a b s t r a c t

Background: Variation in Hospital admission rates of patients presenting to the emergency department (ED) may represent an opportunity to improve practice. We seek to describe national variation in hospital admission rates from the ED and to determine the degree to which variation is not explained by patient characteristics or Hospital factors.

Methods: We conducted a cross-sectional analysis of a nationally Representative sample of ED visits among adults within the 2010 National Hospital Ambulatory Care Survey ED data of hospitals with admission rates from the ED between 5% and 50%. We calculated risk-standardized hospital admission rates (RSARs) from the ED using contemporary hospital profiling methodology, accounting for patients’ Sociodemographic and clinical characteristics.

Results: Among 19831 adult ED visits in 252 hospitals, there were 4148 hospital admissions from the ED. After accounting for patients’ socioDemographic and clinical factors, the median RSAR from the ED was 16.9% (interquartile range, 15.0%-20.4%), and 8.1% of the variation in RSARs was attributable to an institution- specific effect. Even after accounting for hospital teaching status, ownership, urban/rural location, and geographical location, 7.0% of the variation in RSARs from the ED was still attributable to an institution- specific effect.

Conclusions and relevance: There was variation in hospital admission rates from the ED in the United States, even after adjusting for patients’ sociodemographic and clinical characteristics and accounting for hospital factors. Our findings suggest that suggesting that the likelihood of being admitted from the ED is not only dependent on clinical factors but also at which hospital the patient seeks care.

Introduction

A large and increasing proportion of hospitalized patients are admitted via the emergency department (ED) [1,2]. The decision to admit a patient from the ED has substantial consequences not only for each patient but also for the health care system. Often the decision to admit a patient is straightforward because the condition unequivo- cally mandates admission, such as an acute myocardial infarction.

* Corresponding author. University of Colorado, Medical School, Leprino Building, Office number 755, Aurora, CO 80045. Tel.: +1 720 848 4358.

E-mail addresses: [email protected], [email protected] (R. Capp).

¹ RC performed the analysis and manuscript writing while a fellow at Yale University but is an assistant professor at the University of Colorado at the time of this submission.

However, for other conditions or symptoms, the admission decision is discretionary because there may be uncertainty about the marginal benefit of hospitalization [3-5].

Although previous studies have illustrated variation in the utilization of health care services, we know little about variation in hospital admission rates for patients who initially present to the ED. Recently, 1 study by Pines et al [6] found 2.5-fold variation in hospital admissions from the ED and determined that hospital factors, such as teaching status, for-profit ownership, and trauma expertise, were associated with higher hospital admission from ED. However, their study, although useful, lacked detailed clinical data and could not investigate whether variation in patients’ clinical characteristics explained the variation in hospital admission from the ED. In addition, we know little about the role of the importance of individual

http://dx.doi.org/10.1016/j.ajem.2014.03.033 0735-6757/

Sociodemographic variables“>institutions, after accounting for clinical and hospital factors, in explaining any variation.

Using contemporary hospital profiling methodology, similar to that used by the Center for Medicare and Medicaid Services for their mortality and readmission measures, we analyzed data from the ED component of the 2010 National Hospital Ambulatory Medical Care Survey [7-12]. The database provides a representative sample of the nation’s hospitals and possesses detailed clinical data on each patient. Accordingly, we sought to determine the extent of variation in risk-standardized hospital admission rates (RSARs) for patients presenting to the ED, after accounting for patient socio- demographic and clinical characteristics, along with hospital factors. We also sought to determine a specific institutional effect, which can describe the extent to which variation is explained by institutional or performance factors of individual hospitals.

Methods

Study setting and data source

We analyzed data from the 2010 NHAMCS [13]. The NHAMCS is an annual, national probability sample of ambulatory visits made to nonfederal, general, and short-stay hospitals conducted by the Centers for Disease Control and Prevention and National Center for Health Statistics. The multistaged sample design is composed of 3 stages for the ED component: (1) 112 geographic primary sampling units (PSUs) that comprised a probability subsample of PSUs from the 1985-1994 National Health Interview Surveys, (2) approximately 480 hospitals within PSUs, and (3) patient visits within emergency service areas. Sample hospitals are randomly assigned to 16 panels that rotate across thirteen 4-week reporting periods throughout the year to minimize seasonal influences [13].

Hospital staff or Census Bureau field representatives complete a patient record form for each sampled emergency service area visit based on information obtained from the medical record. The data collected include information on patient demographics; reasons for visit; vital signs; cause(s) of injury; diagnoses rendered; diagnostic tests ordered; procedures provided; medications prescribed; pro- viders seen; and disposition, including hospital discharge information if admitted. As part of the quality assurance procedure, a 10% quality control sample of patient record forms is independently keyed and coded. Error rates typically range from 0.3% to 0.9% for various survey items [13].

