Portal de Periódicos da CAPES

Decreasing Length of Stay in the Emergency Department With a Split Emergency Severity Index 3 Patient Flow Model

2013; Wiley; Volume: 20; Issue: 11 Linguagem: Inglês

10.1111/acem.12249

1553-2712

Rajiv Arya, Grant Wei, Jonathan V. McCoy, Jody Crane, Pamela Ohman‐Strickland, Robert Eisenstein,

Healthcare Policy and Management

There has been a steady increase in emergency department (ED) patient volume and wait times. The desire to maintain or decrease costs while improving throughput requires novel approaches to patient flow. The break-out session "Interventions to Improve the Timeliness of Emergency Care" at the June 2011 Academic Emergency Medicine consensus conference "Interventions to Assure Quality in the Crowded Emergency Department" posed the challenge for more research of the split Emergency Severity Index (ESI) 3 patient flow model. A split ESI 3 patient flow model divides high-variability ESI 3 patients from low-variability ESI 3 patients. The study objective was to determine the effect of implementing a split ESI 3 flow model has on patient length of stay (LOS) for discharged patients. This was a retrospective chart review at an urban academic ED seeing over 70,000 adult patients a year. Cases consisted of adults who presented from 9 a.m. to 11 p.m. from June 1, 2011, to December 31, 2011, and were discharged. Controls were patients who presented on the same times and days, but in 2010. Visit descriptors included age, race, sex, ESI score, and first diagnosis. The first diagnosis was coded based on methods used by the Agency for Healthcare Research and Quality to codify International Classification of Diseases, ninth version, into disease groups. Linear models compared log-transformed LOS for cases and controls. A front-end ED redesign involved creating guidelines to split ESI 3 patients into low and high variability, a hybrid sort/triage registered nurse, an intake area consisting of an internal results waiting room, and a treatment area for patients after initial assessment. The previous low-acuity area (ESI 4s and 5s) began to see low-variability ESI 3 patients as well. This was done without additional beds. The intake area was staffed with an attending emergency physician (EP), a physician assistant (PA), three nurses, two medical technicians, and a scribe. There was a 5.9% decrease, from 2.58 to 2.43 hours, in the geometric mean of LOS for discharged patients from 2010 to 2011 (95% confidence interval CI = 4.5% to 7.2%; 2010, n = 20,215; 2011, n = 20,653). Abdominal pain was the most common diagnostic grouping (2010, n = 2,484; 2011, n = 2,464) with a reduction in LOS of 12.9%, from 4.37 to 3.8 hours (95% CI = 10.3% to 15.3%). A split ESI 3 patient flow model improves door-to-discharge LOS in the ED. Ha habido un incremento progresivo en el volumen de pacientes y los tiempos de espera (TDE) en los servicios de urgencias (SU). El deseo de mantener o disminuir el coste mientras se mejora el rendimiento requiere aproximaciones novedosas para el flujo de pacientes. La sesión "Intervenciones para Mejorar los Tiempos de la Atención Urgente" en la conferencia de consenso "Intervenciones para Asegurar la Calidad en un Servicio de Urgencias Saturado" de la Academic Emergency Medicine de 2011 planteó el reto de una mayor investigación del modelo de flujo dividido de pacientes nivel 3 según el Emergency Severity Index (ESI). Un modelo de flujo dividido de pacientes nivel 3 según el ESI separa los pacientes nivel 3 según el ESI en alta y de baja variabilidad. El objetivo fue determinar el efecto que tiene implementar un modelo de flujo dividido en el nivel 3 según el ESI en el tiempo de estancia (TDE) para los pacientes dados de alta. Revisión retrospectiva de historias clínicas de un hospital traumatológico de Nivel 1 que atiende más de 70.000 paciente al año. Los casos consistieron en adultos que acudieron de 9 am a 11 pm desde el 1 de junio de 2011 al 31 de diciembre de 2011 y que fueron dados de alta. Los controles fueron pacientes que acudieron en el mismo periodo de tiempo y días, pero en 2010. Las variables de la visita incluyeron edad, raza, sexo, puntuación ESI y primer diagnóstico. El primer diagnóstico se codificó según los métodos utilizados por la Agency for Healthcare Research and Quality para codificar la Clasificación Internacional de Enfermedades 9ª edición, en grupos de enfermedades. El TDE con transformación logarítmica para los casos y los controles fue comparado mediante modelos lineales. El rediseño pleno del SU incluyó la creación de guías clínicas para dividir los pacientes de nivel 3 según ESI en baja y alta variabilidad, un híbrido de triaje/breve por el enfermero especialista (EE), un área de entrada que consistía en una sala de espera de resultados internos y un área de tratamiento para los pacientes tras la valoración inicial. El área previa de atención baja (ESI 4 y 5) empezó a ver también pacientes ESI 3 de baja variabilidad. Esto se hizo sin camas adicionales. El área de entrada tuvo una plantilla con un urgenciólogo, un asistente médico, tres EE, dos auxiliares médicos y un secretario. Hubo un descenso de un 5,9%, de 2,58 horas a 2,43 horas, en la media del TDE para los pacientes dados el alta desde 2010 a 2011 (IC 95% = 4,5% a 7,2%) (2010, n = 20.215; 2011, n = 20.653). El dolor abdominal