of informed consent because this study used deidentified data.We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Data Sources
This study was conducted using the following data sources for the study period of 2014 to 2017: (1)   GWTG-R, a large, prospective, multisite registry of IHCA events in the United States that includes detailed information on all confirmed IHCA cases submitted by participating hospitals using standardized Utstein-style definitions, 13 (2) Center for Medicare and Medicaid (CMS) Part A data that include hospitalization data on all Medicare beneficiaries, (3) annual case-mix index files for each hospital, which are publicly reported by CMS and include a hospital measure of case-mix severity, and (4) the American Hospital Association data for information on hospital variables.

Data Linkage
First, using Medicare Part A data, we identified all patients aged 65 years and older admitted to an acute care hospital in the US during 2014 to 2017.We chose this time frame as we had access to Medicare data through 2017.For each hospital, we summed the total number of hospitalizations during each year.Next, we merged this data set with the case-mix index data for that year, using the CMS hospital identifier available in both data sets.The case-mix index was based on the relative weight assigned to the diagnosis-related groups of all hospital discharges at each hospital and is a measure of illness severity of admitted patients.Finally, the aforementioned hospital-level data set was merged with hospital-level GWTG-R data and American Hospital Association data.This last step was performed by the Biostatistics Core at University of Pennsylvania (data analytic center for GWTG-R) using a master file that included a crosswalk between the hospital identifiers in the CMS, American Hospital Association, and the GWTG-R data.Upon completion of this linkage, a

Study Cohort
Within patient-level GWTG-R data, we identified 386 hospitals that included 53 295 patients aged 65 years and older with an index, pulseless cardiac arrest in an inpatient location during 2014 to 2017 (Figure 1).Of these 386 hospitals, 260 (67.3%) hospitals (43 161 patients) had a corresponding match in the linked data set, which provided total number of hospitalizations and case-mix index from CMS data.To ensure that the IHCA incidence rates were stable, we excluded 19 hospitals that reported fewer than 10 IHCA events per year, and 71 hospitals that did not participate in GWTG-R for at least 3 consecutive years.Our final study sample consisted of 38 630 adult patients at 170 hospitals representing 4 473 897 admissions of Medicare enrollees.

Study Variables and Outcomes
The primary outcome of interest was hospital incidence rate of IHCA among Medicare beneficiaries, adjusted for case-mix index.The annual incidence of IHCA for each hospital was calculated by dividing the total number of IHCA events in patients aged 65 years or greater (obtained from GWTG-R) by the total number of hospital admissions during the same year (obtained from CMS data).
For hospitals that did not participate in GWTG-R for the entire 4-year study period, we prorated the total number of admissions (denominator) by the number of months of participation and divided by number of years of participation in GWTG-R to calculate their overall annual incidence of IHCA. 14spital-level variables included bed size, geographic census region, rurality, ownership type, teaching status, and trauma center designation from the American Hospital Association data.We also calculated a measure of nurse staffing at each hospital as ratio of the number of full-time nurse equivalents and the total inpatient days available in the American Hospital Association data.14,15 Patient-level demographic and clinical characteristics for IHCA patients were identified from the GWTG-R.Self-reported demographic characteristics included age at hospital admission, gender, and race and ethnicity.Clinical characteristics included presence of comorbid conditions including current or prior heart failure, current or prior myocardial infarction, diabetes, central nervous system depression, acute neurological nonstroke event, acute stroke, hepatic insufficiency, renal insufficiency, respiratory insufficiency, septicemia, pneumonia, hypotension, major trauma, malignancy, and metabolic or electrolyte abnormality.We also collected year of admission, first pulseless rhythm, event location, time of cardiac arrest (weekday, weeknight, or weekend), the use

