1
Hospital Admission Prediction and Patient Outflow at the Emergency Department
Master Thesis, Msc Supply Chain Management University of Groningen, Faculty of Economics and Business
24th of June, 2019
ELIEN KLUITER Student number: 2565269 e-mail: e.f.kluiter@student.rug.nl
Supervisors/ university
dr. M. J. Land & prof. dr. J. T. van der Vaart
2 ABSTRACT
3
TABLE OF CONTENTS
1. Introduction 4
2. Theoretical background 5
2.1 The Emergency Department 5
2.2 Crowding at the ED 6
2.3 Causes of Crowding 6
2.4 Patient Outflow 7
2.5 Solutions to Delayed Outflow 7
2.6 Hospital Admission Predictions 8
2.7 Use of predictions 8
3. Methodology 9
3.1 Research design 9
3.2 Case selection 9
3.3 Data collection 10
3.3.1 Phase 1: ED processes and overall logistics performance 10 3.3.2 Phase 2: Outflow process before the intervention 11 3.3.3 Phase 3: Outflow process during the intervention 11
3.4 Data analysis 13
3.5 Validity and reliability 14
4. Results 14
4.1 Logistics performance ED 14
4.2 Comparison of LOS before and during the intervention 18
4.3 Pre- and post-intervention outflow comparison 20
5. Discussion 26
5.1 Conclusion 28
5.2 Managerial implications 29
5.3 Limitations and future research 30
6. References 31
7. Appendices 37
Appendix A: Lay-out and processes of the ED of the Martini hospital 37
Appendix B: Intervention flowchart 41
Appendix C: Questionnaires for nurses 42
Appendix D: Measurement list 43
Appendix E: Interview protocols 44
4
1. INTRODUCTION
Crowding at emergency departments (EDs) is considered a major problem for public health worldwide (Pines and Griffey, 2015). Due to crowding, non-value adding waiting times are increasing (Pines and Griffey, 2015), which results in lowered patient satisfaction rates and poor health outcomes, such as increased complication and mortality rates (Bernstein et al., 2009; Chang, Lin, Fu, McConnell, & Sun, 2017; Derlet, & Richards, 2000; Schull, Vermeulen, Slaughter, Morrison, & Daly, 2004; Wu, Zhou, Ye, Gan, & Zhang, 2015). In order to reduce waiting times, input, throughput and output interventions have emerged in literature (ter Avest, Onnes, van der Vaart, & Land, M. J., 2018; van der Linden et al., 2013). Output interventions are assumed to be the main contributors to waiting time reduction, since output factors affect the entire ED system (Forero, McCarthy, & Hillman, 2011). However, there is still no clarity on the effectiveness of those interventions (Boyle, 2015). This research aims at determining how an output intervention, predicting whether a patient needs hospital admission and sharing this prediction with inpatient units (IUs) as soon as possible after arrival at the ED, influences ED crowding in practice. In many studies, a delayed transfer of patients from the ED to UIs appears to be a major cause of crowding in EDs (Falvo et al., 2007; Olshaker, & Rathlev, 2006). Abraham and Reddy (2010) found that this is due to a lack of interaction and information sharing between different departments. More specifically, literature suggests that patient flow can be smoothened by sharing information on whether hospital admission is necessary upon a patient’s arrival at the ED, since this allows IUs to prepare and start the transfer process timely (Yen, & Gorelick, 2007). Adding to that, the need for hospital admission can be predicted quite accurately early in the ED process (Barak-Corren, Israelit, & Reis, 2017; Kosowsky, Shindel, Liu, Hamilton, & Pancioli, 2001; Peck, Benneyan, Nightingale, & Gaehde, 2012b). However, there are contradicting perspectives in the literature with regard to the effectiveness of the early sharing of hospital admission predictions. Beardsell and Robinson (2011) state that the aforementioned approach results in inappropriate admissions and resource waste. On the other hand, based on their simulation study Peck, Benneyan, Gaehde, Nightingale, and Boston (2012a) found that the early sharing of hospital admission predictions is beneficial for patient flow through the ED, while Qiu, Chinnam, Murat, Batarse, Neemuchwala, and Jordan (2015) determined quite similar benefits on the basis of data analysis. Although various studies have been conducted in this area, still no research has shown that sharing early information on predictions of the need for hospital admission is effective in practice (Uniken Venema, 2018).
This research aims to determine the effectiveness of early hospital admission prediction sharing in practice. Therefore, the following research question is composed:
5
By answering this question, this research tests the not yet tested proposition from earlier studies that fast sharing of hospital admission predictions smoothens patient flow at an ED. Based on this, it contributes to the literature on ED crowding, since it provides insights into the average durations of outflow steps and into the way in which hospital admission prediction sharing influences crowding. The implementation of such a process is highly relevant for practice, because it could contribute to the improvement of emergency care. For staff, it could diminish work pressure (Nugus, Holdgate, Fry, Forero, McCarthy, & Braithwaite, 2011), frustration (Derlet, & Richards, 2000; Olshaker, & Rathlev, 2006) and dissatisfaction (Rondeau, & Francescutti, 2005) and for patients, it could reduce waiting times and improve health outcomes.
In order to deliver those contributions, a single case study is executed at the ED of the Martini hospital, which is located in the North of the Netherlands. At this ED, an intervention comprising the communication of hospital admission predictions shortly after patients’ arrivals is implemented. Subsequently, the situations before and during the intervention are compared and analysed based on qualitative and quantitative data to determine its influence on ED crowding.
This thesis will be structured as follows. Firstly, an overview of the literature related to crowding at EDs and the prediction of hospital admissions will be provided. Secondly, the research method selection and data collection procedures will be explained. Thirdly, the research results will be described. And lastly, the results will be discussed and conclusions will be drawn.
2. THEORETICAL BACKGROUND
This section will first describe the responsibilities and activities of the ED. Thereafter, the crowding problems at the ED and the related causes will be assessed. Subsequently, the role of outflow in crowding and the possible methods to smoothen patient flow will be discussed. Finally, the potential of hospital admission prediction sharing to reduce crowding will be described.
2.1 The Emergency Department
6 2.2 Crowding at the ED
Although EDs are expected to be available when patients are in need for care (Kellermann, 1991), their ability to do so is limited by crowding (Asplin et al., 2003). Crowding can be described as the situation in which there is insufficient physical or staffing capacity to handle all patients that are waiting for triage or initial evaluation, undergoing treatment or waiting for hospitalization or discharge (Forero et al., 2011).
As a result of crowding, the time to provision of care, non-value adding waiting time and length of stay (LOS) are increasing (Higginson, 2012; Pines, & Griffey, 2015). When waiting times are excessive, EDs are even forced to divert ambulances (Schull, Lazier, Vermeulen, Mawhinney, & Morrison, 2003). This implies that the ED is temporarily not able to provide safe care for new arriving patients (Asplin et al., 2003).