fue el grupo diagnóstico más frecuente (2010, n = 2.484; 2011, n = 2.464) con una reducción en el TDE de un 12,9%, de 4,37 horas a 3,8 horas (IC 95% = 10,3% a 15,3%). Un modelo de flujo dividido de pacientes de nivel 3 según ESI mejora el TDE en el SU. There has been a steady increase in emergency department (ED) patient visits and a decrease in the number of EDs over the past 20 years.1 Recently enacted legislation aimed at reducing costs and increasing access will likely further increase ED demand while driving the necessity for reducing cost per patient. This need to decrease cost and improve throughput requires novel approaches to patient flow. The breakout session "Interventions to Improve the Timeliness of Emergency Care" at the June 2011 Academic Emergency Medicine consensus conference "Interventions to Assure Quality in the Crowded Emergency Department" posed the challenge for more research on the split Emergency Severity Index (ESI) 3 patient flow model and greater efficiency.2, 3 The ESI is reported to be the most commonly adopted triage acuity system in the United States.4, 5 Similar scales are in use in Canada, the United Kingdom, Australia, and other countries. The boards of the American College of Emergency Physicians and the Emergency Nurses Association support the adoption of five-level triage scales. ESI's five-level system is driven by patient acuity and resource utilization.6 Levels 1 and 2 are distinguished by instability of vital signs or severity of presenting complaint, while levels 3, 4, and 5 are considered less acute and thus are differentiated by the number of resources required. Examples of resources include diagnostic testing and interventions. Level 4 and 5 patients require one or no resources, respectively, and are traditionally treated in designated low-acuity areas of the ED such as "fast tracks." A split patient flow model for ESI Level 3 patients segments high-variability and low-variability ESI 3 patients into separate streams with the intent of driving them through a more customized and efficient process. The ultimate goal is to reduce variability in the care of these patients, realizing improvements in metrics such as door-to-doctor times, length of stay (LOS), left without being seen, and patient satisfaction.7 The benefits of reducing variability have been described in fast track settings.8-10 The split-flow model works in a similar manner, reducing variability, reducing LOS, and enhancing service responsiveness. An ever-growing body of literature demonstrates the many negative consequences of the ED overcrowding and boarding crisis.11, 12 Downstream effects include delays in care, decreased patient and physician satisfaction, and increased patient mortality.13-16 Our objective was to determine the effect of implementing a split ESI 3 flow model has on patient LOS for all discharged patients. We conducted a retrospective chart review to examine the effect of a split ESI 3 flow model on the LOS for patients who were discharged from the ED. The process had a go-live date of May 31, 2011. The study protocol was approved by the institutional review board of the medical school. This study was conducted in an urban, academic ED that sees over 70,000 adult patients a year. Cases consisted of adults, age greater than 21 years, who presented from 9 a.m. to 11 p.m. from June 1, 2011, to December 31 2011. Controls were patients who presented the same time and the same calendar days in 2010. Our site implemented a split ESI 3 flow model to efficiently manage patients presenting during the busiest hours of the day, 9 a.m. to 11 p.m. The implementation of a split ESI 3 flow model required six major changes: splitting of ESI 3 patients, infrastructure, staffing, responsibilities, purpose of the area, and a roll-out plan. Each will be described in detail, and then a few patient examples will be presented to demonstrate the application. The intake17 area saw all ESI 4 and 5 patients as well as low-variability ESI 3. Low-variability ESI 3 patients required two or more resources and typically follow a standardized work flow: assessment, diagnostics, medications, and one reassessment prior to disposition. They rarely require significant physician or physician assistant (PA) attention above and beyond the initial assessment and final disposition reassessment. The physician and nursing leadership worked together and created a guide to determine who was considered a low-variability ESI 3. Some of the recommendations are institution-specific; e.g., we are unable to perform a private pelvic examination in the intake area (Table 1). The team determined that the intake area would see about six patients per hour, between a physician and PA, and wanted to limit the patients to those who required no more than 20 minutes of provider time for assessment, orders, and reassessment. Intake was divided into four distinct zones: assessment, treatment area, results waiting room, and the discharge area. The intake assessment area was equipped with six reclining chairs where the staff performed assessments and procedures and administered initial treatments. In addition to the six assessment chairs, a discharge chair was designated to review results with patients and perform the discharge encounters. The assessment area was our prior fast track and required no