Decline in U.S. Emergency Department Admission Rates Driven by Critical Pathway Conditions, 2006–2014

Introduction

Background and Importance

Hospital-based care accounts for one-third of national health expenditures.¹ The adoption of value-based payment has pressured providers to reduce costs associated with hospital-based care by reducing readmissions, emergency department (ED) visits, and hospitalizations originating in the ED.^2,3 While policymakers perceive the ED as an expensive setting for care, EDs play a key role in determining the optimal setting of care for patients. The decision to admit a patient is one of the most important and costly decisions made in the ED setting. However, there is substantial evidence of persistent variation in the decision to hospitalize patients after ED evaluation, with little difference in outcomes.⁴ Prior work has identified common ED conditions that may be amenable to outpatient management after ED evaluation based on the presence of evidence to stratify clinical risk and/ or support expedited diagnostic protocols.⁵ However, little is known regarding recent trends in ED hospitalization patterns for these conditions.

Recent evidence demonstrates a 10% reduction in the rate of hospitalization after an ED visit in 2014 compared to 2006, despite a 20% increase in ED visits over the same period.³ However, it is unknown what clinical and hospital factors account for these trends. The lack of understanding regarding clinical conditions accounting for these trends limits ongoing efforts to improve acute care value. Specifically, the identification of conditions amenable to outpatient pathways with persistently high hospitalization rates could be a target for quality improvement using best practices from conditions for which hospitalization rates have decreased.

Goals of this Investigation

Our primary objective was to describe clinical conditions and hospital factors associated with rates of hospitalization after ED evaluation from 2006–2014. We describe trends in the most commonly admitted conditions and those previously identified as being amenable to outpatient pathways after ED evaluation. We also sought to describe hospital-level variation in condition-specific hospitalization rates to determine whether changes were uniform or driven by select hospitals.

Methods

Study Design, Setting, Participants

We performed a cross-sectional analysis of the 2006–2014 Nationwide Emergency Department Survey (NEDS), an approximate 20-percent stratified sample of U.S. hospital-based EDs.⁶ NEDS is a component of the Healthcare Cost and Utilization Project (HCUP) maintained by the Agency for Healthcare Research and Quality (AHRQ), and contains data from billing records of 26 to 29 million ED visits per year from about 950 annually selected hospitals; a full description of the NEDS is available online.⁶ As a publicly available data set without identifiers, this study was not considered human subjects research by our Institutional Review Board.

We focused on adults, as hospitalization rates for children are lower. We excluded patients who left without being seen, left against medical advice, were transferred to another facility, died on arrival, died in the ED, and those missing data on ED disposition (total excluded = 4.7% of ED visits).

Outcomes

Our primary outcome of interest was the proportion of hospitalizations originating in the ED divided by the number of ED visits for each condition and year, hereafter referred to as “ED admission rate.” We performed stratified analysis of two groups of diagnoses: 1) the top twenty most frequently admitted conditions in 2006, and 2) commonly admitted, low mortality conditions traditionally managed by admission from the ED for which there are evidence-based critical pathways supporting outpatient management and significant hospital-level variation in admission rates based on prior research.^4,5 Conditions were classified using Clinical Classification Software (CCS) code associated with the primary discharge diagnosis for each visit. CCS codes are clinically meaningful groupings of International Classification of Diseases (ICD) codes developed by HCUP and AHRQ for the purposes of research. For hospital-level analyses, we excluded hospitals with fewer than twenty hospitalization events per year to avoid unstable estimates.

Analysis

National estimates of ED visits and ED admission rates were calculated using sampling weights and strata to account for NEDS’ complex sampling design. We report frequencies, patient (age, sex, insurance, income, comorbid conditions) and hospital (teaching status, geographic location, region, annual ED visit volume) characteristics. Comorbid conditions were coded using the Elixhauser score, a previously validated algorithm for use in administrative datasets using the Health Care Utilization Project Elixhauser Comorbidity Software based on ICD-9 diagnosis codes.^7,8 Trends in ED admission rates from 2006 to 2014 were tested using logistic regression with year as independent variable and clustering of standard errors at the hospital level. To show hospital-level variation, we also calculated median hospital-level ED admission rates and interquartile range. For each condition, we used logistic regression with survey weights to compare the odds of ED admission in 2014 compared to 2006. To compare the odds of ED admission for similar patients across years, we used logistic regression adjusting for patient (age, sex, insurance, income, Elixhauser comorbidity score) and hospital (teaching status, geographic location, region, annual ED visit volume) characteristics, with clustering by hospital. All analyses were performed in SAS version 9.4.