Besides delay and rejection of patients, crowding has many other negative consequences. It is proven to increase work pressure (Nugus et al., 2011), frustration (Derlet, & Richards, 2000; Olshaker, & Rathlev, 2006) and dissatisfaction among ED staff (Rondeau, & Francescutti, 2005), which in turn increases staff turnover (Pathman, Konrad, Williams, Scheckler, Linzer, & Douglas, 2002). In addition, crowding leads to lowered patient satisfaction and diminished health outcomes, such as increased complication and mortality rates (Bernstein et al., 2009; Chang et al., 2017; Derlet, & Richards, 2000; Schull et al., 2004; Wu et al., 2015).
2.3 Causes of Crowding
Since the ED is only part of the total health care system, ED crowding is not only caused by the ED itself. Asplin et al. (2003) designed a conceptual model consisting of three components -input, throughput and output- in order to analyse the causes of the crowding problems.
Input factors can be described as all characteristics, conditions and events related to the demand for emergency care (Asplin et a., 2003). Factors that might increase ED crowding are, for example, high volumes of arriving patients and highly urgent and/or complex types of patients (van der Linden et al., 2013; Morris, Boyle, Beniuk, & Robinson, 2012). During the last decades, the proportion of elderly in society has increased, which caused a growing demand for emergency care (Derlet, & Richards, 2000; Morris et al., 2012). Since elderly usually need more extensive investigations, the complexity of demand has also grown (Derlet, & Richards, 2000; George, Jell, & Todd, 2006). Therefore, overall ED crowding has been increased.
7
accessibility of information about the patient (Asplin et al., 2003; Boyle, Beniuk, Higginson, & Atkinson, 2012).
Output factors are related to the discharge of patients from the ED to other hospital departments (Morris et al., 2012). Possible output factors causing crowding are long durations of diagnostics services at IUs, limited availability of inpatient beds, and inefficient patient discharge to other facilities (Asplin et al., 2003; van der Linden et al., 2013).
Although many different factors could cause crowding, not all factors are equally important. Where earlier research assumed that crowding is mainly due to input factors (Gallagher, & Lynn, 1990), current research argues that output factors are the main contributors to crowding, because they have influence on the entire ED system (Boyle et al., 2012; Fatovich, Nagree, & Sprivulis, 2005; Forero et al., 2011). 2.4 Patient Outflow
From the aforementioned output factors, inefficient patient discharge from the ED to other hospital departments seems to be the most important cause of crowding (Boyle et al., 2012; Fatovich, Hughes, & McCarthy, 2009; Schneider, Zwemer, Doniger, Dick, Czapranski, & Davis, 2001). Patient discharge from the ED is mainly inefficient due to boarders, which are patients who occupy beds at the ED after treatment until inpatient beds become available (Derlet & Richards, 2000; King, 2018; Lynn, & Kellermann, 1991; Mustafa et al., 2016). These boarders require a proportion of the ED resources (Handel, Hilton, Ward, Rabin, Zwemer Jr, & Pines, 2010.), whereby insufficient ED capacity is available to treat newly arriving patients (Asplin et al., 2003; Handel et al., 2010; Wiler et al., 2010). Previous research has described several causes for those boarding problems. Morris et al. (2012) state that boarding problems could be caused by insufficient physical bed capacity at IUs. Asplin et al. (2003) add to this that causes could also lie in limited bed availability at IUs, due to insufficient staff availability, patients who require isolation and treatment and discharge delays (e.g. cleaning delays) at IUs. And Abraham and Reddy (2010) found, on the basis of their case study in an academic hospital, that boarding problems might be caused by a lack of interaction and information sharing between the different departments.
2.5 Solutions to Delayed Outflow
8
inpatient bed capacity (Khare et al., 2009). Since discharge process improvement seems to be more effective, this research will be focused on that intervention.
There are many different ways in which the discharge process could be improved. Firstly, the (early) involvement of a senior doctor, specialist or primary care clinician in the ED process has been proven to facilitate the diagnosing process, to reduce the amount of admissions and to improve the flow of patients through the ED (Hoeben, 2017; Jarvis, 2016). Secondly, the implementation of a flow coordination program is assumed to facilitate the admission process (Murphy, Barth, Carlton, Gleason, & Cannon, 2014). Thirdly, the use of policies to accelerate the admission process, whereby specialist assessment and completion of diagnostic testing are postponed until admission, improves patient flow (Amarasingham, Swanson, Treichler, Amarasingham & Reed, 2010; Quinn, Mahadevan, Eggers, Ouyang, & Norris, 2007). Fourthly, rapid assessment of a patient’s hospital admission needs is proposed to improve the discharge process (Yen, & Gorelick, 2007).
2.6 Hospital Admission Predictions
Literature describes that rapid assessments of whether patients require hospital admission are based on predictions, since they are proven to be quite accurate (Kosowsky et al., 2001; Qiu et al, 2015). However, some researchers state that those predictions are not accurate enough. According to Beardsell and Robinson (2011), inappropriate admissions and resource waste are unavoidable when applying rapid assessment.
Several different methods are used to predict patients’ need for hospital admission. A distinction can be made between prediction models and ED staff predictions. Examples of prediction models are the regression model of Sun, Heng, Tay, and Seow (2011), the probabilistic graphical model of Leegon, Jones, Lanaghan, and Aronsky (2005) and the subsequent deepening semantic information model of Li, Guo, Handly, Mai, and Thompson (2012). An example of ED staff predictions is provided by Stover-Baker, Stahlman, and Pollack (2012), who made nurses responsible for the predictions.
Although all of the aforementioned prediction methods are proven to be quite accurate, there appears to be a difference in quality between them. Uniken Venema (2018) and Cameron, Ireland, McKay, Stark, and Lowe (2017) both found that nurses predict hospital admissions more accurately than prediction models. However, nurses seem to be only more accurate when the nurse is certain about the outcome (Cameron et al., 2017) or when the prediction model is not confident about its’ admission prediction (Uniken Venema, 2018). Given that there is some evidence for the higher accuracy of nurses, the focus of this study will be on the nurse prediction approach.
2.7 Use of Predictions
9
timely inpatient bed preparation. Peck et al. (2012a) studied the benefits of predicting whether patients need hospital admission when they enter the ED on the basis of a simulation model. They found that early prediction reduced non-value adding waiting times and improved patient flow through the ED in the simulated hospital. Qiu et al. (2015) analysed whether a bed reservation policy improved patient flow making use of medical centre data. The results suggested that the implementation of the policy might improve patient flow.
Although these previous studies suggest promising benefits, the effectiveness of the intervention is poorly evaluated (Boyle, 2015; Morris et al., 2012). Still no research has empirically tested whether a rapid assessment of whether patients should be admitted to the hospital is effective in practice (Uniken Venema, 2018). Therefore there is a need for research in this field (Asplin et al., 2003, Gelissen, 2016).
3. METHODOLOGY
3.1 Research design
In order to determine how early hospital admission prediction sharing smoothens patient flow in practice, a single case study was performed. Case studies are used to gather an in-depth understanding of a certain phenomenon in its real environment on the basis of multiple data sources (Karlsson, 2016; Robson, 2002). Since this research requires an in-depth understanding of patient flow through the ED, a case study is an appropriate research method. Given that patient flow is the main research topic in this research, the unit of analysis is the patient stream through the ED. This patient stream has been examined based on both qualitative and quantitative data, gathered from interviews, observations, questionnaires and the hospital information system. In order to investigate how patient flow through the ED could be improved an intervention was implemented. This intervention comprised the application of early hospital admission prediction sharing. By comparing and analysing the situations before and during the intervention, the way in which early hospital admission prediction and communication smoothen patient flow could be determined.