construction. The treatment area consisted of seven curtained treatment bays with stretchers, which were previously part of the main ED. During the times intake was open it was used for those patients and returned to the main ED after intake closed. The bays were the closest to the intake assessment area and no construction was required. The results waiting room was defined as an internal waiting room and contained five reclining chairs with a TV, a desk, and a computer. The results waiting room was a conference room that was repurposed to clinical space and only required cosmetic changes. The discharge area was a pediatric triage room near the front entrance of the ED that was no longer in use and was converted to a registration work space. The intake assessment area was staffed with one emergency physician (EP), one PA, one scribe, two nurses, and one medical technician. The results waiting room was staffed by one medical technician. The treatment area was staffed by one nurse, and the discharge kiosk was staffed by a registration clerk who completed the final registration paperwork. There was no increase in the physician or PA staffing hours. Sixteen additional nurse and medical technician hours were added. Each provider was accountable to ensure that each patient flowed smoothly through the process (Table 2). The triage nurse or the assessment nurses would direct-bed the patient based on initial presenting complaint, which was compared to the list created by the team to determine the proper assessment area for the patient. The physician or PA would perform the medical evaluation and assessment. The scribe would prepare the medical record, update the providers on results, and prepare discharge instructions and prescriptions. After placing the patient in a treatment bed, the assessment nurse would place the initial intravenous (IV) line, draw blood, and medicate the patient as needed. The assessment medical technician would transport the patient to radiology and perform general medical technician duties. The treatment nurse would provide ongoing management of the patient and would update the physician and PAs to the patient's progress. The results waiting room medical technician would transport the patient in and out of the room, would observe each patient for change in conditions, and would keep the patient informed as to the status of his or her test results and treatment progress. One physician One PA Two nurses One medical technician One scribe Each area within intake had a unique role relative to the overall flow. The assessment area is where the physician and PA would see the patients and the nurses would perform initial nursing tasks. The treatment area enabled patients who could not stay vertical during their entire clinical encounter (i.e., those medicated with IV opiates) to assume treatment beds while undergoing evaluation and management. The results waiting room provided an additional, more efficient location for patients to await ancillary results, enabling the assessment area to maintain flow and enhancing overall bed capacity. Finally, the discharge area allowed the collection of the financial information and the completion of the registration process to occur in a nonclinical setting, further improving bed capacity. The main ED had bedside registration. In this flow model, patients initially presented to the registration clerk, who performed quick registration (name, date of birth, chief complaint). Based on the chief complaints, patients were pulled into the intake area or waited for a more traditional triage encounter (Figure 1). Those patients pulled directly to the intake area were often seen by the nurse and provider simultaneously, further streamlining the process. For example, a suspected renal colic patient, an ideal example of a low-variability ESI 3 patient, would be seen by the physician and initial medications and fluids started in intake. From there the patient would go to radiology for a computed tomography scan and would return to either the results waiting room or the treatment area, depending on the type of medication he or she was given in intake. High-risk medications, such as IV opiates, required the patient to go to a treatment bed, while patients receiving oral pain relievers, which they could otherwise take at home, could be managed from the results waiting room. For a patient having continued symptoms, the treatment nurse would contact the intake physician for additional medication orders. When all of the diagnostic testing was complete and resulted and the nurse felt the patient was clinically ready for discharge, the treatment nurse would contact the physician to discharge the patient. The ED redesign team was composed of EP and nursing leadership along with three staff physicians, one staff PA, two staff medical technicians, and six staff nurses who worked to tailor the split ESI 3 flow model to our department. A rapid-cycle testing model for process improvement was employed with 10 pilot days that were performed prior to going live on May 31, 2011.18 The test pilots solidified the timing, staffing, responsibilities, patient flow, and education of the splitting of the ESI 3. To differentiate between low- and high-variability ESI 3 patients, the nursing staff composed a list of qualifying chief complaints. Where there