Randomized Clinical Trial of an Emergency Department Observation Syncope Protocol vs. Routine Inpatient Admission

Abstract

Study Hypothesis

Older adults are frequently hospitalized from the emergency department (ED) after an episode of unexplained syncope. Current admission patterns are costly with little evidence of benefit. We hypothesized that an Emergency Department Observation Syncope Protocol would reduce resource use without adversely affecting patient-oriented outcomes.

Methods

This randomized trial at five EDs compared an ED observation syncope protocol to inpatient admission for intermediate-risk adults (≥50 years) presenting with syncope or near-syncope. Primary outcomes included inpatient admission rate and length-of-stay. Secondary outcomes included 30-day and 6-month serious outcomes after hospital discharge, index and 30-day hospital costs, 30-day quality-of-life scores, and 30-day patient satisfaction.

Results

Study staff randomized 124 patients. Observation resulted in a lower inpatient admission rate (15% vs. 92%, 95%CI Difference: −88%, −66%) and shorter hospital length-of-stay (29 vs. 47 hours, 95%CI Difference: −28, −8). Serious outcome rates after hospital discharge were similar for observation vs. admission at 30-days (3% vs. 0%, 95%CI Difference: −1%, 8%) and 6-months (8% vs. 10%, 95%CI Difference: −13%, 9%). Index hospital costs in the observation group were $629 (95%CI Difference: −$1376, −$56) lower than in the admission group. There were no differences in 30-day quality-of-life scores or in patient satisfaction.

Conclusions

An ED observation syncope protocol reduced the primary outcomes of admission rate and hospital length-of-stay. Analyses of secondary outcomes suggest reduction in index hospital costs with no difference in safety events, quality-of-life, or patient satisfaction. Our findings suggest that an ED observation syncope protocol can be replicated and safely reduce resource use.

Introduction

Background

Syncope represents a common and vexing chief complaint in emergency departments. In the United States alone, syncope accounts for 740,000 annual emergency department (ED) evaluations¹ and yearly hospital costs of over $2.4 billion.² Because patients have recovered by the time of ED presentation, it is often difficult to distinguish among the many potential causes that include benign as well as life-threatening conditions. Despite international efforts to develop clinical guidelines^3-7, diagnostic pathways^8-11, and risk prediction tools^12-18, there remains considerable uncertainty about how to optimally manage patients at intermediate risk of adverse outcomes.¹⁹ As a result, providers often hospitalize older adults without a clear cause for syncope for diagnostic evaluation.^1,20-22 However, current admission practices are characterized by low diagnostic yield and significant practice variation ²³, do not clearly improve outcomes²⁴, and are costly.^2,25 These findings have been reported from multiple countries.^26-33 Efforts by the U.S. Center for Medicare and Medicaid Services (CMS) to deny hospital payments for ‘unnecessary’ inpatient admissions have further intensified the need to develop an alternative diagnostic pathway; syncope was recently identified as the top diagnosis associated with payment denials by CMS Recovery Audit Contractors.³⁴

Importance

An Emergency Department Observation Syncope Protocol may safely reduce hospitalizations by expediting and standardizing the evaluation of syncope. A previous single-center randomized evaluation of an ED based, syncope evaluation unit suggested a 55% reduction in hospital admissions without increase in mortality.³⁵ However, these results have not been replicated at other sites, and there is no information about how such an approach may impact costs, non-fatal clinical events, and patient centered outcomes such as quality-of-life and satisfaction. Evaluating the efficiency and safety of this alternative delivery approach has important health delivery implications; 36% of U.S. EDs operate an observation unit and have the potential ability to implement an ED observation protocol.³⁶

Goals of This Investigation

We compared an ED observation syncope protocol vs. routine inpatient admission for intermediate-risk patients after an unrevealing emergency department evaluation for syncope. We tested the primary hypotheses that an ED observation protocol would reduce hospital admissions and hospital length-of-stay.

We originally intended to collect planning data for a definitive non-inferiority trial of safety, costs, and quality-of-life. Because of changes in payer audit and payment policies during the study period³⁴, however, it is unlikely that U.S. hospitals will participate in future randomized studies of ED observation unit care. In exploratory analyses, we assessed the impact of the ED observation protocol on safety, costs, quality-of-life, and patient satisfaction.