3.2 Case selection
10
The ED of the Martini hospital is primarily interesting for this research, because of its improvement drive. In recent years, the hospital has been busy accelerating patient flow through the ED. For example, a fast track has been recently implemented for quick-to-treat patients and a permanent radiologist is located on the ED. In addition, Hoeben (2017) has temporarily implemented a "quick look", which consisted of supervised early assessment of patients in order to be able to timely determine and deploy the required resources for the patient trajectory. This approach appeared to have potential to accelerate the outflow of patients from the ED, which is shown to be a significant part of the overall waiting time at the Martini hospital’s ED (Gelissen, 2016). This research delves deeper into the outflow process and the effects of a specific outflow-related intervention.
3.3 Data collection
In order to be able to analyse patient flow through the ED, data had to be collected. The data collection consisted of three phases: gathering general information on the ED processes and logistics performance, acquiring information on the outflow process before the intervention and obtaining information on the outflow process during the intervention. These three phases will be described below in more detail. 3.3.1 Phase 1: ED processes and overall logistics performance. The first step in this phase consisted of doing fieldwork in the Martini hospital on the basis of observations and asking questions. The activities of several nurses, physicians, specialists and patients of the ED were observed during a number of day shifts in order to get an understanding of the ED processes and related hospital characteristics. Questions were asked to the staff members to clarify or deepen the insights gathered from the observations. The observations and questions were used to find out how the ED process looks like, where there is a delay in the process, what consequences this delay has for the ED and IUs, whether the intervention is considered effective to reduce delay and how the intervention should be designed. The observations and answers to the questions were carefully noted.
11
3.3.2 Phase 2: outflow process before the intervention. In the second phase, data was collected on the basis of questionnaires, observations, nurse consultations and the hospital information system in order examine the patient outflow process and the reasons for delay in this process more specifically. Phase 1 revealed that the following rough steps can be distinguished in the outflow process: making the hospitalization decision, starting the bed arrangement process, completing ED diagnosis and/or treatment and discharging the patient. As HiX did not include information on all of those steps, additional data was collected on the basis of questionnaires. Between May 1st and 10th 2019, nurses were asked to record a number of time points for each patient of the specialisms internal medicine, neurology, cardiology and lung medicine. Only these patients were included, because they have the largest proportions of hospitalizations and (on average) the longest throughput times, which makes them most interesting for this research. The requested time points were the time of the hospitalization decision (of the assistant physician or specialist), the time of the first call for bed arrangement and the time the patient was ready to go to the IU. The questionnaire used to record those time points is included in Appendix C.
In order to gain a deeper understanding of the outflow process, additional observations were made. In Appendix D, the full measurement list that was used to gather data on the different patient processes was included. Nurses were asked several times to explain the collected data. Since ED staff is located in the Central Post of the ED, this place was selected to ask for explanations.
Furthermore, another dataset was retrieved from HiX for the period between May 1st and 10th 2019. In addition to information about the subjects from the phase 1 dataset, this dataset also included the time of IU known for each patient.
All the aforementioned data were collected and merged in Excel. The data gathered in this phase was also used to design the intervention, which will be described in the next phase.
3.3.3 Phase 3: outflow process during the intervention. In the third phase, data was collected on the basis of questionnaires, observations, the hospital information system, and interviews in order examine the patient outflow process and the reasons for delay in this process during the intervention.
12
Phases 1 and 2 showed that the ED process of the Martini hospital can be described roughly as follows: a future inpatient arrives at the ED, goes to a treatment or waiting room, is seen by ED staff members who triage and diagnose the patient and (partly) determine the patient’s trajectory, undergoes treatment (if necessary), is announced to the IU, and leaves the ED by a transfer service when the IU is ready to receive the patient. A more extensive description of the patient process is, as aforementioned, provided in Appendix A.
During this phase, the ED process was changed slightly. During the intervention period, all nurses were asked to record the answers to two questions in a short questionnaire (see Appendix C) after their first contact moment with patients from the four selected specialisms. The first question was: Do you expect that the patient needs to be admitted to the hospital? In case the answer was yes, the nurse was asked to start up the bed arrangement process. This means that the nurse had to communicate to the coordinating nurse that the admission planning should be called as soon as possible. Subsequently, the admission planning tried to arrange a provisional bed at the preferred IU, which was determined on the basis of patient conditions. In order to determine whether the arrangement of a provisional bed has contributed to an accelerated outflow process, the same nurse was asked to answer a second question: At what time is the patient ready to go to the IU? A schematic overview of the intervention process is included in Appendix B.
In order to gain a deeper understanding of the outflow process during the intervention, additional data was collected on the basis of observations. Similar to phase 2, the measurement list of Appendix D was used to gather data on the different patient processes. Furthermore, observations were used to determine to which extent ED staff tried to start IU bed arrangement in an earlier phase of patients’ ED processes, since noncompliance could lead to data limitations.
Furthermore, data was again collected from the hospital information system. The data concerned patients’ times of entrance at the ED, next destinations and times of leaving the ED. This data was merged with the data from the questionnaires and observations in Excel to enable an analysis of whether the intervention has led to reduced waiting time and time spend in the ED.
13 3.4 Data analysis
The data analysis of this research consisted of four stages. First, the overall logistics performance of the ED of the Martini hospital was analysed on the basis of input, throughput and output patterns. Second, the overall throughput times of patients of the measurement periods before and during the invention period were compared. Third, the durations of the outflow process steps were compared for the periods before and during the intervention. Fourth and last, the qualitative data was linked to the quantitative data.
In the first stage, the dataset obtained in phase one was analysed by calculating the patient inflow per day, the length-of-stays (LOS) of all patients and the work-in-process (WIP) per hour. The LOS can be described as the amount of time patients spend at the ED (Qureshi et al., 2011) and can be calculated on the basis of the following formula:
LOS = Time of patient discharge – Time of patient arrival
Corrections were made for patients with unrealistic LOS values, by removing the shortest and largest LOSs from the dataset. For the INT and LUN specialisms, the LOS below 20 minutes were removed, since this is seen as the absolute minimum amount of time needed to diagnose and treat patients of those specialisms. Since NEU and CAR can be hospitalized really quick in case of acute matters, the LOS below 5 minutes were removed. Furthermore, LOS larger than 1000 minutes were also deleted, because in this way all patients with a LOS reaching to infinity were captured. LOS calculations were performed for the entire data set of 2018, but also for selected patients. For example, calculations were made per destination and specialism. In this way, differences between patient groups could be determined. The WIP is the amount of patients in the system during a certain time interval. The formula of WIP is as follows:
WIP = Amount of patients being present at start of the time interval + Amount of patients entering during time interval (input) – Amount of patients leaving during time interval (output)
For 2018, the WIP per hour was calculated. To this end, the daily average number of patients being present at the ED at midnight was calculated. This was done by adding up all patients who left the ED on a different day than they arrived and subsequently, dividing this amount by the number of days of the dataset. Thereafter, the aforementioned formula was used to determine the WIP per hour. The formula shows that when the input is larger than the output, the WIP increases.