The 2010 NHAMCS data sources included, respectively, 373 participating hospitals (unweighted sampling response rate of 90%) and a total of 34936 patient visits to the ED [13]. The NHAMCS includes patient- and hospital-level sampling weights, which have been adjusted by National Center for Health Statistics (NCHS) for survey nonresponse to yield an unbiased national estimate of ED visit occurrences, percentages, and characteristics. As a publicly available data set with no patient identifiers, this study was exempt from review by the Institutional Review Board of Yale University, New Haven, CT.

Study cohort

Among all patient visits to the ED included within NHAMCS, we excluded patients who were younger than 18 years (n = 8015; 23%), Left without being seen or left before completion of examination or left against medical advice (n = 813; 2%), transferred to another facility (n = 626; 1.7%), with missing data on discharge disposition (n = 100; 0.3%), died on arrival or died in the ED (n = 60; 0.2%). Finally, we excluded hospitals with less than 30 visits to comply with reliable relative SEs (n = 307; 0.9%), as recommended by the Centers for Disease Control and Prevention, and hospitals with less than 5% and greater than 50% admission rates (n = 5184; 14.8%) as these are unlikely to represent true admission rates of hospitals across the

United States [13]. The final study cohort included 19831 adult ED visits in 252 hospitals.

Outcomes

Our primary outcome was risk-standardized hospital admission from the ED, which is defined using 2 NHAMCS items: “admit to this hospital” or “admit to observation unit, then hospitalized.” Our secondary outcome was to determine the extent to which the variation in RSAR was due to the performance or hospital factors, rather than patient sociodemographic or clinical factors.

Patient-level variables

Sociodemographic variables. Demographic variables included age, sex, race/ethnicity, and insurance status. The NHAMCS survey tool categorizes race/ethnicity into 4 categories: white, black, Hispanic, and other. Insurance type was classified as private insurance, Medicare, Medicaid, uninsured, and other/unknown.

For socioeconomic status, we used the percent poverty and median income in the patient’s zip code. The percentage of poverty, based on the patient’s zip code and US Census Bureau definitions, is classified by the NHAMCS as missing data, less than 5%, 5% to 9.9%, 10% to 19.9%, and 20% or more.

• Clinical variables. We categorized patient’s vital signs into categories adopted from the modified early warning score [14,15]. We considered the following values as missing because they were implausible and likely data collection errors: respiratory rate less than 4 or greater than 60 breaths per minute, systolic blood pressure less than 50 mm Hg or greater than 260 mm Hg, body temperature less than 90?F or greater than 107?F, and pulse less than 10 beats per minute.

To adjust for clinical case mix, we adjusted for selected chief complaints and severity of illness based on ED primary discharge diagnosis using the International Classification of Diseases, Ninth Revision, codes. We assessed the frequency of each chief complaint and the associated hospital admission rate. To account for frequent chief complaints with high hospital admission rates, we first identified those associated with an admission rate that exceeded 30% and was present in 1% or more of patient visits. Eight chief complaints met both criteria: chest pain and related symptoms, shortness of breath, other symptoms/probably psychologically related, general weakness, la- bored or difficulty breathing, fainting (syncope), unconscious arrival, and other symptoms referable to the nervous system. We used the Clinical Classification Software to group conditions into clinical meaningful categories.

To quantify severity of illness, we used a previously validated

method (Billings criteria) based on primary ED discharge International Classification of Diseases, Ninth Revision, codes provided by NHAMCS [16,17]. This method categorizes severity of illness into 4 categories: nonemergent, emergent, indeterminate, or unclassified. The unclas- sified category is composed of mental health, alcohol, substance abuse, or unclassifiED diseases [17,18]. Although the Billings criteria is controversial in determining which patients seek ED for emergent vs nonemergent conditions, it has been validated as a criteria that can be used to determine high or low probability of hospital admission based on the final ED diagnosis [17]. The NHAMCS also provides information on patient’s means of arrival, such as arrival by ambulance.

Hospital factor variables

The NHAMCS provided information about the region of the country, hospital ownership type, and urban or rural location. We defined a teaching hospital as any facility in which a resident evaluated a patient in at least 1 ED visit [19].

