Emergency Department Flow Performance; What are you waiting for?

(1)

Emergency Department Flow Performance;

What are you waiting for?

Thesis Supply Chain Management &

Technology and Operations Management

EBM720B20/EBM766B20

Version:

3.1 Final (25-06-2016)

Supervisors:

Dr. M.J. Land & Dr. J.T. van der Vaart

(2)

Abstract

The emergency department is a complex unit in the hospital in which a large amount of factors and mechanisms determine healthcare outcomes for patients. Up to this point research with respect to Emergency Department flow performance in the Netherlands is limited to analysis of crowding. The effect of crowding emerges from the temporary or systematic lack of capacity in the ED, which cause obstructions in the system. This results in a reduced flow of patients and increased waiting times. Earlier research showed this to be partially caused by a lack of responsiveness of the ED department. By further investigating the factors and mechanisms that influence flow, this paper found certain recurring patterns in the flow of patients and found that the specialty a patient is assigned to is an important factor for the flow performance.

Work Sampling is used as a method to determine portions of waiting time and its impact on patient flow in the emergency department. By combing earlier findings and the results of this research, the road is paved for implementing improvements with respect to flow of patients in the ED.

(4)

Introduction

The Emergency Department (ED) is a complex unit, where decisions by ED staff lead to healthcare outcomes for ED patients. When the outflow of patients in the ED is lower than the inflow of patients, Work in process (WIP, within this paper: The Amount of patient in the ED) will increase (van Manen 2015). When this state is maintained for a longer period the effect is called crowding (Bellow & Gillespie 2014). Crowding is further elaborated in the next sections. If hospitals want to reduce crowding and it’s associated long waiting times, they need to improve flow through the ED. One important aspect of flow is the decision that ED staff makes with respect to the clinical process. Determining what tests and when to order them will determine the length of stay (LOS) for patients. In order to understand the effect of these clinical decisions on flow the actual cause of the waiting time for patients needs to be determined. The patient pathway is a journey consisting of several stages. This research aims to give more insight into the effect called crowding and waiting times in the ED, by determining what amount of time is spend waiting for each distinctive stage.

A large amount of literature concerning the ED is focused on either the clinical aspect or the effect of crowding. In regards to clinical outcomes, the field of medicine is continuously evolving with respect to new treatments, medication and equipment. This ongoing change demands a lot from ED staff with respect to decision making in a complex situation (Laxmisan et al. 2007; Franklin et al. 2011). On the other side of the spectrum the effect of crowding received a lot of attention (Pines et al. 2011; Asplin et al. 2003), both from the explanatory perspective considering the differences of in- and out-flow (Anneveld et al. 2013; Carter et al. 2014) as from the solution oriented perspective aimed at solving crowding (Boyle et al. 2012; Mumma et al. 2014). Some preliminary results suggest that solutions can be found in analyzing the flow more accurately and observing the influences of responsiveness in the ED (van Manen 2015; Koomen 2015).

A patient that is being treated in the Emergency Department goes through several distinct stages: Entry, triage, initial diagnosis & tests, treatment and departure from the ED. At these distinct stages, decisions with respect to the triage level, diagnosis test, treatment and departure (either home or to another department) are made. These decisions require co-operation between different medical personnel within the ED, e.g. nurses, interns, physician assistants and attending doctors (supervisors). The patient journey and the amount of time a patient will spend in the ED is determined based on the decision made by the medical professional. It is well known that the LOS is not equal to processing time, since some of the patient journey is spend waiting. In some instance waiting is inevitable (for example the results of additional test, or the availability of equipment or personnel) in other instance this waiting is caused by inefficiencies. Up to this point there is a literature gap with respect to what amount of waiting time is inevitable, and which part is not. This information seems essential in determining how to improve flow and where there is most to gain. By determining percentages and quantitatively showing the potential gains, the gap in literature should be filled and show a way to actually reduce waiting time and increasing flow.

In general the aim of this thesis is to distinguish what factors and mechanisms cause flow problems in the ED by focusing on factors and mechanisms that needlessly prolong the LOS of patients in the ED. Since this directly influences flow performance of the ED, the research question within this paper is:

(5)

By means of measuring the amount of time patients spend in distinct stages it is hoped to gain insight into unnecessary waiting time of patients in the ED. By means of this determination, the previously mentioned gap in literature should open up the possibility to determine what improvements have the highest potential for reducing the effects of crowding. By combining previous quantitative research and the insights from this paper there is a possibility that waiting times can be reduced and negative effects of crowding will decrease.

(6)

Background, Theory

This section covers the theoretical background necessary for understanding this research. The first subsection will introduce the Emergency care process and the position of the ED in the healthcare system. The second subsection will discuss flow in the Emergency Department. Flow (or the lack there of) is the main cause of crowding. Understanding the effect of crowding is essential for broader understanding of factors and mechanisms that influence flow.

The third subsection covers Waiting in the Emergency Department, the amount of time spend waiting is not necessarily preventable or wasted. A deeper and broader understanding of the differences between inevitable and preventable waiting time are given in this section.

The final subsection links the previous subsections and gives a summary of the essential understanding necessary to determine what factors and mechanisms can influence flow performance.

The Emergency Department

The Emergency Department is the primary contact point for patients in need of acute care in the hospital. Acute care is by definition unplanned and unpredictable, therefore the necessary resources to treat ED patients are hard to determine and fluctuate (Higginson et al. 2011). Combining this acute care with a broad range of different illnesses and injuries results in the difficulty of determining the necessary resources (Cohen 2013).

Primary acute care is handled by either the General Practitioner (GP) during office hours or an Ambulance Service outside of office hours in the Netherlands. The result of this structure is that most patients that are referred to the Emergency Department, are referred by the GP (Thijssen et al. 2015). A very small percentage of patients is self-referred to secondary acute care (van Manen 2015). The GP assures to a large extent that only patients in need of the extended secondary acute care, access these valuable resources.

(7)

the stage, within diagnosis and/or treatment it is possible that waiting occurs in the center of the stage or at the end.