14
and largest LOSs were removed from the dataset. Subsequently, the calculated LOS values were used to compare both periods. First, the average LOS for patients of all specialisms together as well as patients from the four selected specialisms were compared. Disadvantages of averages are that they may provide distorted insights due to outliers and that they do not provide insights in the spread of values. Therefore, second and last, the LOS distributions for the same patient groups were also compared.
In the third stage, data from the observations and questionnaires were used to compare the outflow processes of the periods before and during the intervention. On the basis of the gathered data, the durations of the time intervals between the outflow process steps were calculated for each patient. This was done by subtracting the start time of a process step from the end time of that step. All calculations were converted into tables and diagrams. Thereafter, the (spread of) durations of process steps of both periods were compared. As it was not possible to collect all time stamps for each patient, there are different numbers of observations per process step.
Fourthly and lastly, the information gathered from observations and interviews was used to substantiate the findings from the quantitative analysis and to provide more depth to the understanding of the effects of the intervention.
3.5 Validity and reliability
The use of various quantitative and qualitative data sources will enhance triangulation (Saunders, Lewis, & Thornhi, 2009) and thereby the reliability and external validity of this research. External validity also slightly enhanced due to the fact that results may be applicable for other hospitals. Internal validity will be reached by frequent presence at the ED for a long period of time and careful notation of observations and interviews. Transparency in the methods used will enhance the reliability of this research.
4. RESULTS
In this section, the results of the data analysis will be described. First, a general overview of the recent logistics performance of the ED of the Martini hospital will be provided. Thereafter, the LOS before and during the intervention will be compared. Finally, more specific attention will be paid to the differences between the two periods in terms of the duration of outflow steps.
4.1 Logistics performance of ED
15
hospital, which allows the treatment of more complex patients. As another hospital in the region is in charge of the most complex patients and repels less complex patients, more patients are assigned to the Martini hospital.
On average, the amount of ED visits was 83 per day in 2018. The amount of patient arrivals varied from 51 to 121. A more extensive overview of the distribution of the amount of ED visits per day is included in Table 4.1.
TABLE 4.1
Variation in the amount of ED visits per day in 2018
Minimum 25% Median 75% Maximum
#ED visits 51 74 83 91 121
The patient arrivals were distributed according to the average daily inflow pattern shown in Figure 4.1. This figure also shows the average daily outflow pattern and the number of patients present at the ED at the beginning of each hourly time interval (Work-In-Process or WIP). Although Figure 4.1 provides insight into the average daily in- and outflow and WIP patterns, there are major differences between the patterns of different days. Those differences occur, for example, because “nobody comes by appointment”, as said by a coordinating nurse. In other words, the patterns vary due to differences in the number of patient arrivals per day (see Table 4.1) and the distribution of arrivals over the day.
Figure 4.1 Average patient flow and WIP of the ED in 2018
16
of patients is quite stable at night and in the afternoon. In the morning, flow is increasing, while in the evening, flow is decreasing. Given that the inflow exceeds the outflow in the morning, there is a high volume of patients present at the ED in the afternoon. This volume decreases again when the outflow exceeds the inflow (i.e. in the evening). The ED attempted to limit the increase in the number of patients present on the ED by setting a 2-hour target at the patient LOS in order to align the outflow as closely as possible to the inflow. However, the figure reveals that there is still room for improvement if it comes to this alignment. Improved alignment can be achieved by keeping the LOS of patients stable during the day. To this end, more insight is needed into the factors that affect the LOS.
In Figure 4.2, the daily average LOS is plotted against the number of patients arriving that day. The daily average LOS seems to correlate significantly to the number of patients arriving that day (r =
.39,
p < .01)
, which confirms the limited correlation described in previous research (Gelissen, 2016; van Manen, 2015).Figure 4.2 Average daily throughput time per daily amount of ED arrivals in 2018
In addition, the LOS of patients seems highly dependent on their conditions and the resulting ED process. One factor determining patients’ ED processes are their destinations after ED discharge. After discharge from the ED, approximately 36% of the patients in 2018 were hospitalized, 59% went home and the remaining 5% went to other destinations like external care institutions or the GP. The average LOS for all patients, home-going patients and hospitalized are included in Table 4.2.
TABLE 4.2
Average LOS 2018 per destination in minutes
All Home Hospitalized
17
The table shows that the difference between the average LOS of patients who went home and patients who were hospitalized is approximately 37 minutes. The average LOS of both patient groups differ significantly (p < .01).
Another factor determining patients’ ED processes is their specialism. The distribution of specialisms among patients is included in Appendix F. Since specialisms differ in, for example, the number and sort of tests required for diagnosis, a difference in LOS between different specialisms is expected. An overview of the average LOS per specialism is provided in Table 4.3.
TABLE 4.3
Proportion of hospitalized patients and average LOS per specialism
Specialism % Hospitalized Average LOS in minutes
Surgery 23% 161 Cardiology 42% 163 Internal medicine 50% 208 Paediatric medicine 33% 140 Lung medicine 53% 191 Orthopaedics 23% 132 Neurology 60% 170
The table above shows that there are considerable differences between the average patient LOS and proportions of hospitalized patients of different specialisms. The specialisms with the highest average LOS also appear to have high proportions of hospitalized patients. Since the specialties internal medicine (INT), neurology (NEU), cardiology (CAR) and lung medicine (LUN) have the highest proportions of hospitalizations and highest average LOS, the reduction of the average LOS of those specialisms seems to contribute most to the reduction of crowding, assuming that the differences in LOS are mainly caused by non-value adding waiting time. Therefore, this research will from now on be mainly focused on patients of the specialisms INT, NEU, CAR and LUN.
Based on figures such as the aforementioned average amount of patient arrivals, distribution of specialisms among patients and the proportion of hospitalizations per specialism, the Martini hospital calculates how much bed capacity is needed at the different IUs. However, this appears to be insufficient to ensure a smooth outflow of patients to be hospitalized and thereby a reduced average LOS, as practice shows that there are various outflow problems.
18
In addition, beds are occupied because patients wait for their medical policy to be determined or for medication verification, while they are ready to go home. Furthermore, beds are assumed to be occupied for an increasing period of time as a result of increasing patient complexity. Another cause is that beds are available, but patients cannot be received by an IU. As mentioned by ED staff, possible reasons for this are that IU staff has a break or a briefing moment, IU staff does not have sufficient knowledge and skills to be able to care for a patient of a different specialism or too many patients have to be hospitalized, whereby the work pressure of IU staff is too high. This may imply that during peak moments there may not be enough staff available in the IUs to receive patients. As mentioned by a nurse “the cleaning service is also a delaying factor. Often beds cannot be changed between noon and 1.30 p.m., whereas this is necessary at those times.” A final cause is that there are insufficient resources (i.e. staff and/or means of transport) available to transfer the patient to the IU.