Table 1

Patient characteristics

Variables Total population (n = 19831) Patients admitted from the ED (n = 4148)

	Weighted %a		Weighted %a	Unadjusted ORa
Sex
Females	58.0%		54.9%	Referent
Males	42.0%		45.1%	1.17
Age, median (IQR)	46.4 (45.8-47.0)		59.6 (58.7-60.6)	1.04
Arrival by ambulance
Yes	20.6%		42.9%	4.56
No	74.6%		51.3%	Referent
Unknown	4.7%		5.8%	2.05
Race/ethnicity
White	64.9%		69.4%	Referent
Black	20.7%		18.2%	0.79
Hispanic	11.6%		9.5%	0.72
Other	2.7%		2.8%	0.96
poverty level by patient’s zip code
b 5%	14.4%		15.8%	Referent
5%-9.99%	24.9%		26.4%	0.96
10%-19.99%	34.6%		33.7%	0.86
N 20%	21.3%		19.0%	0.77
Unknown	4.8%		5.1%	0.98
Insurance
Private insurance	30.3%		24.0%	Referent
Medicare	23.5%		45.0%	3.34
Medicaid	20.2%		16.6%	1.05
Uninsured	21.1%		10.1%	0.56
Unknown or missing	4.9%		4.4%	1.10
Vital signs
Temperature (?F) 90-95	0.3%		0.5%b	1.77
95.1-100.4	93.3%		88.7%	Referent
100.5-107	2.1%		5.0%	3.92
Missing data	4.3%		5.7%	1.57
Heart rate (beats per minute)
10-60	3.4%		3.6%	1.24
61-100	71.7%		61.4%	Referent
N 101	19.9%		27.4%	1.75
Missing data	4.9%		4.9%	1.15
Respiratory rate (breaths per minute) 4-11	45 (0.2)		0.4%b	3.03
12-20	21613 (86.7)		75.0%	Referent
21-60	2418 (9.9)		20.9%	3.51
Missing data	939 (3.2)		3.6%	1.38
Systolic blood pressure (mm Hg)
50-90	1.0%		3.1%	6.74
91-160	84.0%		76.0%	Referent
161-260	13.0%		18.9%	1.87
Missing data	2.0%		2.0%	1.14
Chronic diseases
Congestive heart failure	4.5%		13.5%	6.83
Cerebrovascular disease	3.8%		9.0%	3.92
Diabetes	12.0%		23.0%	3.00
HIV	0.5%		0.8%	2.01
On dialysis	1.1%		3.2%	5.38
Severity of illness
Emergent	17.8%		41.0%	7.42
Nonemergent	43.8%		22.5%	Referent
Indeterminate	1.2%		2.6%	6.24
Alcohol, Mental illness, unclassified	38.2%		33.9%	1.84
Primary chief complaint present on arrival
Chest pain and related symptoms	7.0%		13.2%	2.67
Shortness of breath	3.5%		9.6%	5.45
Other symptoms/probably psychologically related	1.2%		3.7%	6.50
General weakness	1.2%		3.0%	4.21
Labored or difficult breathing	0.8%		1.9%	3.17
Fainting (syncope)				3.09
Unconscious on arrival	0.5%		1.2%	4.98
Other symptoms referable to the nervous system	0.4%		1.1%	4.14

Values may not add up due to rounding.

^a Represents weighted values.

^b Values less than 30, analysis considered unreliable by the Center for National Health Statistics.

Table 2

Distribution of patient mix among hospitals (n = 252)

Patient characteristics used in to calculate RSAR	Median (IQR)
Sex % Females	57.5 (52.3-61.9)
% Males	42.5 (38.1-47.7)
Mean age	46.2 (43.5-49.5)
Arrival by ambulance	19.4 (13.9-26.2)
Race/ethnicity % White	68.6 (42.5-85.7)
% Black	13.0 (4.1-27.9)
% Hispanic	5.4 (2.4-16.4)
% Other	1.2 (0.0-3.9)
Poverty level by patient’s zip code % b5%	8.2 (1.6-21.0)
% 5%-9.99%	21.6 (10.6-34.2)
% 10%-19.99%	31.4 (16.4-51.1)
% N 20%	10.4 (2.2-37.1)
% Unknown	2.3 (0.8-5.7)
Insurance % Private insurance	29.8 (20.3-40.2)
% Medicare	25.0 (17.7-30.4)
% Medicaid	18.0 (10.5-26.1)
% Uninsured	17.5 (10.3-28.3)
% Unknown or missing Vital signs Temperature (?F)	1.5 (0.0-5.0)
% 90-95	0.0 (0.0-0.0)
% 95.1-100.4	94.8 (91.9-96.8)
% 100.5-107	1.9 (1.1-2.9)
% Missing data	2.6 (1.1-5.1)
Heart rate (beats per minute) % 10-60	3.2 (1.7-4.9)
% 61-100	74.3 (70.0-79.3)
% N 101	19.3 (10.9-23.6)
% Missing data	0.0 (0.0-3.3)
Respiratory rate (breaths per minute) % 4-11	0.0 (0.0-0.0)
% 12-20	87.5 (83.1-91.0)
% 21-60	9.5 (6.4-13.1)
% Missing data	1.3 (0.0-3.3)
Systolic blood pressure (mm Hg) % 50-90	0.8 (0.0-1.5)
% 91-160	85.1 (80.6-88.5)
% 161-260	11.9 (8.3-16.2)
% Missing data	1.1 (0.0-2.4)
Chronic diseases % Congestive heart failure	0.0 (0.0-0.1)
% Cerebrovascular disease	0.0 (0.0-1.9)
% Diabetes	11.2 (7.4-15.2)
% HIV	3.9 (1.9-6.4)
% On dialysis Severity of illness	2.2 (0.0-4.2)
% Emergent	16.9 (13.6-20.3)
% Nonemergent	42.3 (37.7-46.5)
% Indeterminate	1.1 (0.0-1.8)
% Alcohol, mental illness, unclassified	38.5 (34.7-43.8)
Primary chief complaint present on arrival % Chest pain and related symptoms	6.0 (4.1-8.6)
% Shortness of breath	3.1 (1.4-4.8)
% Other symptoms/probably psychologically related	1.1 (0.0-2.1)
% General weakness	1.1 (0.0-2.0)
% Labored or difficult breathing	0.0 (0.0-1.5)
% Fainting (syncope)	0.0 (0.0-0.0)
% Unconscious on arrival	0.0 (0.0-1.5)
% Other symptoms referable to the nervous system	0.0 (0.0-0.0)