Figure 1 Emergency Department Process and stages involving waiting

Flow performance

The solution to the problem of crowding is to create a better patient flow within the ED. Flow performance is often defined as: “The speed at which patients are transferred from one step in the care process to the next” (Drupsteen et al. 2013). This definition implies that the higher the speed of the transfer to subsequent processes in the health care process, the better the flow performance. This does however ignore the waiting time within stages, but as long as each waiting process is seen as an individual step in the care path, the definition holds.

There are three main components of flow for an ED system: The inflow measured in patients entering the ED per time unit, the work in process (WIP) measured in patients in the ED at time X and the

outflow measured in patients exiting the ED per time unit (see figure 2). This is in line with general

operations management practices as introduced by Hopp & Spearman (2011). In general, there are some commonly used metrics that quantify the amount of time patients spend in the system and how long they wait between subsequent steps in the entire process. These metrics are Length of stay (LOS) which measures from entry in the ED to discharge from the ED (Qureshi et al. 2011) and Time to Triage (TTT) which is measured from entry to the ED until triage level is determined (van Manen 2015). These and other metrics can be used to determine the amount of time patients spend in the system. The amount of patients that need to leave the ED (given a certain amount of inflow) can be determined by determining the current WIP.

Figure 2 Inflow-Outflow Model

(8)

define crowding as a state where the demand for Emergency Care is higher than the capacity available. This sounds logical but ignores the definition of capacity in this setting. If a patient has to wait longer than average, there would be too little capacity. If the WIP of the system at a certain time is lower than the amount of treatment rooms, one would not call this crowding.

According to Bellow & Gillespie (2014) ED capacity consist among other things of the amount of rooms, this however has proven to have no effect on flow or LOS (Mumma et al. 2014). This suggest that in order to reduce crowding, solutions should aim at improving flow through the ED (Hoot & Aronsky 2008; Cohen 2013). For this reason one could start to wonder which portion of the amount of time spend waiting in the process is within the stage and between two stages. Moreover, the apparent gap in literature actually quantifies the amount of waiting time in a stage and between subsequent stages. It seems essential to determine this amount in order to determine grounds for reduction of crowding.

Waiting in the ED

In the previous sub sections the ED process and flow through this process are described. Professionals aimed at improving the health of the patients run the ED. The patient pathway is determined by decisions of the professionals. Though this would seem a critical point, Emergency healthcare literature on this subject is rather limited (Perimal-Lewis et al. 2014). Most literature with respect to decisions in the ED aims at the right decisions with respect to clinical procedures (Lahr et al. 2013) whilst ignoring the effect of decisions on flow performance. Specifically how do the clinical decisions made in the ED influence the flow and waiting of patients? This perspective suggests that the personnel cannot influence the resulting flow of clinical decisions. The contrary should be true, the flow is always the result of the decisions made by the clinical staff whether conscious or not. These decisions include the type of additional test, the moment to order them and decisions/predictions of the patient path throughout the patient journey.

Waiting patients is most often caused by a lack of resources that are a result of deliberate rationing of these resources (Lucas et al. 2009). In many instances patients have to wait for the process instead of actually being treated or seen by personnel in the ED. The amount of time patients spend with hospital personnel can be as little as 25% of the Length of Stay in the Emergency department (Mccarthy et al. 2014). These values suggest that for some reason the process is not aimed at improving flow but seems to be a result of the process. This suggests there is room for improvement of these processes within the patient journey. Of course by no means the remaining 75% is wasted time, it is well possible that test or results are being performed/interpreted. Knowing what part of a stage is actually value added and what part is waste would seem imperative to improving flow within an ED.

(9)

Flow & Waiting

The findings from the previous subsections would suggest that focusing on determining the amount and fraction of waiting within the ED can lead to an increased understanding of the patient journey. Improvements in flow performance can be realized by focusing on parts of the process and increase the speed of these processes. This could contribute to reduce the effect of crowding. Though limited in its holistic view of the emergency department, Qureshi (2011) found that flow performance can be improved from this perspective, with respect to the decisions by clinicians. The mere fact that so much literature has been devoted to flow and the logistics of hospitals and the emergency department specifically should give sufficient reason to believe there is something to gain by improving the understanding of causal effect of clinical decisions on flow performance.

(10)

Methodology

The aim of the thesis is to create a better understanding of the cause of waiting time for patients within the Emergency department and how this influences flow performance. By means of a single case case-study this will be, to the best of our knowledge, the first paper in the Netherlands that will quantify the proportions of time waited within the ED by patients. The case-study method allows for greater in-depth analysis and allows for the possibility to use multiple sources of evidence and analysis techniques (Voss et al. 2002; Yin 2004). Within this paper three broad stages will be distinguished. The first stage consists of exploration of the general ED characteristics and logistic performance. The second stage consists of developing and using the Work Sampling method to determine proportions of waiting times and the last stage consists of data analysis. By means of work sampling we can distinguish broad stages of waiting activities for patients. This information is currently unknown and could give insight into the potential of activities for reduction of LOS.

Case description

The research takes place in a peripheral hospital in the city of Groningen based in the northern part of the Netherlands. It is the smaller of two teaching hospitals in the city of Groningen. The ED in the Martini Hospital is open for 24 hours per day and 7 days per week. The ED has 16 different treatment rooms varying form cardiac monitoring to pediatric emergency care (figure 3). The OND rooms are the examination rooms and de BEH rooms are the treatment rooms. In order to gather a reasonable amount of data by means of a work sampling method it is imperative that the physical distribution of rooms allows for measurements. The Martini Hospital with its long shape makes it possible to physically observe all rooms. Furthermore, staff is concentrated at a Central Post, which makes it possible to collect and validate measurements. Staff can be quickly consulted and measures with respect to waiting time can be validated, thanks to the central location of the Central Post.

Figure 3 Lay-out of the Martini Hospital ED

(11)

Data collection

Data for the primary stage of the research is based upon a combination of data retrieved from the hospital information system and observation from different perspectives of personnel in the ED. For the secondary stage the Work Sampling method was used to determine proportions of waiting time. The combination of using both quantitative and qualitative data gave more reliability to statements of practitioners and more insight into the quantitative data.

