ASystematicReviewofSimulationStudies Investigating Emergency Department Overcrowding

ASystematicReviewofSimulationStudies Investigating Emergency Department Overcrowding Sharoda A. Paul Madhu C. Reddy College of Information Sciences and Technology, The Pennsylvania State University, University Park, PA 16802-6823, USA {spaul,mreddy}@ist.psu.edu Christopher J. DeFlitch Department of Emergency Medicine, Penn State Hershey Medical Center, 500 University Drive, Hershey, PA 17033-0850, USA cdeflitch@hmc.psu.edu The problem of emergency department (ED) overcrowding has reached crisis proportions in the last decade. In 2005, the National Academy of Engineering and the Institute of Medicine reported on the important role of simulation as a systems analysis tool that can have an impact on care processes at the care-team, organizational, and environmental levels. Simulation has been widely used to understand causes of ED overcrowding and to test interventions to alleviate its effects. In this paper, we present a systematic review of ED simulation literature from 1970 to 2006 from healthcare, systems engineering, operations research and computer science publication venues. The goals of this review are to highlight the contributions of these simulation studies to our understanding of ED overcrowding and to discuss how simulation can be better used as a tool to address this problem. We found that simulation studies provide important insights into ED overcrowding but they also had major limitations that must be addressed. Keywords: emergency department simulations, literature review, emergency department, overcrowding, simulation 1. Introduction The major role of the emergency department (ED) is to provide care for acutely ill and injured patients 24 hours SIMULATION, Vol. 86, Issue 8-9, August-September 2010 559 571 c 2010 The Society for Modeling and Simulation International DOI: 10.1177/0037549710360912 Figure 1appearsincoloronline:http://sim.sagepub.com aday,7daysaweek.theusemergencymedicaltreatment and Active Labor Act (EMTALA) requires that all ED patients must be provided with medical screening and stabilization of their conditions, irrespective of their ability to pay [1]. As a result, EDs care not only for acutely ill patients but also for under-served populations who have no other options for medical care because of socioeconomic barriers [2]. Thus, the ED is the safety net of the healthcare system due to its role in providing care to uninsured, indigent and otherwise vulnerable patients [3, 4]. Volume 86, Numbers 8-9 SIMULATION 559

Paul, Reddy, and DeFlitch One of the main problems facing EDs is overcrowding this problem has now reached crisis proportions [2, 5 8]. Between 1993 and 2003, ED visits in the United States increased by 23.6 million, while at the same time 425 EDs closed and total hospital beds declined by 198,000 [9]. ED overcrowding manifests itself in many different ways: an excessive number of patients in the ED, patients being treated in hallways [8], ambulance diversions [10], long patient wait times, and patients leaving without treatment. ED overcrowding leads to increased medical errors [2], poor patient outcomes [11, 12], high levels of stress, decreased morale among ED staff, and decreased capacity of EDs to respond to mass casualty incidents [9]. Other effects of ED overcrowding are patient dissatisfaction, decreased physician productivity, violence, negative effects on teaching missions in academic EDs, and miscommunication [13]. Research on addressing the problem of ED overcrowding has primarily fallen into three categories: descriptive, predictive and intervention-oriented. Descriptive studies have focused on defining overcrowding [14], examining the causes and effects of overcrowding [2, 7, 8, 10] and proposing models [15] to describe the problem and measures to quantify it [16 23]. Predictive studies have focused on measures [20] to predict when an ED will become overcrowded and development of early warning systems for impending overcrowding episodes [24, 25]. Such predictive models assume that extra resources, such as reserve personnel and auxiliary treatment bays, will be deployed once the ED is alerted to an impending overcrowding episode. However, given the limited availability of resources caused by cuts in hospital funding [13], a third stream of research, intervention-oriented studies, has focused on interventions to optimize available resources and processes. These interventions include monitoring code red hours and patient length of stay (LOS), educating physicians regarding non-ed options for patients [26], and re-designing processes [27] and patient flows [28, 29]. In 2000 and 2001, the Institute of Medicine (IOM) published two reports, To Err is Human [30] and Crossing the Quality Chasm [31], which highlighted the deficiencies in current care processes and urged stakeholders in the American healthcare system to take steps to improve quality and efficiency of care. In response, the National Academy of Engineering (NAE) collaborated with the IOM to report on the important role of systems engineering tools in improving and optimizing care processes [32]. This report stressed that there is a knowledge/awareness divide separating healthcare professionals from their potential partners in the engineering fields and that bridging this gap is key to increasing the quality and productivity of healthcare. It emphasized the utility of simulation as a systems analysis tool that can have a positive impact on care processes at the care-team, organizational, and environmental levels. In the ED, simulation has been has been extensively applied to test what if scenarios to combat overcrowding [33 35]. These simulation studies have proposed several solutions to alleviate overcrowding. In this paper, we present a systematic review of the ED simulation literature from 1970 to 2006 from the fields of healthcare, systems engineering, operations research, and computer science. The goals of this review are to highlight the contributions that simulation studies make to our understanding of the problem of ED overcrowding, and discuss how simulation can be better used as a tool to address this problem. This paper is organized in the following manner. The next section provides background on the role of simulation in understanding ED overcrowding. Then, the Section 3 describes the process we used for selecting studies. Next, we report the findings of the simulation studies in Section 4. In Section 5, we present insights gained from these studies, the limitations of current studies, and directions for future research. We conclude with some final thoughts of how simulation can be used as a tool to address ED overcrowding. 