Statistical analyses

First, we examined the prevalence of the selected variables and the association of the selected variables with hospital admission using descriptive analysis and logistic regression models. Inter-

quartile ranges (IQRs) are reported with median results for continuous variables. Odds ratios (ORs) and 95% confidence intervals (CIs) of hospital admissions are reported for variables used in the logistic regression model. We used PROC SURVEY procedures in SAS to take into account the clustered nature of the sample and the sampling design of the NHAMCS. Although race, poverty level, and 3 chief complaints (syncope, labored breathing, and unconscious on arrival) were not statistically significant, these variables were considered clinically important enough for inclusion in the model.

We developed 2 hierarchical logistic regression models (HLRMs). To examine the variation that exists while accounting for patient and clinical factors, the first HLRM included only patient-level variables. We used this model to calculate RSARs and used the results of this model to study the association between the RSAR and hospital factors available in the NHAMCS survey. For the evaluation of this association, we used the linear regression model based on the corresponding hospital-level data. To assess for an institutional effect, we performed a second HLRM that included patient- and hospital-level variables. The intraclass correlation coefficient of patients within hospitals was calculated from this model. In both HLRMs, patient weight was incorporated in the analyses, and in the linear regression model at the hospital-level data, hospital weight was incorporated in the analyses.

In our study, the hospital RSAR was calculated as a ratio of the number of predicted hospital admissions in the hospital to the number of expected hospital admissions in the hospital. This approach is congruent to the methodology used by the Center for Medicare and Medicaid Services to publicly report hospital risk-standardized rates of mortality and readmission for acute myocardial infarction, heart failure, and pneumonia [10,11,20]. This ratio is then multiplied by the national unadjusted rate of hospital admissions. For each hospital, the number of the predicted hospital admissions is the sum of the predicted hospital admission rates of all patients in the hospital based on the hospital’s performance with its observed case mix; the number of the expected hospital admissions is the sum of the expected hospital admission rates of all patients in the hospital based on performance of the nation’s “average” hospital with this hospital’s case mix. The minimum number of cases required to estimate reliable estimates is 25. We calculated the c-statistic of the HLRM model to assess discrimination. All the analyses were done using SAS 9.2 version (SAS, Cary, NC).

Results

There were 19831 adult ED visits in 252 hospitals within the 2010 NHAMCS ED data included in our study. Patient median age was 46.4 years (IQR, 45.8-47.0), and most were female (58.0%), white (64.9%), and had private insurance (30.3%) (Table 1). There were 4148 patients (20.6%) admitted to the hospital from the ED.

Accounting for patient sociodemographic and clinical characteristics

Patient sociodemographic and clinical characteristics varied among hospitals (Table 2), and several were significantly associated with hospital admission, including chronic diseases, severity of illness, and abnormal vital sign values (Appendix 1). In particular, hospitals’ patient populations varied with respect to the proportion insured by Medicare and Medicaid (median percent, 25.0 [IQR, 17.7-30.4] and 18.0 [IQR, 10.5-26.1]). Furthermore, numerous clinical characteristics also varied among hospitals, such as percent severity of illness: emergent 16.9 (13.6-20.3) vs nonemergent 42.3 (IQR, 37.7-46.5) and chief complaints of chest pain 6.0 (IQR, 4.1-8.6) and shortness of breath 3.1 (IQR, 1.4-4.8) (Table 2).