The Hospital uses a patient data management system (CHIPSOFT) to record data with respect to patients. Though the primary objective of this system is to store medical records and order tests, certain quantitative data with respect to logistic aspects can be retrieved. The aspects include age, sex, origin, specialization, starting time in the ED, movement in the ED, Departure from the ED and the destination after departing the ED for every patient. For the first analysis a dataset with data from January 1 till October 31 (2015) was used. This data set was primarily used to analyze daily patterns and search for differences in days of the week, furthermore since the MZH assigns patients to different specialties (Chirurgic, Internal, and Cardiac etc.) the differences between these were also analyzed. For further analysis a second dataset was used, which contained data from January 1 till December 31 (2015), this set was used for retrieving information about supervisors and LOS.

Data for the secondary stage of the research was gathered by means of a Work Sampling method. The necessary data should contain information about the distinction between the stages patients are in. However, information concerning the amount of time spend within a stage is currently not recorded within the patient management data system. Since the Work Sampling method allows for determination of fractions of time spend within a stage, the method could be translated for use on waiting times in this case. This application shows a way of exploiting a proven method within a new setting. Since the work sampling method has proven to be effective in sampling work time fractions (Robinson 2010), It would seem appropriate to determine the amount of waiting for specific tasks within the patient journey as well.

The primary step in conducting a work sampling study is to determine what activities a patient can wait for during his/her patient journey within the ED. By means of 4 different observation days with; the nurse coordinator, a nurse, an assistant physician of internal medicine and a supervisor in internal medicine these tasks and stages within the patient journey were determined. This resulted in a slight alteration with respect to the theoretical figure 2.

In figure 4 it can be observed that there are 5 distinct waiting moments. The first one is waiting for the 1st_{contact with the physician assistant. The second one is waiting for the moment a Test is performed.}

The third one is waiting for the results of tests are known. The fourth one is waiting for the discussion with the supervising doctor, such that a treatment plan and/or decision about admission is made. The fifth up until the patient has left the ED. Though the figure suggests that all these stages are sequential, this is not always the case for each patient. It is possible to have tests taken before the first contact with the physician. The stages however make a distinction in the journey. The average patient takes and gives distinctively measurable stages necessary for work sampling.

(12)

The next step is designing the work sampling study, which means creating observation forms and determining what to include within a sampling and what to exclude. Since one observer conducted the sampling it was decided to only include patients within the hallway that was observable. This resulted in a form (appendix 1) that gave the possibility to record the waiting stage of each patient in rooms OND1-OND4 & BEH1-BEH8, all other rooms and the waiting room were excluded from the measurement because it would be impossible to physically observe these patients. Combining the observability of the ED with just one observer showed that in order to collect all data for the patients in case of a full ED would take 20-30, for this reason the sampling interval was chosen to be 30 minutes. In order to determine the amount of necessary samples for a maximum 2% absolute error within a 95% confidence interval, an estimation of the largest percentage for a category had to be made. Since this is the first Work Sampling study in an ED aimed at waiting by patients, no previous values could be used. In consultation with the health care professionals it was estimated that with these 5 main stages none of them would really account for more than 20% and therefore the necessary amount of observation should be N=1084. This seems reasonable, as few activities take longer than 1 hour and would thus most likely only be measured once within a patient journey. For a LOS of 130 minutes, between 4 and 5 measures are taken per patient and therefore 20% sounds reasonable. It was roughly estimated that in order to achieve this amount of data between 10 and 15 days of 8-hour measurements would be necessary.

(13)

Data analysis

Primarily the analysis of data should answer the general research questions with respect to factors and mechanisms influencing flow. In order to structure the answering of this question a couple of sub-questions are formulated. As well as defining how these sub-sub-questions were answered within the research, the purpose is stated, assumptions are mentioned and limitations are identified.

Case specific patterns

The first analysis focused on answering the question: “What are the characteristics of the case with respect to flow, and which patterns can be distinguished?” The first dataset was used to determine both LOS and patient amount. In order to have some basic understanding of the patterns that can be observed within the research time-path (04-2016 till 06-2016), only weekly patterns were observed. For this purpose the median, 25th_{, 75}th_{, minimum and maximum values of amount of patients in a}

weekly pattern were generated for the entire dataset. For the LOS only the median, 25th_{and 75}th

percentile were shown.

Since earlier research had shown that amount of patients and average LOS are not correlated, an extra analysis that compares perceptual differences in both LOS and amount of patients per day was compared to the overall median for both. In this last comparison it should be noted that since the absolute values differ greatly a perceptually lower difference is observed for LOS.

Flow performance

The next analysis focused on the flow patterns observed within the ED and tries to answer to following question: “What is currently the flow performance of the Martini ED and what factors seem to influence it?” In order to answer this question, several analyses were performed. Firstly, the flow pattern of the ED over the first dataset was constructed, by means of determining the in- and outflow per hour block, i.e. how many people entered or left the ED within an hour. This value would then be divided over the amount of days within the dataset to determine averages for specific hours. The purpose of this analysis is to quantify flow on a long-term average in order to make comparisons with other hospitals and to understand the daily pattern within the ED. One drawback of this method is that it gives only a long-term average and ignores the variation patterns that occur on a weekly basis. To compensate for this problem flow patterns for busy and calm days were compared, which will be explained later in this section.

Theory showed that the most important metrics for understanding crowding include not only flow but work in process (WIP) as well. In order to see the daily WIP pattern it is essential to determine the average amount of patients that are in the ED at 00:00. This was done by cumulatively determining the amount of patients that enter in one day (before 00:00) and leave the next day (after 00:00) and dividing this number by the amount of days in the dataset. For the next hours the inflow per hour was added and outflow per hour subtracted to determine WIP.

