- Email: [email protected]
Models for Improving Patient Throughput and Waiting at Hospital Emergency Departments
The Journal of Emergency Medicine, Vol. 43, No. 6, pp. 1119–1126, 2012 Copyright Ó 2012 Elsevier Inc. Printed in the USA. All rights reserved 0736-4679/$ - see front matter
http://dx.doi.org/10.1016/j.jemermed.2012.01.063
Administration of Emergency Medicine
MODELS FOR IMPROVING PATIENT THROUGHPUT AND WAITING AT HOSPITAL EMERGENCY DEPARTMENTS Jomon Aliyas Paul, PHD* and Li Lin, PHD† *Department of Economics, Finance and Quantitative Analysis, Kennesaw State University, Kennesaw, Georgia and †Industrial and Systems Engineering, University at Buffalo, Buffalo, New York Reprint Address: Jomon Aliyas Paul, PHD, Department of Economics, Finance and Quantitative Analysis, BB 341 Coles College of Business, Kennesaw State University, Kennesaw GA 30144
, Keywords—discrete event simulation (DES); ED crowding; patient throughput; length of stay
, Abstract—Background: Overcrowding diminishes Emergency Department (ED) care delivery capabilities. Objectives: We developed a generic methodology to investigate the causes of overcrowding and to identify strategies to resolve them, and applied it in the ED of a hospital participating in the study. Methods: We utilized Discrete Event Simulation (DES) to capture the complex ED operations. Using DES results, we developed parametric models for checking the effectiveness and quantifying the potential gains from various improvement alternatives. We performed a follow-up study to compare the outcomes before and after the model recommendations were put into effect at the hospital participating in the study. Results: Insufficient physicians during peak hours, the slow process of admitting patients to inpatient floors, and laboratory and radiology test turnaround times were identified as the causes of reduced ED throughput. Addition of a physician resulted in an almost 18% reduction in the ED Main discharged patient length of stay. Conclusion: The case study results demonstrated the effectiveness of the generic methodology. The research contributions were validated through statistically significant improvements seen in patient throughput and waiting time at the hospital participating in the study. Ó 2012 Elsevier Inc.
INTRODUCTION Overcrowding diminishes the ability of the Emergency Department (ED) to provide immediate access and stabilization to those patients who have an emergent medical condition (1). Increased ED patient length of stay (LOS) across hospitals in the United States is a very good indicator of this ever-growing problem (1). It has led to heightened patient and staff frustration and a surge in the risk for poor medical outcomes and unnecessarily high costs (1,2). Inability to transfer admitted patients to inpatient beds has been reported to be the most serious cause of ED overcrowding (2,3). Long waits in triage, delays in testing or obtaining test results, waiting for the physician, and shortage of nursing staff are among other issues identified in similar studies (4–6). Strategies proposed to overcome these problems are varied. For example, Finamore and Turris recommended the creation of satellite clinics for reducing ED wait times (7). This satellite clinic would take care of patients returning for follow-up diagnostics or treatment who thus would not use up ED resources. Tanabe et al. suggested that inpatient flow could be improved by closing the waiting room and instead sending patients directly to a stretcher or a chair inside the ED (8).
Appendices that accompany this article are available in the web version posted on the journal website: www.jem-journal. com.
RECEIVED: 13 February 2011; FINAL SUBMISSION RECEIVED: 11 July 2011; ACCEPTED: 19 January 2012 1119
1120
J. A. Paul and L. Lin
Miro´ et al. recommended the improvement of internal factors, such as the layout of the work environment, as a possible strategy for improving patient flow through the ED (9). Kulkarni reported on the success achieved by a hospital in reducing time between admission to an inpatient bed and transfer to a ward by 33% by applying Lean principles (10). Although several strategies have been tried with varying degrees of success, the problem still continues to affect hospitals across the United States. This could be explained by the lack of a generic model that would be useful for all EDs regardless of hospital size and operational setup differences. Such a model would not only be able to accurately capture the ED patient flow dynamics and identify improvement strategies, but also provide a quantitative assessment of gains from improvement measures before their actual implementation. However, the ED is a highly complex system, and a valid mathematical model for such a system would in itself be very complex. This makes determination of an analytical solution extremely difficult (11,12). Discrete Event Simulation (DES), a branch of computer simulation science, is an effective tool for studying such complex systems (11,12). Several examples of ED DES models are available in the extant literature. For instance, Draeger developed a model for the three EDs of a Bethesda, MD, hospital to evaluate process performance and the impact of changes in nurse staffing and patient flow on system performance (13). It helped the management to select the best alternatives from those proposed. Similarly, Rossetti et al. developed an ED simulation model for the University of Virginia Medical Center to test the impact of alternative attending physician schedules on patient throughput and resource utilization (14). In another study, Connelly and Bair used DES to model the ED operations at a Level I trauma center to test the effectiveness of two new triage methods (15). The model predicted average patient service times with reasonable accuracy. In this study, we develop a generic methodology to investigate the causes of ED overcrowding and identify strategies to overcome them. We demonstrate its effectiveness via a case study at a participating hospital. In the proposed methodology, first the complex ED operations were captured utilizing DES. Using the DES results, we then developed parametric models for checking the effectiveness of, and quantifying the potential gains from, various improvement alternatives. Finally, we performed a follow-up study that compared the outcomes before and after implementation of the model recommendations. METHODS ED overcrowding, the primary focus of this study, can result either from a lack of resources such as beds (indirectly involved in care delivery) or from a shortage of
resources such as physicians, nurses, or other medical professionals (directly involved in care delivery), or it can be due to inefficiencies in processes like obtaining laboratory and radiology results, admitting patients to inpatient floors, or triage. The phases of the methodology to address these challenges are explained next. Phase 1 involves three steps. Step 1 focuses on obtaining a thorough understanding of the detailed operational logic and procedures performed in the ED using a process flow diagram. This is vital for efficient data collection. Step 2 deals with classification of all the activities identified in Step 1 into one of two categories: waiting (for personnel, laboratory-radiology results, beds) or care delivery. This helps to prioritize the activities that need to be improved or eliminated. The third step focuses on assigning internal or external ED status to the activities and subprocesses. This would not only help the team to scope the modeling effort but also will identify the factors that are within the team’s control and therefore need to be tackled first. Phase 2 involves development of a DES model. This is the logical next step after process flow mapping, because a DES model has the ability to capture activities in a system