At the Intersection of Health, Health Care and Policy Cite this article as: Amber K. Sabbatini, Brahmajee K. Nallamothu and Keith E. Kocher Reducing Variation In Hospital Admissions From The Emergency Department For Low-Mortality Conditions May Produce Savings Health Affairs, 33, no.9 (2014):1655-1663 doi: 10.1377/hlthaff.2013.1318
The online version of this article, along with updated information and services, is available at: http://content.healthaffairs.org/content/33/9/1655.full.html
For Reprints, Links & Permissions: http://healthaffairs.org/1340_reprints.php E-mail Alerts : http://content.healthaffairs.org/subscriptions/etoc.dtl To Subscribe: http://content.healthaffairs.org/subscriptions/online.shtml
Health Affairs is published monthly by Project HOPE at 7500 Old Georgetown Road, Suite 600, Bethesda, MD 20814-6133. Copyright © 2014 by Project HOPE - The People-to-People Health Foundation. As provided by United States copyright law (Title 17, U.S. Code), no part of Health Affairs may be reproduced, displayed, or transmitted in any form or by any means, electronic or mechanical, including photocopying or by information storage or retrieval systems, without prior written permission from the Publisher. All rights reserved.
Not for commercial use or unauthorized distribution Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Emergency Department Use By Amber K. Sabbatini, Brahmajee K. Nallamothu, and Keith E. Kocher 10.1377/hlthaff.2013.1318 HEALTH AFFAIRS 33, NO. 9 (2014): 1655–1663 ©2014 Project HOPE— The People-to-People Health Foundation, Inc.
doi:
Reducing Variation In Hospital Admissions From The Emergency Department For Low-Mortality Conditions May Produce Savings
Amber K. Sabbatini is an instructor of emergency medicine at the University of Washington, in Seattle.
The emergency department (ED) is now the primary source for hospitalizations in the United States, and admission rates for all causes differ widely between EDs. In this study we used a national sample of ED visits to examine variation in risk-standardized hospital admission rates from EDs and the relationship of this variation to inpatient mortality for the fifteen most commonly admitted medical and surgical conditions. We then estimated the impact of variation on national health expenditures under different utilization scenarios. Risk-standardized admission rates differed substantially across EDs, ranging from 1.03-fold for sepsis to 6.55-fold for chest pain between the twenty-fifth and seventy-fifth percentiles of the visits. Conditions such as chest pain, soft tissue infection, asthma, chronic obstructive pulmonary disease, and urinary tract infection were low-mortality conditions that showed the greatest variation. This suggests that some of these admissions might not be necessary, thus representing opportunities to improve efficiency and reduce health spending. Our data indicate that there may be sizeable savings to US payers if differences in ED hospitalization practices could be narrowed among a few of these high-variation, low-mortality conditions. ABSTRACT
A
dmitting a patient to the hospital from the emergency department (ED) is one of the more expensive, routine decisions made in health care. Emergency providers determine whether a patient requires hospitalization or can safely be discharged home about 350,000 times a day1 in the approximately 5,000 US EDs,2 resulting in almost twenty million annual admissions to hospitals.3 The Centers for Medicare and Medicaid Services (CMS) reported that hospital care represented about 30 percent of the $2.7 trillion in total expenditures for 2011—the largest share of health care spending.4 Appropriate admissions from the ED should be both emergent and necessary, dictated by a patient’s diagnosis and clinical presentation.
Brahmajee K. Nallamothu is an associate professor of cardiovascular medicine at the University of Michigan; a core investigator at the Center for Clinical Management Research, Ann Arbor Veterans Affairs Medical Center; a faculty member at the Center for Healthcare Outcomes and Policy; and a faculty member at the Institute for Healthcare Policy and Innovation, all in Ann Arbor. Keith E. Kocher (kkocher@ umich.edu) is an assistant professor in emergency medicine at the University of Michigan; a faculty member at the Center for Healthcare Outcomes and Policy; and a faculty member at the Institute for Healthcare Policy and Innovation.
Certain life-threatening diagnoses, such as acute myocardial infarction, almost always require hospitalization. Yet the majority of admissions from the ED are for intermediate-severity conditions such as acute exacerbations of chronic diseases (for example, asthma exacerbation), with significant variability in presentation.5 In these cases, appropriate dispositions are not always clear, allow for provider and patient discretion, and could be influenced by the local practice culture and resources.6 For example, it is estimated that up to half of patients with congestive heart failure in the ED could be discharged home after a period of observation and treatment. However, many of these patients are admitted, leading to greater resource use.7 As EDs are now the primary source through which paSeptember 2014
33:9
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
H ea lt h A f fai r s
1 655
Emergency Department Use tients with acute illnesses are hospitalized, surpassing direct admissions from ambulatory settings,3,8 differences in admission practices have important implications for the quality and financing of the health care system. Recent studies have demonstrated that allcause admission rates are highly variable across individual providers and EDs.6,9 This variation may ultimately represent the overuse, underuse, or misuse of hospital services. As a result, understanding which conditions show the greatest variation in ED admission practices may lead to improvements in efficiency, as well as cost savings.We therefore examined a national sample of ED visits to compare differences in admission rates for the fifteen most commonly hospitalized medical and surgical conditions. Our goal was to identify conditions that showed the greatest variation and, thus, represent possible targets for standardizing admission practices. We then sought to understand how this variation affects national health spending by exploring potential savings related to narrowing differences in admission rates for each condition.