Statistical Analysis
We used descriptive statistics for summarizing patient and hospital characteristics-mean (SD) or median (IQR) for continuous variables, and number (percentage) for categorical variables.
Categorical variables were compared using χ 2 or Fisher exact test.Trend tests were calculated using Cochran-Armitage or Cochran-Mantel-Haenzei tests, where appropriate.P < .05 was considered statistically significant.Next, we constructed a hierarchical random effects regression model to calculate case-mix adjusted rates of IHCA incidence for each hospital.Hierarchical models account for clustering of patients within hospitals thus reducing overestimation of statistical significance. 16In these models, IHCA events per-hospital were modeled using a logit link and binomial distribution, with hospital site included as a random effect.To ensure that hospital rates of IHCA incidence were not confounded by differences in case-mix severity across hospitals, we adjusted hospital IHCA incidence rates for case-mix index obtained from CMS.Using regression coefficients from this model, we estimated each hospital's incidence rate for IHCA as the ratio of predicted-to-expected incidence multiplied by the overall unadjusted incidence rate during the study period.Compared with the observed-to-expected ratio, the predicted-to-expected ratio does not unfairly penalize small volume hospitals by accounting for the lower precision in survival estimates from such volume hospitals. 17nfidence intervals for risk-adjusted incidence rates were estimated using a bootstrap approach.
Specifically, 1000 hospital samples were generated with replacement, which were then used to estimate the 2.5th and 97.5th percentile risk-adjusted incidence rates over repeated modeling.To quantify the magnitude of hospital-level variation in IHCA incidence, we calculated a median odds ratio from the aforementioned hierarchical model using the variance estimate of the random hospital intercept. 18 categorized study hospitals into quartiles based on the case-mix adjusted IHCA incidence and examined patient-and hospital-variables across quartiles of IHCA incidence using linear regression for continuous variables and χ 2 test for categorical variables.Finally, we evaluated the association between hospital variables and case-mix adjusted IHCA incidence using the same hierarchical regression models previously described.All analyses were performed using SAS software version 9.4 (SAS Institute) from January to December 2021.
After adjusting for case-mix index, the median odds ratio for hospital IHCA incidence was 1.51 (95% CI, 1.44-1.58),suggesting that the relative odds that a hospitalized patient at one randomly selected hospital would experience an IHCA were 51% higher compared with a similar patient admitted at another randomly selected hospital with identical case-mix.In addition to case-mix index, higher nurse staffing was associated with a lower incidence of IHCA (odds ratio [OR], 0.96 [95% CI, 0.92-0.99];P = .007)(Table 3).Hospital teaching status was also associated with IHCA incidence, although the effect estimate was similar for minor teaching and nonteaching hospitals, an association was found only for minor teaching hospitals (OR, 0.74 [95% CI, 0.58-0.93])(Table 3).
The addition of hospital variables to the hierarchical model explained 12% of hospital variation in IHCA, and the median OR after adjusting for hospital variables was 1.47 (95% CI, 1.41-1.54).

