Institute of Industrial and Systems Engineers



GASTROINTESTINAL UNIT ANAYLSIS:

USING DISCRETE SIMULATION TO DETERMINE EFFECTIVE IMPROVEMENTS

Jillian Johnson: jmjohns2@ncsu.edu

Jonathan Woodall: jcwoodal@ncsu.edu

North Carolina State University Department of Industrial and Systems Engineering

ISE 498: Undergraduate Senior Design Project

December 1, 2009

Abstract

There are many process flow challenges facing medical and health care facilities today. Primary concerns from the patient perspective include longer than necessary wait times and delayed/prolonged procedures. From a staffing perspective, resource utilization can be inefficient and create problems in the facility.

The purpose of this project was to analyze the gastrointestinal (GI) units at The University of North Carolina’s (UNC) Memorial Hospital and Meadowmont Center. Specifically, the goal was to address typical inefficiencies that are faced at these two GI units, such as:

• Longer than necessary wait times

• Delayed/prolonged procedures

• Patient schedules

• Staff allocation

The units were modeled with discrete event simulation and all necessary inputs and system background were obtained through onsite visits and email communication.

To create models of the current processes (base models) at both Memorial and Meadowmont, data was compiled from several different sources at UNC, including: time data from ProVation RN, patients’ schedules, staff schedules, and blueprints of the GI units.

The models were used to analyze potential changes in the system to improve performance. The performance measures used were: patient throughput, total time a patient is in the unit based on procedure type, patient wait time and utilization of resources.

This project included examination of several scenarios for potential improvement as provided by the UNC project team and following analysis to the impact of these changes. Scenarios examined include:

1. Optimal mix of procedures with respect to:

a. Examination of the allocation of endoscopists at Memorial

b. Patient procedure schedule analysis

c. Patient going to first available room versus scheduled room

d. Assigning procedure types to specific rooms (e.g. all EGDs room 3)

2. Combining prep and procedure nurse duties

3. Converting prep and recovery bays to multi-functional bays

In addition to evaluating the above scenarios, animations of the base models were developed. The model animations were used for model verification, visualization of process inefficiencies, and troubleshooting.

With the results and knowledge of the system obtained from the scenarios and animation, recommendations were developed to improve the throughput and process inefficiencies.

Recommendations for Memorial include:

1. Use current allocation of endoscopists across procedure rooms (1 in rooms 1 and 2, and 2 floating between rooms 3, 4 &5)

2. There is neither improvement nor negative impact to total times and wait times when combining the prep and procedure nurse duties. If cost of implementation is low and staff would prefer the change, combining them will not affect the unit’s times.

Recommendations for Meadowmont include:

1. Have patients go to the first available procedure room instead of the one they are assigned to in the schedule.

2. There is a slight improvement to total times and wait times when combining the prep and procedure nurse duties. If cost of implementation is low we recommend combining them.

3. Total times decrease up to 20 minutes when converting prep and recovery bays to multi-functional bays and therefore we recommend further evaluation of potential implementation of this design change.

Current State Model

Overview of Base Data

To complete a discrete event simulation model of the current system, base data had to be collected to calibrate the model. Data was compiled from several different sources at UNC. These included:

• Time data from ProVation RN

• Patients’ schedules

• Staff schedules

• Blueprints of the GI units

The time data from ProVation RN is a spreadsheet of time stamps for each of the steps that a patient encounters during their visit to the GI unit. Examples of steps are: check in, sedation start, colon end etc. With the different time stamps, the data was summarized to determine a group of times for each of the steps making up the overall process of patient flow through the GI units. Using Arean’s Input Analyzer the data was fit to statistical distributions using statistical best-fit procedures. Below is an example of colonoscopy procedure times at Meadowmont best-fit:

[pic]

The best fit for this example was determined to be a lognormal distribution with the expression of, 8+LOGN(14.1,11.5). Using this distribution produces a square error of 0.004124. The resulting distributions in the model were used so the time it takes a patient to go through each process is synchronized with the actual time it takes according to the ProVation RN data. This allows the simulation model to emulate the natural variation that is present in activity times such as a colonoscopy procedure or recovery.