2. The Role of Simulation in ED Overcrowding Several solutions have been proposed to alleviate the effects of ED overcrowding including providing patients with better access to clinics, expanding ED square footage and beds, improving support by radiology, laboratory, and consultant services, and reducing incoming transfers to the ED during busy periods [13]. Fatovich and Hirsch [10] proposed stop-gap measures such as increasing ED capacity, increasing human and physical resources, and improving discharge processes to deal with overcrowding. Most of these proposed solutions call for increasing capacity and resources. However, because of economic constraints, most hospitals do not have extra resources to deploy. Therefore, there is a need to focus on optimizing existing resources and processes. Systems analysis tools can play an important role in this process. Systems analysis tools are used by engineers to understand how complex systems operate, how well these systems meet operational goals, and how they can be improved [32]. Such tools can be used for enterprise management, financial engineering and risk analysis, and knowledge discovery. Simulation, an important systems analysis tool, provides great flexibility in testing scenarios, hypotheses, policies, and re-engineering ideas in healthcare settings. It can be used as research tool, education device, decisionmaking tool and planning model [36]. Pritsker [37] defined simulation as the development of a mathematical/logical model of a system and the experimental manipulation of the model on a computer. To study a system, a model, which is a set of mathematical or logical assumptions about the system, is created. The model is then either solved via mathematical methods (i.e. an analytic solution) or evaluated numerically using a computer (i.e. a simulation) [38]. Simulation has been used in an array of healthcare settings ranging from hospital sub-systems and outpatient 560 SIMULATION Volume 86, Numbers 8-9

A SYSTEMATIC REVIEW OF SIMULATION STUDIES INVESTIGATING EMERGENCY DEPARTMENT OVERCROWDING Figure 1. Identified simulation studies by year departments [39] to national healthcare systems [40]. In the late 1970s, England and Roberts [41] reported 21 areas of application of simulation in healthcare including hospitals, ambulatory care, manpower planning and forecasting, community, regional and national health systems, and education. In hospitals, simulation research has been applied to areas such as admission control systems, bed planning, ambulance and emergency services, labs and radiology, and surgery. Simulation studies have also focused on understanding the care planning process [42], tracing the spread of diseases and epidemics, and virtual reality simulations for training [43]. The earliest efforts in simulating emergency services date back to the mid-1960s [44]. Starting with Bolling s study [45] in 1972, simulation studies of EDs have provided valuable insight into factors and reasons for overcrowding [46]. 3. Methods We used a two-phase approach to identify simulation studies relevant to ED overcrowding. In the first phase, we searched the databases of Proquest, PubMed, ACM, IEEE, and the Systems Dynamics Conference from 1970 to 2006. These databases are the comprehensive sources of literature in computer science, operations management, healthcare, and engineering fields. We used the search phrases emergency department simulation, emergency department patient flows and other combinations of these phrases (e.g. emergency department flow simulation ). We defined relevant documents as those studies that used simulation to understand the problem of overcrowding, effects of overcrowding (e.g. long patient wait times), and/or proposed solutions to the overcrowding problem. We did not include studies that had merely modeled the ED but did not perform computer simulations of the model. Using the various search phrases, we found 37 relevant documents. In the second phase, we examined the references of these 37 documents for additional studies that met our criteria and found 6 additional studies. After this phase, we had a total of 43 simulation studies. Most of these studies were conducted after 1990 (Figure 1 shows the distribution of studies by year). Of the 43 simulation studies, 24 (56%) studies were published in computer science venues, 9 (21%) in medical and health sciences venues, 5 (12%) in operations research and management venues, 4 (9%) in industrial engineering venues, and 1 (2%) in other venues. Of these studies, we were able to access 32 via online and print sources. Some of the early studies were not available online and we did not have access to the print versions. 4. Results We analyzed the simulation studies with respect to: (1) their motivation and goals (2) the modeling techniques used (3) the data sources and collection methods (4) patient classification and patient flows and (5) study findings. 4.1 Motivation and Goals Few studies mentioned explicitly that their motivation was the desire to create a general model of overcrowding in EDs [47], or to decrease levels of overcrowding [48]. Instead, the motivations for most studies were related to costs and competition, efficiency, re-engineering, and quality of service (Table 1). Given the motivations to cut costs and increase efficiency, one of the primary goals of these studies were to examine causes of inefficiencies in processes and resource utilization. Therefore, studies examined patient flows [49] and bottlenecks in flows [34, 50], causes of excessive wait times [49, 51], and patient throughput [35]. A few studies also evaluated the effects of introducing fast care for low acuity patients [33, 52], Volume 86, Numbers 8-9 SIMULATION 561