Study Data And Methods Data Source And Study Population We used the 2010 Nationwide Emergency Department Sample (NEDS) database to conduct the analysis.10 NEDS is part of the Healthcare Cost and Utilization Project (HCUP), a family of longitudinal databases on hospital care in the United States, sponsored by the Agency for Healthcare Research and Quality. It is the largest all-payer database of ED visits in the United States, capturing information on over twenty-nine million ED visits from twenty-eight states, with weights to produce national estimates. All ED visits except for patients who died in the ED were included in the analysis. For the purposes of this study, we considered visits resulting in transfer as admissions, given the high likelihood of admission at the receiving hospital.11 Observation-status cases were not included in this analysis because they are not tracked in NEDS.12 Clinical Classifications Software categories13 identified ED visits for the fifteen most commonly admitted medical and surgical conditions. NEDS also includes linked inpatient information for ED visits that resulted in admission, from which we derived total inpatient charges and in-hospital mortality. Outcome Measures Primary outcomes for this study were unadjusted and risk-standardized admission rates. To ensure valid results, we excluded EDs with fewer than thirty cases for each condition from our calculation of admission rates, as small sample sizes provide inade1656
Health A ffairs
September 2014
33:9
quate representations of health care use. Survey weights were used to generate national estimates of the number of ED visits and admissions, observed inpatient mortality, mean inpatient charges, and total national charges associated with hospitalizations for each condition. Observed inpatient mortality rates were examined to provide clinical context for the average associated severity of each condition. Data Analysis Mixed-effects, hierarchical logistic regression was used to adjust for the severity of patient case-mix and clustering of admissions among hospitals. Covariates in the model included age, sex, Elixhauser comorbidities (a list of thirty comorbid conditions associated with inpatient hospital mortality),14 primary payer (uninsured, private, Medicaid, Medicare, other), and median income of the ZIP code in which the patient resides. This approach assumes that after adjustment for patient case-mix, the remaining variation is due to institutional or community factors that influence an ED’s propensity to admit. Risk-standardized admission rates for each condition were derived from regression models by dividing the number of predicted admissions for each ED, given that institution’s specific casemix, by the expected number of admissions had those patients been treated at the average ED. This predicted-to-expected ratio was then multiplied by the mean admission rate for the representative sample to determine the risk-standardized admission rate. This method of indirect standardization has been well described in the literature as a means for case-mix adjustment when comparing outcomes among hospitals.15,16 Variation is expressed as the ratio of observed and risk-standardized admission rates for EDs at the seventy-fifth to twenty-fifth percentiles. The seventy-fifth and twenty-fifth percentiles were used instead of ninetieth and tenth percentiles to provide a more conservative estimate of variation. Variation between the ninetieth and tenth percentiles is available in the online Appendix.17 We also calculated the coefficient of variation for the distribution of risk-standardized admission rates, which is a unitless measure of dispersion generated by dividing the standard deviation by the mean rate, allowing for comparison between conditions with widely differing mean admission rates. The correlation between the coefficient of variation and observed inpatient mortality was then examined, to assist in understanding the relationship between variation in admission practices and risk of poor clinical outcomes. To illustrate the impact of variation in ED admission rates on national health care spending, we estimated annual national charges under
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
three different utilization scenarios that narrow variation: (1) if EDs with risk-standardized admission rates above the median hypothetically decreased their hospitalization rates to the median for each condition; (2) if EDs with riskstandardized admission rates in the top quartile decreased admissions to the seventy-fifth percentile; and (3) if EDs with risk-standardized admission rates in the top quartile decreased admissions to the seventy-fifth percentile while EDs with risk-standardized admission rates in the bottom quartile increased admissions to the twenty-fifth percentile. This latter scenario is to account for the possibility that some EDs may also be inappropriately underadmitting patients for their case-mix. NEDS contains information on hospital charges only and does not provide hospitalspecific cost-to-charge ratios. Therefore, our analysis is largely limited to charge data. However, we also estimated national health spending by generating an overall cost-to-charge ratio of 0.30 for all US hospitalizations. This ratio is calculated from health expenditure data provided by HCUP18 by dividing total national costs ($371.7 billion) by total national charges ($1.224 trillion) for hospital care. Data management and analysis were performed using Stata software (version 12.1 MP). Additional technical details regarding the NEDS database, the approach to modeling, and calculation of health expenditures are in the online Appendix.17 The Institutional Review Board of the University of Michigan evaluated this study and classified it under “not regulated” status. Limitations Our results should be interpreted in the context of the following limitations. First, this analysis was descriptive and intended to provide a broad picture of variation in ED admissions and, in particular, the implications on health care spending associated with these differences. Our consideration of health spending under the three utilization scenarios does not imply that there is an optimal rate of ED admissions that can be inferred from this study. Second, our description of variation and health spending is at the ED (hospital) level after controlling for patient case-mix. Many factors not assessed in this study can influence an ED’s admission rate in addition to patient factors, such as social conditions, market forces, the medicolegal environment, local access to care,6 and providers’ training and experience. Addressing all of these factors might not be practical to target with policy or quality improvement efforts. Third, although we controlled for severity of case-mix in our estimates by using accepted techniques to produce risk-standardized admission
rates, we recognize that there is likely some portion of unmeasured severity of illness inherent in using administrative data. For example, patients who are admitted may have a greater number of comorbidities listed than patients who are discharged from the ED, which could affect risk adjustment. As a result, certain EDs may be appropriately high admitters and would be unable to reduce hospitalizations without adversely affecting patient outcomes and safety. Conversely, some EDs may be inappropriately low admitters, discharging patients who would have benefited from hospitalization. Any interventions to address systematic overuse, underuse, or misuse of hospital admissions will necessarily have consequences for health spending. Ultimately, variation in admissions must be interpreted in the context of longitudinal outcomes such as patient mortality, unanticipated return visits to the ED, or readmissions, which this study was not designed to evaluate. Finally, our estimates of national health spending are not meant to serve as a formal economic analysis of net expenditures and do not account for increases in outpatient resource utilization that occur by shifting care to ambulatory settings, such as costs associated with observation, office, and home-based care. We are also only able to report estimates derived from hospital charges, adjusted for an overall cost-tocharge ratio for US hospitalizations, which may overestimate the impact on national expenditures. Therefore, if ED admission rates could be reduced for certain conditions, true spending reductions to the health system will be less than our current estimates. However, this work provides an important starting point to use in understanding the scale of national health spending that could be affected by narrowing variation.