were disagreements, the nursing design team reviewed selected triage encounters in a blinded fashion, each rating the patient encounter as intake or main. The cases where there was a definite majority opinion were incorporated into the final criteria, and the staff was subsequently educated on these criteria (Table 1). An electronic medical record (EMR) was used to extract patient demographics, clinical data, arrival and discharge times, and the final disposition. The EMR recorded a time stamp for each significant patient milestone. The main outcome variable, LOS for discharged patients, was calculated as the difference in time between when the patient was entered (arrival time) and removed (discharge time) from the EMR. Patients were entered into the EMR by the quick registration clerk. All patients were manually removed from the EMR by one of the following individuals: nurse, physician, PA, registration clerk, or scribe. Visit descriptors included age, race, sex, ESI score, and first diagnosis. The first diagnosis was coded based on methods used by the Agency for Healthcare Research and Quality to codify International Classification of Diseases, ninth version, into disease groups.19 Descriptive statistics summarized the distributions of both the outcome of interest LOS for each visit and the visit descriptors. Independent-sample t-tests and chi-square tests were used, as appropriate, to compare distributions of the visit descriptors across years. Since LOS was positively skewed, a log-transformation was taken to stabilize the variance before more formal inferential statistics were conducted. Additional histograms confirmed that this resulted in bell-shaped distributions for the LOS, with homogeneous variances between cases and controls. Linear models compared log-transformed LOS for cases and controls, both unadjusted as well as adjusted for age, race, and sex. These analyses included a random effect for patient to account for correlations between visits by the same patient. All analyses were conducted using SAS for Windows, version 9.3 (SAS Institute Inc., Cary, NC). Results are expressed as geometric means and percent changes in these means, calculated as the exponential of the least-square means of the adjusted means of the log-transformed values or the regression coefficients summarizing the difference between adjusted means, respectively. Use of the log-transformed values and geometrics means to summarize results ensures that the overall results are not strongly influenced by the extreme values in the tail of the positively skewed distribution. Additional exploratory analyses evaluated whether ESI score or diagnosis code (where analyses were restricted to the 12 most common diagnoses) modified the difference between cases and controls. In these analyses, indicator variables were used to represent the categories of ESI or diagnosis code. Interactions between case/control status and the modifiers were used to assess effect modification. With two modifying variables of interest, it was noted whether they were significant while maintaining a familywise error rate of 0.05 using the Bonferroni correction. We analyzed 20,653 cases from 2011 and 20,215 controls from 2010 for the study period. ESI level differed significantly across years. No other visit descriptors differed significantly across years (Table 3). Linear models found significant differences between cases and controls, both unadjusted and adjusted for sex, age, and race (p < 0.0001). Using the unadjusted regression coefficients, we estimate that there was a 5.9% decrease (calculated as [exp(beta for year) – 1) × 100%]) in the geometric mean of LOS from controls relative to cases (95% confidence interval [CI] = 4.5% to 7.2%). The adjusted estimates were similar. We found that ESI significantly modified the difference in LOS between cases and controls (p = 0.0086). That is, the difference varied by levels of ESI. Thus, these results were further stratified by ESI and first diagnosis code (12 most frequent; Table 4). For visits with ESI 1 (most critical), the difference between cases and controls was not significant (p = 0.40). For ESI 2, the difference was marginally significant (p = 0.033); however, the lower endpoint of the 95% CI was very close to zero. For ESI 3 or 4, LOS for cases were significantly lower than those for controls (p = 0.0011 and 0.0060). In particular, the cases had LOS that were estimated to be 3.0 and 2.9% smaller than those for controls, respectively. The difference in LOS for patients with ESI values of 5 was similar to those with values of 3 or 4, but not significantly. Diagnosis was also found to significantly modify the difference between cases and controls (p < 0.0001; Table 4). Among the top 12 most common diagnoses, the largest effects were seen for the following visits: headache/migraine (%diff = –17.8%; p < 0.0001), abdominal pain (%diff = -12.9%, p < 0.0001), skin infection (%diff = –9.0%, p = 0.022), sprains (%diff = –10.8%; p < 0.0001), and superficial injury (%diff = -8.7%, p = 0.0004). In other words, for visits in which the diagnosis was headache/migraine, the LOS for cases was estimated to be 17.8% less than for controls. Differences between cases and controls were not significant for the remaining seven of the 12 most common diagnoses (back problem; chest