as a network of interdependent and discrete events (11). The DES model enacts the events using the prioritybased rules and decision-making that drive the actual ED operations. Once the DES model is built, it can be numerically applied to evaluate the impact of changes to inputs on the output measures of performance (12). Before the DES model is used to identify improvement strategies, a validation of the model results is necessary. This ensures the reliability of the DES model recommendations. This forms the third phase of our methodology. Validation can be done by comparing critical performance measures obtained from the model to the actual data. From an ED perspective, this can be achieved by comparing variability and central tendency of patient LOS (including the duration of various subprocesses) obtained from the model to the actual data for different patient types. Methods such as comparison of means and goodness of fit can be used for these purposes. Once the DES model is validated, the fourth phase involves the use of the model to study the effect of resource availability and process improvements on patient throughput and waiting time. Effect of resources can be studied with the help of utilization graphs that record the utilization of a resource by the hour of day. Utilization can be defined as the percentage of time the resource is involved in delivering patient care out of the total time it is present in the ED. Further, it is possible to measure the impact of improvement alternatives suggested by the project team, using the model. This can be accomplished with the help of parametric models developed from simulation results that quantify the potential gains from each
Improving Patient Throughput at Hospital EDs
improvement alternative proposed by the ED management. Based on these results, the most appropriate recommendations can be determined. In the fifth and last phase of the methodology, implementation of the recommendations is an effective final validation of the model contributions. Next, we demonstrate its application at the hospital participating in the study. The hospital ED in this study has a census of approximately 65,000 patients per year, about 20% of whom are admitted. In general, approximately 55% of the admissions come from the ED, thus making it a major entry point of inpatients to the hospital. The ED has two sections: the ED Main that handles high-severity patients (43% of total ED population); and the Fast Track section (57% of total ED population) that takes care of the relatively lowseverity patients. Incoming patients are categorized into five severity levels: Severity 1 is assigned to patients with the most critical conditions and Severity 5 is assigned to patients with the least severe conditions. This classification system is based on the chief complaint and severity of the patient’s presenting symptoms. The ED Main primarily treats Severity 1, 2, and 3 patients and some fraction of Severity 4 patients. The remaining patients were sent to Fast Track. These Fast Track patients were discharged from the ED the same day. Patients seen at the ED Main could be either discharged (11% of ED Main population) or admitted (89% of ED Main population) to inpatient floors based on the Emergency Physician’s evaluation of their condition. Patients presenting to the ED could come either on their own or with their family, in an ambulance, or in a helicopter. The first stop for the patient is the triage station, where the triage nurse, if available at the time, evaluates the patient. If a triage nurse is not available, the patient remains in the waiting room. The analysis by the triage nurse is used as the initial diagnosis of the patient’s illness, and the triage nurse assigns a severity level to the patient based on this initial diagnosis. Patients are then treated at either the ED Main section or in the Fast Track area, depending on the severity of the patient’s condition. If the chief complaint or symptoms of an incoming patient are severe, the patient is taken immediately to an ED Main examination room. If there is no room available for this patient, a common protocol is to move a relatively low-severity patient (Severity 4 or 5) from an examination room to the hallway, placing the higher-severity patient in the examination room. Patients with severe conditions are given the highest priority and generally require the most attention and resources. Registration for these patients is done at the bedside. If a patient’s condition is not severe and no bed is available, the patient is sent back to the waiting room after completion of registration and triage. Once the patient is in the ED examination room bed, appropriate diagnostic tests are
1121
ordered based on the protocol associated with the chief complaint of the patient or initial evaluation by the physician. To save time, sometimes samples are taken and standard laboratory tests performed while the patients wait in the waiting room for a bed assignment. An emergency physician reviews these results and recommends either that the patient be discharged or that the patient might possibly be admitted to the hospital. Once the physician recommends the patient be admitted, a call is made to the patient’s primary care physician or to the attending physician of the relevant inpatient unit or a hospitalist who is on the hospital staff to issue orders for admittance. This physician evaluates the patient and then issues the admit orders. If a bed is available and the unit is ready, the patient is sent to the unit, otherwise the patient is held in the ED. Patients who did not require admission to inpatient floors are discharged or, if the triage diagnosis indicates that the patient’s illness is minor (all Severity 5 and some Severity 4), the patient is sent to the Fast Track for treatment and eventually discharged. The data reported in this study were obtained in part from the ED patient database. The patient flow described above is presented in Figure 1. Once the process flow was developed, with the help of discussions with ED staff we were able to classify the activities into one of two categories: either ‘‘waiting’’ or ‘‘care delivery,’’ and assign an internal or external-to-ED status to all the subprocesses. This exercise provided valuable insights to the data collection and analysis and also helped us determine the extent of modeling effort required. A detailed breakdown of the ED process flow for the study hospital is presented in the Appendices, available only on the website; see Appendix 1. To accurately represent the patient flow in our model, we obtained historical data from the ED patient database. Additional details on the data and the data collection process are provided in Appendix 2. Institutional Review Board approval was obtained for this study, as a retrospective study. Only retrospective ED data were collected, and no patient identifiers were collected from the ED patient database. Patient volume by day of the week and time of day, distribution of patient volume by severity of illness level, turnaround times for treatment processes performed by physicians and nurses, staffing levels and changes, and laboratory and radiology turnaround times were obtained either from recorded historical data or staff interviews. Analysis of ED patient data indicated that the two processes that affected LOS most significantly were ‘‘Evaluation to Disposition’’ by the physician for both ED Main discharged and admit patients, and ‘‘Disposition to ED departure’’ for admit patients. Additional analysis and discussion with ED staff indicated that ‘‘Evaluation to Disposition’’ was mainly influenced by the time required to receive laboratory and radiology test results, as well as availability of the physician who evaluated the test
1122
J. A. Paul and L. Lin
Figure 1. Emergency Department flow chart.
