Allen J. Giannakopoulos, PhD and Matthew Kadrie share the simulation model they used in the redesign of an Emergency Department.
Allen J. Giannakopoulos, PhD and Matthew Kadrie built a model of an Emergency Department that clearly defined and thoroughly tested each process.
Hospital Emergency Departments are constantly faced with possible gridlock due to an unexpected increase in the number of patients that arrive. This can be further aggravated by an absence of processes to improve patient flow. What change can be made to improve the process? How will that change affect the patient’s length of stay? Will that change affect the door to doctor time? These are all questions that have been asked by Emergency Departments. Their patient flow process is not performing up to standards, so how do we improve it? Well, there is always the option of making the change in real life and “hoping for the best”. Or, a more rational and cost effective way is to simulate the processes. With the help of a large number of staff from diverse areas, a computer simulation model was designed that mimics a real life, real time Emergency Department (ED). The amount of work involved with the development indeed took a “village”, realizing that no single area alone can accomplish such a task.
An ED with separate silos for patients means possible gridlock for patient flow. The leadership team of the ED realized this problem and was searching for a solution, when computer simulation was brought to their attention. This is a tool that helps simulate their patient flow processes, saving them both time and money. So how did we create a computer simulation for one of the busiest, most hectic areas of a hospital?
While building a model for this ED, the team discovered that it is vital to clearly define the exact issue being addressed. Since a computer simulation model will be a perpetual function, it is necessary to have a target in place to get accurate results. During our initial build of the ED, we had a vague idea of a few different problems that were in need of a solution, but had to target each one individually. How to expedite the lower acuity patients, nurse staffing hours, and accelerating the patient admit process are examples of some of the pain points brought to our attention.
Process flows of the area were created to gain an understanding of how the ED operates. After multiple meetings with various staff members of the ED, we determined what processes were required. The processes mapped out were based on the different acuity levels of a patient. The higher acuity patients require more diagnostic testing as well as higher priority over lower acuity patients. We also included the staff member(s) assigned to each activity in the process flows.
The remaining data points were researched throughout the model building phase. The data collected for each step in the process was obtained from multiple sources; we started with the arrival rates of the patients so we could accurately represent the patient input during a busy day in the ED. Physicians, nurses, technicians, phlebotomists, and many other areas within the ED provided the amount of time each step takes in the process. We also collected data on the staffing schedules, when staff went on break, and when they took a lunch. Every factor that may interrupt a staff member from direct patient care had to be recognized and documented in the processes.
Next, we needed to determine what elements to measure against or compare the simulation to. When it was time to validate the simulation, we needed specific metrics to serve as a baseline. These metrics would be what the department, in this case the ED, uses to measure their results. Length of stay, door to doctor, door to nurse, and disposition to departure are some sample metrics that we used in our ED simulation. In performing this part of the project, it indeed took a “village” to accurately identify the elements and gather the data, while working through the team differences in approaching the decisions of what data to use - which is a crucial point.
Building the Model
As preliminary data gathering wrapped up, model construction started. By utilizing the floor plan of the ED we were able to start building the main structures. The ED’s floor plan provided us with the infrastructure (location of the walls, nurse stations, bathrooms, etc.). Once the walls and other structures were built into the model, we made several in-person walkthroughs of the multiple areas located within the ED. The walkthroughs supplied a thorough look at the areas, providing the locations of certain items, such as medical equipment, computers, and patient beds. After the walkthroughs were concluded and were built into the model, we had a visual representation that mimicked the ED.
With this representation of the ED as a foundation, we started makingit functional by inputting the numerous processes. We entered what was going to happen, when it was going to happen, how long it was going to happen, and who was going to make it happen. After frequent meetings with ED staff, we were able to detail the processes and include as many of the “what if” scenarios as possible. For example, if a patient arrives and there is an empty bed, they will be escorted directly to that empty bed to be triaged. If a bed is not available, the patient will wait in the waiting room until the dedicated triage room becomes available. All of the processes built into the model were “as is” or what was currently happening. As in any simulation, we began by simulating real life rather than what should be happening. This is another aspect of “it takes a village” as now we were dealing with multiple processes with dozens of data elements and staff from every facet of ED life. While sounding simple, it is far more complex performing team initiatives and coming to consensus on the various process decision points than one realizes.
Real life data was added into the processes to make them functionally accurate. The most crucial part of any computer simulation is data accuracy; it drives the simulation and allows us to see an accurate depiction of real-time events. One can build a picture-perfect model that mimics reality, but running the model is useless if the data is flawed. Gathering data was the most essential and prolonged phase.
Data was extracted from several applications utilized in the ED. While some would assume the data was accurate but in fact the data we received was only as good as the data entered. We encountered several issues with data that may have resulted from either improper data entry, data being entered at the wrong time, or staff members documenting differently. Management became aware of the discrepancies and they are working on standardizing documenting procedures. As a result, the corrupt data could not be utilized and the team determined the best fit statistical distribution based on staff experience.