When patients cannot flow out of the ED, the number of patients being present at the ED to becomes sometimes so high that the ED of the Martini hospital has insufficient capacity in terms of ED beds, staff, or other resources (e.g. cardiac monitors) available to treat the patients. One coordinating nurse mentioned that in case of insufficient capacity “safe care can no longer be delivered and the ED has to take a time-out”. This means that the ED does not accept non-urgent patients for a maximum of 120 minutes. In time-out periods, non-urgent patients therefore have to go to other hospitals. According to ED staff, time-outs occur multiple times a week. In order to be able to cancel the time-out status, “specialists are called upon to empty the ED as quickly as possible”, said a coordinating nurse. In this way an attempt is made to accelerate the outflow of patients. This calling process takes a lot of time, which in the eyes of an ED physician “could be better spent on other tasks”. Since the ED physician is busy making calls, no time could be spend on patient care, as a result of which patient LOS may be increased. In addition, this approach leads to many patients entering the IUs at the same time, increasing the work pressure at the IUs.
4.2 Comparison of LOS before and during the intervention
19 TABLE 4.4
LOS 1-10 May and 15-24 May in minutes
ALL INT NEU CAR LUN
Pre-implementation 203,1 245,2 163,4 185,1 206,9
Post-implementation 175,4 226,5 114,6 151,3 187,9
Increase -13,6% -7,6% -29,9% -18,2% -9,2%
From Table 4.4 appears that during the intervention period, the average LOS of patients from all specialisms decreased with 18 to 48 minutes compared to the pre-implementation measurement period. In order to gather more insights into the LOS difference between the pre-implementation measurement period (1-10 May) and the post-implementation measurement period (15-24 May), the frequency distribution of the patient LOS is provided in Figure 4.3 for both periods. The LOS frequency distribution of hospitalized INT, NEU, CAR and LUN patients of 2018 is also included, in order to determine whether the data gathered in the measurement periods is representative for the population.
Figure 4.3 Frequency distributions of patient LOS of 2018, 1-10 May, 15-24 May and 2018
The figure shows that there are some differences in the frequency distributions of all periods. A difference that stands out is the high proportion of the post-implementation patient LOS being below 60 minutes. Furthermore, the patterns of 2018 and the post-intervention period behave about the same. Another outstanding difference is the pattern of the pre-intervention period. The patient LOS distribution of this period seems to be shifted to the right compared to 2018 and the post-intervention, since there were higher proportions of LOS values in this period.
20 TABLE 4.5
Quartile distribution LOS 1-10 May and 15-24 May in minutes
N Min 25% Med 75% Max
ALL Pre 58 35 143 193 254 703 Post 98 22 115 166 206 492 Increase 69% -20% -14% -19% INT Pre 17 142 198 232 283 387 Post 31 23 160 201 289 492 Increase 82% -19% -13% 2% NEU Pre 10 53 67 167 227 329 Post 18 44 60 102 150 274 Increase 80% -10% -39% -34% CAR Pre 20 35 129 148 193 703 Post 27 22 121 153 183 269 Increase 35% -6% 3% -5% LUN Pre 11 92 167 210 241 338 Post 13 115 141 182 205 329 Increase 18% -16% -13% -15%
Looking at Table 4.5, almost all patient LOS quartiles seemed to decrease after implementation of the intervention. The differences between the LOS values of both periods appear to be significant for patients of the four selected specialisms together and neurology patients (p < 0.1), which was determined on the basis of an unpaired T-test. The distribution of LOS for hospitalized patients, therefore, appears to have changed for the four selected specialisms compared to before the intervention.
Despite that the LOS frequency distribution of the measurement period before the intervention deviated considerably from 2018, it cannot be said that the LOS differences between pre- and post-implementation measurement periods are entirely due to the intervention. As the amount of hospitalized patients after the implementation of the intervention was higher than before, the change in LOS cannot be explained by a reduction in patient volume, which was assumed to decrease LOS (see Figure 4.2). Also the ED nurse occupation is not assumed to have caused the LOS change, since no clear differences between the staff planning of both periods can be distinguished. According to a coordinating nurse, factors that may have affected the LOS are “the occupation of doctors and the availability of IU beds”. 4.3 Pre- and post-intervention outflow comparison
21
hospitalized patients of the pre-implementation period and 40 (of the 89) hospitalized patients of the post-implementation period were included, since only reliable data was collected for those patients. The results are shown in Figure 4.4. Although the time intervals are shown in a sequence in Figure 4.4, this sequence may not be the same for all patients. For instance, for some patients the bed arrangement process is started even before the hospitalization decision is made.
Figure 4.4 Average duration of outflow process steps 1-10 May and 15-24 May
22
measurement method, mainly times of patients who both arrived and left during the day shift were collected, which means that patients with long LOS may be under-represented in the sample.
The foregoing provided some general insight into the outflow process. However, there are major differences between the processes of patients. Therefore, several times and time intervals are discussed in more detail.
Although the average time to the start of the bed arrangement process appeared to be shortened in the intervention period, the bed arrangement process should ideally be started just as early as the prediction was made. The delay between the hospitalization decision and the start of bed arrangement can be explained in different ways. A first reason might be that some predictions were incorrect, as a result of which the bed arrangement process was not started at an earlier stage for some patients. A second reason is that nurses might be not confident enough about their predictions. Several nurses mentioned that they only arrange a bed on the basis of a prediction when they are quite sure about their prediction. The interviews showed that they tend to be more confident about their predictions when it concerns an acute patient or a patient with a DVT (deep venous thrombolysis leg), since the former is almost always admitted to hospital and the latter is almost never admitted to hospital. Although several ED nurses indicate that hospitalizations cannot be predicted in all cases, the majority of them has the impression that hospitalizations can be predicted accurately. This is confirmed by the intervention data. During the intervention, 79 predictions were made, which consisted of 23 times no, 16 times maybe and 40 times yes. Only in case the prediction concerned yes, the bed arrangement process was started as soon as possible. When the maybe predictions were considered incorrect, 68% of the predictions were correct. When the maybe predictions are disregarded, 86% of the predictions were correct. A third reason is that coordinating nurses differ in their pro-activity. According to a coordinating nurse, “One nurse is more proactive than another in communicating hospitalizations. Sometimes hospitalizations are already communicated before the medical policy is known”. A fourth and last reason could be that ED staff was not yet used to the intervention, as were strongly inclined to wait for the medical policy, before a bed was arranged. In particular during busy periods at the ED, they tend to tick to old procedures. However, “Cultural change takes time”, as stated by an ED manager.
23
Figure 4.5 Overview of logical order of outflow process steps
In Figure 4.6, the distribution of the time a patient is waiting for an IU bed, which is the time interval α of Figure 4.5, is shown.
Figure 4.6. Frequency distribution duration of bed arrangement 1-10 May and 15-24 May
Figure 4.6 shows that during the intervention period, the waiting time for a bed was more frequently zero or lower than before the intervention period. The difference in the durations of this process steps between both periods appears to differ significantly (p < .05). Thus, during the intervention, IUs were more often able to receive patients immediately. However, both Figure 4.4 and Figure 4.6 show that there were still patients waiting for a bed. In this context, it is important to note that in practice patients are sometimes only prepared to go to the IU when the patient can be received at the IU. This implies that patients who can be received at the IU first and then are ready to go to the IU may as well have waited for a bed.