Methods

Study Design and Setting

We conducted a randomized clinical trial at five emergency departments (EDs) from March 1, 2010 to October 1, 2011 (ClinicalTrials.gov Identifier: NCT01003262). Study staff completed participant follow-up on April 31, 2012. We include the trial protocol and CONSORT checklist as online supplements.

The study sites represent a diversity of hospital characteristics, geography, and patient populations. (eTable 1) All ED observation units are located in a distinct physical space adjacent to the main ED, supervised by attending emergency physicians, and staffed by mid-level providers.

The IRBs of the coordinating center and all enrolling sites approved this study. An independent safety monitor reviewed all data on clinical events.

Selection of Participants

Patients aged 50 years or greater were prospectively screened in the emergency department for a complaint of syncope or near-syncope. Syncope was defined as a sudden, transient loss of consciousness. Near-syncope was defined as a sensation of imminent loss of consciousness, without actual syncope.

Using specialty society guidelines, the study team developed risk stratification guidelines for short-term, dangerous clinical events after syncope .^4,6,37 (Figure 1) We included additional feedback from enrolling physicians to ensure that the guidelines were feasible and acceptable at the study sites. Treating physicians used these criteria to categorize patients as high, intermediate, or low risk. Patients at intermediate risk were eligible for study enrollment. Although we considered objective risk scores, none have been validated for routine clinical use³⁸ and were not felt to be feasible at our enrolling sites.

Figure 1.

Risk Stratification Guidelines

We excluded patients with a serious condition identified during the ED, including symptomatic arrhythmias, myocardial infarction, pulmonary embolism, acute pulmonary edema, stroke, severe anemia or blood loss requiring blood transfusion, sepsis, and major traumatic injury. Additional exclusion criteria included: seizure, head trauma, or intoxication as the reason for loss of consciousness; new or baseline cognitive impairment; do-not-resuscitate or do-not-intubate status; active chemotherapy for cancer; and inability to speak either English or Spanish.

Randomization Assignment

After providing informed consent, patients were block randomized (n=4) by site in a 1:1 ratio to either the observation protocol or routine inpatient admission. A computer generated the study arm assignment at the time of randomization, and no research personnel had advance knowledge of study arm assignment. We could not blind this health service intervention to patients, providers, or research personnel. All study patients received an initial ED evaluation consisting of a directed history, physical examination, standardized laboratory tests, and a 12-lead electrocardiogram.

Intervention

We used professional society guidelines^4,6,37 to design the ED observation protocol intervention. All observation protocol patients received continuous cardiac monitoring for at least 12 hours. The ED treating team ordered at least two serial cardiac troponin tests, approximately six hours apart, to exclude acute myocardial infarction. The study sites used different troponin assays; we defined a normal troponin threshold as the 99^th percentile value for a reference population for each site’s assay. The ED treating team ordered a rest echocardiogram for patients with a cardiac murmur on chest auscultation, if a prior echocardiogram had not been performed in the prior six months. The ED treating teams could perform additional testing at their discretion.

The maximum stay in the ED observation unit could not exceed 24 hours. Observation protocol patients who were diagnosed with a serious condition, had persistent symptoms of syncope or near-syncope, were felt by the treating physician to be unable to be safely discharged home because of functional reasons (e.g. inability to ambulate), or had pending tests at 24 hours were admitted to the hospital. All other patients were eligible for discharge. The treating ED team made the final decision to admit or discharge observation patients.

An inpatient medicine service managed patients randomized to routine inpatient admission. The study protocol did not guide the care of patients randomized to this arm. Contamination between the study arms was minimized because both groups were managed in distinct physical spaces by different clinical services. Although it is possible that some patients in the routine inpatient admission arm were classified as “observation status” for billing purposes, the physical setting, providers, and processes were indistinguishable from “inpatient” care.

Outcomes

Primary outcomes included inpatient admission rates (%) and hospital length-of-stay (hours) at the index ED visit. Prior studies of health service interventions for syncope have used admission rate as a primary outcome.^8-11,35 Because hospital services may be similar in inpatient, observation, and ED settings, we also assessed hospital length-of-stay as a primary outcome.