Results

Characteristics of Study Subjects

We examined 788 million adult ED visits from 2006–14; patient and hospital characteristics are shown in Appendix 1. Patients of older age, with Medicare insurance, and those with more comorbid chronic illness accounted for largest reductions in unadjusted ED admission rate in 2014 compared to 2006 (−6.4% for those age 80+ years, −6.4% for those with Medicare insurance, and −14.7% for those with 3 or more Elixhauser conditions, respectively). Metropolitan teaching hospitals accounted for a greater proportion of ED visits in 2014 compared to relative to non-metropolitan and non-teaching hospitals. We observed similar reductions in ED admission rates for hospitals regardless of teaching status or metropolitan location. The largest decrease in unadjusted ED admission rates from 2006 to 2014 occurred at hospitals in the Midwest region (−3.1%) and those with lower ED visit volumes (−4.1%).

Condition-specific Trends

The top three conditions most likely to result in ED admission in 2006 were septicemia; acute myocardial infarction; and acute cerebrovascular disease (Table 1). The conditions with the greatest relative change in ED admission rates were chest pain (66% relative reduction) and syncope (52%). The absolute and relative change in ED admission rates, and the unadjusted and adjusted change in odds of ED admission, for all conditions are shown in Table 1, and trends in ED admission rates over time by condition are in Figure 1. After adjusting for patient and hospital characteristics, the odds of hospitalization after any ED visit was less than half in 2014 compared to 2006 (OR 0.49, CI 0.45, 0.52). The conditions with the largest decrease in adjusted odds of ED admission were chest pain (OR 0.16, CI 0.14–0.19) and congestive heart failure (OR 0.17, CI 0.15–0.2).

Table 1.

ED Visits and Admissions for Most Frequently Admitted Conditions and Conditions Amenable to Outpatient Pathways, 2006–2014. (Based on most frequently admitted primary CCS diagnoses in 2006).

Condition	Admissions from ED, 2006 (N)	ED Admission Rate, 2006	ED Admission Rate, 2014	Absolute change, 2006–2014	Relative % change, 2006–2014	Unadjusted change in probability of ED admission, 2006–2014, OR (CI)	Adjusted change in probability of ED admission, 2006–2014 OR (CI)
All conditions	17,377,131	19.4%	17.5%	−1.9%	9.8%	0.88 (0.84, 0.93)	0.49 (0.45, 0.52)
Congestive heart failure*	180,347	84.4%	82.3%	−2.1%	−2.5%	0.86 (0.78, 0.94)	0.17 (0.15, 0.2)
Pneumonia*	180,297	65.9%	58.9%	−7.0%	−10.6%	0.741 (0.7, 0.79)	0.49 (0.45, 0.54)
Nonspecific chest pain*	154,630	21.4%	7.3%	−14.1%	−66%	0.29 (0.25, 0.33)	0.16 (0.14, 0.19)
Septicemia	108,853	98.4%	98.6%	0.2%	0.2%	1.14 (0.94, 1.39)	0.84 (0.68, 1.03)
Cardiac dysrhythmias	104,431	42%	37%	−5%	−11.8%	0.81 (0.77, 0.86)	0.45 (0.41, 0.49)
Acute myocardial infarction	99,397	96.4%	95.9%	−0.5%	−0.6%	0.87 (0.69, 1.08)	0.54 (0.41, 0.7)
Chronic obstructive pulmonary disease*	99,267	31.6%	29.3%	−2.2%	−7.03%	0.90 (0.84, 0.97)	0.52 (0.48, 0.57)
Acute cerebrovascular disease	97,607	93.4%	92.6%	−0.7%	−0.8%	0.89 (0.79, 1.00)	0.56 (0.49, 0.63)
Coronary atherosclerosis and other heart disease	94,273	81.3%	65.8%	−15.5%	−19.1%	0.44 (0.39, 0.5)	0.28 (0.25, 0.32)
Urinary tract infections*	85,176	17.4%	14.7%	−2.7%	−15.4%	0.82 (0.77, 0.87)	0.51 (0.46, 0.55)
Conditions Amenable to Outpatient Pathways
Skin and subcutaneous tissue infections	79,308	14.81%	16.2%	1.42%	9.54%	1.11 (1.04, 1.19)	0.63 (0.57, 0.69)
Gastrointestinal hemorrhage	62,231	58.1%	53.6%	−4.5%	−7.74%	0.84 (0.79, 0.88)	0.60 (0.56, 0.65)
Syncope	51,241	28.3%	13.5%	−14.7%	−52.1%	0.4 (0.35, 0.45)	0.23 (0.20, 0.27)
Asthma	48,336	20.9%	18.3%	−2.6%	−12.7%	0.85 (0.78, 0.92)	0.49 (0.45, 0.55)
Transient ischemic attack	36,867	64.1%	52.5%	−11.6%	−18.1%	0.62 (0.56, 0.69)	0.4 (0.35, 0.45)
Abdominal Pain	31,068	75.73%	69.1%	−6.68%	−8.83%	0.72 (0.66, 0.78)	0.50 (0.45, 0.56)
Deep venous thrombosis	20,727	66.3%	50.6%	−15.7%	−23.7%	0.52 (0.46, 0.59)	0.29 (0.26, 0.33)
Pulmonary embolism	3,688	95.8%	89%	−6.9%	−7.2%	0.35 (0.23, 0.41)	0.26 (0.23, 0.31)