pain; open wound, extremity; other, connective tissue; other, joint disease; other, lower respiratory; and other injury). Figure 2 shows these estimated differences with their 95% CIs. The diagnoses for which significant effects were found accounted for 30.7 and 30.3% of all controls and cases, respectively, while the diagnoses for which nonsignificant effects were found accounted for 25.3 and 25.4% of controls and cases. Note that both interactions were significant at the 0.025 significance level, the level that would be required to maintain a familywise error rate of 0.05 using the Bonferroni correction for the effect modifications. We found an 9% to 18% (9 to 34 minutes) reduction in LOS for five of the 12 most common complaints (the five account for about one-third of all ED visits) and a 5.9% reduction in LOS for all patients. It is important to note that this improvement occurred in the setting of a significant increase in volume of 3.94% (from 40,564 to 42,162) during the study period. The central premise of splitting ESI 3 visits into high and low variability is to take advantage of the segmented flow process of a traditional fast track area. Segmenting patient flow processes is beneficial when there are significant differences in patient characteristics, number or types of activities in the process, or provider demands that can potentially result in enhanced throughput in one or both of the segmented streams. The low-variability ESI 3 patients all followed a standardized work flow: assessment, diagnostics, medications, and reassessment to disposition. The higher-variability ESI 3 patients had more complex initial presentations and required multiple reassessments throughout their stays by various staff members. The intake area provided a more efficient, fast track–like setting for a stream of patients who are not traditionally considered appropriate for this type of process. Our results show that there are other presenting patient streams that can benefit from an enhanced process design that is similar to a fast track environment. These patient streams have predictable, low-variation evaluation and treatment pathways with low physician and nursing demands. These patients are relatively well and unlikely to require admission. Staff and resources can be better aligned to the patient arrivals during specific time periods based on expected service times for the nurse, physician, and the treatment space (recliner chair). The results waiting room allowed us to minimize the occupancy of treatment spaces by patients who are clinically well and reduced the effect of laboratory and radiology turnaround times and consultant response time on the capacity of ED treatment beds. The discharge area permitted the registration to occur in a nonclinical space, improving bed capacity in a similar manner. This conservation of bed capacity likely explains the improved LOS of low-variability level 3 and level 4 patients. Level 5 patients also had an improvement, although the number was too small to be statistically significant. This study adds evidence to the effectiveness of "vertical patient flow" models described in a recent survey of ED medical directors by Liu et al.,20 which found that 29% had adopted these in response to the American College of Emergency Physicians Task Force on Boarding. By placing lower-variability ESI 3 patients in intake, we decreased the total number of patients going to the urgent area of the main ED. Decompressing the urgent area of the main ED, where all level 3 patients were previously evaluated, predictably resulted in improved LOS of high-variability level 3 patients treated in this area in the new process. This result can likely be attributed to the reduced staff workload and decreased the use of treatment spaces in this area in the improved process. Improved flow can affect not only the quantity but also the quality of care. For example, Sun et al.21 recently found increased LOS, cost, and mortality in a claims-based review of patients admitted through California EDs during times of high crowding. Delays in antibiotic or pain medication administration and increased errors associated with increased crowding have also been previously reported.21-24 While this study did not focus on these outcomes, its presentation of a possible means to adapt to overcrowding can be beneficial.25 There was not a significant change in ESI 1 or 2 discharged LOS. While this result is complicated by the presence of many confounding variables, we believe that this is due to the very high demand of admitted patients, resulting in a nursing bottleneck that was not affected by the process changes. ESI 1 or 2 patients were typically treated in the emergent area of the ED, which has the highest number of admissions, the sickest patients, and little utility for vertical patient flow. Thus the demand for treatment spaces in this area did not change as it did in the urgent area of the main ED, and this may represent a location where crowding as described by Pines25 is more a result of hospital crowding than ED processes. The LOS of ESI 5 patients was likely improved, but the small sample size prevented us from demonstrating a statistically significant improvement. Also, our historic ESI 5 LOS was already quite short, about 1 hour, so to see a statistically significant