results. The laboratory and radiology test turnaround times were: 1 h 23 min for Severity 2, 1 h 15 min for Severity 3, 39 min for Severity 4, and 10 min for Severity 5 patients. Similarly, additional analysis of the ‘‘Disposition to Departure’’ process for admit patients indicated that it was made up of three key subprocesses: ‘‘Disposition to Admit order,’’ ‘‘Admit order to Room available,’’ and ‘‘Room available to Departure.’’ Data analysis revealed that the average durations of these three subprocesses were 1 h 7 minutes, 40 min, and 55 min, respectively. Moreover, we found that, ‘‘Disposition to Admit order’’ and ‘‘Admit order to Room available’’ were affected by factors that were external to the ED and therefore not under the ED’s control. This included availability of primary care physician, hospitalist, or attending unit physician to create admit orders and for arrangements to be made for a bed on the inpatient floor to which a patient could be assigned. The entire ‘‘Disposition to Departure’’ process average duration for the admit patients was equal to 2 h and 42 min. A detailed breakdown of the ED patient LOS time period is presented in Appendix 3. Although the data analysis revealed the effect of laboratory and radiology turnaround times and patient admission to inpatient floors on patient LOS; to discover
the true impact of changes in any of the process turnaround times or physician availability, or other resources like beds or nurses, the support of a DES model was required. This was also vital for identification of improvement strategies. The state-of-the-art simulation software ProModel (Orem, UT) was used for this purpose (16). (Discrete Event Simulation modeling is explained in significant detail in Discrete-Event Simulation: Modeling, Programming, and Analysis by George S. Fishman) (11). The following assumptions apply to our DES model: probability distributions based on data analysis results using ProModel were used to represent various care delivery processes and patient arrival; patient interarrival times were independent and identically distributed (iid); care delivery process lead times were iid random variables and independent of inter-arrival times; certain variables, such as number of patients in the waiting room or number of rooms available, change at discrete points in time; personnel such as physicians, nurses, and transporters performed their duties efficiently; finally, to simplify the analysis, the waiting room was assumed to have an infinite capacity. We used the patient LOS in the ED by patient category as a measure for validation of our model. The results (Table 1) indicated that the simulation results were
Improving Patient Throughput at Hospital EDs
1123
Table 1. Validation of the ‘‘Baseline’’ Model Average Length of Stay (h:min)
Admit ED discharged Fast track
Median Length of Stay (h:min)
Simulation
Data Collected
Simulation
Data Collected
5:31 3:28 1:21
5:23 3:20 1:18
5:19 3:21 1:20
5:10 3:27 1:15
ED = Emergency Department.
consistently close to the real patient LOS in terms of average values and medians. The difference in the average LOS between the simulation and the collected data was within 5%. The difference between the medians was also small, indicating that data collected and simulation results had similar distributions. The above findings were further substantiated by hypothesis testing of differences between mean times for all the main patient categories (ED Main admit, ED Main discharge, and Fast Track). The p-values for these tests were as follows: 0.469 for ED Main admit patients, 0.417 for ED Main discharge patients, and 0.643 for Fast track; all three indicating statistically nonsignificant differences between the model results and real data, thus validating the simulation model. As discussed above, the process bottlenecks that were negatively impacting the overall patient LOS were identified via data analysis. However, to identify the resources that were negatively affecting the throughput and LOS, additional analysis was necessary. Logically, utilization of various resources, such as ED beds, physicians, and nursing staff, would be of interest. We used the DES model to generate utilization graphs for the resources that were part of the ED operations. The hourly bed utilization graph (Figure 2A) indicated that ED Main beds seemed to be fully occupied during the entire day and were particularly busy in the evening.
The green line (upper line) and the red line (lowest line) represent the upper and lower bounds, respectively, of the 95% confidence interval of ED bed utilization. These bounds indicate that the probability of having a higher or lower utilization than within the bounds is 5% (i.e., 1–95%). As utilization of resources can never exceed 100%, the upper bound above the 100% line represented situations in which ED beds were completely full. As bed utilization was observed to be consistently high, adding beds was identified as an improvement strategy. Similarly, Figure 2B presents the utilization of ED Main physicians at various hours of the day. The number of attending physicians in the ED is represented by the line labeled ‘‘Staffing.’’ The graph indicated that the ED Main physicians were constantly in high demand, as their average utilization was consistently around 90%, and at times even higher. Therefore, adding physicians was identified as one of the improvement strategies. The hours when an additional ED physician would be especially helpful were also identified with the help of these utilization graphs. Next, we studied the impact of various improvement measures, including the addition of resources (physicians, nurses, and beds based on findings from the utilization graphs discussed in the previous subsection) and reduction in process times (admit time [1] and laboratory and radiology turn-around time [2]). It was found that adding nurses or beds did not have a significant impact on the throughput. On the other hand, the addition of an Emergency Physician, and a reduction in laboratory and radiology turnaround times in the admission process resulted in significant improvements in patient throughput and reduced waiting times. Based on these findings, we simulated a combined scenario of adding an Emergency Physician, reducing admit time (1), and reducing laboratory time (2). Using simulation results, we developed parametric regression models. The relationships for admit patient LOS (Ta) and discharged patient LOS (Td), generated using SAS IML (SAS Institute Inc., Cary, NC), are
Figure 2. (A) Emergency Department (ED) Main bed utilization. (B) ED Main physician utilization.
1124
J. A. Paul and L. Lin
revealed below. All the regression coefficients were significant at the 95% confidence level. Ta ¼ 287:27 0:67x1 0:51x2 þ 0:0014x22
(1)
Td ¼ 149:99 0:62x2
(2)
Equation 1 indicates that a reduction in admission process wait time impacts admit patient LOS more significantly than that due to laboratory and radiology time improvements. On the other hand, only laboratory and radiology improvements have a significant impact on the discharged patient LOS. This information is valuable for prioritization of improvement measures. RESULTS Based on the DES model recommendations, the hospital added a new physician to the ED Main between 11:00 a.m. and midnight because these were generally the peak hours of the day. In a follow-up effort, ED performance was compared on days with and without an additional physician to check for improvements. Specifically, we compared the average daily patient throughput and the average patient LOS for these two time periods. The results from this analysis are presented in Table 2. As can be noted from Table 2, addition of the new physician significantly improved the daily patient throughput. In addition, a statistically significant improvement was seen in average patient LOS for ED Main discharged patients. However, the same could not be said about admit patients. To accurately estimate the impact of adding one physician on admit patient outcomes, we performed some additional analysis. The discrepancy could be due to skewness and kurtosis in LOS data collected on days when the new physician was present, or due to a significant difference in patient throughput. For days with and without the new physician, the skewness was found to be .62 and .53, respectively. Similarly, kurtosis was 3.55 and 3.40, respectively, for these two time periods. This was due to longer LOS values in the data on days with the new physician. A detailed LOS
Table 2. Comparison of Outcomes with and without the New Physician Variable Average daily patient throughput Td in minutes Ta in minutes * p < 0.05. ** p > 0.05.