Secondary outcomes included 30-day and 6-month serious clinical events, index and 30-day hospital costs, 30-day changes in quality-of-life after study enrollment, and 30-day patient satisfaction.

We defined potential safety events as serious clinical events that occurred after discharge from the index hospital visit. Clinical events that occurred during the hospitalization were not considered safety events, as these were felt to represent appropriate recognition of serious illness during the diagnostic evaluation. A multi-specialty panel of emergency physicians, internists, geriatricians, and cardiologists previously defined syncope-related, serious clinical events.²⁰ These include death, ventricular arrhythmias, Mobitz II or complete heart block, sick sinus syndrome, sinus pause greater than 3 seconds, symptomatic supraventricular tachycardia (heart rate>100 beats per minute), symptomatic bradycardia (heart rate <60 beats per minute), major cardiac intervention such as permanent pacemaker placement, myocardial infarction, stroke, pulmonary embolism, aortic dissection, non-traumatic intracranial hemorrhage, internal hemorrhage or anemia requiring blood transfusion, and major traumatic injury (intracranial bleed, bone fracture, or thoraco-abdominal visceral injury) associated with syncope, near-syncope, or a fall occurring after the index visit. We also measured recurrent episodes of syncope or near-syncope resulting in an ED visit.

We modeled hospital costs for all 30-day acute care services in the ED, observation unit, and inpatient settings by imputing 2011 Medicare national mean payments for procedures and observation facility fees. The eMethods and eTable 2 describe the cost model in detail.

We administered the Quality of Well-being Scale³⁹ to measure general health utility and the Syncope Functional Status Questionnaire^40,41 to measure symptom specific quality-of-life. The Quality of Well-being Scale ranges from 0-1, with 0 indicating worst possible health and 1 indicating optimum health. The Syncope Functional Status Questionnaire ranges from 0-100, with 0 indicating no syncope- related impairment and 100 indicating maximum impairment.

Finally, we measured patient satisfaction using the Consumer Assessment of Healthcare Providers and Systems – Hospital overall rating of care⁴², which ranges from 0 (worst) −10 (best).

Baseline and Follow-up Measures

At baseline, all patients received 12-lead electrocardiogram, cardiac troponin, basic metabolic panel, and hematocrit testing during the index ED evaluation. Research staff extracted laboratory and vital sign data from the ED chart. Research assistants obtained demographic data, including race/ethnicity, directly from patients and administered baseline quality-of-life instruments. Treating physicians completed a survey about patient history of presentation, pre-existing co-morbidities, physical exam, and electrocardiogram interpretation. We used clinical data to estimate a Syncope Risk Score, which is a case mix measure derived from a population of older adults with syncope.⁴³ Research staff extracted information on inpatient admission and length-of-stay from administrative data.

Outcomes ascertainment after patient enrollment included direct patient phone interviews by research assistants and medical chart abstraction by physician-reviewers. Research staff called patients at 30-days to determine vital status, identify subsequent hospital visits occurring at non-enrolling site facilities, and administer follow-up quality-of-life instruments. Physician reviewers abstracted records from all ED visits and hospitalizations within 30 days, including those that occurred outside of the enrolling sites, to identify serious clinical events and to quantify hospital-based health service use. Research staff recorded disposition (discharge from ED, observation, inpatient admission) and major diagnostic and therapeutic procedures. We repeated direct patient phone interviews and chart reviews at 6-months to identify additional serious clinical events and episodes of recurrent syncope. Finally, we verified the vital status of all patients at 6-months after enrollment using the Social Security Death Masterfile.⁴⁴

A second physician-reviewer who was blinded to the original reviews reabstracted charts for the 40 enrolled patients (first 10 enrollments at each of the 4 sites). There was 100% agreement on the occurrence of serious clinical events and good agreement on whether patients received any of 25 diagnostic or therapeutic procedures (K=0.63-1).