(14)

Earlier research in the northern part of the Netherlands with respect to ED flow showed that responsiveness is a major factor contributing to the lack of flow and crowding (van Manen 2015; Koomen 2015). The papers however did not show the potential gain that could be achieved if responsiveness was increased. In order to actually quantify this gain a desired flow was constructed. The desired outflow represents a time of two hours after the inflow of patients starting at 09:00. The inflow is trailed by the desired outflow with a lag of two hours starting at 9:00 . Based on this desired outflow a desired WIP is constructed by adding the inflow for each hour and subtracting the desired outflow. The lag of two hours was chosen since this closely mimics the LOS of an average patient in the Martini ED. The assumption with this simulation is that the LOS remains a constant 2 hours over the day. In figure 6 the expected gain can be visualized, if the outflow is at an earlier moment in time, LOS is smaller than when outflow is at a later moment in time.

The Martini Hospital assigns each patient to a specific specialty (e.g. Surgical, Internal and Cardiac), based on the primary described condition. Since this distinction in patient mix is very distinguishable, it would seem imperative to see whether there are differences between the specialties that show how the buildup of the average patterns of flow & WIP are, when looking at specialty as a variable. Since the specialty could be one of the factors influencing flow, the flow patterns of the largest 3 specialties are compared. In order to make a proper comparison in pattern, the lines are scaled with a factor that assures the maximum value to be 1. This limits comparability with respect to absolute amounts of flow, but allows for greater comparison in patterns.

Since a weakness of the simulation of desired flow for the ED, was the assumption of a constant Length of stay for patients. It seemed important to determine whether this assumption holds under current flow. By means of cumulative flow it was possible to determine an Expected LOS for patients entering the ED at any given moment in time. By splitting the LOS values for the specific patient groups once again, it is possible to distinguish between patterns for this variable.

Since the primary flow patterns are based on large dataset averages, some of the factors within the dataset are ignored. Since the problem of crowding is most apparent on busy days, it would seem important to distinguish between busy days and calm days. In order to make this distinction, the dataset was split over the median of amount of patient inflow in the morning. In this way all days with more than the median amount of patients inflow between 09:30 and 12:30, were called busy days and all days with an equal or less amount than the median were called calm days. Though this seemed promising, it ignored the distribution of amount of patients per day, i.e. weekend days have a lot less patients to start with. Since capacity is different in the weekends as well, it was chosen to make the split of calm and busy days based on only weekdays, as this allows for a fairer comparison in patters of flow. As with the previous determination, WIP was determined by using the average amount of patients in the ED at 00:00, adding hourly inflow and detracting hourly outflow.

(15)

Supervision Capacity

The Martini hospital has a teaching structure, which implies that all doctors are under the supervision of an attending doctor (supervisor). Primarily it was insinuated (by hospital staff) that the person supervising might be correlated with the throughput time of patients being treated whilst this doctor has supervision duties. In order to determine whether this is true we would like to answer the following question: “How is a specific supervisor related to the LOS of a patient?”

In order to answer this question, the department provided a list of supervisors that were thought to be more physically present at the ED and thus have lower throughput times for patients, as the waiting for this supervisor would be shorter.

Extra data cornering the supervisor was retrieved from the system and the median, averages and normality of distributions of throughput times for patient with a specific supervisor were determined. Next, the values were tested by means of unpaired-sample t tests in order to determine statistical significance in the differences recorded.

Work Sampling

The amount of outflow determines the amount of time patients spend in the ED, it thus seems imperative to understand the fractions of time patients spend during their LOS as this is closely tied to outflow. The current patient data management system was unable to produce information about the division of time spend within the ED. For this reason Work Sampling was performed with the goal to answering the following question: “What is currently the proportion of LOS spend within a predefined stage?” Answering this question could point into fields of improvement with higher potential than others.

The work sampling data was used to determine portions of waiting times for specific stages in the ED stay. In order to determine whether different patterns would emerge the dataset was split at 13:30 to distinguish between the distribution in the morning and the afternoon. This pattern seems interesting as the daily pattern of flow and WIP show different values for this distinction of morning and afternoon as well.

Next to these distinctions based upon time, the dataset was split over the largest patient specialties in order to understand how the earlier distinguished differences would translate to the differences within the Work Sampling results.

Furthermore, for each category it was determined how often it occurred that the same stage was recorded more than once. This gives insight into fractions of length for a specific stage and could thus be translated to a potential improvement for patients that go through this stage and thus deals with the interaction effect of the results.

Qualitative support

(16)

Results

This section will discuss the results. First, the datasets and alterations are discussed. Next, the case specific patterns are shown in order to build a base for comparison with other cases. Next, the flow performance is gradually shown in increasing complexity in order to create a clear overview of relevant aspects and factors influencing flow. The supervisor and the associated LOS for patients are shown next. Finally we conclude with the Work Sampling Analysis. First, the distinctive stages are discussed and the primary result is presented. Next, the differences between morning and afternoon are presented and finally the occurrence of repeated samples is shown. In order to understand how the figures and results were obtained we refer back to the Data Analysis subsection of the Methodology section.

Data Analysis

Since analysis was performed on different datasets with data retrieved from the same hospital information system, each dataset will be discussed briefly. The first dataset (D1) contained data between 01-01-2015 and 31-10-2015. This dataset had a count of 20680 patients distributed over an amount of 293 days. This dataset was provided by earlier research done and was adjusted for error values (van Manen 2015). Though mostly complete, some triage dates were not recorded within the system. After deletion of the 2,5% of longest Lengths of stay, which captured mostly errors in throughput time, 20.560 patients visiting the ED were included in the data set.

The second dataset (D2) contained data from 01-01-2015 till 01-04-2016, in order to make future comparability on yearly basis easier and remove bias over the first quarter of a year, data between 01-01-2015 and 31-12-2015 was used. From the entire dataset patients with throughput times larger than 1000 minutes were deleted, this was the case in 7 instances. The entire dataset contained 27430 patients distributed over an amount of 365 days.

For different types of analyses either one of both datasets were used for the analysis. No analyses were repeated after the availability of dataset D2. Though some differences in the most important metrics were found (see table 1), it is believed that the gain in accuracy would not significantly increase the understanding and accuracy of the conclusions drawn upon the different datasets.