Days with Days without Net New Physician New Physician Change 199 (17 days)
182 (14 days)
17*
182 (544) 320 (166)
221 (167) 316 (119)
39* 4**
breakdown for the admit patients is presented in Appendix 4. Finally, to justify the addition of the physician, we performed a simple financial analysis. Based on the data provided to us by the accounting department, it was estimated that hospital revenue would increase by approximately $5000 a day ($300 per patient, on average). This would justify expenditure on any direct or indirect job-performing resource in the long run, given the average patient volume increase of 17 on days that the physician was added. In addition to changes internal to the ED, the model also recommended changes external to the ED. These were in laboratory and radiology departments due to their significantly long turnaround times, as well as the slow admission process of patients to inpatient units. Two additional projects focusing on improvement of the process of patient admission to the inpatient floor were carried out based on findings from our project. Software for tracking patient beds was installed across the hospital. This led to additional improvements in patient LOS in the ED and also helped better streamline the process for ED patient admission to the inpatient units. DISCUSSION This study demonstrates that computer simulation is useful in modeling all operational details in a complex system such as an ED, with a great deal of modeling flexibility. The DES model is able to efficiently enact actual events using historical data that represent patients, staff, laboratory and imaging studies, and associated resources. The logic of patient moves and waiting for resources is captured by the simulation model so that resource utilization and patient LOS data are obtained. Additional programming effort makes it possible to represent ED dynamics on an hourly basis, 24 h a day. In addition to average resource utilization for each hour, representation of the fluctuation of patient volumes at different hours of the day can be achieved via statistical confidence intervals. This study shows the importance of understanding the roles played by different types of resources in the ED and shows the impact of process improvement on ED overcrowding (rather than assuming or blaming a lack of a bed as the root cause without supporting statistical evidence). The generic methodology proposed in this article would be valuable to hospitals in achieving improvement of throughput or decreasing waiting time because each hospital is different and only a detailed analysis such as that made possible by our methodology could reveal the true bottlenecks. LIMITATIONS The models in this study made assumptions regarding the participating hospital, including triage efficiency, bed
Improving Patient Throughput at Hospital EDs
allocation, and patient disposition processes by physicians and nurses. Inefficiency in such key processes could lead to patients staying longer in the ED than required for their care. A potential improvement might be to add a level of detail in recording the chief complaints, as the current model captures only severity of patient illness. Patient symptoms, treatment protocol, chance of getting admitted, and even correct diagnosis depend, to a significant extent, on the chief complaint. Further, our model and the subsequent analysis could improve by studying the disparate effects of leaving without being seen and leaving against medical advice on problems faced by the ED. In the current study, we have not differentiated between these two patient groups. Finally, effect of the mode of patient arrival on ED problems has not been studied. Any preference given to a patient based on mode of arrival at the ED in terms of severity assignment and subsequent bed allocation could significantly affect ED operations and outcomes. CONCLUSIONS We developed a generic methodology to address the overcrowding issues in the ED. We demonstrated the applicability of our methodology at a study participating hospital. With the help of the DES model results, we were able to identify the key areas of improvement internal to the ED, and also propose new projects in areas external to ED. The model also provided us with the capability to check the effectiveness of various proposed improvement strategies through the use of parametric models developed using simulation results. The results from the quantitative analysis of the crucial relationships between the factors provided the basis for additional analysis, such as financial justification on the return of investment in the improvements. Follow-up actions that measured the benefits of the model recommendations validated the usefulness of the models developed in this research.
1125 2. Solberg LI, Asplin BR, Weinick RM, Magid DJ. Emergency department crowding: consensus development of potential measures. Ann Emerg Med 2003;42:824–33. 3. Howell E, Bessman ES, Rubin HR. Hospitalists and an innovative emergency department admission process. J Gen Intern Med 2004;19:266–8. 4. Messner ED. Quality of care and patient satisfaction—the improvement efforts of one emergency department. Top Emerg Med 2005; 27:132–41. 5. Asplin BR, Rhodes KV, Crain L, Camargo CA. Measuring emergency department crowding and hospital capacity. Acad Emerg Med 2002;9:366–7. 6. McMahon MM. ED triage—is a five-level triage system best? Am J Nurs 2003;103:61–3. 7. Finamore SR, Turris SA. Shortening the wait: a strategy to reduce waiting times in the emergency department. J Emerg Nurs 2009; 35:509–14. 8. Tanabe P, Gisondi MA, Medendorp S, Engeldinger L, Graham LJ, Lucenti MJ. Should you close your waiting room? Addressing ED overcrowding through education and staff based participatory research. J Emerg Nurs 2008;34:285–9. 9. Miro´ O, Sa´nchez M, Espinosa G, Coll-Vinent B, Bragulat E, Milla´ J. Analysis of patient flow in the emergency department and the effect of an extensive reorganization. Emerg Med J 2003;20:143–8. 10. Kulkarni RG. Going lean in the emergency department: a strategy addressing emergency department overcrowding. Med Gen Med 2007;9:58. 11. Fishman GS. Springer series in operations research. Discrete-event simulation: modeling, programming, and analysis. New York: Springer; 2001. 12. Law AM. McGraw-Hill series in industrial engineering and management science. Simulation modeling and analysis. New York: McGraw-Hill; 2007. 13. Draeger MA. An emergency department simulation model used to evaluate alternative nurse staffing and patient population scenarios. Proceedings of the 1992 Winter Simulation Conference. Arlington, VA; 1992:1057–63. 14. Rossetti MD, Trzcinski GF, Syverud SA. Emergency department simulation and determination of optimal attending physician staffing schedules. Proceedings of the 1999 Winter Simulation Conference. Phoenix, AZ; 1999:1532–40. 15. Connelly LG, Bair AE. Discrete event simulation of emergency department activity: a platform for system-level operations research. Acad Emerg Med 2004;11:1177–85. 16. ProModel. Available at: www.promodel.com. 2005. Accessed January 2005.
SUPPLEMENTARY DATA REFERENCES 1. Derlet RW, Richards JR. Overcrowding in the nation’s emergency departments: complex causes and disturbing effects. Ann Emerg Med 2000;35:63–8.
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10. 1016/j.jemermed.2012.01.063.
Appendices that accompany this article are available in the web version posted on the journal website: www.jem-journal.com.