Study Results Study Population There were 28,539,883 visits among 961 hospital-based EDs in 2010 (Exhibit 1). Overall, 15.4 percent of the ED visits resulted in admission, with a mean accompanying charge of $34,826. Just over half of the visits (55.6 percent) were by females, with an average age of 38.8 years and 21.6 percent carrying two or more comorbidities. The majority of visits were for patients who had private insurance (30.8 percent), followed by Medicaid (25.6 percent), and then Medicare (20.9 percent). Uninsured patients made up 18.1 percent of the visits. Clinical conditions in the top fifteen admitted medical and surgical diagnoses ranged from chest pain (965,432 total visits and an average of 1,002 cases per ED) to acute renal S e p t em b e r 2 0 1 4
33:9
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Health Affa irs
1657
Emergency Department Use Exhibit 1 Emergency Department (ED) Visit- And Hospital-Level Characteristics Of The Study Population, 2010 a
Characteristic ED disposition admitted Mean inpatient charge (SD)
ED visit level (N = 28,539,883) 15.4% $34,826 (53,046)
Hospital levelb (N = 961) 12.4% $31,856 (17,138)
Patient characteristics Mean age, years (SD) Percent female Percent with 2 or more comorbidities Primary payer (%) Private Medicare Medicaid Uninsured Other
38.8 (24.2) 55.6 21.6
39.9 (5.6) 55.1 19.3
30.8 20.9 25.6 18.1 4.6
31.1 23.0 24.7 16.7 4.7
Household income (%) $39,999 or less $40,000–$49,999 $50,000–$65,999 $66,000 or more
32.9 27.8 22.1 17.2
35.4 30.9 19.1 12.3
Primary diagnosis [CCS category]
Number of visits (% of total visits)
Average number of cases per ED
Chest pain [102] Soft tissue infections [197] Asthma [128]
965,432 (3.4) 755,649 (2.6) 429,853 (1.5)
1,002 813 481
COPD [127] Urinary tract infection [159] Fluid and electrolyte disorders [55]
425,803 (1.5) 699,940 (2.5) 204,658 (0.7)
461 749 231
Biliary tract disease [149] Cardiac dysrhythmias [106] Diabetes with complications [50]
149,161 (0.5) 311,992 (1.1) 176,884 (0.6)
196 348 216
Pneumonia [122] Congestive heart failure [108] Stroke [109]
377,213 (1.3) 215,305 (0.8) 132,585 (0.5)
406 254 185
Acute renal failure [157] Acute myocardial infarction [100] Sepsis [2]
82,595 (0.3) 117,403 (0.4) 182,844 (0.6)
137 165 272
SOURCE Authors’ analyses of data from the 2010 Nationwide Emergency Department Sample database. NOTES SD is standard deviation. CCS is Clinical Classifications Software. COPD is chronic obstructive pulmonary disease. aUnweighted data. bHospitallevel characteristics represent averages per hospital of the visit-level information.
failure (82,595 total visits and an average of 137 per ED). Admission Practices There was substantial variation in unadjusted ED admission rates (see online Appendix Exhibit 1)17 and riskstandardized admission rates (see online Appendix Exhibit 2)17 for many of the conditions studied. Observed differences in admission rates ranged from a low of 1.02 for sepsis to 2.60 for chest pain at the seventy-fifth and twentyfifth percentiles (Exhibit 2). After risk adjustment, the degree of variation between EDs widened. Those conditions showing the greatest variation included chest pain (risk-standardized admission rate ratio: 6.55), soft tissue infection (risk-standardized admission rate ratio: 3.40), and asthma (risk-standardized admission rate 1658
Health Affairs
S ep t e m b e r 20 1 4
33 : 9
ratio: 3.07). In contrast, higher-severity conditions with less diagnostic and prognostic uncertainty, such as sepsis (risk-standardized admission rate ratio: 1.03), acute myocardial infarction (risk-standardized admission rate ratio: 1.06), and acute renal failure (risk-standardized admission rate ratio: 1.10), showed the smallest differences. There was a strong inverse correlation (Pearson correlation: −0.71) between observed inpatient mortality and the magnitude of variation in risk-standardized admission rates as quantified by the coefficient of variation. For example, chest pain demonstrated wide variation (risk-standardized coefficient of variation: 1.04) and low observed inpatient mortality (0.05 percent). Condition-specific variation between the ninetieth and tenth percentiles is
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Exhibit 2 Variation In Emergency Department Admissions Among Hospitals, By Clinical Condition, 2010
Clinical condition Chest pain Soft tissue infections Asthma
Observed inpatient mortality rate (%) 0.05 0.32 0.26
Observed admission rate ratioa 2.60 2.27 2.01
Risk-standardized admission rate ratioa,b 6.55 3.40 3.07
Risk-standardized coefficient of variationb,c 1.04 0.92 0.86
COPD Urinary tract infection Fluid and electrolyte disorders
1.19 0.77 1.29
2.10 2.06 1.72
2.84 2.82 2.62
0.66 0.87 0.57
Biliary tract disease Cardiac dysrhythmias Diabetes with complications
0.49 1.00 0.56
1.55 1.54 1.48
2.09 1.94 1.78
0.47 0.46 0.42
Pneumonia Congestive heart failure Stroke
3.08 2.83 7.28
1.43 1.23 1.09
1.75 1.39 1.14
0.38 0.24 0.15
4.03 5.07 14.72
1.06 1.04 1.02
1.10 1.06 1.03
0.09 0.13 0.05
Acute renal failure Acute myocardial infarction Sepsis
SOURCE Authors’ analyses of data from the 2010 Nationwide Emergency Department Sample database. NOTES Conditions are presented in descending order from most to least variable by their risk-standardized admission rate ratio. COPD is chronic obstructive pulmonary disease. aRatio compares the admission rates at the seventy-fifth to twenty-fifth percentile of emergency departments. bAdjusted for age, sex, comorbidities, primary payer, and income. cPearson correlation between the riskstandardized coefficient of variation and observed inpatient mortality shows strong inverse correlation (−0.71).