Dataset 1 Dataset 2 Difference

Median LOS 130 minutes 129 minutes 0,8%

Average LOS 143 minutes 141 minutes 1,1%

Standard deviation LOS 83 minutes 76 minutes 8%

CV 0,58 0,54 6,9%

(17)

Patient variability

Since patient input is highly variable in the clinical dimension, this subsection focuses on determining variability in order to create understanding of the patient variability within the Martini Hospital. With respect to the specialty of patients the largest group falls within surgical medicine with 40% of all patients being assigned here. The three largest patient groups (Surgical, Cardiac and Internal medicine) account for roughly 69% of all patients in the emergency department (Appendix 2).

The origin of the patients is by a majority of 49% the GPs’ office and 18% 911 calls. The amount of self-referrals is limited to 5% of all patient visits. The remaining 28% have other origins (Appendix 3). After entering the emergency department the patients are triaged by means of the Manchester triage system in order to determine the acuity of the patient. Most patients are triaged yellow, which means they have an average acuity of 3 on the 5-point scale and can thus wait up to 60 minutes before being seen by ED staff (Appendix 4).

After being treated within the ED, on average 52% of patients leave the ED and return home while 48% is hospitalized to either one of departments or the Observatory in close proximity to the ED (Appendix 5).

Amount of patients per day

The inflow of patients cannot be planned, because of the sudden onset for the need of ED services. It is however possible to distinguish a pattern in the amount of arrivals on a daily basis. This makes for both unplannable arrival of patients but not completely unknown arrival.

In figure 7 it can be observed that compared to the total median of arrivals over all days, on average the median for both Mondays and Fridays are roughly 10 patients higher. The calm days occur in the weekends when not only the median but also the 75th_{and 25}th_{percentile are much lower.}

(18)

Furthermore, it can be observed that the spread of values differ among the days. This was in line with calculation of both the standard deviations and the CV (coefficient of variation) values. For the entire dataset μ=71 and σ=13,4 giving a CV=σ/μ=0,19. For the other weekday values of σ ranged between 9,0 on Sunday and 11,9 on Tuesday. The CV values were counter intuitively smallest for the busiest days with CV=0,12 on both Monday and Friday.

These insights can be used to make accurate predictions of the amount of patients per day and thus the capacity needed. Because of the earlier determined 48% hospitalization of patients, the amount of free beds needed within the hospital per day should be estimated fairly precisely as well.

Lengths of stay per patient per day

Comparable to the findings of amount of patients per day are the throughput times or LOS (lengths of stay) of patients within the ED. In figure 8 it can be seen that though the deviations of median values over the week differ insignificantly, the spread of values is larger. The standard deviation of LOS was found to be σ=75 minutes over the entire dataset giving a CV=0,53. For the other weekdays, values of σ ranged between 76 minutes on Wednesday and 69 minutes on Sunday. Though the spread and thus CV is higher for LOS, the differences in both LOS and Standard deviation do not differ a lot between the different days.

(19)

Correlation Amount of patients and Length of stay

In order to understand whether there is a relation between the amount of patients and the throughput time, the differences between them are shown in figure 9. The blue bars show the perceptual difference in median value of patients for that day compared to the median of the total data set. The orange bars show the percentage difference in median values of LOS for that day compared to the median of the total data set. From this figure we can confirm once again that the Monday and Friday are the busiest days, but that there seems to be little correlation between the amount of patients and the LOS of patients for that specific day. These findings are in line with the findings of earlier research done in the Martini Hospital (van Manen 2015).

(20)

Flow performance

Within this subsection the focus is on understanding flow performance and the factors influencing it. By looking at inflow, outflow and WIP under certain circumstances, some of these factors are recognized.

Flow

In figure 10 the daily average flow pattern of the Martini Hospital is shown. The time brackets are shown on the horizontal axis. The amount of flow is shown in patients per hour on the vertical axis. Within the figure it can be observed that inflow of patients is divided in four rough stages. The first stage is at night with relatively low in and outflow between 00:00 and 08:00. The second stage is when the ED starts to fill up between 08:00 and 12:30. The third stage is between 12:30 and 19:00 where both in-and outflow remain fairly constant and the fourth stage between 19:00 and 00:00 where both in- and outflow decrease.

Figure 10 Long term average flow pattern Martini ED

(21)

Figure 11 Long term average Flow & WIP pattern Martini ED

In previous research in the EDs in the north of the Netherlands, most of the associated analyses were aimed at the responsiveness of the ED to follow the patterns of inflow as directly as possible in order to reduce the buildup of patients in the ED.

In figure 12 the daily flow pattern and WIP in the entire ED is combined with the flow and WIP in case of following more responsively to inflow. This figure adds the desired outflow to the graph at lagging 2 hours behind the inflow of patients, comparable to the median LOS. This greater responsiveness leads to a reduction of the amount of patients within the ED. For the third stage, which in general is seen as the busiest time of the day, this reduction of WIP is on average 23%. The WIP values were determined by determining the amount of patients on average in the system after 00:00 and adding the hourly inflow and detracting the hourly outflow as previously described within the methodology section.

(22)

Patient Specialty

Patient specialty plays an important role within the Martini Hospital. All patients that are announced are immediately assigned to a specialty based upon preliminary determination of illness. The patient pathway is determined to a large extent by this distinction. The assigned capacity available per specialty differs since the amount of patient per specialty differs. Within table 2, the most important flow metrics are given per specialty. The rest of this subsection will focus on de differences between the specialties in both flow and LOS in order to determine how this factor influences flow performance.

Total (TOT) Surgical (CHI) Internal (INT) Cardiac (CAR)

Average LOS 143 141 168 141 Median LOS 130 126 160 130 Standard Deviation 83 79 117 63 Patient Percentage 100% 40% 14% 14%

Table 2 Differences between patient specialties

In order to distinguish between patterns, the inflow of patients with a specialty is shown compared to the total inflow in figure 13. This figure clearly shows that there are differences between the patterns. The total pattern shows the largest similarity with the surgical (CHI) pattern. The Internal pattern seems to gradually build to one peak 15:00 and decrease gradually after this peak whilst cardiology builds steeply up to a peak at 11:00 and slowly decrease as the day progresses. Though this scaled figure allows for comparability, the absolute values explain the average patterns of surgical medicine in the total pattern.