1126
J. A. Paul and L. Lin
ARTICLE SUMMARY 1. Why is this topic important? Emergency Department (ED) crowding has been a serious problem in the nation’s hospitals for more than a decade, with approximately 90% of hospital directors reporting it to be a constant issue. 2. What does this study attempt to show? In this study, we sought to develop a generic methodology to address overcrowding and patient throughput issues in an ED that can be used at other hospitals regardless of differences in size and operational framework. 3. What are the key findings? The major issues leading to ED crowding and throughput problems were: a shortage of physicians; increased duration of patient admission to inpatient units; and slow laboratory and radiology test turnaround times. Counter to conventional wisdom, a shortage of beds was not an issue in this study. 4. How is patient care impacted? The addition of an extra Emergency Physician during specific shifts resulted in an almost 18% reduction in length of stay for patients discharged from the ED Main. The bed tracking software installed in the hospital, which was based on the Discrete Event Simulation model recommendations developed in this study, resulted in additional reductions in ED patient length of stay and also streamlined the admission process.
http://dx.doi.org/10.1016/j.jemermed.2012.01.063
Administration of Emergency Medicine
MODELS FOR IMPROVING PATIENT THROUGHPUT AND WAITING AT HOSPITAL EMERGENCY DEPARTMENTS Jomon Aliyas Paul, PHD* and Li Lin, PHD† *Department of Economics, Finance and Quantitative Analysis, Kennesaw State University, Kennesaw, Georgia and †Industrial and Systems Engineering, University at Buffalo, Buffalo, New York Reprint Address: Jomon Aliyas Paul, PHD, Department of Economics, Finance and Quantitative Analysis, BB 341 Coles College of Business, Kennesaw State University, Kennesaw GA 30144
, Keywords—discrete event simulation (DES); ED crowding; patient throughput; length of stay
, Abstract—Background: Overcrowding diminishes Emergency Department (ED) care delivery capabilities. Objectives: We developed a generic methodology to investigate the causes of overcrowding and to identify strategies to resolve them, and applied it in the ED of a hospital participating in the study. Methods: We utilized Discrete Event Simulation (DES) to capture the complex ED operations. Using DES results, we developed parametric models for checking the effectiveness and quantifying the potential gains from various improvement alternatives. We performed a follow-up study to compare the outcomes before and after the model recommendations were put into effect at the hospital participating in the study. Results: Insufficient physicians during peak hours, the slow process of admitting patients to inpatient floors, and laboratory and radiology test turnaround times were identified as the causes of reduced ED throughput. Addition of a physician resulted in an almost 18% reduction in the ED Main discharged patient length of stay. Conclusion: The case study results demonstrated the effectiveness of the generic methodology. The research contributions were validated through statistically significant improvements seen in patient throughput and waiting time at the hospital participating in the study. Ó 2012 Elsevier Inc.
INTRODUCTION Overcrowding diminishes the ability of the Emergency Department (ED) to provide immediate access and stabilization to those patients who have an emergent medical condition (1). Increased ED patient length of stay (LOS) across hospitals in the United States is a very good indicator of this ever-growing problem (1). It has led to heightened patient and staff frustration and a surge in the risk for poor medical outcomes and unnecessarily high costs (1,2). Inability to transfer admitted patients to inpatient beds has been reported to be the most serious cause of ED overcrowding (2,3). Long waits in triage, delays in testing or obtaining test results, waiting for the physician, and shortage of nursing staff are among other issues identified in similar studies (4–6). Strategies proposed to overcome these problems are varied. For example, Finamore and Turris recommended the creation of satellite clinics for reducing ED wait times (7). This satellite clinic would take care of patients returning for follow-up diagnostics or treatment who thus would not use up ED resources. Tanabe et al. suggested that inpatient flow could be improved by closing the waiting room and instead sending patients directly to a stretcher or a chair inside the ED (8).
Appendices that accompany this article are available in the web version posted on the journal website: www.jem-journal. com.
RECEIVED: 13 February 2011; FINAL SUBMISSION RECEIVED: 11 July 2011; ACCEPTED: 19 January 2012 1119
1120
J. A. Paul and L. Lin
Miro´ et al. recommended the improvement of internal factors, such as the layout of the work environment, as a possible strategy for improving patient flow through the ED (9). Kulkarni reported on the success achieved by a hospital in reducing time between admission to an inpatient bed and transfer to a ward by 33% by applying Lean principles (10). Although several strategies have been tried with varying degrees of success, the problem still continues to affect hospitals across the United States. This could be explained by the lack of a generic model that would be useful for all EDs regardless of hospital size and operational setup differences. Such a model would not only be able to accurately capture the ED patient flow dynamics and identify improvement strategies, but also provide a quantitative assessment of gains from improvement measures before their actual implementation. However, the ED is a highly complex system, and a valid mathematical model for such a system would in itself be very complex. This makes determination of an analytical solution extremely difficult (11,12). Discrete Event Simulation (DES), a branch of computer simulation science, is an effective tool for studying such complex systems (11,12). Several examples of ED DES models are available in the extant literature. For instance, Draeger developed a model for the three EDs of a Bethesda, MD, hospital to evaluate process performance and the impact of changes in nurse staffing and patient flow on system performance (13). It helped the management to select the best alternatives from those proposed. Similarly, Rossetti et al. developed an ED simulation model for the University of Virginia Medical Center to test the impact of alternative attending physician schedules on patient throughput and resource utilization (14). In another study, Connelly and Bair used DES to model the ED operations at a Level I trauma center to test the effectiveness of two new triage methods (15). The model predicted average patient service times with reasonable accuracy. In this study, we develop a generic methodology to investigate the causes of ED overcrowding and identify strategies to overcome them. We demonstrate its effectiveness via a case study at a participating hospital. In the proposed methodology, first the complex ED operations were captured utilizing DES. Using the DES results, we then developed parametric models for checking the effectiveness of, and quantifying the potential gains from, various improvement alternatives. Finally, we performed a follow-up study that compared the outcomes before and after implementation of the model recommendations. METHODS ED overcrowding, the primary focus of this study, can result either from a lack of resources such as beds (indirectly involved in care delivery) or from a shortage of