Discussion
In this large cohort study of 170 GWTG-R hospitals linked with Medicare data, we found that the median risk-adjusted incidence of IHCA among Medicare beneficiaries was 8.5 per 1000 admissions.
Importantly, we found substantial variation in the incidence of IHCA across study hospitals, ranging from 2.4 per-1000 to 25.5 per-1000 admissions with a median odds ratio of 1.51.In adjusted analyses, The x-axis represents each hospital arranged in the ascending order of IHCA incidence, and the y-axis represents the IHCA incidence expressed as per-1000 admissions.Error bars represent 95% CIs.The horizontal dotted line represents the median adjusted IHCA incidence (8.5) for our study.The graph is color coded for dark blue = quartile 1, cyan = quartile 2, brown = quartile 3, and orange = quartile 4.
we found that higher levels of nurse staffing and minor teaching status was associated with lower incidence of IHCA.Furthermore, adjustment for differences in hospital variables including case-mix explained only 12% of the variation in IHCA incidence, suggesting that unmeasured differences in hospital care processes likely account for the large variation in IHCA incidence.Our findings are important and merit further discussion.
The overall incidence of IHCA in our study (8.5 per1000 admissions) was higher than the reported incidence of 4.02 per 1000 admissions in an earlier GWTG-R study that used 2000 to 2009 data. 14Although recent studies have reported an increase in the incidence of IHCA, 1 we attribute the higher incidence in the current study to the inclusion of only Medicare beneficiaries who represent an older, higher risk population.Although the incidence rate of IHCA in our study may not be <.001 Acute CNS nonstroke event 2515 ( studies have obtained data on total admissions from the American Hospital Association data, which does not differentiate adult from pediatric admissions.Therefore, IHCA incidence rates are likely to be underestimated in these studies especially at hospitals that treat a large number of pediatric patients. 1,19Second, linkage with Medicare files also provided data on case-mix severity in the denominator population of admitted patients.Adjustment for case-mix in the calculation of IHCA incidence overcomes an important limitation of prior studies because case-mix index information is not available in the American Hospital Association data. A notable finding of our study was the large variation in IHCA incidence across hospitals, which highlights that some hospitals are able to achieve a lower incidence of IHCA compared with others, adjusted for case-mix.Numerous studies have shown that a large proportion of IHCA events in hospitalized patients may be preventable and may occur because of inappropriate triage, delays in recognition, diagnosis and treatment of life-threating conditions, or development of new problems or complications among admitted patients (ie, failure to rescue). 10,20,21In adjusted analyses, we found higher nurse staffing was associated with lower incidence of IHCA, which could be due to nurse surveillance of at-risk patients and prompt activation of hospital emergency response teams for patients experiencing deteriorating conditions.These findings are consistent with prior studies that have shown higher levels of nurse staffing to be associated with lower hospital mortality and higher quality of care. 22,23Unlike other hospital structural variables such as teaching status, which are usually not modifiable, nurse staffing is modifiable.Future studies are needed to confirm or refute whether improving nurse staffing at hospitals with higher IHCA incidence are associated with a reduction in IHCA and other adverse patient outcomes.
We also found that adjustment for hospital variables and case-mix accounted for a small proportion of the variation in IHCA incidence across hospitals, which suggests that some hospitals may have developed innovative care processes that may enable earlier recognition and treatment of patients experiencing deteriorating conditions before they progress to IHCA.In a recent study, we found substantial differences across hospitals in the organization and structure of rapid response teams-a key hospital intervention that is focused toward preventing IHCA. 24Likewise, others have developed the electronic Cardiac Arrest Risk Triage system that uses real-time data in the electronic health record to forewarn bedside nurses and physicians about the risk of impending IHCA. 25,26rrently, neither GWTG-R nor other existing hospital databases collect information on these innovative processes of care and treatment strategies.Future research focused on hospitals that have achieved exceptionally low incidence of IHCA may yield important insights regarding care processes and organizational variables (best practices) for prevention of IHCA among hospitalized patients. 27

Limitations
Our study findings should be in the context of the following limitations.First, due to linkage with Medicare data, we were only able to calculate IHCA incidence among patients aged 65 years and older.Although the IHCA incidence rates may not be generalizable to younger patients, there is no reason to believe that hospitals in the highest quartile of IHCA for Medicare patients will be a lower quartile for younger patients given that processes of care for IHCA prevention do not differ for Medicare and non-Medicare patients at a hospital.Second, information on hospital-level variables was limited to only hospital structural variables, and we lacked information on modifiable hospital process for IHCA (eg, triage processes, composition, and structure of rapid response teams, quality improvement efforts, etc.) that possibly underlie variation in IHCA incidence.Based on our findings,

Figure 2 .
Figure 2. Hospital Variation in the Case-Mix Adjusted Incidence of IHCA Among Medicare Beneficiaries Per 1000 Admission Variation Across Hospitals in In-Hospital Cardiac Arrest Incidence Among Medicare Beneficiaries JAMA Network Open.2022;5(2):e2148485.doi:10.1001/jamanetworkopen.2021.48485(Reprinted)February 28, 2022 2/12 Downloaded From: https://jamanetwork.com/ on 09/16/2023deidentified data set was provided to the study team for analysis to ensure that the study authors remained blinded to the identity of the hospitals included in the study.

Table 2 .
Patient Characteristics by Quartiles of Case-Mix Adjusted IHCA Incidence a

on 09/16/2023 generalizable
to younger patients, there are several strengths of our study that are important to emphasize.First, linkage with Medicare files provided accurate data on the total number of admissions, which was the denominator used for calculation of IHCA incidence.In contrast, prior Downloaded From: https://jamanetwork.com/

Table 2 .
Patient Characteristics by Quartiles of Case-Mix Adjusted IHCA Incidence a (continued) a Table shows the association between patient-level characteristics and case-mix adjusted IHCA incidence divided into quartiles.Categorical variables were compared using χ 2 or Fisher exact test.Trend tests were calculated using Cochran-Armitage or Cochran-Mantel-Haenzei tests, where appropriate.

Table 3 .
Association of Hospital Variables With Incidence of IHCA Abbreviations: FTEs, full-time equivalents; IHCA, in-hospital cardiac arrest; OR, odds ratio.