24
Figure 4.7 Frequency distribution duration of bed arrangement 1-10 May and 15-24 May
Figure 4.7 shows that in the post-implementation period, the duration of the bed arrangement process was more frequently below 45 minutes. This may imply that there were more nursed beds available at the IUs during the post-implementation period. The figure also shows that in limited cases, the bed arrangement process takes more than an hour. During the measurement periods, a case occurred in which the process even lasted for several hours. When this process takes long, due to the unavailability of beds, the benefits of the intervention might be limited. Given that a patient is possibly still waiting for a bed to be ready at the end of the process, the LOS does not necessarily have to be shortened by the intervention. This is confirmed by a nurse who said that the intervention is only beneficial “when IU beds are available”.
The distribution of the outflow durations, which is the time interval γ of Figure 4.5, is provided in Table 4.6 for the period before and during the intervention.
TABLE 4.6
Outflow duration 1-10 May and 15-24 May in minutes
N Min Max Median Average
Pre-implementation 29 0 165 22 41,5
Post-implementation 28 0 92 10 19,7
25
Figure 4.8. Frequency distribution outflow duration 1-10 May and 15-24 May
Figure 4.8 confirms the insights gathered from Table 4.6. In addition, the figure shows that about 40% of the cases of both periods exceed 15 minutes outflow time. Given that only 3 to 4% of the patients waits more than 15 minutes for a bed (see Figure 4.6), this could imply that there are more delaying factors in the outflow process. A factor that might explain delay in the outflow process is the aforementioned insufficiency of resources (i.e. staff and/or means of transport) available to transfer the patient to the IU.
In addition to the figures above, that imply that the intervention can save time, the intervention is also seen as a welcome change by the ED nurses, as revealed from discussions and interviews. According to a coordinating nurse, the implementation of the intervention can be seen as a refresher, since a quite similar logistics view has been implemented in the past, which may have subsided at the ED. Furthermore, some ED nurses mentioned that the timely communication of those predictions could lead to time savings and more overview. IU nurses add to this that early communication of (provisional) hospital admissions offers them the opportunity to work ahead and prepare for hospitalizations. However, a coordinating nurse feels like the intervention could be even more effective, since the nurse stated that “there are also other influencing factors”.
26
In addition, other negative effects of the intervention have been mentioned. An ED nurse mentioned that hospital admission prediction sharing requires additional tasks to be performed. For example, an extra call has to be made to communicate the provisional admission and communication among ED staff has to be more extensive. Another negative effect mentioned was that the intervention possibly increases the work pressure at the ED, as rooms will fill up again immediately.
5. DISCUSSION
In this section, the results of the data analysis will be discussed in order to show how hospital admission prediction sharing could influence ED crowding. This will be done by linking the results to each other and to the literature.
The research results showed that the ED outflow process takes a lot of time. On average, patients in the sample had to wait more than 40 minutes between the moment they were ready to leave the ED and the moment they left the ED, which is a substantial part of the total ED process. This is consistent with the findings of Gelissen (2016), which suggest that the majority of the waiting time in the ED process is in patient discharge from the ED. According to the ED nurses, the main problem in the outflow process of the ED of the Martini hospital is that patients cannot go to the IU immediately after treatment at the ED is completed. In addition, ED staff mentioned that because no patient outflow was possible, sometimes no capacity was available in the ED or the Martini hospital to treat patients. This is in line with the literature, which describes that inefficient patient discharge from the ED to other hospital departments is seen as the most important outflow factor that causes ED crowding (Boyle et al., 2012; Fatovich et al., 2009; Schneider et al., 2001).
With the aim to prevent for ED crowding, the use of hospital admission prediction sharing was investigated in this research, as proposed by the literature (Khare et al., 2009; Yen, & Gorelick, 2007). This research shows that nurses have the impression that they can correctly predict hospital admissions in the majority of cases, but not in all. Their (79) admission predictions were correct in 68% of the cases when the "maybe" predictions were considered incorrect and in 86% of the cases when the "maybe" predictions were omitted. Assuming that some of the "maybe" predictions would have been correct, the accuracy rate would come close to the accuracy rates of 75% to 80% found in other studies (Cameron et al., 2017; Uniken Venema, 2018). However, the accuracy rates of the different studies are not fully comparable, since other studies only allowed conclusive predictions (i.e. yes or no).
27
coordinating nurses differed in pro-activity and in some cases staff tended to stick to old procedures. The 10-day period during which hospital admission prediction sharing was implemented, therefore, turned out not to be long enough to optimize effectiveness. This can be explained, as organizational change requires a lot of effort and time (Kotter, & Schlesinger, 1979).
IU nurses indicated that they appreciate to be previously informed about admissions, as this provides them with time to prepare for admissions and to work ahead. ED nurses also see hospital admission prediction sharing as a welcome change, since it reminds them of the importance of logistics, it provides them with more overview and it results in time savings. The quantitative results also implied time savings, because during the intervention patients had to wait less often for an IU bed and the average duration of the outflow process was decreased. The benefits of hospital admission prediction sharing mentioned by the nurses correspond to the literature that indicates that this approach allows for timely inpatient bed preparation, whereby non-value adding waiting time is reduced, the outflow process is accelerated and patient flow is smoothened (Peck et al. al., 2012a; Qiu et al., 2015; Yen, & Gorelick, 2007). However, this approach only seems to reduce waiting time for a bed when nursed IU beds are available, which contradicts the statement by Hoeben (2017) that this bottleneck (i.e. waiting for a bed) can be completely removed from the ED process.
Whether the implementation of hospital admission prediction sharing has caused the difference in LOS between the two measurement periods of this research is unclear, since the ED process is complex and many factors have influence on the LOS (e.g. the occupation of ED staff and the availability of nursed IU beds). This implies that the implementation of hospital admission prediction sharing could solve some, but definitely not all flow problems. Although there are different other factors that could improve flow, many of them seem to be outside of the scope of the ED. An example of such a factor is the availability of nursed IU beds, which can be caused by insufficient staff availability, patients who require isolation and treatment and discharge delays (e.g. cleaning delays and insufficient transfer resources) at IUs, as shown in both the results and the literature (Asplin et al, 2003; Morris et al, 2012). The research results implied that more IU beds were available during the intervention period, given that the duration of the bed arrangement process decreased, while more patients were hospitalized during this period. However, more information about the occupation on the IUs is needed to be able to state that this has caused the LOS difference.
28
In addition to advantages, the implementation of hospital admission prediction sharing also has disadvantages. The results have shown that preparing IU beds may be at the expense of time for already hospitalized patients, which is in line with the statement of Beardsell and Robinson (2011) that resource waste is unavoidable when this approach is used. Another disadvantage that is indicated is that it increases the work pressure, which is contradictory to the literature that suggests that this approach could diminish work pressure (Nugus et al, 2011).
5.1 Conclusion
This research was aimed at answering the following research question:
How does the sharing of hospital admission predictions between the ED and IUs early in the ED process influence crowding?
This research showed that the sharing of hospital admission predictions between the ED and IUs early in the process has led to more focus on logistics among nurses. Based on this focus, nurses have indications of patients’ treatment plans earlier in the ED process, allowing the necessary resources to be positioned in a more proactive way. For example, the IU bed arrangement process can be started at an earlier stage of the ED process. In this way, IUs are provided with more time to make capacity available and prepare for admissions (or in other words to work ahead), as a result of which IUs may be ready to receive patients in an earlier phase of the ED process. It follows that patients have to wait less for an IU bed, which might lead to a shortened outflow process for patients to be admitted to the hospital. This implies that the sharing of hospital admission predictions between the ED and IUs early in the ED process could improve patient flow and reduce ED crowding.