Data Analysis

We created intent-to-treat regression models with study sites as fixed-effects to analyze all outcomes. The predictor variable was randomization assignment. We analyzed binary outcomes using logistic regressions. To correct for skewness, we applied a log transformation to cost outcomes. The difference in cost outcomes between observation and routine care was estimated based on a log-scale analysis and then transformed back to the original scale, and the corresponding 95% confidence interval was calculated using bootstrapping with 1000 iterations. Finally, we used the log-rank test to analyze survival without a safety event during the first six months after initial hospital discharge, and Kaplan-Meier curves for safety-events were plotted.

The sample size was designed to achieve a power of 80% to detect a 22% reduction in inpatient admission rate in the observation group (two sided α=0.05), assuming an inpatient admission rate of 85% in the standard care group. A safety monitor performed an interim analysis at 6-months with a pre-defined stopping threshold for an absolute difference of 8% or greater in serous outcomes between the two study arms.

Non-Inferiority of Safety

We performed a post-hoc, non-inferiority analysis of one-month serious outcomes after index hospital discharge. We defined a pre-specified, non-inferiority threshold of 4% absolute difference in 30-day safety event rates using an economic justification. (See Trial Protocol) If the observation protocol resulted in a 4% worse safety event rate than routine care, then prior cost data² suggest that it would cost $135,000 to potentially avoid one safety event by admitting all patients rather than observing them. This threshold assumes that inpatient admission reduces downstream morbidity and mortality related to syncope safety events, which has never been proven²⁴, and this margin is more expensive than commonly suggested criteria for cost-effective care (e.g., $100,000 / Quality-Adjusted Life Year).⁴⁵

In contrast to frequentist procedures which seek to disprove the null hypothesis, a Bayesian approach estimates the likelihood that a study hypothesis is true. Bayesian analyses may be particularly helpful for small studies, when there may be insufficient data to exclude the null hypothesis.⁴⁶ We used Bayesian logistic regression models to estimate the posterior probability that the absolute difference in one-month safety events was within the pre-specified 4% non-inferiority margin. We used non-informative prior distributions for the models. Bayesian parameter estimates were generated with Markov chain Monte Carlo methods, and Bayesian model inferences were summarized as a point estimate and an interval containing the true parameter with some probability (i.e. the 95% Credible Interval). As recommended by CONSORT guidelines for non-inferiority analyses⁴⁷, we analyzed: (1) all enrolled patients (intent-to-treat approach) and (2) all patients who received the assigned treatment (per-protocol approach). The Bayesian modeling was implemented using the WinBUGS software (publicly available; version 1.4.3).⁴⁸

Protocol Modifications

This study differs from the original trial registration and protocol in the following aspects: 1. We lowered the age criterion from the original trial proposal (≥60 years) after four months of enrollment to improve patient recruitment; the study sponsor and study site IRBs approved this change; 2. The original trial protocol proposed collection of 6-month quality-of-life data; however, this was dropped because of participant complaints about survey length and burden; 3. We proposed a formal cost effectiveness analysis (i.e., comparison of the ratio of cost to quality-adjusted-life-year). We dropped this analysis due to lack of observed differences in the denominator (see Results below). A formal cost-effectiveness analysis would yield no additional information compared to an assessment of cost-difference only; 4. We added satisfaction as a secondary outcome because of its importance to patient centered care; 5. Finally, the Bayesian non-inferiority assessment of safety events was an exploratory analysis and not proposed in the original protocol.

Results

Characteristics of Study Subjects

Figure 2 describes screening, eligibility, and randomization. Of 2,724 emergency department patients screened for syncope or near syncope, there were 1,235 who were potentially eligible prior to risk stratification. Treating physicians excluded an additional 315 patients for low risk and 633 patients for high risk. Compared to intermediate risk patients, high risk patients were older (70 vs 66; 95%CI difference 2.4, 5.7); there were no differences by gender (67% vs 70% female; chi-square p=0.26). eTable 3 lists specific reasons for high-risk exclusions.

Figure 2.