improvement may have required drastic change. In fact, our team anticipated a potential increase in LOS for ESI 5 patients as these patients were segmented into our fast track in our historical process, and they would now "compete" with the low-variability ESI 3 patients for resources in intake. If our ESI 5 LOS had increased significantly, we would then have considered treating these patients in an optimized environment such as a "super track."26 Flow for some diagnoses improved more compared to others. Throughput times for headache/migraine, abdominal pain, and sprains were statistically improved compared to those of open wounds, chest pain, and back pain. Intake facilitated a standardization of treatment and augmented workflow for some conditions that was not available before. For instance, abdominal pain and sprain evaluations typically involve treatment and/or imaging. In intake, the whole team knew the general approach even if not the physician's specific preferences. For example, during an abdominal pain or kidney stone evaluation, the physician would see the patient, the intake nurse would place the IV, and the technician would transport the patient to imaging on the way to the treatment area for IV fluids or medications that the physician by then would have ordered. The patient moved to the various processes that, in essence, were ready and waiting for him or her, rather than the various processes finding the patient and completing serial assessments prior to task completion. We saw improvements for order-to-in-lab times for phlebotomy, and improvement in order-to-radiology completion for plain films, as the staff were ready and assigned to those particular tasks. The intake nurse only performed initial assessments, gave medications, and gathered diagnostics. The same nurse, when working in the main ED, would be required to perform the entire nursing task on the patient, e.g., one patient may require medications while the other waits for phlebotomy. The approach to these patients also changed in terms of streamlined treatment. Traditionally, most patients with abdominal pain received IV fluids. However, many of these patients can take fluids by mouth so they do not need IV access, another opportunity for improved turnaround time. LOS for diagnoses like open wounds, chest pain, and back pain improved less, perhaps because they required more lengthy treatment procedures such as suturing that could not be overcome by improved transport flow to imaging. Also, while their evaluations may have been relatively standard, treatment modalities used may have had much greater variability so that staff could less successfully anticipate providers' plans. Further study is needed to assess the relationship between a resource use–based triage system (ESI) and diagnoses. While controversial, clinicians may consider generating consensus for evaluation and treatment of the most common diagnoses to reduce variability. We had 10 pilot days over a 3-month period to adjust and hardwire our process, but the group felt that we may have still had a learning curve for both nurses and providers that may have blunted the early results. There was an initial resistance to placing ESI 3 patients into a physical area that had formerly been a fast track. Despite the rollout of the new paradigm, there was discussion of specific policy stating the ESI 3 patients could go to this area, and many patients who should have been ESI 3 were given acuities of ESI 4 until the policy was changed (Table 3). This may have resulted in a significant difference in the ESI distributions between the study year and the control year. Prior to, and in conjunction with, this implementation the nurse and physician staffing were aligned and an improved inventory and regional supply management plan was created. We believe that this may have improved the process for patients going to the main ED, but do not believe this had a meaningful effect on those patients managed in intake. Given the retrospective nature of this study, we acknowledge there may be other unmeasured changes that influenced the results. However, we do not believe that there were any other significant process or policy changes during the study period. Staff were not blinded during the study; however, this effect may be minimal due to the retrospective nature of data collection and the fact that there we were not anticipating publishing our work. Finally, the discharge times were generated when each patient was removed from the EMR, and that was done by the staff soon after the patient left the ED. This could have artificially increased the patient LOS. This part of the process was not changed between the 2 years, and as such we do not believe that it affected the results. The use of a split Emergency Severity Index level 3 triage flow system reduced length of stay for five of the 12 most common ED complaints by 9% to 18% without increasing length of stay for the most or least ill, despite a 3.9% increase in overall ED patient volume. Separating high- and low-variability ESI 3 visits improved throughput and reduced length of stay in our ED. This technique is recommended for EDs experiencing excessive length of stay for middle- and low-acuity patients or those EDs that have bed capacity constraints and have sufficient volumes to warrant segmentation.