However, when no beds are available, the contribution of this approach to the reduction of ED crowding is limited. Moreover, most benefits associated with the approach are limited to a select group of patients, since time savings in the waiting time for IU beds might only be achieved when we are dealing with patients who are hospitalized and were also expected to be hospitalized. Given that those patients are only a proportion of the total amount of ED patients that have to be hospitalized, the benefits of the approach for ED crowding is restricted. In addition, the sharing of hospital admission predictions between the ED and IUs early in the ED process entails additional tasks (e.g. more calls and communication between nurses), which influences nurses' availability for patient care. This might have a slight adverse effect on crowding.
29
research adds to patient experience, since the reduction of non-value adding waiting times could also have implications for their health outcomes. Since the reduction of waiting time for one patient may influence the processes of other patients, this research might also have benefits for society.
5.2 Managerial implications
This research has shown that the sharing of hospital admission predictions between the ED and IUs early in the ED process can have several advantages, such as increased logistics focus and overview in the ED, increased admission preparation time for IUs and reduced waiting time for patients. Based on this, it is recommended to the ED of the Martini hospital to implement this approach.
In order to optimize its effectiveness, a number of improvements are recommended. First, assistant physicians and specialists should be involved more closely. Some specialists and assistant physicians are responsible for the bed arrangement process themselves. Since, in that case, nurses do not tend to start the bed arrangement process on the basis of their own predictions, specialists and assistant physicians should also be informed about the benefits of hospital admission prediction sharing and be involved in its implementation. Second, the agreements with IUs should be clarified. During the intervention period, the ED and IUs did not agree on in which situations beds could be reserved and how to deal with cancelled admissions. The different parties should agree on such uncertainties to optimize the effectiveness of hospital admission prediction sharing. Third, effort must be made to ensure that nurses arrange provisional beds as soon as possible. During the intervention some staff tended to avoid or postpone the arrangement of provisional beds. Since this might be detrimental to effectiveness of hospital admission prediction sharing, staff should receive more extensive training in the new approach. Fourth and last, it is recommended to extend the approach to patients from other than the four selected specialisms. Additional research has shown that within the surgical and orthopaedic specialisms the LOS of home-going patients and hospitalized patients differ the most, which implies that those specialisms may also be suitable for the implementation of hospital admission prediction sharing.
Furthermore, the ED should be aware of the limitations and possible detrimental effects of hospital admission prediction sharing mentioned below. First, the approach seems to be only effective when beds are available. Second, preparing IU beds (for patients that are not hospitalized in the end) could be harmful to already hospitalized patients. Third and last, the approach requires extra tasks and possibly increases the work pressure at the ED.
30 5.3 Limitations and future research
A first limitation of this research is that it is based on a single case study. The research is mainly based on case-specific information, which means that the results cannot be directly generalized to other hospitals. The effectiveness of the use of hospital admission prediction planning could differ between hospitals, since hospitals differ in, for example, their size, care profile and work processes. Future research in different hospitals could, therefore, add to the understanding of the effectiveness of the approach.
Another limitation was the small size of the samples used in this research. During the intervention, only a limited number of patients of the selected specialties were seen. In addition, a certain amount of data about these patients has been missing, as a result of which some patients could not be included in the sample. One reason for this was that it was not possible for a single researcher to observe all outflow-related actions and collect questionnaires outflow-related to those actions. During busy moments, nurses were not present at the central post, as a result of which it was not possible to observe them or ask them to fill in questionnaires. Furthermore, due to the absence of the researcher after the day shifts and to staff transfers, it was difficult to collect reliable data from patients who arrived during and left after the day shift. Another reason for missing data was that the staff had to get used to the new work process, causing non-compliance. Due to the foregoing, the sample sizes of this research turned out to be small. However, sample sizes should be sufficiently large to draw reliable statistical conclusions over the population. Therefore, in future research the data collection process should be managed more strictly. The extension of the data collection period or the amount of specialisms included, would allow for sufficient data to be gathered. Additionally, the deployment of multiple observers could contribute to more complete data. A third limitation of this research is the design of the measurement instruments. For reliability reasons, only nurses were asked to fill in questionnaires. They were ask to fill in various answers, including the time the hospitalization decision for a patient was made. Given that this decision was made by specialists or assistant physicians, it had to be communicated to the nurse first. As a result, the times of hospitalization decisions may not be completely reliable, as they represent the times that the nurses were informed about the hospitalization decisions. Moreover, the recording of events or activities in a questionnaire or in HiX by nurses happened in some cases a while after occurrence. This may have resulted in slight deviations in the collected times.
31
6. REFERENCES
Abraham, J., & Reddy, M. C. 2010. Challenges to inter-departmental coordination of patient transfers: A workflow perspective. International Journal of Medical Informatics, 79(2): 112–122.
Ahmed, M. A., & Alkhamis, T. M. 2009. Simulation optimization for an emergency
department healthcare unit in Kuwait. European journal of operational research, 198(3): 936-942. Amarasingham, R., Swanson, T. S., Treichler, D. B., Amarasingham, S. N., & Reed, W. G. 2010. A rapid admission protocol to reduce emergency department boarding times. Quality and Safety in Health Care, 19(3): 200-204.
Asplin, B. R., Flottemesch T. J., & Gordon B. D. 2006. Developing models for patient flow and daily surge capacity research. Academic Emergency Medicine, 13(11): 1109-1113.
Asplin, B. R., Magid, D. J., Rhodes, K. V., Solberg, L. I., Lurie, N., & Camargo Jr, C. A. 2003. A conceptual model of emergency department crowding. Annals of emergency medicine, 42(2): 173-180.
ter Avest, E., Onnes, B. T., van der Vaart, T., & Land, M. J. 2018. Hurry up, it’s quiet in the emergency department. Netherlands Journal of Medicine, 76(1): 32–35.
Barak-Corren, Y., Israelit, S. H., & Reis, B. Y. 2017. Progressive prediction of hospitalisation in the emergency department: uncovering hidden patterns to improve patient flow. Emergency Medicine Journal, 34(5): 308-314.
Beardsell, I., & Robinson, S. 2011. Can emergency department nurses performing triage predict the need for admission? Emergency Medicine Journal, 28(11): 959-962.
Bernstein, S. L., Aronsky, D., Duseja, R., Epstein, S., Handel, D., Hwang, U., McCarthy, M., McConnell, K. J., Pines, J. M., Rathlev, N., & Schafermeyer, R. 2009. The effect of emergency department crowding on clinically oriented outcomes. Academic Emergency Medicine, 16(1): 1-10.
Boyle, A., 2015. Crowding in emergency departments: guidance from CEM emphasises system-wide solutions. Emergency Medicine Journal, 32(2): 92.
Boyle, A., Beniuk, K., Higginson, I., & Atkinson, P. 2012. Emergency department crowding: time for interventions and policy evaluations. Emergency medicine international: 1-8.