In the multivariate logistic regression analysis patients with Medicare (OR 1.2; 95% CI, 1.0-1.4) and Medicaid (OR 1.2 95% CI, 0.9-1.4) were

Table 3

Crude hospital admission rates from the ED and RSARs

Description Hospital admission rates from the ED^a RSARs

nonprofit (18.0%; 95% CI, 16.9%-19.0%) and private (17.8%; 95% CI,

16.1%-19.5%) when compared with government, nonfederal hospitals

(16.6%; 95% CI, 13.9%-19.4%) (Table 4), as they were among teaching

100 percentile 95 percentile	50.0% 36.1%	40.6% 26.3%	hospitals (18.7%; 95% CI, 17.2%-20.1%) and hospitals located in urban areas (19.2%; 95% CI, 18.4%-20.1%) when compared with nonteaching
90 percentile	31.5%	24.6%	hospitals (17.2%; 95% CI, 15.9%-18.5%) and hospitals located in rural
75 percentile	23.2%	20.4%	areas (15.5%; 95% CI, 13.6%-17.4%), respectively.

50 percentile 17.3% 16.9%

25 percentile 12.3% 15.0%

10 percentile 8.6% 12.1%

5 percentile 6.3% 10.5%

0 percentile 5.0% 7.5%

IQR 11 5.5

Mean 18.7% 17.8%

^a Weighted values.

more likely to be admitted from the ED when compared with those who have private insurance (Appendix 1). Furthermore, numerous clinical characteristics were also associated with higher likelihood of hospital admission from the ED: emergent conditions (OR, 4.3; 95% CI, 3.6-5.0) when compared with nonemergent conditions as well as chief complaints of chest pain (OR, 1.8; 95% CI, 1.5-2.1) and shortness of breath (OR, 2.0; 95% CI, 1.5-2.6). Percentage of patients with Abnormal vital signs also varied among hospitals (Table 2) and was associated with higher odds of ED hospital admission (Appendix 1). We also categorized ED visits to low (15% ED admissions), medium (15%-20% ED admissions), and high (N 20% ED admissions) to demonstrate the variation of the top 15 primary ED diagnosis for hospitals in these 3 categories (Appendix 2). After accounting for patients’ sociodemographic and clinical factors, there was wide variation in hospital RSARs. The median RSAR from the ED was 16.9% (IQR, 15.0%-20.4%) (c-statistics, 0.84; Table 3; Fig.). The intraclass correlation coefficient of patients within hospitals was 0.08, indicating that 8% of the variation in RSARs from the ED is attributable to an institution-

specific effect.

Accounting for hospital factors

Hospital factors varied across hospitals, several of which were associated with higher RSARs (Table 4). Risk-standardized hospital admission rates were higher among hospitals that were voluntary

After accounting for hospital teaching status, ownership, urban/ rural location, and geographical location, in addition to patient sociodemographic and clinical characteristics, there remained wide variation in hospital RSARs. The median RSAR from the ED was 19.0% (IQR, 16.1%-21.4%; c-statistic, 0.84). Moreover, the intraclass correla- tion coefficient of patients within hospitals was reduced only to .07, indicating that 7% of the variation in RSARs from the ED is attributable to an institution-specific effect.

Discussion

In our study, we found that there was variation among institutions in RSARs from the ED. Although we are unable to determine if this variation in ED admissions is high or low, we believe that there is room for improvement. Patient sociodemographic and clinical characteristics and hospital factors were both strong predictors of admission, but some variation remained. Moreover, we identified an institutional effect, determining that the likelihood of being admitted from the ED is dependent, to a certain extent, on the hospital where patients seek care as opposed to the patient’s clinical or demographic characteristics.

There are a number of potential explanations for our finding that much of the variation in admission rates from the ED resides at the level of the individual hospital, including aspects of care both extrinsic and intrinsic to the hospital. Extrinsic hospital factors, such as the presence of an efficient network of outpatient providers with prompt follow-up care or highly effective community social services, are associated with low ED hospital admissions [6,21]. Intrinsic hospital factors, such as higher intensity Treatment practices or poor adherence to clinical practice guidelines, may also contribute to the variation in RSAR [21,22]. There may also be a contribution of the local culture that influences the propensity to admit patients as well as factors related to bed availability and incentives based on time, convenience, and payment.

Fig. Risk-standardized hospital admission rates from the ED.

Table 4

Hospital factors and differences in RSAR.