Paul, Reddy, and DeFlitch Table 1. Motivation of the simulation studies Costs and competition Rising costs [33, 64] Motivations Decrease or control costs of operation [65 67] High costs of building, equipment, and staffing [62] Reduce staffing levels without decreasing efficiency [68] Rising competition [59, 65] Reduced patient visits [69] Increase corporate customer base without decreasing quality [54] Efficiency Inefficiencies [33] Increase efficiency [67] More efficient staff utilization [59] Overloaded ED staff [70] Staff scheduling to meet unpredictable workload patterns [60] Doctors had the highest average utilization, were not appropriately assigned and were the bottlenecks of the system [55] Develop a general tool for evaluating policy changes for improving productivity and efficiency in the ED [53] Re-engineering Increase in ED size and separation of ambulance patients from outpatients [49] Improve the ER process [71] New ED with lab and X-ray facilities [72]. Quality of service Excessive patient wait times [34, 50, 51, 56, 63, 66, 70, 73, 74] Long patient wait times for low acuity patients [52, 73] High LOS of patients [35] Lack of ED capacity [70, 74] High withdrawal rates of patients [70] ED on ambulance diversion status often [35] Increased patient dissatisfaction [69] not providing care to low acuity patients [53], and redesigning processes to reduce patient LOS [54]. A second major goal was to optimize resources such as staff and beds. Several studies examined alternative staff schedules [55 57], assessed the effect of different staff schedules on wait times [58, 59], and quality of service [60] in order to recommend cost-effective schedules. Since beds are an important resource in the ED, studies examined critical bed requirements [61, 62] and the impact of bed availability on wait times of admitted patients [50, 63]. 4.2 Modeling Techniques The studies utilized a variety of modeling techniques including discrete-event [47 49, 51, 55, 58, 60, 67, 73, 75], systems dynamics [63], and conceptual [64, 71] and mathematical modeling [68]. Asimulationmodelcanbedeterministic (if it does not contain any random variables) or stochastic (if it contains one or more random variables) [38]. Stochastic processes are governed by probabilistic laws and have been applied to study various aspects of health systems since the early 1950s [76]. We found that most often the ED was modeled as a stochastic system since the inter-arrival times and service times of patients are considered random variables. The ED was also primarily modeled as a discrete system. Discrete event simulation has been used extensively in examining patient flows and allocation of resources in healthcare clinics [77] and was the most popular simulation technique in the studies we reviewed. Queuing models are discrete-event models used to represent customers queuing to gain access to limited resources and have been used to simulate unscheduled patient arrivals in EDs, operating rooms (ORs), intensive care units (ICUs), blood clinics, and X-ray departments. For instance, Liyanage and Gale [74] used queuing theory to develop a simulation to find the optimal number of resources that would minimize the mean operating cost of the ED. Some discrete-event models [53, 57, 70] were written in SLAM (Simulation Language for Alternative Modeling), a process-oriented simulation language developed by Pegden and Pritsker (see [38]). Using SLAM, the ED can be pictorially represented as a network of nodes and branches through which patients flow. Other discrete-event modeling tools include Arena [34, 35, 49, 51, 54, 56, 59, 65, 68], Extend [48, 64], SIMAN [60, 73], and SIMUL-8 [58]. Another modeling technique used was systems dynamics [78]. This technique is used to model the complexity in large systems and has been successfully applied to business modeling [79]. The systems dynamics tool ithink was used for such models [63]. While discrete-event simulations can be used to create detailed models of subsystems within healthcare, systems dynamics enables a systemic view of the interactions of patient flows and information. Finally, conceptual modeling was used in some studies to create process maps and documentation. 4.3 Data Sources and Collection Methods The simulation studies used a wide variety of data sources as inputs to their models. The data collection techniques included interviews with care providers and management, observations, historical data from ED databases, patient charts, time and motion studies, and using automated datetime stamping machines (Table 2). Hospital databases and information systems, patient charts and medical records all play an important role in obtaining data on arrival patterns, time spent on different activities by care providers, and inter-arrival times and LOS distributions. Data obtained about inter-arrival times and service times [34, 35, 52], volume and mix of patients 562 SIMULATION Volume 86, Numbers 8-9

A SYSTEMATIC REVIEW OF SIMULATION STUDIES INVESTIGATING EMERGENCY DEPARTMENT OVERCROWDING Table 2. Data collection sources for modeling the ED Data source Hospital databases information and systems Medical records Data obtained Historical patient data Arrival patterns of patients and number in each priority category [58] Time spent for each activity [58] Distribution of patients arrival times by time of day and day of the week [59] Determine inter-arrival time and LOS distributions [61] Arrival rates and service times Observations Shift patterns [73] Patient charts Arrival times, mode of arrival, tests performed, discharge time [69] Surveys Identify patient flows and common CCU configurations [61] ED logs Patient volume and mix data [75] Interviews ED staff activity data [75] Time studies Registration time of patients [60] Paid ED staff Bed management Times of entry, service, exit etc. Data on admissions [50], staffing levels [60, 74], and types and duration of treatment [54] were used to determine model inputs and outputs. 