Samples of anonymous patient schedules for both facilities were provided for use in the model. Once the data was transposed to Excel, it was able to then be input to the Arena simulation model so that patients arrived according to the actual scheduled times for each procedure. The model assumes that patients arrive on time, therefore making the arrival schedule deterministic.

The model we used staffing schedules for each of the GI units to define staff availability as a limiting resource. The model assumes that every employee shows up on time.

Blueprints of both units were provided to give a general idea of the unit flow. These were helpful in the development of the animation.

Overview of Base Models

A discrete event simulation model was developed for the GI units at Memorial Hospital and Meadowmont. The model allows the evaluation of scenarios to compare them on the basis of performance measures. Performance measures of interest which the models estimate include the following:

• Throughput

• Total time a patient is in the unit based on procedure type

• Wait time

• Utilization of resources

For starters, a conceptual model of the two units broken into the main parts of the process was developed: arrive/prep, procedure room, and recovery.

Below shows the arrive/prep part of the system:

For Memorial and Meadowmont there are five prep bays for outpatients. At Memorial there are two additional prep bays for inpatients. For both units a Registered Nurse (RN) has to complete some of the prep but an endotech can help to complete the rest. After prep at Memorial the patient has a choice of five rooms. If the patient is getting an Endoscopic Ultrasound (EUS) they are required to go to room one. If the patient is getting an Endoscopic Retrograde Cholangiopancreatography (ERCP) they must go to room two. If the patient is an inpatient, they will most likely go to room five. In Meadowmont the patient has a choice of four rooms to go to with no restriction on procedure type. For both units the patient will wait in the prep bay until a room is available.

Below shows the procedure room part of the system:

For both GI units the procedure room requires two staff members to be in the room during sedation, at least one being an RN. After the procedure, the patient waits in the procedure room until a recovery bay is available (Five bays in Memorial, three bays in Meadowmont).

Below shows the recovery part of the system:

There is one RN assigned to recovery at both Memorial and Meadowmont. At Memorial, if general anesthesia is used two RNs are needed, this typically happens on Wednesdays. After recovery the patient leaves the unit.

Preliminary Results/Validation

The models simulated over a two week period using the schedules that were given as the patient arrival times. The various statistics collected included: the maximum time a patient has to wait to go to prep, the maximum waiting time in prep to a procedure room, and the average total time a patient is in the unit based on procedure type. Below are the base case results for the total time it takes a patient to get through the system based on procedure time:

|Procedure Type |Memorial |Meadowmont |

|Colon |152.59 min |154.7 min |

|Double |165.57 min |166.89 min |

|EGD |142.35 min |148.64 min |

|Flex Sig |129.03 min |137.21 min |

|IRC |- |146.03 min |

|ERCP |184.81 min |- |

|EUS |141.63 min |- |

|Others |170.43 min |- |

The first part of the validation process was expert validation. The results were reviewed with the project team at UNC that experience the process daily. There was a general consensus that the results appeared consistent with past experience and expectations.

The second part of the validation was to compare the outputs of the model to those from ProVation RN. The data from the model and ProVation RN are statistically different for some of the procedures. Since the primary use of the model is to compare alternative scenarios it is the ranking of scenarios that is more important than the raw values from the model and thus the statistical difference in values was not a major issue or concern. Given that the first part of the validation passed, and that any difference in raw values is uniform across all compared models, the team considered the model validated.