shown in online Appendix Exhibit 3.17 Impact On Health Spending Average charges per hospitalization for each condition ranged from $18,162 for chest pain to $64,086 for acute myocardial infarction (Exhibit 3). There were an
estimated $266.6 billion in total charges for admissions related to the fifteen conditions studied, representing $80.0 billion in national health costs (assuming an overall estimated cost-tocharge ratio of 0.30). For the top five conditions
Exhibit 3 National Estimates Of Emergency Department (ED) Visits And Charges Related To ED Admissions, By Clinical Condition, 2010 Admissions from ED
National charges ($ billions)
Clinical condition Chest pain Soft tissue infections Asthma
Number 703,115 487,001 343,814
Percent 16.2 14.3 17.8
Average charge per admission ($) 18,162 22,772 19,770
All hospitalsa 10.3 10.4 6.3
COPD Urinary tract infection Fluid and electrolyte disorders
601,153 529,007 381,105
31.7 16.9 41.2
25,681 22,123 20,210
14.6 11.0 7.1
4.8 3.1 2.1
9.8 7.9 5.0
Biliary tract disease Cardiac dysrhythmias Diabetes with complications
357,188 603,619 446,367
53.9 42.7 55.7
38,819 30,353 28,597
13.1 16.9 12.1
3.9 5.2 3.1
9.2 11.7 9.0
Pneumonia Congestive heart failure Stroke
929,515 823,067 556,514
54.3 84.7 92.8
31,476 34,394 47,034
26.7 26.5 22.3
12.2 7.5 7.3
14.5 19.0 15.0
Acute renal failure Acute myocardial infarction Sepsis
347,470 508,526 816,853
94.5 96.7 98.4
34,479 64,086 63,876
11.4 28.0 49.9
4.9 10.3 20.1
6.5 17.7 29.8
Low-admitting EDsb 4.0 2.6 1.6
High-admitting EDsc 6.3 7.8 4.7
SOURCE Authors’ analyses of data from the 2010 Nationwide Emergency Department Sample (NEDS) database. NOTES Results were calculated from survey weights provided in the 2010 NEDS. Conditions are presented in descending order from most to least variable by their risk-standardized admission rate ratio. COPD is chronic obstructive pulmonary disease. aTotal national charges related to all hospitalizations from the emergency department. bCharges related to EDs that admit below the median risk-standardized admission rate. cCharges related to EDs that admit above the median risk-standardized admission rate.
S ep t e m b e r 20 1 4
33 : 9
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Health Affairs
1659
Emergency Department Use exhibiting the greatest variation in riskstandardized admission rates, we estimated total charges of $52.6 billion ($15.8 billion in costs). Asthma was the least-expensive condition ($6.3 billion in national charges) and sepsis the most ($49.9 billion in national charges). Exhibit 4 shows estimated reductions in national charges based on three utilization scenarios. Under the first scenario—if higheradmitting hospitals were to admit at the median rate—we estimated that there would have been $16.9 billion less in charges (for a cost savings of $5.1 billion) to US payers in 2010 for the five most variable conditions. Under the second scenario, if hospitals in the top quartile reduced admissions to the seventy-fifth percentile, we estimated that there would have been $7.0 billion less in charges (for a cost savings of $2.1 billion) for these same five conditions. Under the third scenario, if hospitals below the bottom quartile also increased admissions to the twenty-fifth percentile, the reduction in charges would be estimated at $2.8 billion (for a cost savings of $0.8 billion). Most conditions studied achieved cost savings under all three scenarios. However, for a handful of conditions, with chest pain being the only high-variation one, raising the bottom quartile and lowering the top quartile of hospital risk-
standardized admission rates actually resulted in net increases in spending.
Discussion Among a national sample of EDs, we found substantial variation in risk-standardized hospital admission rates for many commonly admitted conditions. In particular, chest pain, soft tissue infections, asthma, chronic obstructive pulmonary disease, and urinary tract infections showed the greatest variation and, therefore, may represent the best opportunity to improve the efficiency of ED admission practices and result in potential cost savings. In contrast, diagnoses such as sepsis, acute myocardial infarction, and stroke showed markedly less variation across EDs, representing consistent practice patterns, probably in response to these conditions’ high mortality and less diagnostic ambiguity. These data also highlight the magnitude of health spending that could be affected with a change in ED practices. We found national charges to payers in excess of $266 billion per year for the fifteen conditions studied, with highmortality, time-sensitive diagnoses such as sepsis and acute myocardial infarction representing the greatest cost burden to the health care system. However, these conditions also presented
Exhibit 4 Potential Annual Reductions In National Charges For Emergency Department (ED) Admissions, By Clinical Condition, Under Different Utilization Scenarios Annual reduction in national charges ($ billions) Scenario 2: top quartile at 75th percentileb 1.3 1.8 0.9
Scenario 3: top quartile at 75th percentile and bottom quartile at 25th percentilec,d +0.8 1.3 0.7
Clinical condition Chest pain Soft tissue infections Asthma
Current national charges ($ billions) 10.3 10.4 6.3
Scenario 1: high-admitting at mediana 3.3 3.9 2.2
COPD Urinary tract infection Fluid and electrolyte disorders Biliary tract disease Cardiac dysrhythmias Diabetes with complications
14.6 11.0 7.1 13.1 16.9 12.1
3.9 3.6 1.8 2.7 3.3 2.2
1.3 1.7 0.6 1.1 1.2 0.9
0.4 1.2 0.2 0.5 0.6 0.6
Pneumonia Congestive heart failure Stroke
26.7 26.5 22.3
3.2 2.3 0.6
0.9 0.5 0.1
+0.3 0.2 +0.2
Acute renal failure Acute myocardial infarction Sepsis
11.4 28.0 49.9
0.2 0.3 0.3
The online version of this article, along with updated information and services, is available at: http://content.healthaffairs.org/content/33/9/1655.full.html
For Reprints, Links & Permissions: http://healthaffairs.org/1340_reprints.php E-mail Alerts : http://content.healthaffairs.org/subscriptions/etoc.dtl To Subscribe: http://content.healthaffairs.org/subscriptions/online.shtml
Health Affairs is published monthly by Project HOPE at 7500 Old Georgetown Road, Suite 600, Bethesda, MD 20814-6133. Copyright © 2014 by Project HOPE - The People-to-People Health Foundation. As provided by United States copyright law (Title 17, U.S. Code), no part of Health Affairs may be reproduced, displayed, or transmitted in any form or by any means, electronic or mechanical, including photocopying or by information storage or retrieval systems, without prior written permission from the Publisher. All rights reserved.