4.4 Patient Classification and Flows There was no single approach to classifying patients. Different studies categorized patients along different dimensions. The three main dimensions of categorization were mode of arrival, level of acuity, and case type. The mode of arrival of patients was helicopter, ambulance, or walkin. In the emergency care domain, the most common way to categorize patients according to acuity is emergency, urgency, and non-urgency [72]. However, different EDs used different terms to track levels of acuity including degrees (e.g. first, second), levels [60, 80] (e.g. I V), trauma levels (e.g. major, minor), and ESI-5 levels [33]. Similarly, there was a variety of ways to categorize patients by case type, including by chief complaint [69] (e.g. abdominal pain, laceration), specialty (e.g. internal, surgical), and even treatment areas [50] (e.g. fast-track patients, observation room patients). Some studies categorized patients by combining these dimensions, such as Clark and Waring [57] who combined mode of arrival with level of acuity (e.g. critical walk-in), and Sinreich and Marmor [67] who combined mode of arrival with case type (e.g. walk-in surgical). In most studies, different patient flows were modeled for each category of patients. The wide variety of patient categories lead to a variety of patient flows across studies. Patient flows were based on patient entry mode [52, 75], patient acuity [60, 73], number and types of resources needed [72], and the need for auxiliary services such as labs and X-rays [55]. Some flows were based on a combination of these factors. 4.5 Study Findings The scenarios tested by the simulation studies can be broadly categorized as resource-related, process-related, and environment-related (Table 3). Resources in the ED were human, equipment, and space resources. Resource-related scenarios focused on changing levels of resources, allocation, and reallocation of resources. Process-related scenarios focused on changing processes in the ED, including how processes were performed, as well as when certain processes were done. Environment-related scenarios focused on variables external to EDs such as demand patterns and characteristics of hospital units which interface with the ED. 4.5.1 Resource-Related Findings 4.5.1.1 Space Space in the ED, defined in terms of beds or rooms, is an important resource. During periods of overcrowding, patients experience their most significant delay waiting for an ED bed [81]. Takakuwa and Shiozaki [49] found that 59% of the waiting time in the ED was for beds. Komashie and Mousavi [51] tested two different scenarios regarding beds in the minor treatment area or minors of their ED. In scenario 1, they added an extra bed to the minors area, but, in scenario 2, they added six extra beds to the area. Surprisingly, they found very little improvement in LOS in scenario 2 as compared with scenario 1. They did find that queuing time for beds went down 83% in scenario 2 however, there was a significant increase in wait times for nurses and doctors. Their study highlighted that adding extra beds merely shifted the queues from the waiting room to the bed. Samaha et al. [35] also found that adding beds or square footage to the ED did not shorten LOS. These results are corroborated by recent findings that increase in ED bed capacity does not decrease ambulance diversions (an indicator of overcrowding) and might even lead to an increase in LOS [82]. Studies also examined the effects of re-using space or rooms in the absence of adequate beds. Kirtland et al. [69] found that placing patients in the treatment area when beds were not available instead of sending them back to waiting area saved 14.1 minutes. McGuire [66] found that having aseparateholdingareaforadmittedpatientswaitingfor test results saved 22 minutes per patient on average, and Kirtland et al. [69] found that using an internal waiting room for patients awaiting lab and X-ray results would be useful when the ED is very busy. Volume 86, Numbers 8-9 SIMULATION 563

Paul, Reddy, and DeFlitch Table 3. Categorization of scenarios tested Resource-related Process-related Space Varying the number of beds or rooms available [49, 51, 62, 63, 70] Having a single holding area instead of one room per patient [71] Using an internal waiting room for patients waiting for lab results [66, 69] Human resources Alternative staff scheduling [55 60, 68, 71, 75] Varying the number of ED staff available [34, 35, 69, 70, 80] Varying resident availability [35, 73] Varying the number of non-ed staff [49] Adding a dedicated triage nurse [34] Addition of a registration clerk during peak hours [66] Estimating the optimal number of servers that will minimize the mean operating cost of the system [74] Equipment Varying the number of implements in the drip room and stretchers [49] Installing lab and X-ray facilities in ED [72] Procedural Addition of fast-track [35, 52, 69, 71] Take patient back to open treatment rooms instead of keeping in waiting rooms [69] Change triage protocols so triage nurse can order certain tests [69, 71] Allowing nurse to order testing/treatment without participation of physician [71] Changing criteria used for sending patients to fast-track areas [66] Priority given to only Category 1 patients, rest treated on a first-come-first-served basis [80] Not serving Category 5 patients [80] Triaging patients into different categories [72] Scheduling non-emergency patients so as to smooth demand [72] Temporal Reducing lab turn-around time [64, 66, 69] Initiate search for admission room earlier [69] Discharging inpatients earlier [64] Extend hours of fast-track and pediatric clinic [66] Environment-related Varying patient demand [54, 63, 70, 72] Varying percentage of true emergency patients [72] Adding inpatient beds [64] Varying number of beds in different locations or units of hospital [34, 50] Reduce time for bed notifications from Medical Telemetry Unit (MTU) to ED, decrease number of patients being admitted to MTU [34] Varying queue discipline [73] 4.5.1.2 Human and Equipment Resources Simulation studies focused on two major resources other than beds: human and equipment resources. One cause of inefficiency in EDs is that due to the sporadic demand, the staff are idle at times and overworked at other times [72]. Hence, several simulation studies were interested in examining the effects of alternative staff schedules on waiting times and LOS [55 60, 68, 71, 75]. Rossetti et al. [59] simulated 18 attending staff schedules and identified a schedule that decreased average patient time in the ED by 14.5 minutes/patient. Based on the simulation, they found that this schedule also decreased physician utilization and percentage of long visits. Coats and Michalis [58] compared different shift patterns via simulation and found that the doctor shift pattern that best matched the patient arrival pattern gave the shortest wait times. Queuing analysis techniques were used to match staffing patterns to ED demand [83]. Tan et al. [55] used preliminary queuing analysis to develop an alternative doctor schedule and compared it with the current schedule via simulation. The results identified the doctor s station as the bottleneck. The new schedule increased the capacity of the bottleneck and hence reduced patient time in the system. Clark and Waring [57] tested whether doctor and nurse scheduling would affect the time spent waiting 564 SIMULATION Volume 86, Numbers 8-9