^*Clinical conditions that are both most frequently admitted in 2006 and amenable to outpatient pathways

Figure 1.

a Unadjusted ED Admission Rates for Top 5 Most Frequently Admitted Conditions, 2006–2014.

b Unadjusted ED Admission Rates for Conditions Amenable to Outpatient Pathways, 2006–2014.*

X axis = year, Y axis = ED admission rate.

*Select conditions with the greatest relative (syncope (−) 52.1%, DVT (−) 23.7%, TIA (−)18.1%) and absolute change (DVT (−) 15.7%, syncope (−) 14.7%, TIA (−) 11.6%) and largest increase (cellulitis, relative increase of 9.54%, absolute 1.42%) in ED admission rates.

Hospital-Level Variation

Hospital-level variation in ED admission rates is illustrated in Figure 2, and shows a decrease in the median all-cause ED admission rate and a widening interquartile range (IQR) across all hospitals, from 11.8 to 13.8% in magnitude. However, for chest pain and syncope, the conditions with the greatest relative reduction in ED admissions, we observe a decrease in hospital-level variation in ED admission rates, with a reduction in IQR magnitude from 21.6 to 6.4% for chest pain, and 23.6 to 12.5% for syncope (Appendix 2). For congestive heart failure and pneumonia, which appear on the list of most frequently admitted conditions, there appears to be little change in the hospital-level variation and IQR over time.

Limitations

Our study has limitations common to administrative data; for example, the NEDS does not include clinical data such as vital signs; however, we adjusted for number of comorbid conditions. Patients and hospitals are not individually identified, so some patients may have accounted for multiple visits within the sample; however, this likely represents a small proportion of visits. Given lack of hospital identifiers, we are unable to directly compare hospitals across years; however, the large sample size allows us to produce a nationally representative sample. Further studies are needed to determine if reductions in hospitalization are associated with outcomes and mortality. The lower adjusted odds of ED admission may be attributed in part to more intense coding;⁹ however, the reduction in adjusted ED admission rates is less likely to be explained solely by documentation, because older patients and those with Medicare accounted for a greater proportion of ED visits over time. Additionally, ED visits with more than 10 comorbid conditions in 2014 accounted for only 0.02% of all visits in 2014. NEDS does not include observation care, which may have resulted in an underestimate of ED admission rate. While there are few sources of all-payer observation trend data, as per the National Hospital Ambulatory Medical Care Survey (NHAMCS), only 1.3% of ED visits in 2014 were associated with observation care, an increase from 1.1% in 2006; thus, the 0.2% absolute increase in observation encounters is less than the 1.9% absolute reduction in hospitalization found in our analysis.^10,11 We did not individually evaluate pyelonephritis and atrial fibrillation, as they are included in the CCS category for urinary tract infections and cardiac dysrhythmias, respectively, as per Sabbatini et al., which may decrease the specificity of our findings.⁴ Finally, changes in practice patterns influencing the way ICD codes are assigned—for example, given the increased diagnosis and regulatory attention focused on sepsis—may have contributed to observed findings.

Discussion

After adjusting for patient and hospital characteristics, the odds of being admitted to the hospital after an ED visit for any condition was over 50% lower in 2014 compared to 2006, with substantial variation across conditions and hospitals.