Not for commercial use or unauthorized distribution Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Emergency Department Use By Amber K. Sabbatini, Brahmajee K. Nallamothu, and Keith E. Kocher 10.1377/hlthaff.2013.1318 HEALTH AFFAIRS 33, NO. 9 (2014): 1655–1663 ©2014 Project HOPE— The People-to-People Health Foundation, Inc.
doi:
Reducing Variation In Hospital Admissions From The Emergency Department For Low-Mortality Conditions May Produce Savings
Amber K. Sabbatini is an instructor of emergency medicine at the University of Washington, in Seattle.
The emergency department (ED) is now the primary source for hospitalizations in the United States, and admission rates for all causes differ widely between EDs. In this study we used a national sample of ED visits to examine variation in risk-standardized hospital admission rates from EDs and the relationship of this variation to inpatient mortality for the fifteen most commonly admitted medical and surgical conditions. We then estimated the impact of variation on national health expenditures under different utilization scenarios. Risk-standardized admission rates differed substantially across EDs, ranging from 1.03-fold for sepsis to 6.55-fold for chest pain between the twenty-fifth and seventy-fifth percentiles of the visits. Conditions such as chest pain, soft tissue infection, asthma, chronic obstructive pulmonary disease, and urinary tract infection were low-mortality conditions that showed the greatest variation. This suggests that some of these admissions might not be necessary, thus representing opportunities to improve efficiency and reduce health spending. Our data indicate that there may be sizeable savings to US payers if differences in ED hospitalization practices could be narrowed among a few of these high-variation, low-mortality conditions. ABSTRACT
A
dmitting a patient to the hospital from the emergency department (ED) is one of the more expensive, routine decisions made in health care. Emergency providers determine whether a patient requires hospitalization or can safely be discharged home about 350,000 times a day1 in the approximately 5,000 US EDs,2 resulting in almost twenty million annual admissions to hospitals.3 The Centers for Medicare and Medicaid Services (CMS) reported that hospital care represented about 30 percent of the $2.7 trillion in total expenditures for 2011—the largest share of health care spending.4 Appropriate admissions from the ED should be both emergent and necessary, dictated by a patient’s diagnosis and clinical presentation.
Brahmajee K. Nallamothu is an associate professor of cardiovascular medicine at the University of Michigan; a core investigator at the Center for Clinical Management Research, Ann Arbor Veterans Affairs Medical Center; a faculty member at the Center for Healthcare Outcomes and Policy; and a faculty member at the Institute for Healthcare Policy and Innovation, all in Ann Arbor. Keith E. Kocher (kkocher@ umich.edu) is an assistant professor in emergency medicine at the University of Michigan; a faculty member at the Center for Healthcare Outcomes and Policy; and a faculty member at the Institute for Healthcare Policy and Innovation.
Certain life-threatening diagnoses, such as acute myocardial infarction, almost always require hospitalization. Yet the majority of admissions from the ED are for intermediate-severity conditions such as acute exacerbations of chronic diseases (for example, asthma exacerbation), with significant variability in presentation.5 In these cases, appropriate dispositions are not always clear, allow for provider and patient discretion, and could be influenced by the local practice culture and resources.6 For example, it is estimated that up to half of patients with congestive heart failure in the ED could be discharged home after a period of observation and treatment. However, many of these patients are admitted, leading to greater resource use.7 As EDs are now the primary source through which paSeptember 2014
33:9
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
H ea lt h A f fai r s
1 655
Emergency Department Use tients with acute illnesses are hospitalized, surpassing direct admissions from ambulatory settings,3,8 differences in admission practices have important implications for the quality and financing of the health care system. Recent studies have demonstrated that allcause admission rates are highly variable across individual providers and EDs.6,9 This variation may ultimately represent the overuse, underuse, or misuse of hospital services. As a result, understanding which conditions show the greatest variation in ED admission practices may lead to improvements in efficiency, as well as cost savings.We therefore examined a national sample of ED visits to compare differences in admission rates for the fifteen most commonly hospitalized medical and surgical conditions. Our goal was to identify conditions that showed the greatest variation and, thus, represent possible targets for standardizing admission practices. We then sought to understand how this variation affects national health spending by exploring potential savings related to narrowing differences in admission rates for each condition.