A SYSTEMATIC REVIEW OF SIMULATION STUDIES INVESTIGATING EMERGENCY DEPARTMENT OVERCROWDING to see a doctor, the total time in the system, the utilization rate of doctors, and the utilization rate of nurses. They found that scheduling of physicians will have a significant effect on waiting times. However, the nurses schedule did not have the same impact. Evans et al. [56] tested alternative schedules containing different numbers of nurses and technicians but the same number of doctors as the actual system and found only a 5-minute reduction in average LOS. These finding, which seem to indicate that nurse scheduling does not significantly impact waiting times, are interesting given the large amount of research focusing on optimizing nursing allocation in various parts of the hospital [76]. Although equipment is an important resource, only a few studies examined the effects of adding equipment to the ED. Hannan et al. [72] found that installation of lab and X-rays in the ED had the same effect as hiring an additional nurse and physician. Takakuwa and Shiozaki [49] looked at the effects of varying the number of stretchers and implements in the drip room. Some studies aimed at determining the optimal resources required to minimize the mean operating cost of the system [74] and successfully support the patient demand [70]. Takakuwa and Shiozaki [49] adjusted the number of rooms, internists, surgeons, pediatricians, implements in the drip room, stretchers, etc., and found that there was no optimal configuration of these resources which can lead to the lowest waiting time. 4.5.2 Process-Related Findings The simulation studies examined several procedural changes to alleviate the effects of ED overcrowding. Most studies found that establishing a fast-track path for low acuity patients was effective in decreasing wait times without negatively impacting quality of care [84]. Samaha et al. [35] simulated all routine patients being directed to fast-track and found a considerable reduction in LOS. Similarly, Kirtland et al. [69] found that fast-track saved 15.5 minutes in the ED. Pallin and Kittell [71] simulated fast-tracking by eliminating return visits and found a 50% reduction in staff and resources due to the fast-tracking. Garcia et al. [52] found that taking one nurse and bed from the ED and using them in a fast-track would significantly lower flow time for low acuity patients. McGuire [66] also found that extending the hours of the fast-track led to a 16- minute decrease in LOS. However, they found that waiting time was not affected by initiating search for admission rooms earlier in the process. Samaha et al. [35] found that bedside registration would not reduce LOS and would be costly to implement but, a recent empirical study found that including bedside registration in the process did decrease LOS [85]. Some procedural scenarios looked at changing standard protocols in the ED to reduce wait-times. Studies found that changing triage protocols so triage nurses could order tests and X-rays saved 3.6 minutes [69]. Pallin and Kittell [71] tested a protocol to allow nurses to order testing/treatment without participation of physician but did not mention the results of simulating this scenario. Laboratory turnaround times add to waiting times in the ED and can be decreased significantly with point-of-care lab testing [86]. McGuire [66] found that reducing lab turnaround time decreased LOS. Kirtland et al. [69] found that using I-stat machines in the ED for point-of-care testing saved 8.4 minutes. 4.5.3 Environment-Related Findings Many simulation studies also focused on the effects of factors external to EDs that can cause overcrowding. A major cause of overcrowding is the unavailability of inpatient beds and inpatient bed occupancy has been found to be strongly correlated with ED LOS [87]. Miller et al. [64] found that adding 30 inpatients beds would cut LOS by half. Gonzalez et al. [70] found that increasing the number of beds to which ED patients can be admitted would maximize profits and minimize waiting time. Lane et al. [63] examined how reductions in bed capacity in the hospital wards affected patient waiting times in the ED. They examined the ED as part of the larger hospitalwide system and considered emergency patients and elective treatment patients. They found that removing hospital beds did not increase ED waiting time, but did cause more cancellations of elective procedures. Elbeyli and Krishnan [50] found that adding beds to step-down units and other specialized units decreased the average time of patients waiting to be admitted from the ED. Blasak et al. [34] investigated how the interface between the ED and the Medical Telemetry Unit (MTU) affected wait times in the ED. They found that they needed to reduce the time for bed notifications from MTU to ED, decrease number of non-ed patients being admitted to MTU and decrease LOS in MTU [34] to reduce waiting times for ED patients. Two of the primary uncontrollable external features of the ED environment are patient demand and mix of patient types. Studies have examined the effects of changes in patient demand on wait times. Hannan et al. [72] tested the effects of increased demand and increased percentage of true emergency patients. They also examined scheduling non-emergency patients to reduce demand. They found that when demand increased above 20%, the waiting times for emergency patients did not change much, but the nonemergency patients had to wait longer. Baesler et al. [54] used their simulation to find the maximum demand that the ED would be able to handle without the average patient time exceeding 100 minutes. They found that this would happen when the demand increased by 130%. To handle this increase in demand and keep the wait time under 100 minutes, the ED would need four full-time doctors and one half-time doctor. Volume 86, Numbers 8-9 SIMULATION 565