Screening and Enrollment Flowchart

Of 287 eligible patients after risk stratification, study staff randomized 124 patients. eTable 4 describes reasons why eligible patients were not randomized. In the final study cohort, there was a higher proportion of patient with abnormal electrocardiograms in the control group; there were otherwise no major imbalances in demographic and clinical features between the groups. (Table 1) Protocol violations included four patients who left against medical advice after randomization and two patients discharged by the treating physician despite randomization to the admission arm.

Table 1.

Characteristics of Study Cohort

Characteristic	Observation (n=62)	Routine Admission (n=62)
Age, years (Mean, SD)	65 (11)	64 (11)
Chief complaint of syncope* (n,%)	46 (74%)	38 (61%)
Male (n,%)	29 (47%)	32 (52%)
White/non-Hispanic (n,%)	27 (44%)	24 (39%)
Past Medical History (n,%)  Congestive heart failure
	1 (2%)	2 (3%)
Coronary artery disease	8 (13%)	5 (8%)
Arrhythmia	5 (8%)	4 (6%)
Syncope in prior year	10 (16%)	13 (21%)
Index Syncope History (n,%)  Occurred while supine
	8 (13%)	8 (13%)
Associated with chest pain	8 (13%)	7 (11%)
Associated with shortness of breath	11 (18%)	4 (6%)
Associated with palpitations	6 (10%)	4 (6%)
No warning signs	18 (29%)	13 (21%)
Baseline quality-of-life scores  Quality of Well Being Scale (Mean, SD)  Quality of Well Being Scale (median, IQR)
	0.55 (0.15)	0.55 (0.14)
	0.55 (0.45, 0.65)	0.55 (0.47, 0.66)
Syncope Functional Status Questionnaire (Mean, SD)	29 (25)	25 (26)
Syncope Functional Status Questionnaire (median, IQR)	26 (7.1, 41)	17 (4.8, 39)
Triage systolic blood pressure (Mean, SD)	140 (24)	141 (24)
Hematocrit <30 (n,%)	5 (8%)	3 (5%)
Troponin >99% normal reference (n,%) †	5 (8%)	5 (8%)
Abnormal initial electrocardiogram (n,%) ‡	11 (18%)	23 (37%)
Syncope Risk Score (Mean, SD)||	0.76 (0.84)	0.76 (0.67)

^*vs. near-syncope

^†>99% normal reference range for site-specific troponin assay

^‡Presence of non-sinus rhythm; multiple PVCs (>1 on standard 12-lead tracing); sinus bradycardia<40; left or right ventricular hypertrophy; left or right axis deviation; complete left or right bundle branch block; first degree block (>20mS); short PR interval (<10 mS); prolonged QRS (>1 OmS); prolonged QTc (>450mS);

Q/ST/T changes consistent with acute or chronic ischemia

^||Risk prediction score developed for older adults with syncope⁴³

Primary and Secondary Outcomes

We describe study outcomes in Table 2 and missing outcome data in eTable 5. In the routine admission arm, protocol violations accounted for the 8% of patients who were not hospitalized. Compared to routine admission, observation was associated with absolute reductions of 77% in inpatient admission rate (relative risk ratio: 0.16; 95%CI: 0.09, 0.29; p<0.0001) and 18 hours in hospital length-of-stay (p=0.0003).

Table 2.