Study Data And Methods Data Source And Study Population We used the 2010 Nationwide Emergency Department Sample (NEDS) database to conduct the analysis.10 NEDS is part of the Healthcare Cost and Utilization Project (HCUP), a family of longitudinal databases on hospital care in the United States, sponsored by the Agency for Healthcare Research and Quality. It is the largest all-payer database of ED visits in the United States, capturing information on over twenty-nine million ED visits from twenty-eight states, with weights to produce national estimates. All ED visits except for patients who died in the ED were included in the analysis. For the purposes of this study, we considered visits resulting in transfer as admissions, given the high likelihood of admission at the receiving hospital.11 Observation-status cases were not included in this analysis because they are not tracked in NEDS.12 Clinical Classifications Software categories13 identified ED visits for the fifteen most commonly admitted medical and surgical conditions. NEDS also includes linked inpatient information for ED visits that resulted in admission, from which we derived total inpatient charges and in-hospital mortality. Outcome Measures Primary outcomes for this study were unadjusted and risk-standardized admission rates. To ensure valid results, we excluded EDs with fewer than thirty cases for each condition from our calculation of admission rates, as small sample sizes provide inade1656
Health A ffairs
September 2014
33:9
quate representations of health care use. Survey weights were used to generate national estimates of the number of ED visits and admissions, observed inpatient mortality, mean inpatient charges, and total national charges associated with hospitalizations for each condition. Observed inpatient mortality rates were examined to provide clinical context for the average associated severity of each condition. Data Analysis Mixed-effects, hierarchical logistic regression was used to adjust for the severity of patient case-mix and clustering of admissions among hospitals. Covariates in the model included age, sex, Elixhauser comorbidities (a list of thirty comorbid conditions associated with inpatient hospital mortality),14 primary payer (uninsured, private, Medicaid, Medicare, other), and median income of the ZIP code in which the patient resides. This approach assumes that after adjustment for patient case-mix, the remaining variation is due to institutional or community factors that influence an ED’s propensity to admit. Risk-standardized admission rates for each condition were derived from regression models by dividing the number of predicted admissions for each ED, given that institution’s specific casemix, by the expected number of admissions had those patients been treated at the average ED. This predicted-to-expected ratio was then multiplied by the mean admission rate for the representative sample to determine the risk-standardized admission rate. This method of indirect standardization has been well described in the literature as a means for case-mix adjustment when comparing outcomes among hospitals.15,16 Variation is expressed as the ratio of observed and risk-standardized admission rates for EDs at the seventy-fifth to twenty-fifth percentiles. The seventy-fifth and twenty-fifth percentiles were used instead of ninetieth and tenth percentiles to provide a more conservative estimate of variation. Variation between the ninetieth and tenth percentiles is available in the online Appendix.17 We also calculated the coefficient of variation for the distribution of risk-standardized admission rates, which is a unitless measure of dispersion generated by dividing the standard deviation by the mean rate, allowing for comparison between conditions with widely differing mean admission rates. The correlation between the coefficient of variation and observed inpatient mortality was then examined, to assist in understanding the relationship between variation in admission practices and risk of poor clinical outcomes. To illustrate the impact of variation in ED admission rates on national health care spending, we estimated annual national charges under
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
three different utilization scenarios that narrow variation: (1) if EDs with risk-standardized admission rates above the median hypothetically decreased their hospitalization rates to the median for each condition; (2) if EDs with riskstandardized admission rates in the top quartile decreased admissions to the seventy-fifth percentile; and (3) if EDs with risk-standardized admission rates in the top quartile decreased admissions to the seventy-fifth percentile while EDs with risk-standardized admission rates in the bottom quartile increased admissions to the twenty-fifth percentile. This latter scenario is to account for the possibility that some EDs may also be inappropriately underadmitting patients for their case-mix. NEDS contains information on hospital charges only and does not provide hospitalspecific cost-to-charge ratios. Therefore, our analysis is largely limited to charge data. However, we also estimated national health spending by generating an overall cost-to-charge ratio of 0.30 for all US hospitalizations. This ratio is calculated from health expenditure data provided by HCUP18 by dividing total national costs ($371.7 billion) by total national charges ($1.224 trillion) for hospital care. Data management and analysis were performed using Stata software (version 12.1 MP). Additional technical details regarding the NEDS database, the approach to modeling, and calculation of health expenditures are in the online Appendix.17 The Institutional Review Board of the University of Michigan evaluated this study and classified it under “not regulated” status. Limitations Our results should be interpreted in the context of the following limitations. First, this analysis was descriptive and intended to provide a broad picture of variation in ED admissions and, in particular, the implications on health care spending associated with these differences. Our consideration of health spending under the three utilization scenarios does not imply that there is an optimal rate of ED admissions that can be inferred from this study. Second, our description of variation and health spending is at the ED (hospital) level after controlling for patient case-mix. Many factors not assessed in this study can influence an ED’s admission rate in addition to patient factors, such as social conditions, market forces, the medicolegal environment, local access to care,6 and providers’ training and experience. Addressing all of these factors might not be practical to target with policy or quality improvement efforts. Third, although we controlled for severity of case-mix in our estimates by using accepted techniques to produce risk-standardized admission
rates, we recognize that there is likely some portion of unmeasured severity of illness inherent in using administrative data. For example, patients who are admitted may have a greater number of comorbidities listed than patients who are discharged from the ED, which could affect risk adjustment. As a result, certain EDs may be appropriately high admitters and would be unable to reduce hospitalizations without adversely affecting patient outcomes and safety. Conversely, some EDs may be inappropriately low admitters, discharging patients who would have benefited from hospitalization. Any interventions to address systematic overuse, underuse, or misuse of hospital admissions will necessarily have consequences for health spending. Ultimately, variation in admissions must be interpreted in the context of longitudinal outcomes such as patient mortality, unanticipated return visits to the ED, or readmissions, which this study was not designed to evaluate. Finally, our estimates of national health spending are not meant to serve as a formal economic analysis of net expenditures and do not account for increases in outpatient resource utilization that occur by shifting care to ambulatory settings, such as costs associated with observation, office, and home-based care. We are also only able to report estimates derived from hospital charges, adjusted for an overall cost-tocharge ratio for US hospitalizations, which may overestimate the impact on national expenditures. Therefore, if ED admission rates could be reduced for certain conditions, true spending reductions to the health system will be less than our current estimates. However, this work provides an important starting point to use in understanding the scale of national health spending that could be affected by narrowing variation.