Paul, Reddy, and DeFlitch 5. Discussion After analyzing the simulation studies, we found that they provided useful insights into the problems of ED overcrowding. However, at the same time, we found major limitations to the studies that must be addressed if they are to help us to better understand and alleviate ED overcrowding. 5.1 Insights from Simulation Studies The simulation studies highlighted a number of important issues that we must consider as we try to address ED overcrowding. First, the conventional solution to ED overcrowding is increasing available ED space (e.g. increase the number of ED beds). However, the simulation studies have clearly shown that while increasing ED space may provide shortterm relief by allowing more patients to be admitted to the ED, this would not necessarily reduce patient wait times. The patient queues would merely shift from the waiting rooms to the bedside. Therefore, although beds are an important resource, a more critical resource are the physicians because they are often the bottleneck of the system and the most utilized resource [55]. Most improvements in waiting times and LOS resulted from more effective scheduling of doctors [55, 57, 59], eliminating non-patient care duties from doctors duties [73] and using doctors in the fast-track. Blake and Carter [73] came to an interesting but counter-intuitive finding that while most processes with service problems are improved by adding manpower, the performance of the ED is negatively affected by the addition of residents since the time spent by attending physicians in resident education as opposed to direct patient care decreases the benefits of adding more manpower. Second, many studies found that the problems in the ED were process-related [35]. Improvements in processes, such as fast-tracking and reducing lab turn-around times, reduced wait times. Changing procedures and protocols in the ED such as placing patients in separate areas when they are waiting for test results, and having the triage nurse order tests and X-rays, also saved time. However, although afewprocessessuchasfast-trackingprovidedpositiveresults across studies, most process improvements were EDspecific. For instance, Kirtland et al. [69] found no significant time saving for using an internal waiting room for patients waiting for lab and X-ray results except when the ED was very busy but McGuire [66] found that having a separate holding area for admitted patients waiting for test results saved 22 minutes per patient. Similarly, Blasak et al. [34] found that adding a dedicated triage nurse would create a bottleneck but Gonzalez et al. [70] recommended the addition of a nurse to perform administrative work such as following up room availability. Therefore, while there is consensus that process-related issues are critical to an ED s ability to handle overcrowding, there is little consensus on the specific process changes that can be applied across all EDs. Rather, each ED must identify its own unique characteristics and processes when trying to deal with the problem of overcrowding. Third, the ED is part of a larger and more complex hospital system and is affected by many external factors. As the recent IOM report on ED overcrowding states [9], overcrowding is a system-wide issue that must be addressed across multiple hospital units and care settings. Hence, it is important to understand the relationships of the ED with other units of the hospital. Environmentrelated changes, such as variable patient demand, are outside the control of EDs. However, simulation of various demand patterns can help EDs predict resource levels needed to meet those demands. In the simulation studies, adding beds to units that interfaced with the ED invariably led to decrease in LOS for ED patients waiting to be admitted to those units. Reducing time for bed notifications from other units to the ED also improved wait times. These results indicate that there is a strong interconnectedness between the ED and the rest of the hospital. Lane et al. s study [63] revealed that the effects of changes in the ED may not be reflected within the ED but in other parts of the hospital. For instance, they found that reductions in hospital beds did not have an impact on ED wait times but resulted in cancellations of elective procedures. They concluded that policy changes must be based on an understanding of how EDs connected to pre-hospital services, to the rest of the hospital, and the surrounding community. 5.2 Limitations of Simulation Studies Although the simulation studies provided important insights into the problem of ED overcrowding, they had limitations that affected their usefulness in helping deal with the problem. First, patient flows were viewed in these studies in an overly simplistic manner. There are two aspects of patient flows in healthcare: clinical and operational [88]. From a clinical perspective, a patient flow is the progression of a patient s health status from an operational perspective, a patient flow is the movement of a patient through various locations or stations in a hospital. In the ED, the clinical and operational aspects of flows are often intertwined since the patient s health status (case type and level of acuity) typically determines which treatment stations they visit. This leads to a large variety of patient flows in the ED. For instance, Takakuwa and Shiozaki [49] found 70 patterns of patient flows for 9 patient categories in a single ED. The process of modeling this variety of patient flows is difficult and time-consuming. Therefore, for the sake of simplicity, most flows were modeled as linear, that is, patients moved in a sequential manner from station to station. However, in reality patients might undergo several care processes at the same time. The overlaps and interde- 566 SIMULATION Volume 86, Numbers 8-9