The processing times for all activities used in the ED simulation were based on a statistical distribution curve. The use of statistics was a vital component in providing a variation in the process, just as each process varies in real life. A nurse exam does not always take five minutes and not every patient will have the same experience. The simulation tool can actually calculate the statistical distribution. By inputting the data set, the tool will provide the best fitting statistical distribution based on that data. This an enormous asset with the tool as it improves the accuracy of the simulation, removing the laborious task of trial and error in using some of the 100+ distribution models built into the tool. For data points that we could not obtain a valid set of numbers, we used a triangular distribution. The triangular distribution takes into account the maximum, minimum, and the mode to distribute the probability accordingly. Note that it is still possible to change these types of distribution after the simulation has been completed in order to measure the changes that the final data sets show.
Some data could not be obtained from an application, so we used our next best option, “expert estimation”. What is “expert estimation”? This is when the process owners, or the ones who are actually performing the process, told us how long the process takes. They provided the best case scenario, the worst case scenario, and the average. However, there were times when we would get mixed responses, for example, management reporting that an activity only took five minutes; when in fact, after talking to a nurse, it actually took fifteen minutes. The information provided by the process owners, rather than management, was more likely to be reliable since they were the ones actually performing the process.
Building the simulation required numerous model builds. After presenting the model to the ED staff, we performed several recommended changes. Some changes were as simple as the color of their scrubs but others were major process changes, such as, where a patient would go after receiving treatment in a certain area. While accurate, a perfect solution is not realistic due to the amount of variation in an ED.
Once again, the “village” of staff united and assembled a simulation that combined all of the dynamic processes involved in the ED with agreement to the data inputs. Next, we had to validate that the simulation was producing data comparable to real life. The model was set up to display length of stay times, door to doctor times, door to nurse times, and many other metrics that the ED uses as a benchmark. A customized dashboard displays the metrics used to analyze data produced by the simulation.
Next, we ran the simulation for 30 replications and viewed the output data in the dashboard. The metrics were compared to data that the ED produces in real life, to confirm that the simulation is behaving in a manner aligned with reality. The first run presented an issue with the lower acuity patients; the length of stay in the simulation for these patients well exceeded reality. We went back to the “village” and the problem was identified in the patient registration process. The simulation naturally treats the higher acuity patients with priority, since they are in a more serious condition and typically require precedence. This continued to postpone the registration of the lower acuity patients, something that was not actually happening in the ED. After confirming this with the ED staff, we modified the registration process in the simulation. It turns out that in most cases, a lower acuity patient is seen faster than a higher acuity patient with the intent of speeding up their discharge. We ran the simulation again for 30 replications. This time the performance data numbers were more precise, validating that we built a working model mimicking a real life ED.
Experiment and Analyze
Innovation provided a functioning model similar to a real ED, but what would we do with it? This was the prime motive for building the simulation model, having the ability to make changes in the model, and analyzing the results that are produced. By running different scenarios using computer simulation, we were able to forecast possible outcomes, prevent adverse patient outcomes, and be fiscally responsible by staying within the department’s budget.
The computer simulation tool allows us to experiment with different scenarios. We modified only one variable per scenario, since modifying many variables could impact each scenarios’ results. Then, we performed an experiment that ran each scenario many times. The data produced from each scenario was presented on the dashboard, allowing us to compare and analyze the results.
The ED staff clearly identified the issues that needed to be addressed. One example involved the hours an area within the ED was open for patient care. At the start of building this simulation, the ED was operating this area from 11:00 am to 11:00 pm, their busiest hours. But, what would happen if these beds were available 24 hours? How would that affect the patient’s length of stay and the time that it takes a patient to see a physician? This was a perfect scenario to run in the simulation. The appropriate variables were modified in the model to keep the area open for 24 hours rather that only 12 hours. We ran this scenario 30 times along with the original “as is” scenario. The two were compared and, as a result of keeping this area open 24 hours, the length of stay for all patients decreased considerably. The results were presented to the ED management team and it was left in their hands to decide whether or not to change the hours of this area. Computer simulation made a decision less complicated, with a greater chance of success.
The ED simulation is a perpetual function, utilized by the staff on an as-needed basis. Changes are typically led by the front line of the ED with the “village” of nurses, technicians, and physicians. Management provides support for these decisions and helps align them into place. Until now, our ED was on their own for making their decision based on “what if” scenarios only. Now, they have an innovative tool that can help support or oppose a “what if” scenario. This is the future of reengineering healthcare processes. And, it takes a “village”.
About the Authors
Allen J. Giannakopoulos, PhD is the Corporate Director for Reengineering and Redesign at Baptist Health South Florida in Miami. His duties include process reengineering and computer simulation of processes in clinical and business departments; knowledge reports development; auditing for ePHI and HIPAA; and the management of processes for Role Based Security. Dr. Giannakopoulos earned his academic credentials from the State University of New York in Brockport, BS in Business; University of Rochester, MBA in Business and Marketing; and his PhD in Health Administration from Kennedy - Western University. Dr. Giannakopoulos been published in over 50 health care journals and publications and has been a featured speaker and presenter over the past 25 years in health care, quality improvement and process simulation.
Matthew Kadrie is a Process Management Engineer at Baptist Health South Florida (BHSF). His tasks include process analysis, process simulation, and performing and presenting research for current Information Technology trends. Matthew holds a Bachelors of Computer Science from St. John Fisher College and a Masters of Business Administration from Florida International University (FIU).
A special thank you to Gary May for his support on the article.