Study Results Study Population There were 28,539,883 visits among 961 hospital-based EDs in 2010 (Exhibit 1). Overall, 15.4 percent of the ED visits resulted in admission, with a mean accompanying charge of $34,826. Just over half of the visits (55.6 percent) were by females, with an average age of 38.8 years and 21.6 percent carrying two or more comorbidities. The majority of visits were for patients who had private insurance (30.8 percent), followed by Medicaid (25.6 percent), and then Medicare (20.9 percent). Uninsured patients made up 18.1 percent of the visits. Clinical conditions in the top fifteen admitted medical and surgical diagnoses ranged from chest pain (965,432 total visits and an average of 1,002 cases per ED) to acute renal S e p t em b e r 2 0 1 4
33:9
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Health Affa irs
1657
Emergency Department Use Exhibit 1 Emergency Department (ED) Visit- And Hospital-Level Characteristics Of The Study Population, 2010 a
Characteristic ED disposition admitted Mean inpatient charge (SD)
ED visit level (N = 28,539,883) 15.4% $34,826 (53,046)
Hospital levelb (N = 961) 12.4% $31,856 (17,138)
Patient characteristics Mean age, years (SD) Percent female Percent with 2 or more comorbidities Primary payer (%) Private Medicare Medicaid Uninsured Other
38.8 (24.2) 55.6 21.6
39.9 (5.6) 55.1 19.3
30.8 20.9 25.6 18.1 4.6
31.1 23.0 24.7 16.7 4.7
Household income (%) $39,999 or less $40,000–$49,999 $50,000–$65,999 $66,000 or more
32.9 27.8 22.1 17.2
35.4 30.9 19.1 12.3
Primary diagnosis [CCS category]
Number of visits (% of total visits)
Average number of cases per ED
Chest pain [102] Soft tissue infections [197] Asthma [128]
965,432 (3.4) 755,649 (2.6) 429,853 (1.5)
1,002 813 481
COPD [127] Urinary tract infection [159] Fluid and electrolyte disorders [55]
425,803 (1.5) 699,940 (2.5) 204,658 (0.7)
461 749 231
Biliary tract disease [149] Cardiac dysrhythmias [106] Diabetes with complications [50]
149,161 (0.5) 311,992 (1.1) 176,884 (0.6)
196 348 216
Pneumonia [122] Congestive heart failure [108] Stroke [109]
377,213 (1.3) 215,305 (0.8) 132,585 (0.5)
406 254 185
Acute renal failure [157] Acute myocardial infarction [100] Sepsis [2]
82,595 (0.3) 117,403 (0.4) 182,844 (0.6)
137 165 272
SOURCE Authors’ analyses of data from the 2010 Nationwide Emergency Department Sample database. NOTES SD is standard deviation. CCS is Clinical Classifications Software. COPD is chronic obstructive pulmonary disease. aUnweighted data. bHospitallevel characteristics represent averages per hospital of the visit-level information.
failure (82,595 total visits and an average of 137 per ED). Admission Practices There was substantial variation in unadjusted ED admission rates (see online Appendix Exhibit 1)17 and riskstandardized admission rates (see online Appendix Exhibit 2)17 for many of the conditions studied. Observed differences in admission rates ranged from a low of 1.02 for sepsis to 2.60 for chest pain at the seventy-fifth and twentyfifth percentiles (Exhibit 2). After risk adjustment, the degree of variation between EDs widened. Those conditions showing the greatest variation included chest pain (risk-standardized admission rate ratio: 6.55), soft tissue infection (risk-standardized admission rate ratio: 3.40), and asthma (risk-standardized admission rate 1658
Health Affairs
S ep t e m b e r 20 1 4
33 : 9
ratio: 3.07). In contrast, higher-severity conditions with less diagnostic and prognostic uncertainty, such as sepsis (risk-standardized admission rate ratio: 1.03), acute myocardial infarction (risk-standardized admission rate ratio: 1.06), and acute renal failure (risk-standardized admission rate ratio: 1.10), showed the smallest differences. There was a strong inverse correlation (Pearson correlation: −0.71) between observed inpatient mortality and the magnitude of variation in risk-standardized admission rates as quantified by the coefficient of variation. For example, chest pain demonstrated wide variation (risk-standardized coefficient of variation: 1.04) and low observed inpatient mortality (0.05 percent). Condition-specific variation between the ninetieth and tenth percentiles is
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Exhibit 2 Variation In Emergency Department Admissions Among Hospitals, By Clinical Condition, 2010
Clinical condition Chest pain Soft tissue infections Asthma
Observed inpatient mortality rate (%) 0.05 0.32 0.26
Observed admission rate ratioa 2.60 2.27 2.01
Risk-standardized admission rate ratioa,b 6.55 3.40 3.07
Risk-standardized coefficient of variationb,c 1.04 0.92 0.86
COPD Urinary tract infection Fluid and electrolyte disorders
1.19 0.77 1.29
2.10 2.06 1.72
2.84 2.82 2.62
0.66 0.87 0.57
Biliary tract disease Cardiac dysrhythmias Diabetes with complications
0.49 1.00 0.56
1.55 1.54 1.48
2.09 1.94 1.78
0.47 0.46 0.42
Pneumonia Congestive heart failure Stroke
3.08 2.83 7.28
1.43 1.23 1.09
1.75 1.39 1.14
0.38 0.24 0.15
4.03 5.07 14.72
1.06 1.04 1.02
1.10 1.06 1.03
0.09 0.13 0.05
Acute renal failure Acute myocardial infarction Sepsis
SOURCE Authors’ analyses of data from the 2010 Nationwide Emergency Department Sample database. NOTES Conditions are presented in descending order from most to least variable by their risk-standardized admission rate ratio. COPD is chronic obstructive pulmonary disease. aRatio compares the admission rates at the seventy-fifth to twenty-fifth percentile of emergency departments. bAdjusted for age, sex, comorbidities, primary payer, and income. cPearson correlation between the riskstandardized coefficient of variation and observed inpatient mortality shows strong inverse correlation (−0.71).