The greatest relative reductions in ED admission rates were observed among a select number of high-variation, high-volume conditions amenable to outpatient critical pathways such as chest pain and syncope.^4,5 These conditions are distinguished by the potential for optimal ED management and care coordination to reduce the need for hospitalization. For some conditions, the decline in ED admissions may be attributed to ED clinicians’ adoption of evidence-based protocols promoting outpatient management. For example, chest pain ED admission rates may have decreased due to risk stratification using the HEART score,¹² rapid cycling of troponin blood tests, and use of observation status. For other conditions, lower ED admission rates may be attributable to advances in medical science and technology—for example, direct oral anticoagulants have reduced the need for inpatient heparinization of clinically stable acute venous thromboembolism, atrial fibrillation, and pulmonary embolism. Advanced diagnostic imaging may have increased diagnostic accuracy and reduced the need for hospitalization, such as computerized tomography (CT) imaging for abdominal pain, and magnetic resonance imaging (MRI) for transient ischemic attacks.

While there was no change in the unadjusted ED admission for congestive heart failure, the adjusted odds of ED admission were over 80% lower (OR 0.17, 95% CI 0.15–0.2) in 2014 compared to 2006 after accounting for hospital and patient characteristics. In other words, a patient with the same demographics and same comorbid conditions presenting to the same type of hospital would have over 80% lower odds of ED admission in 2014 compared to 2006. The adjusted odds are likely explained by care pathways, improved care coordination, and more aggressive documentation after implementation of the Hospital Readmissions Reduction Program (HRRP) in 2012, which penalizes hospitals for excess readmissions within 30 days for select conditions including CHF.¹³

The reduction in ED admission rates can also be attributed to in part to other changes in federal payment policy during this period. The 2010 Centers for Medicare and Medicaid Services (CMS) Recovery Audit Contractor (RAC), a program designed to identify and collect Medicare overpayments, increased scrutiny of short-stay hospitalizations and costly retrospective reimbursement denials.¹⁴ In order to clarify which short-stay hospitalizations warranted the most scrutiny, CMS implemented the “Two-Midnight Rule” in 2014, which deems hospitalizations “reasonable and necessary” only if the hospitalization is expected to span two consecutive midnights, increasing the reliance on observation and outpatient pathways.¹⁵ The 2009 HITECH Act mandated the adoption of electronic health records, which could have affected ED admission rates by improving access to medical records in the ED.¹⁶ Such access can sometimes help prevent hospitalizations, for example, obtaining catheterization or stress test results on a patient in the ED with chest pain.

However, the downward trend in ED admission rates began prior to the implementation these policies, suggesting federal payment policy and enforcement do not completely explain the observed decrease in ED admission rates (Figure 1a, 1b). Further, the impact of federal policy changes would be expected to impact almost all hospitals proportionally (with the possible exception of critical access hospitals due to their exclusion from certain programs). Yet we do not observe equal change across all hospitals. Instead, we observe a widening interquartile range and bimodal distribution of hospital-level ED admission rates in the most recent years (Figure 2a). The increased variation in hospital-level ED admission rates suggests that hospitals are not changing their delivery of acute care at similar rates. Assuming no difference in outcomes, hospitals with the greatest reductions in ED admission rates may have important lessons for policymakers and leaders aiming to reduce low-value hospitalizations.

Figure 2. Hospital-Level Variation in ED Admission Rates (All Conditions, Most Frequently Admitted, and Greatest Relative Change), 2006–2014.

* All plots are for hospitals with at least 20 or more. CHF = Congestive heart failure.

Only two conditions examined did not see reduced ED admission rates: sepsis and skin and soft tissue infections. For sepsis, this may be due to policy changes to reduce sepsis mortality, which have led to earlier recognition, more thorough documentation, and more aggressive treatment of sepsis in the ED, resulting in both a lower threshold for hospitalization among ED clinicians and more visits being labelled as sepsis (see Appendix 2 indicating a nearly threefold increase in ED visits).¹⁷ For skin and soft tissue infection, increased hospitalization rates may be due to increased prevalence of community methicillin-resistant Staphylococcus aureus and intravenous opioid use, which increase virulent soft tissue infections requiring hospitalization.^{18, 19}

After accounting for hospital and patient characteristics, the odds of being hospitalized dropped 50% between 2006 and 2014. However, reductions in ED admission rates were not evenly distributed across conditions or hospitals. Recent reductions in ED admission rates are largely attributable to conditions amenable to outpatient critical pathways. Focusing on reducing ED admission rates for conditions amenable to outpatient pathways among hospitals with persistently high ED admission rates may reduce hospital expenditures without affecting quality and improve the value of acute unscheduled care.