resources such as physicians, nurses, or other medical professionals (directly involved in care delivery), or it can be due to inefficiencies in processes like obtaining laboratory and radiology results, admitting patients to inpatient floors, or triage. The phases of the methodology to address these challenges are explained next. Phase 1 involves three steps. Step 1 focuses on obtaining a thorough understanding of the detailed operational logic and procedures performed in the ED using a process flow diagram. This is vital for efficient data collection. Step 2 deals with classification of all the activities identified in Step 1 into one of two categories: waiting (for personnel, laboratory-radiology results, beds) or care delivery. This helps to prioritize the activities that need to be improved or eliminated. The third step focuses on assigning internal or external ED status to the activities and subprocesses. This would not only help the team to scope the modeling effort but also will identify the factors that are within the team’s control and therefore need to be tackled first. Phase 2 involves development of a DES model. This is the logical next step after process flow mapping, because a DES model has the ability to capture activities in a system as a network of interdependent and discrete events (11). The DES model enacts the events using the prioritybased rules and decision-making that drive the actual ED operations. Once the DES model is built, it can be numerically applied to evaluate the impact of changes to inputs on the output measures of performance (12). Before the DES model is used to identify improvement strategies, a validation of the model results is necessary. This ensures the reliability of the DES model recommendations. This forms the third phase of our methodology. Validation can be done by comparing critical performance measures obtained from the model to the actual data. From an ED perspective, this can be achieved by comparing variability and central tendency of patient LOS (including the duration of various subprocesses) obtained from the model to the actual data for different patient types. Methods such as comparison of means and goodness of fit can be used for these purposes. Once the DES model is validated, the fourth phase involves the use of the model to study the effect of resource availability and process improvements on patient throughput and waiting time. Effect of resources can be studied with the help of utilization graphs that record the utilization of a resource by the hour of day. Utilization can be defined as the percentage of time the resource is involved in delivering patient care out of the total time it is present in the ED. Further, it is possible to measure the impact of improvement alternatives suggested by the project team, using the model. This can be accomplished with the help of parametric models developed from simulation results that quantify the potential gains from each
Improving Patient Throughput at Hospital EDs
improvement alternative proposed by the ED management. Based on these results, the most appropriate recommendations can be determined. In the fifth and last phase of the methodology, implementation of the recommendations is an effective final validation of the model contributions. Next, we demonstrate its application at the hospital participating in the study. The hospital ED in this study has a census of approximately 65,000 patients per year, about 20% of whom are admitted. In general, approximately 55% of the admissions come from the ED, thus making it a major entry point of inpatients to the hospital. The ED has two sections: the ED Main that handles high-severity patients (43% of total ED population); and the Fast Track section (57% of total ED population) that takes care of the relatively lowseverity patients. Incoming patients are categorized into five severity levels: Severity 1 is assigned to patients with the most critical conditions and Severity 5 is assigned to patients with the least severe conditions. This classification system is based on the chief complaint and severity of the patient’s presenting symptoms. The ED Main primarily treats Severity 1, 2, and 3 patients and some fraction of Severity 4 patients. The remaining patients were sent to Fast Track. These Fast Track patients were discharged from the ED the same day. Patients seen at the ED Main could be either discharged (11% of ED Main population) or admitted (89% of ED Main population) to inpatient floors based on the Emergency Physician’s evaluation of their condition. Patients presenting to the ED could come either on their own or with their family, in an ambulance, or in a helicopter. The first stop for the patient is the triage station, where the triage nurse, if available at the time, evaluates the patient. If a triage nurse is not available, the patient remains in the waiting room. The analysis by the triage nurse is used as the initial diagnosis of the patient’s illness, and the triage nurse assigns a severity level to the patient based on this initial diagnosis. Patients are then treated at either the ED Main section or in the Fast Track area, depending on the severity of the patient’s condition. If the chief complaint or symptoms of an incoming patient are severe, the patient is taken immediately to an ED Main examination room. If there is no room available for this patient, a common protocol is to move a relatively low-severity patient (Severity 4 or 5) from an examination room to the hallway, placing the higher-severity patient in the examination room. Patients with severe conditions are given the highest priority and generally require the most attention and resources. Registration for these patients is done at the bedside. If a patient’s condition is not severe and no bed is available, the patient is sent back to the waiting room after completion of registration and triage. Once the patient is in the ED examination room bed, appropriate diagnostic tests are
1121
ordered based on the protocol associated with the chief complaint of the patient or initial evaluation by the physician. To save time, sometimes samples are taken and standard laboratory tests performed while the patients wait in the waiting room for a bed assignment. An emergency physician reviews these results and recommends either that the patient be discharged or that the patient might possibly be admitted to the hospital. Once the physician recommends the patient be admitted, a call is made to the patient’s primary care physician or to the attending physician of the relevant inpatient unit or a hospitalist who is on the hospital staff to issue orders for admittance. This physician evaluates the patient and then issues the admit orders. If a bed is available and the unit is ready, the patient is sent to the unit, otherwise the patient is held in the ED. Patients who did not require admission to inpatient floors are discharged or, if the triage diagnosis indicates that the patient’s illness is minor (all Severity 5 and some Severity 4), the patient is sent to the Fast Track for treatment and eventually discharged. The data reported in this study were obtained in part from the ED patient database. The patient flow described above is presented in Figure 1. Once the process flow was developed, with the help of discussions with ED staff we were able to classify the activities into one of two categories: either ‘‘waiting’’ or ‘‘care delivery,’’ and assign an internal or external-to-ED status to all the subprocesses. This exercise provided valuable insights to the data collection and analysis and also helped us determine the extent of modeling effort required. A detailed breakdown of the ED process flow for the study hospital is presented in the Appendices, available only on the website; see Appendix 1. To accurately represent the patient flow in our model, we obtained historical data from the ED patient database. Additional details on the data and the data collection process are provided in Appendix 2. Institutional Review Board approval was obtained for this study, as a retrospective study. Only retrospective ED data were collected, and no patient identifiers were collected from the ED patient database. Patient volume by day of the week and time of day, distribution of patient volume by severity of illness level, turnaround times for treatment processes performed by physicians and nurses, staffing levels and changes, and laboratory and radiology turnaround times were obtained either from recorded historical data or staff interviews. Analysis of ED patient data indicated that the two processes that affected LOS most significantly were ‘‘Evaluation to Disposition’’ by the physician for both ED Main discharged and admit patients, and ‘‘Disposition to ED departure’’ for admit patients. Additional analysis and discussion with ED staff indicated that ‘‘Evaluation to Disposition’’ was mainly influenced by the time required to receive laboratory and radiology test results, as well as availability of the physician who evaluated the test
1122
J. A. Paul and L. Lin
Figure 1. Emergency Department flow chart.