Examination of Allocation of Endoscopists

Overview

In this scenario, a comparison of the various allocations of the three providers at Memorial was done to determine which were most efficient. Current practice has one provider for rooms one and two, and the other two providers rotating between the remaining three rooms. Other potential combinations of provider distribution were evaluated with the assumption that there always is one provider for rooms one and two. The list of provider allocation evaluated includes:

1. Two providers for Rooms 3-5

2. One endoscopist in room three and one endoscopist in rooms four and five.

3. One endoscopist in room four and one endoscopist in rooms three and five.

4. One endoscopist in room five and one endoscopist in rooms three and four.

5. One endoscopist in room three, and both endoscopist float between rooms four and five.

6. One endoscopist in room four, and both endoscopist float between rooms three and five.

7. One endoscopist in room five, and both endoscopist float between rooms three and four.

Model Construction and Inputs

To construct this scenario for Memorial, the way providers were seized in the simulation needed to be changed. This was modeled by creating three distinct providers (instead of three interchangeable providers) and having rooms one and two use Provider 1. With the other two providers (2 & 3) a set was created to pull from for the other three procedure rooms. For the models with non-floating providers, the appropriate provider was merely pulled into the procedure room. For the models with floating providers, we the set of providers in the floating procedure rooms was used with the room using the first available provider.

Results

With the seven different provider allocations the differences in the total time, wait times in the process, and utilization of resources were evaluated. No statistical difference between these allocations was found for any of the measures evaluated. Below is a graph of the total time a patient was in the unit based on procedure type for the seven different allocations:

[pic]

Recommendations

After evaluating the data the current setup is recommended as best as best, one provider in rooms one and two and the other two providers floating between rooms three, four and five. The scenarios where the endoscopist can float between any rooms (5-7) are acceptable. Having one endoscopists in one room solely is not recommended.

Procedure Schedule Analysis

Overview

This scenario looks at the different schedules for the units to determine the “best” schedules out of the 16 that were given for the days of 9/14/2009-9/25/2009, 11/9/2009-11/11/2009, 11/18/2009-11/20/2009. To determine the “best” schedules, an analysis of throughput, waiting times and utilization of the resources was done. Figuring out which schedules delivered the best results will help with scheduling in the future but also will be useful in the next two scenarios.

Model Construction and Inputs

The only change in the model was the schedules the simulation was pulling. The original model was using the schedules from all ten days. For this scenario the simulation is for one day at a time so each model will be for one individual day. The models can then be compared to each other to determine the optimal schedules.

Results

The schedules were first narrowed down by throughput. Over the 16 days the throughput ranged from 19 to 34 at Memorial and 13 to 29 at Meadowmont. The cutoff for the “best” schedules was set at 24 for Memorial and 18 at Meadowmont. Then using the wait times that the simulation generated, graphs of the wait times for each schedule at Memorial and Meadowmont were created. Below is the graph for Meadowmont:

[pic]

With these graphs the throughput is listed after the scenario number. These graphs were used as a visual representation to help decide which schedules were best. The goal was to increase throughput and decrease wait time. The schedules chosen as “best” were days: 5, 6, 11, 12 and 13 for Memorial and days: 3, 7 13, and 14 for Meadowmont. These schedules were used for the scenarios to follow.

Patient to First Available Room vs. Room Scheduled

Overview

For this scenario the difference between sending patients to the first available room and sending patients to the room they were assigned to on the schedule was tested. The “best” schedules obtained from the previous scenario were run them through the base model to do this analysis. Having patients go to the room as assigned allows the nurses to know what procedures they will have and hence what to expect for the work day. The purpose of this scenario is to analyze the impact of sending patients to the scheduled room in terms of wait times, utilization and throughput.

Model Construction and Inputs

In order to analyze this scenario for Meadowmont, the base model needed to be changed to assign specific procedures to the room on the schedule. Before it did not matter what room the patient went to, now the model has to filter by room number from the schedule. This was done by creating a decide model to split the model into four branches for the procedure area, representing each of the four rooms.