(23)

The same analysis is performed for the outflow of patients (figure 14). The total pattern and the surgical pattern show great similarity. Internal medicine peaks at 18:00, roughly one LOS period behind the peak inflow at 15:00. Both figures clearly show differences between the patterns and thus, influencing flow performance in general should look at how the flow influences each of these specialties.

Figure 14 Scaled outflow of patients per specialty

Because of the differences in flow patterns for different specialties, the WIP levels for the specialties are shown as well (figure 15). As expected, the figures show comparable patterns to the flow for each specialty.

(24)

In figure 15 it can be noted that there exists a comparable pattern between surgical medicine and the total pattern with a fairly steady high level between 12:00 and 17:00. An early peak characterizes cardiology with a graduate decrease over the day and a peak for internal medicine at 16:00. In general it can be concluded that the specialty and variability of patients plays an enormous role in the inflow pattern of patients, but outflow and WIP patterns as well.

For previous determination of a desired outflow and WIP, the assumption was made that LOS after 09:00 would be 120 minutes. In order to test this assumption and determine differences in LOS buildup, the expected LOS for a patient entering at time X are calculated. In figure 16 it is observable that in no way the expected LOS is constant. Most noticeably, all specialties have a scaled but similar pattern with a peak between 05:30 and 07:00. This peak is explained by the handovers of the night shift and the day shift occurring at 08:00. All patients that start a treatment before 08:00 have a 30 minute longer LOS. Furthermore each line shows peaks that occur at 12:00 and 18:00. A possible explanation for these peaks is the regular moments when GPs do their rounds at patients, and thus the inflow increases. Remarkably, these patterns do not differ significantly whilst the flow patterns do. The simulation of desired WIP was not adjusted for these finding as it would marginally influence the conclusion drawn from it and lies beyond the scope of this research.

Figure 16 Expected LOS for patients entering at time X

Busy and Calm days

(25)

Figure 17 Differences in flow on buys and calm days

In figure 18 the differences in WIP buildup between busy and calm days can be observed. This figure assumes that the WIP levels at 00:00 are equal for both days. Though this assumption is wrong, the determination of the actual amount would give little extra insight into the buildup of WIP. The most noticeable pattern is the relatively small difference between the maximum WIP. Since the sets were split on inflow in the morning, the busy patterns shows a larger WIP earlier in the morning and remains fairly constant throughout the day, whilst the calm figure slowly builds up to the peak at 17:00.

(26)

Supervising capacity

One of the contributing factors the hospital recognized as being influential for the LOS and patient journey is the amount of time the physician (assistant) has to wait for its supervisor. This is because in the Martini Hospital all patients regardless of their specialty are always assigned to a supervisor within their respective specialty.

The assistant physicians & ED-physician therefore have to consult this supervisor for every patient in order to finish the patient pathway (either by phone or physically). Since every patient has to go through this stage, the question arose whether there is a link between the specific supervisor and the median and average LOS of patients being treated when a specific supervisor was in charge. In order to determine differences that would have the largest impact, the Internal Medicine was chosen to determine the impact as they have the longest average LOS.

Internal medicine has 14 distinct supervisors. All supervisors have shifts in which they accept all patients for their specialty. This assures there is little clinical variability between the patients for each of these supervisors. The average amount of patients seen by a supervisor within a 16-month period was 366 patients (range 230-447). the Average LOS was 164 minutes and the Median 161 minutes. In figure 12 the differences in LOS per patient per Supervisor are shown. The smallest median value is 147 minutes and the largest 187 minutes. For the sets independent samples test were performed to assure statistical significance and for all dataset they were found to be statistically significant with P<0.05.

The difference in LOS is explained by the availability of the Supervisor during his/her shift. Often supervisors have to perform multiple tasks during the shift and therefore have to divert attention between the ED and other obligations. It does however show that a large gain in reduction of LOS can be achieved by means of determining the factors contributing to reduced LOS for certain Supervisors.

(27)

Work Sampling

In order to determine what percentages of time patients are waiting within a specific stage in their treatment, work sampling is used. The work sampling method yields insight into this division. The work sampling was performed between the 15th_{of April and 18}th_{of May 2016. A total of 1210 samples were}

taken within this period, of these 1093 had reliable data concerning the waiting activity. In the other 117 cases it was impossible to measure within a frame of 10 minutes from the onset of sampling what the patient was waiting for and therefore is seen as an invalid data point. After removing this invalid data, these adhere to the patient journey as described in figure 4. The corresponding amount of data and percentages are given in figure 20. For elaborate explanation of the stages, we refer to the Appendix (page 38).

For the amount of given data points an absolute maximum error percentage of 1,8% with unknown data and 1,99% without unknown data was achieved according to standard sample size determination (Robinson 2010).

Figure 20 Results work sampling patient waiting categories

Since the measurements tell in what stage of waiting the patients are during the sampling, this can be used to determine what percentage of waiting is spend in a specific stage. It can be observed in figure 20 that outflow accounts for the largest percentage of average waiting in categories. This is partially explained by a lack of available beds in the hospital that cause patients to have to wait for this availability within the ED.

(28)

Though these limitations exist, the work sampling method allows for the current division of stages for patients and gives a base to build interventions and have a (quantitative) measurement, which can be used as a comparison for the (possible) intervention.

Morning vs. Afternoon

Because of the size of the dataset of work sampling, it was possible to distinguish between morning and afternoon. Since the flow patterns, LOS and WIP levels are shown to be different for the morning and the afternoon.