Characteristics	Hospitals, n = 252	Mean, n = 252	95% CI of	P
	weighted %a	RSAR a	RSAR
Ownership Voluntary	78.5%	18.0%	16.9%-19.0%	.53
nonprofit Government,	11.8%	16.6%	13.9%-19.4%
nonfederal Proprietary	9.75%	17.8%	16.1%-19.5%
Teaching status Yes	41.5%	18.7%	17.2%-20.1%	b.14
No Urban or rural status Urban	58.5% 61.8%	17.2% 15.5%	15.9%-18.5% 13.6%-17.4%	b.0002
Rural Region	38.2%	19.2%	18.4%-20.1%	b.0001
Northeast	17.1%	19.1%	17.5%-20.6%
Midwest	26.2%	18.0%	15.9%-20.0%
South	39.8%	17.4%	15.8%-19.0%
West	16.9%	17.2%	14.3%-20.1%

^a Weights were applied.

Several studies on overall hospital admissions, not just those from the ED, suggest that supply-sensitive care, the delivery of which has been shown to be associated with the supply of resources, plays a significant role in variations of hospital admissions [23,24]. For example, a recent study by Pines et al

[6] focused on investigating which supply side factors are most correlated with variations in hospital admission rates from the ED. They found factors such as higher number of inpatient beds, trauma center status, lower ED volumes, and higher hospital occupancy rates were associated with higher hospital admission rates from the ED.

Our study adds to Pines et al [6], in that we were able to account for important clinical factors that can determine whether a patient needs to be hospitalized. Our data showed that clinical factors play a role in variation of ED hospital admission rates, and contrary to previous studies, our estimates accounted for key patient characteristics, including clinical characteristics, such as chief complaints and vital signs, and demonstrated good discrim- ination. However, despite accounting for clinical and hospital factors, a significant amount of variation in hospital admissions from the ED remained.

This study has several implications for practice and policy. We found that although clinical characteristics account for some of the variation in hospital admissions through the ED, the hospital where an individual seeks care also plays an important role on whether an individual gets admitted through the ED. The biggest contributor to hospital readmission rates for acute myocardial infarction, congestive heart failure, and pneumonia is the hospital’s overall admission rate [25]. Thus, efforts in reducing hospital admissions may need to come, at least in part, from within each hospital, as individual hospitals may have unique needs and assets that it can use to achieve that goal. If future measures were directed at the provider or community level, it would obscure the unique institutional effect that may reveal true opportunities for improvement.

This study has several limitations. We are unable to assess for certain hospital factors that may play a role in ED RSARs, such as ED crowding, number of Hospital beds, number of Previous ED visits, and hospital trauma status, as these factors were not available in our data set. The number of patients per hospital is small, and there is the possibility that some hospitals have higher number of diagnoses that are often associated with hospital admission; however, we have adjusted for several clinical factors that have a strong association with hospital admission through the ED. One recent article discusses issues with accuracy of hospital admissions to through the ED in NHAMCS; however, comparisons of NHAMCS with the National Emergency Department Sample (the largest all payer Claims data set of EDs) showed similar numbers of hospital admissions through the ED [2,26]. Finally, our sample did not contain the longitudinal data necessary to explain the relationship between ED RSARs and downstream outcomes, such as length of hospital stay, mortality, Hospital readmissions, and unscheduled return ED visits. Finally, the goal of this study was to evaluate the variation of ED hospital admissions at the hospital level while risk standardizing for patient and clinical factors. We did not seek to evaluate the associations of

Appendix 1. Multivariate logistic regression analysis odds ratio and P

values for variables used in the hierarchical logistic regression model

ED hospital admitted patients (n = 4148)