For Memorial, models already existed for each of the daily schedules tested in the procedure schedule analysis scenario where patients automatically went to the room they were originally assigned to (with the exception of rooms three and four which were used as interchangeable). Therefore, the best schedules could be taken and the data used as the base results for schedules where the patient whet to the room on the schedule. With the new model the EUS procedures still had to go to Room 1 and ERCP procedures to Room 2 because of the specialized equipment required for these procedures which are only located in their respective rooms. All the other procedures went to the first available room.

Results

For Meadowmont, from arrival to discharge the average times based on type of procedure were:

• Colon: 182.12 minutes for Day 3

o Base 154.7 minutes

• EGD: 324.47 minutes for Day 3

o Base 148.64 minutes

For Memorial the waiting times to go to the different procedure rooms are inflated for rooms one and two. This is because one of the limitations of the model is that if two patients leave prep, one a few seconds before the other, and if the first available room is room one that first patient will go there. The problem occurs when that second patient, who is only a couple seconds behind, is an EUS patient. That patient has to go to room one and therefore waits the entire length of the first patient’s procedure. In the actual unit the charge nurse will be able to tell much earlier and could hold that room for the EUS patient and have the first patient go to the next available room. The average times from arrival to discharge averaged over the 5 ‘best’ schedules were:

• Colon: 152.1 minutes

o Base: 152.59 minutes

• Double: 167.55 minutes

o Base: 165.57 minutes

• EGD: 136.01 minutes

o Base: 142.35 minutes

• ERCP: 174.19 minutes

o Base: 184.81 minutes

• EUS: 181.98 minutes

o Base: 141.63 minutes

• FlexSig: 127.31 minutes

o Base:129.03 minutes

• Others: 175.03 minutes

o Base: 170.43 minutes

Recommendations

For Meadowmont this scenario simulated the patients going to the room on the schedule versus to the first available room. With a minimum of nine extra minutes waiting for a procedure room each day and the increase in the total procedure time, this is not a recommended change for this unit.

In the Memorial unit patients were able to go to the first available procedure room instead of the room that they were scheduled for. The total times and wait times were too variable in comparison to the base data. However, based on the aforementioned limitation in terms of potentially having an EUS patient waiting for an entire procedure, it seems like there is a possibility of a positive impact but further analysis would have to be done in order to make any recommendations.

Dedicating Procedure Types to a Specific Room

Overview

The purpose of this scenario is to assess the impact of dedicating certain procedures to certain procedure rooms. In other words, as much as possible, the goal is to have all the colonoscopies in Room 1, all the EGD’s in Room 2, all the IRC’s in Room 3, etc. Dedicated schedules were created from the “best” schedules to run through the new model.

Model Construction and Inputs

In order to model this scenario, some of the existing sample schedules were used and procedure room assignments were moved around within these schedules so that procedures with the same type would be in the same room (as much as possible). The amount of time each procedure took was considered so as to not overbook a time slot and evenly distribute procedures across all rooms. After all the schedules to test were developed, they were read into the Arena model from the Procedure Schedule Analysis Scenario.

Results

For Meadowmont out of the four “best” schedules the maximum average waiting time to go to prep was 4.8032 minutes. The maximum average waiting time in prep to a procedure room was 25.3662 minutes. From arrival to discharge the average times based on the type of procedure were:

• Colon: 171.12 minutes

o Base: 154.7 minutes

• EGD: 191.72 minutes

o Base: 148.64 minutes

• Double: 182.01 minutes

o Base: 166.89 minutes

• IRC: 185.15 minutes

o Base: 146.03 minutes

• Flex Sig: 217.06 minutes

o Base: 137.21 minutes

For Memorial the maximum average waiting time over the five “best” schedules was 1.678 minutes. The average total times for the five different schedules based on the type of procedure were:

• Colon: 162.83 minutes

o Base: 152.59 minutes

• Double: 167.32 minutes

o Base:165.57 minutes

• EGD: 144.81 minutes

o Base: 142.35 minutes

• ERCP: 164.89 minutes

o Base: 184.81 minutes

• EUS: 147.03 minutes

o Base: 141.63 minutes

• FlexSig: 150.53 minutes

o Base: 129.03 minutes

• Others: 178.58 minutes

o Base: 170.43 minutes

Recommendations

For Meadowmont and Memorial, the comparison of dedicating certain procedures to certain rooms provided mixed results relative to the current scheduling. It is difficult to determine the impact of dedicated procedures on the system. There does not seem to be enough evidence one way or the other to make a recommendation here. In order to make a recommendation, the scenario would have to be looked at more be analyzed more fully.