Figure 21 Difference in waiting time between morning and afternoon

Though figure 21 shows some differences in the patterns between the morning and the afternoon it should be noted that the sample size for half the data has an absolute error margin of 2,8% for the largest category. This means that though patients appear to have to wait longer for supervision in the morning the difference in percentages could be errors in the measurement. This is all within a 95% confidence interval and therefore it is possible to draw conclusion based upon the differences. Overall it can be seen that the five major stages that can be influenced by the ED (1st_{contact and}

supervision) seem to decrease whilst the other three stages (test, results of tests and outflow) seem to increase in the afternoon.

The stage of “no waiting” decreased in the afternoon, however since LOS increases in the afternoon the absolute amount of not waiting should remain fairly constant.

Specialty

Since the differences in flow patterns for specialties showed this variable to be a very important factor in flow performance, there is more than sufficient reason to believe there are differences in the buildup of waiting between the specialties as well. In figure 22 the differences between the specialties can be observed. In figure 22 it should be noted that the absolute error margin is greater for the specialties (maximum value for cardiology 4,8%).

(29)

these tests are performed, they take longer. The outflow of patients (or the lack there of) is mostly caused by the lack of available beds in the rest of the hospital for these types of patients. These findings seem very likely as 71% of cardiac patients are hospitalized compared to an average of 50%.

Another difference that occurs is the increase in the category of additional test for surgical patients. This is most likely caused by the amount of Sonography, by means of Echo. These test were found to take a large amount of time before the test is performed.

Figure 22 Difference in waiting time between specialties

Occurrence

It should be noted that in figure 20, 21 and 22, there is an interaction effect. This in essence means that though the average amount of measurements (e.g. 4,8% for consult total) is the average waiting time. So patients who have a consultation will most likely have a larger amount of their LOS dedicated to consult than the average. The same holds, for patients who did not need a consult and therefore have 0% of waiting for consult.

In order to examine the effect of these interactions, figure 23 shows what percentage of categories within the work sampling study (total amount of patients n=472) occurs more than once. Furthermore it shows the amount of measures per patient.

(30)

(31)

Discussion & Conclusion

This research focused on the factors and mechanisms explaining flow performance. Factors like daily flow patterns were investigated in order to determine possible causes of waiting and crowding in the ED. In order to further quantify the proportions of waiting in the ED, this research used a first modification of standard work sampling for this purpose. The insights gained throughout this research are discussed and concluded within this section.

In general the Emergency Department is a complex environment where a lot of variables determine both the clinical outcomes of patients and the logistical performance associated with it. Throughout this paper several of these variables (factors & mechanisms) are discussed, ranging from time of day, responsiveness, capacity and specialty. Each of these factors shows its influence in the logistical performance. This research found that the most important factor is the subdivision of patients into specialties. This deliberate division of patients with similar clinical problems into distinct groups led to specific characteristics for these groups. These characteristics were evident in the patterns of flow and the difference in waiting time components.

The factors influencing flow within the ED sadly do not point to one mayor culprit for the problems of crowding. It does however show that several factors each have their own influence on flow and waiting in the ED. This research not only expanded upon previous research performed in the northern Netherlands but quantified the extent of waiting in the ED into distinct stages as much as possible. The insights gained throughout this research, should form a solid ground for further research to move from analysis to design-science and build/improve upon the current mechanisms used within the ED. The next step in solving or at least reducing emergency department crowding is to actually find proven methods that can lead to the increase of outflow. This research has provided an extensive before picture and re-developed the tool (Work Sampling) to determine the effects of possible interventions within the ED. Generally the variability of the patient by means of the specialty assigned is of large influence on the journey the patient will experience within the ED. By assuring that the proper solution is implemented within the proper specialty the entire ED can benefit and beyond this level even the entire hospital can benefit.

Emergency Care is one of the most important components of health care and assuring the patients get what they deserve from both a clinical and logistical perspective should not be mutually exclusive. Patient care should never suffer under the logistical consequences, but neither should extensive care be an excuse for bad logistical performance. There are no simple solutions for ED crowding and if there were, they would have been exploited already.

This paper gives a base by explaining factors and mechanisms for the performance. At this point, it is up to practice to determine what aspects to use and how healthcare in the ED can become more logistically oriented toward creating flow.

(32)

Recommendations

This final section will conclude this paper by looking at practical recommendations aimed at the martini hospital and suggestions for further research in the field of ED logistics.

Martini Hospital

Since this hospital is the peripheral of two hospitals, (in general) less urgent patients visit the ED. Because of this lowered acuity it might be possible to assign the patients arrival slots in order to try and influence the inflow of patients into the ED.

Since Logistics and crowding in the ED are becoming increasingly important in the ED, now should be the moment to actively involve the practitioners within the field of ED flow performance and logistics in general. Though logistics should never have priority over patient safety it could be argued that improved logistics, improve patient satisfaction and safety as waiting times can be reduced. Furthermore the practitioners could benefit from the decrease in workload as a result of better logistical performance within the ED.

As mentioned in the previous section, the stage of investigation into effects should be sufficient to start implementing changes that could actually improve patient flow and thus improve overall ED performance. By means of work sampling, awareness within the ED was created and practitioners have become more aware of their influence on logistics and the concept of time with respect to the patient journey. This should be materialized and put to use in actual changes within the ED to actually see whether the predicted gains can be achieved.

Further Research

Within this paper, the work sampling method was found to be a great tool to measure patient waiting proportions. Though some limitations were found within the design of the study (sample size for specialty, amount of rooms measured and interval of measurement) the potential seems great. For future research these shortcomings could be solved and the true potential of the method might lead to new insights.

This research was performed within a medium sized peripheral hospital in the northern part of the Netherlands. It would be interesting to see whether size or location influence the findings with respect to outcome for waiting time distributions. Furthermore, it would seem beneficial to see the differences between the two hospitals within the city of Groningen in order to determine best practices or assign division.

(33)

Literature

Achteren, G. van, 2014. Care for waiting: the effect of emergency department responsiveness on patient waiting times. University of groningen.

Anneveld, M. et al., 2013. Measuring emergency department crowding in an inner city hospital in The Netherlands. International journal of emergency medicine, 6(1), p.21. Available at: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3711920&tool=pmcentrez&render type=abstract.