shown in online Appendix Exhibit 3.17 Impact On Health Spending Average charges per hospitalization for each condition ranged from $18,162 for chest pain to $64,086 for acute myocardial infarction (Exhibit 3). There were an
estimated $266.6 billion in total charges for admissions related to the fifteen conditions studied, representing $80.0 billion in national health costs (assuming an overall estimated cost-tocharge ratio of 0.30). For the top five conditions
Exhibit 3 National Estimates Of Emergency Department (ED) Visits And Charges Related To ED Admissions, By Clinical Condition, 2010 Admissions from ED
National charges ($ billions)
Clinical condition Chest pain Soft tissue infections Asthma
Number 703,115 487,001 343,814
Percent 16.2 14.3 17.8
Average charge per admission ($) 18,162 22,772 19,770
All hospitalsa 10.3 10.4 6.3
COPD Urinary tract infection Fluid and electrolyte disorders
601,153 529,007 381,105
31.7 16.9 41.2
25,681 22,123 20,210
14.6 11.0 7.1
4.8 3.1 2.1
9.8 7.9 5.0
Biliary tract disease Cardiac dysrhythmias Diabetes with complications
357,188 603,619 446,367
53.9 42.7 55.7
38,819 30,353 28,597
13.1 16.9 12.1
3.9 5.2 3.1
9.2 11.7 9.0
Pneumonia Congestive heart failure Stroke
929,515 823,067 556,514
54.3 84.7 92.8
31,476 34,394 47,034
26.7 26.5 22.3
12.2 7.5 7.3
14.5 19.0 15.0
Acute renal failure Acute myocardial infarction Sepsis
347,470 508,526 816,853
94.5 96.7 98.4
34,479 64,086 63,876
11.4 28.0 49.9
4.9 10.3 20.1
6.5 17.7 29.8
Low-admitting EDsb 4.0 2.6 1.6
High-admitting EDsc 6.3 7.8 4.7
SOURCE Authors’ analyses of data from the 2010 Nationwide Emergency Department Sample (NEDS) database. NOTES Results were calculated from survey weights provided in the 2010 NEDS. Conditions are presented in descending order from most to least variable by their risk-standardized admission rate ratio. COPD is chronic obstructive pulmonary disease. aTotal national charges related to all hospitalizations from the emergency department. bCharges related to EDs that admit below the median risk-standardized admission rate. cCharges related to EDs that admit above the median risk-standardized admission rate.
S ep t e m b e r 20 1 4
33 : 9
Downloaded from content.healthaffairs.org by Health Affairs on December 5, 2014 at UCSF Library & CKM RECS Mgmt
Health Affairs
1659
Emergency Department Use exhibiting the greatest variation in riskstandardized admission rates, we estimated total charges of $52.6 billion ($15.8 billion in costs). Asthma was the least-expensive condition ($6.3 billion in national charges) and sepsis the most ($49.9 billion in national charges). Exhibit 4 shows estimated reductions in national charges based on three utilization scenarios. Under the first scenario—if higheradmitting hospitals were to admit at the median rate—we estimated that there would have been $16.9 billion less in charges (for a cost savings of $5.1 billion) to US payers in 2010 for the five most variable conditions. Under the second scenario, if hospitals in the top quartile reduced admissions to the seventy-fifth percentile, we estimated that there would have been $7.0 billion less in charges (for a cost savings of $2.1 billion) for these same five conditions. Under the third scenario, if hospitals below the bottom quartile also increased admissions to the twenty-fifth percentile, the reduction in charges would be estimated at $2.8 billion (for a cost savings of $0.8 billion). Most conditions studied achieved cost savings under all three scenarios. However, for a handful of conditions, with chest pain being the only high-variation one, raising the bottom quartile and lowering the top quartile of hospital risk-
standardized admission rates actually resulted in net increases in spending.
Discussion Among a national sample of EDs, we found substantial variation in risk-standardized hospital admission rates for many commonly admitted conditions. In particular, chest pain, soft tissue infections, asthma, chronic obstructive pulmonary disease, and urinary tract infections showed the greatest variation and, therefore, may represent the best opportunity to improve the efficiency of ED admission practices and result in potential cost savings. In contrast, diagnoses such as sepsis, acute myocardial infarction, and stroke showed markedly less variation across EDs, representing consistent practice patterns, probably in response to these conditions’ high mortality and less diagnostic ambiguity. These data also highlight the magnitude of health spending that could be affected with a change in ED practices. We found national charges to payers in excess of $266 billion per year for the fifteen conditions studied, with highmortality, time-sensitive diagnoses such as sepsis and acute myocardial infarction representing the greatest cost burden to the health care system. However, these conditions also presented
Exhibit 4 Potential Annual Reductions In National Charges For Emergency Department (ED) Admissions, By Clinical Condition, Under Different Utilization Scenarios Annual reduction in national charges ($ billions) Scenario 2: top quartile at 75th percentileb 1.3 1.8 0.9
Scenario 3: top quartile at 75th percentile and bottom quartile at 25th percentilec,d +0.8 1.3 0.7
Clinical condition Chest pain Soft tissue infections Asthma
Current national charges ($ billions) 10.3 10.4 6.3
Scenario 1: high-admitting at mediana 3.3 3.9 2.2
COPD Urinary tract infection Fluid and electrolyte disorders Biliary tract disease Cardiac dysrhythmias Diabetes with complications
14.6 11.0 7.1 13.1 16.9 12.1
3.9 3.6 1.8 2.7 3.3 2.2
1.3 1.7 0.6 1.1 1.2 0.9
0.4 1.2 0.2 0.5 0.6 0.6
Pneumonia Congestive heart failure Stroke
26.7 26.5 22.3
3.2 2.3 0.6
0.9 0.5 0.1
+0.3 0.2 +0.2
Acute renal failure Acute myocardial infarction Sepsis
11.4 28.0 49.9
0.2 0.3 0.3