Combine Prep/Procedure Nurse Duties

Overview

This scenario involves combining prep and procedure nurses’ roles. Currently a patient coming into either unit goes to prep and is seen by one nurse. They are then taken to a procedure room and are taken care of by another nurse. With this scenario a patient will go into prep and be seen by one nurse and then taken to a procedure room and the nurse that saw them in prep will stay with them through their procedure. Not only will this add continuity of care but also will give nurses a sense of responsibility for the patient.

Model Construction and Inputs

To model this scenario the nurses in the simulation were changed from prep and procedure to a more general term. When a patient enters the model they will choose from a group of the seven nurses and one endotech. The endotech will not take part in the rotation, but will still be used to lessen prep time

Results

At Meadowmont, the average time before a patient goes into prep is 0.065 minutes. The average time a patient waits in prep to go to a procedure room is 3.94 minutes. Below is a graph showing the time it takes in this scenario from arrival to discharge based on procedure compared to the base data:

[pic]

For Memorial the average time before a patient goes into prep is 0.0909 minutes for outpatients but 4.298 for inpatients. Below is a graph showing the total time by procedure that patient is in the unit for this scenario compared to the base case:

[pic]

Recommendations

As see in the Meadowmont graph, this scenario revealed an improvement in total times by about five minutes. However, the wait times do not significantly improve. With this in mind, combining prep and procedure nurses is recommended if doing so is an easy transition and low in cost. It makes an improvement, but it needs to be enough of an improvement to counterbalance any cost of implementation.

For Memorial, as the graph indicates, this scenario did not show a significant difference in both total times and wait times. It did not affect the system negatively and therefore if there is a want for the change and a low cost of implementation it would be a management call.

Multi-functional Bay Analysis

Overview

In this scenario, the affect on the suite was examined if the prep and recovery bays were eliminated and converted to multi-functional bays. The idea is that by using the same bays for both prep and recovery, there is more flexibility in times of high volume.

Model Construction and Inputs

To construct this scenario at the two units the resources of prep bays and recovery bays were combined to 13 multi-funcitonal bays at Memorial and 8 multi-functional bays at Meadowmont. A secondary model was developed in the case of keeping the inpatient prep bays at Memorial (leaving 11 mulit-functional bays). In the simulation, whenever a bay was seized before, whether prep or recovery, now the patient will seize a multi-functional bay.

Results

For Meadowmont, the average waiting time to go to prep in this scenario was 0.2619 minutes. The average waiting time in prep to a procedure room was 2.3248 minutes in this scenario. From arrival to discharge the average times based on the type of procedure compared to the base case are shown in the graph below:

[pic]

At the Memorial unit the average wait time for prep was 0.3772 minutes during the first scenario where the inpatient bays were kept, and 0.766 minutes during the second. The average total time a patient was in the unit based on the procedure types and compared to the base case are shown in the graph below:

[pic]

Recommendations

At Meadowmont, as seen in the graph the data improves significantly for total times. Wait times also improve across the system if the prep and recovery bays are turned into multi-functional bays. The net improvement in total times across almost all patient types was 20 minutes, and the wait times were improved by two to five minutes in the procedure. For the aforementioned reasons, it is strongly recommend that the prep and recovery bays be turned into multi-functional bays at Meadowmont.

For Memorial the total times (seen in the graph) and the wait times for the two parts of this scenario were not much different from each other or the base data. The scenarios seem to lean towards having a negative effect on the total times and wait times in the unit. There seems to be no reason why this cannot be done, but there is also no evidence of why it should be done and therefore further study is needed in order to make a recommendation for change in the unit.