Variables Odds 95% confidence HRLM^b model P

ratio^a interval^a value

Sex .007

Females Reference

Males 1.1 1.0-1.2

Age 1.0 1.0-1.0 b.001

Arrival by ambulance b.001

Yes 2.4 2.1-2.7

No Reference

Unknown 1.8 1.4-2.3

Race/ethnicity b.001

White Reference

Black 0.9 0.8-1.1

Hispanic 1.1 0.9-1.3

Other 1.0 0.7-1.3

Poverty level by patient’s zip code .03

b5% Reference

5%-9.99% 1.0 0.8-1.2

10%-19.99% 0.9 0.7-1.3

N 20% 0.8 0.6-1.0

Unknown 1.1 0.8-1.6

Insurance b.001

Private insurance Reference

Medicare 1.2 1.0-1.4

Medicaid 1.2 0.9-1.4

Uninsured 0.7 0.6-0.8

Unknown or missing 1.0 0.8-1.4

Vital signs

Temperature (?F) b.001

90-95 1.0 0.5-2.0

95.1-100.4 Reference

100.5-107 3.1 2.4-4.1

Missing data 1.2 0.9-1.8

Heart rate (beats per minute) b.001

10-60 0.8 0.6-1.1

61-100 Reference

N 101 1.7 1.5-1.9

Missing data 1.2 0.7-1.9

Respiratory rate (breaths per minute) b.001 4-11 2.4 1.2-4.7

12-20 Reference

21-60 1.8 1.6-2.2

Missing data 1.4 1.0-2.1

Systolic blood pressure (mm Hg) b.001 50-90 3.1 2.2-4.9

91-160 Reference

161-260 1.1 0.9-1.2

Missing data 0.8 0.4-1.4

Chronic diseases

Congestive heart failure 1.7 1.4-2.2 b.001

Cerebrovascular disease 1.5 1.0-2.1 b.001

Diabetes	1.5	1.3-1.7	b.001
HIV	1.9	1.0-3.3	.004

On dialysis 2.7 1.8-4.0 b.001 Severity of illness b.001

Emergent 4.3 3.6-5.0

Nonemergent Reference

Indeterminate 2.2 1.4-3.6

Alcohol, mental illness, unclassified 1.5 1.3-1.7

Primary chief complaint present on arrival

Chest pain and related symptoms 1.8 1.5-2.1 b.001 Shortness of breath 2.0 1.5-2.6 b.001

patients and clinical characteristics and hospital admissions from

the ED.

Other symptoms/probably

psychologically related

5.0 3.2-7.9 b.001

In this study, we found that there was variation in ED RSARs across the United States, even after adjusting for patient sociodemographic and clinical characteristics as well as hospital factors associated with

General weakness 1.9 1.3-2.9 b.001

Labored or difficult breathing 1.1 0.6-1.8 .87

Fainting (syncope) 1.6 0.7-3.8 .01

Unconscious on arrival 1.55 0.9-3.6 .48

hospitalization. Notably, although patient and hospital factors explained some variation in hospital admission rates between EDs, 7% of the variation remains unexplained at the hospital level,

Other symptoms referable to the nervous system

C-STAT 0.820

^a Represents weighted values.

2.05 0.7-2.0 .03

suggesting that the likelihood of being admitted from the ED is not only dependent on clinical factors but also at which hospital the patient seeks care.

^b Represents hierarchical model with patient sociodemographic and clinical factors.

Appendix 2. Primary ED diagnosis for ED visits classified by low, medium, high ED hospital admission percentage

	Hospitals with b15% ED admissions			Hospitals with >=15% and <=20% ED	admissions		Hospitals with N 20% ED admissions
	CCS diagnosis	Weighted percent (n = 3150)		CCS diagnosis	Weighted percent (n = 9292)		CCS diagnosis	Weighted percent (n = 7389)
	Abdominal pain	6.5195		Abdominal pain	5.0093		Abdominal pain	6.3717
	Superficial injury	5.6425		Sprain	4.9225		Chest pain	5.617
	Sprain	4.6956		Superficial injury	4.6019		Sprain	3.9166
	Back problem	4.0875		Chest pain	4.7338		Superficial injury	3.8689
	Chest pain	4.3474		Back problem	4.3828		Back problem	3.5321
	Headache/migraine	3.1295		skin infection	2.6216		Skin infection	3.6638
	Other upper respiratory infection	2.8809		Headache/migraine	2.8713		Other injury	2.4756
	Skin infection	2.511		UTI	2.7876		UTI	2.7249
	UTI	2.7833		Other upper respiratory infection	2.7467		Headache/migraine	2.1115
	Open wound extremity	2.9488		Open wound extremity	2.4121		Other lower respiratory	2.2254
	Other injury	2.6637		Other injury	1.9981		Other upper respiratory infection	1.8814
	Other connective tissue	2.3014		Other lower respiratory	2.0193		Teeth diagnosis	1.8632
	Teeth diagnosis	2.257		Other connective tissue	2.2176		Open wound extremity	1.6766
	Other lower respiratory	1.9144		Teeth diagnosis	2.1588		Other connective tissue	1.804
	Other joint diagnosis	1.654		COPD	1.9162		Unclassified	1.5235