A SYSTEMATIC REVIEW OF SIMULATION STUDIES INVESTIGATING EMERGENCY DEPARTMENT OVERCROWDING pendencies between components of patient flows were not modeled in the simulation studies. Second, many studies did not incorporate information flows when modeling patient flows. The healthcare process can be viewed as a series of informationprocessing steps starting from the initial collection of data about the patient s condition to forming a hypothesis and testing it by collection of more data [32]. Given the variety of information required at each step of the patient flow, and the multitude of information and communication technologies used in modern EDs [32], information flows are closely linked to patient flows. These information flows are a crucial aspect of the modeling in an ED but have not been addressed by the simulation literature. Also, simulation studies have not considered the role of information and communication technologies (ICTs) within EDs. ICTs such as electronic medical record (EMR) systems, electronic dashboards, radio frequency identification (RFID), wireless registration, and mobile computing devices are being used in EDs to help with clinical documentation, decision-support, information management, and coordination of patient flows [9]. These ICTs have the potential to significantly impact ED overcrowding. By not incorporating ICTs into the simulations, the studies have failed to capture an important resource. Third, the lack of standardization of workflow, care practices, patient categories, and patient flows across EDs makes it hard to design a generic model of an ED for use in a simulation. Sinreich and Marmor [67] developed a generictoolflexibleenoughtomodelanyed.they classified EDs into four basic types based on two factors: ED physician type, i.e. whether the ED physician specialized in emergency medicine or other areas, and patients condition, i.e. how the ED processed acute and ambulatory patients. However, given the wide variety of processes, this classification may not be sufficient. As a result, most simulation studies had to create ED-specific models, which in turn lead to ED-specific solutions that could not be generalized to other EDs. Fourth, the purpose of data acquisition is to estimate the parameters of the system and to validate the model. The time, cost, and difficulty associated with obtaining empirical data for simulation models has been a challenge [36]. Although arrival patterns and patient volume/mix data can usually be obtained from information systems and medical charts, service times can only be obtained through observation and time/motion studies. However, placing researchers in overcrowded EDs is often difficult due to the fast-paced nature of the environment and patient privacy issues. One approach is to pay ED staff for data collection [72] or to use self-reported work sampling techniques to gather data [59]. However, the success of these techniques depends on the busy care-giver to collect the data. The difficulty of collecting this type of data was highlighted in the following study. Rossetti et al. [59] obtained a list of 16,250 standardized elements, i.e. operations that a patient goes through or which a member of the staff performs, from five hospitals it took 1,350 man-hours to conduct time and motion studies to measure service times for all elements. A simulation model is only as accurate as the data used to build it. Therefore, the difficulty in capturing reliable data can lead to inaccurate simulation results. 5.3 Future Research Directions We have three suggestions for future research directions for simulation studies. First, simulation models need to capture human behavior. In 1975, Valinsky [36] noted that there had been little work on modeling the human elements of the healthcare system, such as the patient, medical staff, and administrators within the health field. More than 30 years later, this is still the case. The simulation models we reviewed have not examined the physiological, psychological, and social aspects of patient care in EDs. Providers and patients behavior is directed by beliefs, attitudes, and expectations. Therefore, an important question to answer is how these beliefs and attitudes can be modeled or if this is an aspect of patient care that is irrelevant to simulation modeling. To incorporate elements of human behavior in healthcare simulations, simulation research can draw on the fields of human computer interaction (HCI) and computer-supported cooperative work (CSCW) which have studied the cognitive and social aspects of human behavior, such as emotion [89], communication [90], collaboration [91, 92], and the social organization of work [93] in healthcare settings. Simulation researchers can apply findings from these fields about the behavior of patients and healthcare providers to model human interactions as part of the ED. Second, we need to study the ED as part of a larger system. In simulation modeling, the choice of system depends on the objectives of the study [38]. In most studies, the objective was to improve efficiency and cut costs of operation and the ED was studied as an isolated unit. The interactions/interfaces between ED and other services such as EMS, labs, and the rest of the hospital were not modeled in most studies. However, factors external to the ED, such as hospital bed occupancy, strongly affect ED length of stay [87]. Research has also shown that interventions aimed at factors external to the ED have been most successful in reducing ED overcrowding [26]. Therefore, simulation studies need to focus more on the role of the ED with the context of the larger hospital system. This can be done by incorporating other hospital units that interface with the ED as part of the model. Simulation studies can draw on research from fields such as CSCW which have examined the effects of patient flows between the ED and other hospital units [94]. Incorporating external factors might lead to complex models and certain types of simulation are better suited for such models. Discrete-event simulation, the most popular technique in the reviewed studies, is not well suited Volume 86, Numbers 8-9 SIMULATION 567