32
Chang, A. M., Lin, A., Fu, R., McConnell, K. J., & Sun, B. 2017. Associations of Emergency Department Length of Stay With Publicly Reported Quality‐of‐care Measures. Academic Emergency Medicine, 24(2): 246-250.
Derlet, R. W., & Richards, J. R. 2000. Overcrowding in the Nation’s Emergency Departments: Complex Causes and Disturbing Effects. Annals of emergency medicine, 35(1): 63–68.
Falvo, T., Grove, L., Stachura, R., Vega, D., Stike, R., Schlenker, M., & Zirkin, W. 2007. The opportunity loss of boarding admitted patients in the emergency department. Academic Emergency Medicine, 14(4): 332-337.
Fatovich, D. M., Hughes, G., & McCarthy, S. M. 2009. Access block: it’s all about available beds. The Medical Journal of Australia, 190(7): 362-363.
Fatovich, D. M., Nagree, Y., & Sprivulis, P. 2005. Access block causes emergency department overcrowding and ambulance diversion in Perth, Western Australia. Emergency Medicine Journal, 22(5): 351–354.
Forero, R., McCarthy, S., & Hillman, K. 2011. Access block and emergency department overcrowding. Critical Care, 15(2): 216.
Gelissen, J. H. 2016. Emergency Department Flow Performance; What are you waiting for? Msc thesis, University of Groningen, Groningen.
George, G., Jell, C., & Todd, B. S. 2006. Effect of population ageing on emergency department speed and efficiency: a historical perspective from a district general hospital in the UK. Emergency Medicine Journal, 23(5): 379-383.
Gallagher, E. J., & Lynn, S. G. 1990. The etiology of medical gridlock: causes of emergency department overcrowding in New York City. The Journal of emergency medicine, 8(6): 785-790.
Handel, D. A., Hilton, J. A., Ward, M. J., Rabin, E., Zwemer, Jr, F. L., & Pines, J. M. 2010. Emergency department throughput, crowding, and financial outcomes for hospitals. Academic Emergency Medicine, 17(8): 840-847.
Higginson, I., 2012. Emergency department crowding. Emergency Medicine Journal, 29(6): 437-443.
Hirshon, J. M., Risko, N., Calvello, E. J., de Ramirez, S. S. D., Narayan, M., Theodosis, C., & O'Neill, J. 2013. Health systems and services: The role of acute care. Bulletin of the World Health Organization, 91(5): 386-388.
33
Jarvis, P. R. E. 2016. Improving emergency department patient flow. Clinical and Experimental Emergency Medicine, 3(2): 63-68.
Karlsson, C. 2016. Research methods for operations management. London: Routledge. Kellermann, A. L. 1991. Too sick to wait. Journal of the American Medical Association, 266(8): 1123-1125.
Khare, R. K., Powell, E. S., Reinhardt, G., & Lucenti, M. 2009. Adding more beds to the emergency department or reducing admitted patient boarding times: which has a more significant influence on emergency department congestion? Annals of emergency medicine, 53(5): 575-585.
King, B. M. 2018. Causes and Adverse Effects from Overcrowding of Emergency Departments: The Solution. Bsc thesis, Murray State University, Murray.
Kosowsky, J. M., Shindel, S., Liu, T., Hamilton, C., & Pancioli, A. M. 2001. Can emergency department triage nurses predict patients' dispositions? The American journal of emergency
medicine, 19(1): 10-14.
Kotter, J. P., & Schlesinger, L. A. 1979. Choosing strategies for change. Harvard business review, 57(2): 106-114.
Leegon, J., Jones, I., Lanaghan, K., & Aronsky, D. 2005. Predicting hospital admission for Emergency Department patients using a Bayesian network. AMIA Annual Symposium Proceedings, 1022
Li, J., Guo, L., Handly, N., Mai, A. A., & Thompson, D. A. 2012. Semantic-enhanced models to support timely admission prediction at emergency departments. Network Modeling Analysis in Health Informatics and Bioinformatics, 1(4): 161-172.
van der Linden, C., Reijnen, R., Derlet, R.W., Lindeboom, R., Van Der Linden, N., Lucas, C., & Richards, J.R. 2013. Emergency department crowding in The Netherlands: managers’ experiences. International journal of emergency medicine, 6(1): 41.
Lowthian, J. A., Curtis, A. J., Cameron, P. A., Stoelwinder, J. U., Cooke, M. W., & McNeil, J. J. 2011. Systematic review of trends in emergency department attendances: an Australian perspective. Emergency Medicine Journal, 28(5): 373-377.
Lynn, S. G., & Kellermann, A. L. 1991. Critical decision making: managing the emergency department in an overcrowded hospital. Annals of emergency medicine, 20(3): 287-292.
34
McConnell, K. J., Richards, C. F., Daya, M. R., & Lowe, R. A. 2004. Effect of increased ICU capacity on length of stay in the emergency department. Annals of Emergency Medicine, 44(4): 8.
Morris, Z. S., Boyle, A., Beniuk, K., & Robinson, S. 2012. Emergency department crowding: towards an agenda for evidence-based intervention. Emergency Medicine Journal, 29(6): 460-466.
Murphy, S. O., Barth, B. E., Carlton, E. F., Gleason, M., & Cannon, C. M. 2014. Does an ED flow coordinator improve patient throughput? Journal of Emergency Nursing, 40(6): 605-612.
Mustafa, F., Gilligan, P., Obu, D., O'kelly, P., O'hea, E., Lloyd, C., Kelada, S., Heffernan, A., & Houlihan, P. 2016. ‘delayed discharges and boarders’: a 2-year study of the relationship between patients experiencing delayed discharges from an acute hospital and boarding of admitted patients in a crowded Ed. Emergency Medicine Journal, 33(9): 636-640.
Nugus, P., Holdgate, A., Fry, M., Forero, R., McCarthy, S., & Braithwaite, J. 2011. Work pressure and patient flow management in the emergency department: findings from an ethnographic study. Academic Emergency Medicine, 18(10): 1045-1052.
Olshaker, J.S., & Rathlev, N.K. 2006. Emergency department overcrowding and ambulance diversion: the impact and potential solutions of extended boarding of admitted patients in the emergency department. The Journal of emergency medicine, 30(3): 351-356.
Pathman, D. E., Konrad, T. R., Williams, E. S., Scheckler, W. E., Linzer, M., & Douglas, J. 2002. Physician Job Satisfaction, Dissatisfaction, and Turnover. Journal of Family Practice, 51(7): 593-601.
Peck, J. S., Benneyan, J. C., Gaehde, S. A., Nightingale, D. J. & Boston, V. H. A. 2012a. Models for using predictions to facilitate hospital patient flow. In Healthcare Systems Process Improvement Conference: 1-6.
Peck, J. S., Benneyan, J.C., Nightingale, D. J., & Gaehde, S.A. 2012b. Predicting emergency department inpatient admissions to improve same‐day patient flow. Academic Emergency Medicine, 19(9): 1045-1054.
Pines, J. M., & Griffey, R. T. 2015. What We Have Learned from a Decade of ED Crowding Research. Academic Emergency Medicine, 22(8): 985–987.