Outcomes by Study Group^*

	Observation	Routine Admission	Difference
OUTCOMES	(n=62)	(n=62)	(95%CI)
Primary Outcomes
Inpatient Admission (n,%)	9 (15%)	57 (92%)	−77% (−88%, −66%)
Length-of-Stay, hours (Mean, SD)	29 (15)	47 (34)	−18 (−28, −8)
Secondary Outcomes
Costs †
Hospital costs, at index visit, in $US (Mean, SD)	1400 (1220)	2420 (3930)	−629 (−1376, −56)
Hospital costs, at index visit, in $US (Median, IQR)	1190 (870, 1550)	1570 (870, 2370)
Hospital costs, within 30 days, in $US (Mean, SD)	1800 (2150)	2520 (3980)	−479 (−1230, 198)
Hospital costs, within 30 days, in $US (Median, IQR)	1210 (948, 1660)	1580 (870, 2390)
Serious Clinical Outcomes During Hospital Visit (n,%)	5 (8%)	3 (5%)‡	3% (−6%, 12%)
Death	0	0
Arrhythmia	2 (3%)	2 (3%)
Pacemaker Insertion	1 (2%)	1 (2%)
Syncope/Fall With Bone Fracture	2 (3%)	1 (2%)
30-Day Recurrent Syncope (n,%)	1 (2%) 32	1 (2%)	0% (−4%, 4%)
30-Day Serious Outcomes After Hospital Discharge (n,%)	2 (3%)	0	3% (−1%, 8%)
6-Month Serious Outcomes After Hospital Discharge (n, %)	4 (8%)	5 (10%)	−2% (−13%, 9%)
Quality-of-life
Change in Quality of Well Being score (Mean, SD)	0.00 (0.20)	0.03 (0.18)	−0.02 (−0.10, 0.06)
Change in Syncope Functional Status Questionnaire score (Mean, SD)	−7.6 (20.1)	−2.4 (26.3)	−5.2 (−15.2, 4.8)
Patient Satisfaction (Mean, SD)	8.9 (1.4)	9.3 (0.9)	−0.46 (−0.95, 0.026)

^*% denominator based on complete outcomes data- see eTable 3 for missing outcomes data; all site fixed effects were non-significant (p>0.05)

^†95%CI for differences based on back-transformation of log-transformed analysis; see text

^‡1 patient experienced symptomatic arrhythmia, cardiac ablation, and pacemaker placement

eTable 6 describes all patients with potential safety events. There were no significant differences in the proportion of patients who were diagnosed with a serious clinical condition after the initial ED evaluation at 30-days or 6-months. Figure 3 displays the Kaplan-Meier curve for safety events at six months (p=0.8).

Figure 3.

Kaplan-Meier Curves for Safety Events

The ED observation syncope protocol was associated with an absolute cost reduction of $629 for hospital services associated with the index visit, and a non-significant trend towards lower 30-day hospital costs. We found no differences in diagnostic testing rates between the groups (eTable 7); therefore the cost differences were attributable to differences in hospital length-of-stay. There were no significant differences in general health utility, syncope-specific quality-of-life, or patient satisfaction.

Non-Inferiority of 30-day Safety Events

eFigures 1 and 2 illustrate the posterior distributions of rate-differences. For the intent-to-treat analysis, the posterior estimate of difference in safety event rates at one month between observation and admission was 3.1% (95% CI: 0.3%, 9.0%). Using a 4% non-inferiority margin, the posterior probability of the observation protocol being non-inferior to routine care was 0.72, given the observed data.

For the per-protocol analysis, the posterior estimate of difference in safety event rates at one month between the two arms was 3.3% (95% CI: 0.3%, 9.4%). The posterior probability of the observation protocol being non-inferior to routine admission was 0.70.

Limitations

Strengths of our study include patient-level randomization and replication of the protocol at EDs with a diversity of structural and patient characteristics. However, we acknowledge potential limitations. First, this study was not a priori powered to assess non-inferiority of secondary outcomes. Given external payer pressures to reduce inpatient syncope admissions³⁴, additional efforts to randomize patients are unlikely to succeed in U.S. settings. However, our Bayesian analyses suggest that the observation protocol is likely to be non-inferior to routine inpatient admission for safety events.

Second, there is no existing consensus for an acceptable non-inferiority margin for safety events. We used an economic approach using a conventional cost/ quality-life year threshold and highly conservative assumptions about the potential benefit of hospitalization over observation. However, the clinical acceptability of our proposed safety margin is unknown and requires future investigation.

Third, we found that almost half of potentially eligible patients were excluded as ‘high-risk’; half of those were based on the clinical judgment of treating physicians. Because published risk scores have not been validated for routine clinical use³⁸, we developed risk stratification guidelines that were acceptable to enrolling site physicians and consistent with prior research.³⁵ However, conservative risk stratification and physician discomfort with the observation protocol may have limited potential enrollment.⁴⁹ Validated risk prediction instruments may broaden the safe application of the observation protocol.⁴³ Although there is the potential for selection bias towards less severely ill patients, the overall 30-day event rate (6%) observed in our trial is almost identical to the reported event rate in a community cohort of intermediate-risk, older adults presenting to an ED with syncope.⁴³