Paul, Reddy, and DeFlitch to studying complex integrated systems in healthcare because of the high level of complexity and data requirements of such simulations, as well as the time and cost associated [77]. However, it provides excellent micro-level analysis of the ED. Systems dynamics techniques can be used to understand the inter-relations between the ED and the rest of the healthcare system but do not assist managers in micro-level analysis. Therefore, researchers need to examine ways to combine different techniques such as discrete-event and systems dynamics techniques to provide multi-level views of the problem [95]. Third, we need to focus on the individual level of care and incorporate the patient perspective. The IOM report [32] states that the ultimate purpose of systems tools should be to ensure that the system is responsive to patient needs. However, systems tools have not been widely applied at the individual level of care because the focus of these tools has been at the team and organization level. Therefore, systems tools may need to be combined with other individual level tools such as quality function deployment, to design processes that meet the level of service a patient/customer wants, and human factors engineering to improve the patient provider interactions. Furthermore, researchers have focused on the problem from the perspective of the healthcare manager instead of the patient. In studying ED overcrowding, little attention has been paid to how overcrowding affects quality of care and patient outcomes [96]. Therefore, measures of overcrowding have only been weakly associated with quality of care [18]. The simulation studies reflected the same bias by focusing mostly on improving efficiency, cutting costs, and optimizing processes and resources. Only a few studies were concerned with the direct impact on patient care or were motivated by reducing patient wait times, LOS, and dissatisfaction with care. Therefore, we need to examine how to incorporate patient care needs into the simulation models. 6. Conclusions ED overcrowding is a serious and growing problem threatening the safety net of the healthcare system. Simulation tools provide an important method to investigate overcrowding issues and explore solutions to this problem. Through a review of the last 30 years of simulation research focusing on ED overcrowding, we have identified a variety of features that these studies have in common. Although simulation has been useful in identifying critical resources and process improvements that can alleviate overcrowding, these studies still have severe limitations that must be addressed. Most interventions proposed by the simulation studies have been EDspecific and are not generalizable. Future simulation efforts must incorporate a patient perspective, the role of information and communication technologies, and environmental features in order to develop solutions to ED overcrowding. Simulation is a powerful tool that researchers can deploy to confront the problem of ED overcrowding. 7. References [1] Lee, T.M. 2004. An EMTALA primer: the impact of changes in the emergency medicine landscape on EMTALA compliance and enforcement.annalsofhealthlaw, 13:145 178. [2] Trzeciak, S. and E.P. Rivers. 2003. Emergency department overcrowding in the United States: an emerging threat to patient safety and public health. EmergencyMedicineJournal, 20:402 405. [3] Altman, S.H. 2000. Statement from the Chair. Committee on the Changing Market, Managed Care, and the Future Viability of Safety Net Providers. Available at: http://www.iom.edu. Accessed 1 April 2007. [4] Institute of Medicine Committee on the Changing Market and the Future Viability of Safety Net Providers. 2000. America s Health Care Safety Net: Intact but Endangered, ed.m.lewinands. Altman. Washington, DC: National Academies Press. [5] Richardson, L. and U. Hwang. 2001. America s health care safety net: intact or unraveling. AcademicEmergencyMedicine, 8(11): 1056 1063. [6] American Academy of Pediatrics Committee on Pediatric Emergency Medicine. 2004. Overcrowding crisis in our nation s emergency departments: is our safety net unraveling? Pediatrics 114: 878 888. [7] Derlet, R.W. 2002. Overcrowding in emergency departments: increased demand and decreased capacity. AnnalsofEmergency Medicine, 39(4): 430 432. [8] Derlet, R.W., J. R. Richards and R.L. Kravitz. 2001. Frequent overcrowding in U.S. emergency departments. AcademicEmergency Medicine, 8(2): 151 155. [9] Committee on the Future of Emergency Care in the United States Health Care System/Board on Health Care Services. 2006. Hospital-based Emergency Care: At the Breaking Point. Washington, DC: National Academies Press. [10] Fatovich, D.M. and R.L. Hirsch. 2003. Entry overload, emergency department overcrowding, and ambulance bypass. Emergency Medicine Journal, 20:406 409. [11] Derlet, R. and J. Richards. 2002. Emergency department overcrowding in Florida, New York, and Texas. SouthernMedicalJournal, 95(8): 846 849. [12] Sprivulis, P., J. Da Silva, I. Jacobs, et al. 2006. The association between hospital overcrowding and mortality among patients admitted via Western Australian emergency departments. The Medical Journal of Australia, 184(5): 208 212. [13] Derlet, R.W. and J.R. Richards. 2000. Overcrowding in the nation s emergency departments: complex causes and disturbing effects. Annals of Emergency Medicine, 35(1): 63 68. [14] Hwang, U. and J. Concato. 2004. Care in the emergency department: how crowded is overcrowded? Academic Emergency Medicine, 11(10): 1097 1101. [15] Asplin, B.R., D.J. Magid, K.V. Rhodes, et al. 2003. A conceptual model of emergency department crowding.annalsofemergency Medicine, 42(2): 173 180. [16] Weiss, S.J., R. Derlet and J. Arndahl. 2004. Estimating the degree of emergency department overcrowding in academic medical centers: results of the National ED Overcrowding Study (NEDOCS). Academic Emergency Medicine, 11: 38 50. [17] Reeder, T.J. and H.G. Garrison. 2001. When the safety net is unsafe: real-time assessment of the overcrowded emergency department. Academic Emergency Medicine, 8(11): 1070 1074. [18] Bernstein, S.L., V. Verghese, L. Leung, et al. 2003. Development and validation of a new index to measure emergency department crowding.academicemergencymedicine, 10(9): 938 942. 568 SIMULATION Volume 86, Numbers 8-9