results. The laboratory and radiology test turnaround times were: 1 h 23 min for Severity 2, 1 h 15 min for Severity 3, 39 min for Severity 4, and 10 min for Severity 5 patients. Similarly, additional analysis of the ‘‘Disposition to Departure’’ process for admit patients indicated that it was made up of three key subprocesses: ‘‘Disposition to Admit order,’’ ‘‘Admit order to Room available,’’ and ‘‘Room available to Departure.’’ Data analysis revealed that the average durations of these three subprocesses were 1 h 7 minutes, 40 min, and 55 min, respectively. Moreover, we found that, ‘‘Disposition to Admit order’’ and ‘‘Admit order to Room available’’ were affected by factors that were external to the ED and therefore not under the ED’s control. This included availability of primary care physician, hospitalist, or attending unit physician to create admit orders and for arrangements to be made for a bed on the inpatient floor to which a patient could be assigned. The entire ‘‘Disposition to Departure’’ process average duration for the admit patients was equal to 2 h and 42 min. A detailed breakdown of the ED patient LOS time period is presented in Appendix 3. Although the data analysis revealed the effect of laboratory and radiology turnaround times and patient admission to inpatient floors on patient LOS; to discover
the true impact of changes in any of the process turnaround times or physician availability, or other resources like beds or nurses, the support of a DES model was required. This was also vital for identification of improvement strategies. The state-of-the-art simulation software ProModel (Orem, UT) was used for this purpose (16). (Discrete Event Simulation modeling is explained in significant detail in Discrete-Event Simulation: Modeling, Programming, and Analysis by George S. Fishman) (11). The following assumptions apply to our DES model: probability distributions based on data analysis results using ProModel were used to represent various care delivery processes and patient arrival; patient interarrival times were independent and identically distributed (iid); care delivery process lead times were iid random variables and independent of inter-arrival times; certain variables, such as number of patients in the waiting room or number of rooms available, change at discrete points in time; personnel such as physicians, nurses, and transporters performed their duties efficiently; finally, to simplify the analysis, the waiting room was assumed to have an infinite capacity. We used the patient LOS in the ED by patient category as a measure for validation of our model. The results (Table 1) indicated that the simulation results were
Improving Patient Throughput at Hospital EDs
1123
Table 1. Validation of the ‘‘Baseline’’ Model Average Length of Stay (h:min)
Admit ED discharged Fast track
Median Length of Stay (h:min)
Simulation
Data Collected
Simulation
Data Collected
5:31 3:28 1:21
5:23 3:20 1:18
5:19 3:21 1:20
5:10 3:27 1:15
ED = Emergency Department.
consistently close to the real patient LOS in terms of average values and medians. The difference in the average LOS between the simulation and the collected data was within 5%. The difference between the medians was also small, indicating that data collected and simulation results had similar distributions. The above findings were further substantiated by hypothesis testing of differences between mean times for all the main patient categories (ED Main admit, ED Main discharge, and Fast Track). The p-values for these tests were as follows: 0.469 for ED Main admit patients, 0.417 for ED Main discharge patients, and 0.643 for Fast track; all three indicating statistically nonsignificant differences between the model results and real data, thus validating the simulation model. As discussed above, the process bottlenecks that were negatively impacting the overall patient LOS were identified via data analysis. However, to identify the resources that were negatively affecting the throughput and LOS, additional analysis was necessary. Logically, utilization of various resources, such as ED beds, physicians, and nursing staff, would be of interest. We used the DES model to generate utilization graphs for the resources that were part of the ED operations. The hourly bed utilization graph (Figure 2A) indicated that ED Main beds seemed to be fully occupied during the entire day and were particularly busy in the evening.
The green line (upper line) and the red line (lowest line) represent the upper and lower bounds, respectively, of the 95% confidence interval of ED bed utilization. These bounds indicate that the probability of having a higher or lower utilization than within the bounds is 5% (i.e., 1–95%). As utilization of resources can never exceed 100%, the upper bound above the 100% line represented situations in which ED beds were completely full. As bed utilization was observed to be consistently high, adding beds was identified as an improvement strategy. Similarly, Figure 2B presents the utilization of ED Main physicians at various hours of the day. The number of attending physicians in the ED is represented by the line labeled ‘‘Staffing.’’ The graph indicated that the ED Main physicians were constantly in high demand, as their average utilization was consistently around 90%, and at times even higher. Therefore, adding physicians was identified as one of the improvement strategies. The hours when an additional ED physician would be especially helpful were also identified with the help of these utilization graphs. Next, we studied the impact of various improvement measures, including the addition of resources (physicians, nurses, and beds based on findings from the utilization graphs discussed in the previous subsection) and reduction in process times (admit time [1] and laboratory and radiology turn-around time [2]). It was found that adding nurses or beds did not have a significant impact on the throughput. On the other hand, the addition of an Emergency Physician, and a reduction in laboratory and radiology turnaround times in the admission process resulted in significant improvements in patient throughput and reduced waiting times. Based on these findings, we simulated a combined scenario of adding an Emergency Physician, reducing admit time (1), and reducing laboratory time (2). Using simulation results, we developed parametric regression models. The relationships for admit patient LOS (Ta) and discharged patient LOS (Td), generated using SAS IML (SAS Institute Inc., Cary, NC), are
Figure 2. (A) Emergency Department (ED) Main bed utilization. (B) ED Main physician utilization.
1124
J. A. Paul and L. Lin
revealed below. All the regression coefficients were significant at the 95% confidence level. Ta ¼ 287:27 0:67x1 0:51x2 þ 0:0014x22
(1)
Td ¼ 149:99 0:62x2
(2)
Equation 1 indicates that a reduction in admission process wait time impacts admit patient LOS more significantly than that due to laboratory and radiology time improvements. On the other hand, only laboratory and radiology improvements have a significant impact on the discharged patient LOS. This information is valuable for prioritization of improvement measures. RESULTS Based on the DES model recommendations, the hospital added a new physician to the ED Main between 11:00 a.m. and midnight because these were generally the peak hours of the day. In a follow-up effort, ED performance was compared on days with and without an additional physician to check for improvements. Specifically, we compared the average daily patient throughput and the average patient LOS for these two time periods. The results from this analysis are presented in Table 2. As can be noted from Table 2, addition of the new physician significantly improved the daily patient throughput. In addition, a statistically significant improvement was seen in average patient LOS for ED Main discharged patients. However, the same could not be said about admit patients. To accurately estimate the impact of adding one physician on admit patient outcomes, we performed some additional analysis. The discrepancy could be due to skewness and kurtosis in LOS data collected on days when the new physician was present, or due to a significant difference in patient throughput. For days with and without the new physician, the skewness was found to be .62 and .53, respectively. Similarly, kurtosis was 3.55 and 3.40, respectively, for these two time periods. This was due to longer LOS values in the data on days with the new physician. A detailed LOS
Table 2. Comparison of Outcomes with and without the New Physician Variable Average daily patient throughput Td in minutes Ta in minutes * p < 0.05. ** p > 0.05.