Fourth, our cost analysis did not include outpatient facility or patients’ costs. We were unable to collect reliable data on outpatient visits, and it is possible that there were difference in outpatient health service utilization between the treatment arms. In addition, there is increasing concern about large and unanticipated patient co-payments associated with observation care.⁵⁰ Although we did not have data on patient co-payments, it is likely that Medicare covered patients in our trial would have paid less under observation status than as an inpatient admission. The 2011 Medicare inpatient deductible for hospital services was $1,132⁵¹; at the outpatient Medicare co-pay rate of 20%, Medicare covered patients in our observation arm on average would have paid less than $300 for hospital services.

Fifth, the original protocol proposed to study patients aged 60 years and older because of higher incidence of adverse events and health service use.^52,53 The age criterion was lowered because of enrollment challenges, and it is possible that observation has a differential impact in patients above and below 60 years of age. In post-hoc analysis, we assessed whether there was any interaction effect between treatment assignment and age (dichotomized as <60 and ≥ 60 years). We found no significant interaction effects (p>0.3) for any of the primary or secondary outcomes.

Sixth, there was an imbalance in patients with abnormal ECGs between the two treatment arms, likely due to the small sample size. Similar Syncope Risk Stratification Scores (which incorporate ECG abnormalities) suggest that overall risk in the groups were comparable prior to randomization. In post-hoc analysis which included abnormal ECG, the difference in index hospital costs became insignificant (p=0.07); there were no qualitative changes in analyses of any other primary or secondary outcome.

Finally, we had incomplete data for quality-of-life and satisfaction ratings (eTable 3) due to participant refusal, which is likely related to length of the survey instruments. There is the potential for response bias, although we did not find evidence of differential response rates by study arm.

Discussion

We found that an ED observation protocol substantially reduced hospital inpatient admissions, length-of-stay, and index hospital costs in older patients with intermediate-risk syncope. We observed similar rates of potential safety events, changes in quality-of-life scores, and ratings of overall care between the study arms. We did note that the cost advantage of an ED observation protocol was attenuated at 30-days, and this finding may potentially be related to subsequent hospital visits that were unrelated to the index visit for syncope. Our findings suggest that a structured observation protocol, based in the emergency department, is a safe and a cost-saving alternative to hospital inpatient admission for the evaluation of intermediate-risk syncope.

Several European studies suggest that structured decision pathways and specialized diagnostic units may safely reduce resource use in the evaluation of syncope.^8,9 However, these studies were not randomized and the findings of benefit have not been uniform.^10,11 There previously has been a single randomized trial (n=103) of an ED based syncope unit, performed at an academic center serving a predominantly white population.³⁵ This prior trial reported a 55% absolute reduction in hospital admissions and a 54% relative reduction in hospital bed-days. There were no significant differences in death or recurrent syncope at 2-year follow-up. Our study confirms and generalizes these previous findings to sites with diverse hospital and patient characteristics. We also assessed costs, non-fatal safety events, quality-of-life, and patient satisfaction which previously have not been studied. Our trial suggests that an ED observation protocol can be successfully implemented in a variety of practice settings with an existing emergency department observation unit.

Several mechanisms may explain our findings. The immediate availability of testing services and attending providers at all times of the day may have facilitated rapid decision-making in the observation group compared to the patients admitted to the inpatient setting. The study protocol is a structured pathway and may reduce variance in the diagnostic evaluation. Similar post-discharge event rates for both groups suggest that 12-24 hours of cardiac monitoring in an observation unit is safe; there may be diminishing diagnostic value in extending monitoring beyond 24 hours.⁵⁴

In summary, we found that an ED-based observation protocol substantially reduced resource use at the index visit for intermediate-risk patients with syncope, without evidence of worse clinical, quality-of-life, or satisfaction outcomes. This protocol can easily be adapted for emergency departments with existing observation units and represents a cost-effective and safe alternative to routine inpatient admission. Future research should confirm our preliminary findings of safety in external cohorts and develop objective risk prediction instruments to identify a broader set of patients who may be eligible for observation unit care.