Asplin, B.R. et al., 2003. A conceptual model of emergency department crowding. Annals of Emergency Medicine, 42(2), pp.173–180.

Bellow, A.A. & Gillespie, G.L., 2014. The evolution of ED crowding. Journal of Emergency Nursing, 40(2), pp.153–160.

Boyle, A. et al., 2012. Emergency department crowding: time for interventions and policy evaluations. Emergency medicine international, 2012, p.838610.

Carter, E.J., Pouch, S.M. & Larson, E.L., 2014. The relationship between emergency department crowding and patient outcomes: A systematic review. Journal of Nursing Scholarship, 46(2), pp.106–115.

Cohen, S., 2013. Perspectives on Emergency Department Throughput. Journal of Emergency Nursing, 39(1), pp.61–64. Available at: http://dx.doi.org/10.1016/j.jen.2012.10.002.

Drupsteen, J., Vaart, J.T. Van Der & Donk, D.P. Van, 2013. Integrative practices in hospitals and their impact on patient flow. International Journal of Operations & Production Management, 33(7), pp.912–933.

Epstein, S.K. et al., 2012. Emergency department crowding and risk of preventable medical errors. Internal and Emergency Medicine, 7(2), pp.173–180.

Franklin, A. et al., 2011. Opportunistic decision making and complexity in emergency care. Journal of Biomedical Informatics, 44(3), pp.469–476. Available at: http://dx.doi.org/10.1016/j.jbi.2011.04.001.

Higginson, I., Whyatt, J. & Silvester, K., 2011. Demand and capacity planning in the emergency department: how to do it. Emergency medicine journal : EMJ, 28(2), pp.128–35. Available at: http://www.ncbi.nlm.nih.gov/pubmed/21030542.

Hoot, N.R. & Aronsky, D., 2008. Systematic Review of Emergency Department Crowding: Causes, Effects, and Solutions. Annals of Emergency Medicine, 52(2).

Hopp, W. & Spearman, M., 2011. Factory Physics 3rd ed., New York: Waveland Press.

Koomen, A., 2015. The influence of emergency department responsiveness on throughput times. University of Groningen.

Lahr, M.M.H. et al., 2013. The chain of care enabling tPA treatment in acute ischemic stroke: A comprehensive review of organisational models. Journal of Neurology, 260(4), pp.960–968. Laxmisan, A. et al., 2007. The multitasking clinician: Decision-making and cognitive demand during and

after team handoffs in emergency care. International Journal of Medical Informatics, 76(11-12), pp.801–811.

(34)

van Manen, V., 2015. Emergency Care Waiting Times; The Impact of Dynamics and Room Utilization. University of Groningen.

Mccarthy, D.M. et al., 2014. Original Contributions PATIENT – PROVIDER CONVERSATIONS DURING AN EMERGENCY. Journal of Emergency Medicine, 47(5), pp.513–519. Available at: http://dx.doi.org/10.1016/j.jemermed.2014.06.056.

Mumma, B.E. et al., 2014. Effects of emergency department expansion on emergency department patient flow. Academic Emergency Medicine, 21(5), pp.504–509.

Perimal-Lewis, L., Hakendorf, P.H. & Thompson, C.H., 2014. Characteristics favouring a delayed disposition decision in the Emergency Department. Internal medicine journal, pp.1–17. Available at: http://www.ncbi.nlm.nih.gov/pubmed/25370171.

Pines, J.M. et al., 2011. International perspectives on emergency department crowding. Academic Emergency Medicine, 18(12), pp.1358–1370.

Qureshi, A. et al., 2011. The impact of an acute care emergency surgical service on timely surgical decision-making and emergency department overcrowding. Journal of the American College of

Surgeons, 213(2), pp.284–293. Available at:

http://dx.doi.org/10.1016/j.jamcollsurg.2011.04.020.

Robinson, M.A., 2010. Work Sampling : Methodological Advances and New Applications. , 20(1), pp.42–60.

Thijssen, W.A.M.H. et al., 2015a. Complaints and Diagnoses of Emergency Department Patients in the Netherlands: A Comparative Study of Integrated Primary and Emergency Care. Plos One, 10(7), p.e0129739. Available at: http://dx.plos.org/10.1371/journal.pone.0129739.

Voss, C., Tsikriktis, N. & Frohlich, M., 2002. Case research in operations management. Journal of Operations & Production Management, 22(2), pp.195 – 219.

Wang, J., Li, J. & Howard, P.K., 2013. A system model of work flow in the patient room of hospital emergency department. Health Care Management Science, 16(4), pp.341–351.

(35)

Appendix

(36)

Appendix 2Patient Distribution

(37)

Appendix 4 Patient Urgency

(38)

Work Sampling Categories Explained

1. 1st_{contact physician: In this stage the patient is waiting to be seen for the first time by a} physician in the ED, it is well possible that the patient has already been seen by a (triage) nurse. 2. (additional) Tests: In this stage the patient is waiting to have additional test performed, so either blood needs to be drawn by a nurse for lab work, the patient needs to be moved to an echo or CT machine. It is possible that additional testing in necessary after results of preliminary test are received.

3. Results (additional) tests: In this stage the patient has had the additional tests performed but is waiting for the results of those tests.

4. Policy determination: In this stage the patient is waiting to have his policy determined. This means he/she has had all test necessary for diagnosis performed and the results are in, but the physician has not yet determined what to do with the patient.

5. Presence physician: In this stage the patient is waiting for the presence of a physician, because this physician was not physically present at the Ed during the Work Sample. This was mostly recorded during the handovers in the morning.

6. Consult: in this stage the patient is waiting for a consultation by a physician from another specialty than the one he/she was originally assigned to. This request for consult comes from the primary physician.

7. No waiting: In this stage the patient is not waiting as the physician is physically in the treatment room of the patient.

8. Supervision: In this stage the physician has all results and determined a policy to execute for his patient, but does need to consult his supervisor to accept the policy determined by the physician.

9. Policy execution: In this stage the patient is waiting for the healthcare providers to execute the policy already determined by the physician.

Emergency Department Flow Performance; What are you waiting for?