References

1. Schuur JD, Venkatesh AK. The growing role of emergency departments in hospital admissions. N Engl J Med 2012;367(5):391-3.
2. Brignole M, Alboni P, Benditt D, Bergfeldt L, Blanc JJ, Block PE, et al. Emergency department care in the United States: a profile of national data sources. Ann Emerg Med 2010;56(2):150-65.
3. Brignole M, Alboni P, Benditt D, et al. Guidelines on management (diagnosis and treatment) of syncope. Eur Heart J 2001;22(15):1256-306.
4. Koopman MM, Prandoni P, Piovella F, Ockelford PA, Brandjes DP, Van Der Meer J. Treatment of venous thrombosis with intravenous unfractionated heparin administered in the hospital as compared with subcutaneous low-molecular- weight heparin administered at home. The Tasman Study Group. N Engl J Med 1996;334(11):682-7.
5. Levine M, Gent M, Hirsh J, Leclerc J, Anderson D, Weitz J, et al. A comparison of low-molecular-weight heparin administered primarily at home with unfractio- nated heparin administered in the hospital for proximal deep-vein thrombosis. N Engl J Med 1996;334(11):677-81.
6. Pines JM, Mutter RL, Zocchi MS. Variation in emergency department admission rates across the United States. Med Care Res Rev 2013;70(2):218-31.
7. Ross JS, Mulvey G, Stauffer B, Bernheim SM, Keenan PS, Krumholz HM. Statistical models and patient-level predictors of readmission for heart failure: a systematic review. Arch Intern Med 2008;168(13):1371-86.
8. Lindenauer PK, Bernheim SM, Grady JN, Lin Z, Wang Y, Wang Y, et al. The performance of US hospitals as reflected in risk-standardized 30-day mortality and readmission rates for Medicare beneficiaries with pneumonia. J Hosp Med 2010;5(6):E12-8.
9. Bernheim SM, Grady JN, Lin Z, Wang Y, Wang Y, Savage SV, et al. National patterns of risk-standardized mortality and readmission for acute myocardial infarction and heart failure. Update on publicly reported outcomes measures based on the 2010 release. Circ Cardiovasc Qual Outcomes 2010;3(5):459-67.
10. Krumholz HM, Wang Y, Mattera JA, Wang Y, Han LF, Ingber MJ, et al. An administrative claims model suitable for profiling hospital performance based on 30-day mortality rates among patients with heart failure. Circulation 2006;113(13):1693-701.
11. Krumholz HM, Wang Y, Mattera JA, Wang Y, Han LF, Ingber MJ, et al. An administrative claims model suitable for profiling hospital performance based on 30-day mortality rates among patients with an acute myocardial infarction. Circulation 2006;113(13):1683-92.
12. Keenan PS, Normand SL, Lin Z, Drye EE, Bhat KR, Ross JS, et al. An administrative claims measure suitable for profiling hospital performance on the basis of 30-day all- cause readmission rates among patients with heart failure. Circ Cardiovasc Qual Outcomes 2008;1(1):29-37.
13. Center for Disease Control and Prevention. NHAMCS Downloadable Documenta- tion. http://www.cdc.gov/nchs/ahcd/ahcd_questionnaires.htm; 2009 . [Accessed 11/01/2011, 2011].
14. Subbe CP, Kruger M, Rutherford P, Gemmel L. Validation of a modified Early Warning Score in medical admissions. QJM 2001;94(10):521-6.
15. Burch VC, Tarr G, Morroni C. Modified early warning score predicts the need for hospital admission and Inhospital mortality. Emerg Med J 2008;25(10):674-8.
16. Billings J, Parikh N, Mijanovich T. Emergency department use in New York City: a survey of Bronx patients. Issue Brief (Commonw Fund) 2000(435):1-5.
17. Ballard DW, Price M, Fung V, Brand R, Reed ME, Fireman B, et al. Validation of an algorithm for categorizing the severity of hospital emergency department visits. Med Care 2010;48(1):58-63.
18. Billings J, Mijanovich T, Frenkel T, Cantor J. Analysis of selection effects in New York City’s Medicaid managed care population prior to mandatory enrollment. J Urban Health 2000;77(4):625-44.
19. Schuur JD, Venkatesh AK, Bohan JS. Variation in the use of Observation status evaluation in Massachusetts acute care hospitals, 2003-2006. Int J Emerg Med 2010;3(4):367-72.
20. Bratzler DW, Normand SL, Wang Y, O’Donnell WJ, Metersky M, Han LF, et al. An administrative claims model for profiling hospital 30-day mortality rates for pneumonia patients. PloS one 2011;6(4):e17401.
21. Bynum JP, Andrews A, Sharp S, McCollough D, Wennberg JE. Fewer hospitaliza- tions result when primary care is highly integrated into a continuing care retirement community. Health Aff (Millwood) 2011;30(5):975-84.
22. Cresswell J. A hospital war reflects a bind for doctors in the U.S. The New York Times; 2012 [Business].
23. Wennberg JE, Freeman JL, Shelton RM, Bubolz TA. Hospital use and mortality among Medicare beneficiaries in Boston and New Haven. N Engl J Med 1989;321(17):1168-73.
24. Wennberg JE, Freeman JL, Culp WJ. Are hospital services rationed in New Haven or over-utilised in Boston? Lancet 1987;1(8543):1185-9.
25. Epstein AM, Jha AK, Orav EJ. The relationship between hospital admission rates and rehospitalizations. N Engl J Med 2011;365(24):2287-95.
26. Green SM. Congruence of disposition after emergency department intubation in the national hospital ambulatory medical care survey. Ann Emerg Med 2013;61

    (4):423-6 [e428].