Days with Days without Net New Physician New Physician Change 199 (17 days)
182 (14 days)
17*
182 (544) 320 (166)
221 (167) 316 (119)
39* 4**
breakdown for the admit patients is presented in Appendix 4. Finally, to justify the addition of the physician, we performed a simple financial analysis. Based on the data provided to us by the accounting department, it was estimated that hospital revenue would increase by approximately $5000 a day ($300 per patient, on average). This would justify expenditure on any direct or indirect job-performing resource in the long run, given the average patient volume increase of 17 on days that the physician was added. In addition to changes internal to the ED, the model also recommended changes external to the ED. These were in laboratory and radiology departments due to their significantly long turnaround times, as well as the slow admission process of patients to inpatient units. Two additional projects focusing on improvement of the process of patient admission to the inpatient floor were carried out based on findings from our project. Software for tracking patient beds was installed across the hospital. This led to additional improvements in patient LOS in the ED and also helped better streamline the process for ED patient admission to the inpatient units. DISCUSSION This study demonstrates that computer simulation is useful in modeling all operational details in a complex system such as an ED, with a great deal of modeling flexibility. The DES model is able to efficiently enact actual events using historical data that represent patients, staff, laboratory and imaging studies, and associated resources. The logic of patient moves and waiting for resources is captured by the simulation model so that resource utilization and patient LOS data are obtained. Additional programming effort makes it possible to represent ED dynamics on an hourly basis, 24 h a day. In addition to average resource utilization for each hour, representation of the fluctuation of patient volumes at different hours of the day can be achieved via statistical confidence intervals. This study shows the importance of understanding the roles played by different types of resources in the ED and shows the impact of process improvement on ED overcrowding (rather than assuming or blaming a lack of a bed as the root cause without supporting statistical evidence). The generic methodology proposed in this article would be valuable to hospitals in achieving improvement of throughput or decreasing waiting time because each hospital is different and only a detailed analysis such as that made possible by our methodology could reveal the true bottlenecks. LIMITATIONS The models in this study made assumptions regarding the participating hospital, including triage efficiency, bed
Improving Patient Throughput at Hospital EDs
allocation, and patient disposition processes by physicians and nurses. Inefficiency in such key processes could lead to patients staying longer in the ED than required for their care. A potential improvement might be to add a level of detail in recording the chief complaints, as the current model captures only severity of patient illness. Patient symptoms, treatment protocol, chance of getting admitted, and even correct diagnosis depend, to a significant extent, on the chief complaint. Further, our model and the subsequent analysis could improve by studying the disparate effects of leaving without being seen and leaving against medical advice on problems faced by the ED. In the current study, we have not differentiated between these two patient groups. Finally, effect of the mode of patient arrival on ED problems has not been studied. Any preference given to a patient based on mode of arrival at the ED in terms of severity assignment and subsequent bed allocation could significantly affect ED operations and outcomes. CONCLUSIONS We developed a generic methodology to address the overcrowding issues in the ED. We demonstrated the applicability of our methodology at a study participating hospital. With the help of the DES model results, we were able to identify the key areas of improvement internal to the ED, and also propose new projects in areas external to ED. The model also provided us with the capability to check the effectiveness of various proposed improvement strategies through the use of parametric models developed using simulation results. The results from the quantitative analysis of the crucial relationships between the factors provided the basis for additional analysis, such as financial justification on the return of investment in the improvements. Follow-up actions that measured the benefits of the model recommendations validated the usefulness of the models developed in this research.
1125 2. Solberg LI, Asplin BR, Weinick RM, Magid DJ. Emergency department crowding: consensus development of potential measures. Ann Emerg Med 2003;42:824–33. 3. Howell E, Bessman ES, Rubin HR. Hospitalists and an innovative emergency department admission process. J Gen Intern Med 2004;19:266–8. 4. Messner ED. Quality of care and patient satisfaction—the improvement efforts of one emergency department. Top Emerg Med 2005; 27:132–41. 5. Asplin BR, Rhodes KV, Crain L, Camargo CA. Measuring emergency department crowding and hospital capacity. Acad Emerg Med 2002;9:366–7. 6. McMahon MM. ED triage—is a five-level triage system best? Am J Nurs 2003;103:61–3. 7. Finamore SR, Turris SA. Shortening the wait: a strategy to reduce waiting times in the emergency department. J Emerg Nurs 2009; 35:509–14. 8. Tanabe P, Gisondi MA, Medendorp S, Engeldinger L, Graham LJ, Lucenti MJ. Should you close your waiting room? Addressing ED overcrowding through education and staff based participatory research. J Emerg Nurs 2008;34:285–9. 9. Miro´ O, Sa´nchez M, Espinosa G, Coll-Vinent B, Bragulat E, Milla´ J. Analysis of patient flow in the emergency department and the effect of an extensive reorganization. Emerg Med J 2003;20:143–8. 10. Kulkarni RG. Going lean in the emergency department: a strategy addressing emergency department overcrowding. Med Gen Med 2007;9:58. 11. Fishman GS. Springer series in operations research. Discrete-event simulation: modeling, programming, and analysis. New York: Springer; 2001. 12. Law AM. McGraw-Hill series in industrial engineering and management science. Simulation modeling and analysis. New York: McGraw-Hill; 2007. 13. Draeger MA. An emergency department simulation model used to evaluate alternative nurse staffing and patient population scenarios. Proceedings of the 1992 Winter Simulation Conference. Arlington, VA; 1992:1057–63. 14. Rossetti MD, Trzcinski GF, Syverud SA. Emergency department simulation and determination of optimal attending physician staffing schedules. Proceedings of the 1999 Winter Simulation Conference. Phoenix, AZ; 1999:1532–40. 15. Connelly LG, Bair AE. Discrete event simulation of emergency department activity: a platform for system-level operations research. Acad Emerg Med 2004;11:1177–85. 16. ProModel. Available at: www.promodel.com. 2005. Accessed January 2005.
SUPPLEMENTARY DATA REFERENCES 1. Derlet RW, Richards JR. Overcrowding in the nation’s emergency departments: complex causes and disturbing effects. Ann Emerg Med 2000;35:63–8.
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10. 1016/j.jemermed.2012.01.063.
Appendices that accompany this article are available in the web version posted on the journal website: www.jem-journal.com.
1126
J. A. Paul and L. Lin
ARTICLE SUMMARY 1. Why is this topic important? Emergency Department (ED) crowding has been a serious problem in the nation’s hospitals for more than a decade, with approximately 90% of hospital directors reporting it to be a constant issue. 2. What does this study attempt to show? In this study, we sought to develop a generic methodology to address overcrowding and patient throughput issues in an ED that can be used at other hospitals regardless of differences in size and operational framework. 3. What are the key findings? The major issues leading to ED crowding and throughput problems were: a shortage of physicians; increased duration of patient admission to inpatient units; and slow laboratory and radiology test turnaround times. Counter to conventional wisdom, a shortage of beds was not an issue in this study. 4. How is patient care impacted? The addition of an extra Emergency Physician during specific shifts resulted in an almost 18% reduction in length of stay for patients discharged from the ED Main. The bed tracking software installed in the hospital, which was based on the Discrete Event Simulation model recommendations developed in this study, resulted in additional reductions in ED patient length of stay and also streamlined the admission process.