Animation

Purpose

The primary purpose of animating a simulation model is to allow for visual inspection, to see what is going on while the model is running. Visual inspection is helpful for two reasons. One, it allows for verification that the model is running as it was designed to run; it is much easier to see if the model does what it should in the animation than to look through the model itself. The other aspect of the verification process the animation is helpful with is troubleshooting where problems with the model may be. Two, visual inspection is helpful in that it provides a great picture for the viewer to see what is going on in the system in terms of process flow, what works well and not so well, and what holds the system up.

Components and Construction

The first thing animated was the patients who move through the system. Pictures were created for each patient type such that when a patient movies through the system, the picture moves for them. Below are the pictures for each patient type:

[pic]

Other components of the animation model can be seen below. The large gray box represents the entire suite at Meadowmont, with a clock to the right that indicates the time of day throughout the simulation. Process flow is indicated by the multi-colored lines and ovals in the animation. The ovals represent stations or areas where the patient must originate from or go to. The lines represent the path they follow as they move throughout the animation. The patients enter the system at the upper left corner of the animation (where the orange oval is) and travel to the Prep Bay area. Prep bays are indicated by the rectangle boxes with the letter “P” inside. If all prep bays are occupied, the blue line above the prep bays on the left indicates the queue to seize a prep bay. All patients will wait in this queue until a bay is available. The white boxes in the prep bays are white when the bay is unoccupied, and turn green once the bay is occupied. Finally, the blue and white rectangle at the bottom left of the animation is a level that shows the percent of bays occupied. The blue represents how many prep bays are occupied.

Recovery bays and procedure rooms operate similarly to the prep bays. Recovery bays are the rectangular boxes with the letter “R” in them, and procedure rooms are the larger rectangular boxes that have people in them. When the bay or procedure room becomes occupied, the white box inside it then turns green. The only caveat is for the procedure rooms. In this case of the model, the procedure rooms are interchangeable. Therefore, the box in each room is not specific to that room. What happens in this particular model is that when one procedure room is occupied, the bottom box turns green; when two are occupied, the one above it turns green as well, etc. The Memorial model is different, as the model has separate resources for each procedure room, and thus, each white box can be specifically identified with the room. The blue and white rectangle beside the recovery bays operates the same way as the level for the prep bays, the only difference being that it measures the usage of the recovery bays instead of the prep bays. Finally, the queues for the recovery bays and procedure rooms can respectively be found in the bottom right corner and in the center box.

[pic]

Future Work

After presenting the recommendations and animation to the project team at UNC, several steps will follow. A presentation will be made for all staff at the two GI units, in particular showing them the animation. Also, the discussion of how to implement the two scenarios from Meadowmont with positive results will take place. In addition, further analysis will take place to develop the scenarios at Memorial further.

Other recommendations for future work include:

1. Adapt the model to include no-shows and late/early patients and look into ways to deal with these issues (e.g., over booking)

2. Do a sensitivity analysis on resources to determine if there are ways to allocate resources that can optimize efficiency.

3. Create a scheduling optimization tool or an interface that allows the scheduler to test whether a schedule is good or not under the day’s conditions.

Acknowledgements

There are several people to which we must acknowledge because without them this project would have not existed. Dr. Brian Denton, North Carolina State University, was the project advisor and Bjorn Berg was the graduate student co-advisor. Marvetta Walker from the University of North Carolina was our key contact and supplied us will all the data that we needed to support our simulation model. Susan Phillips and Glen Spivak were also integral members of the project team at The University of North Carolina.

Biographical Sketch

This project was completed by Jillian Johnson and Jonathan Woodall students at North Carolina State University. Both will graduate in December 2009 with their B.S. in Industrial and Systems Engineering. Starting in the Spring of 2010 both will continue on to complete their master’s degree in the same field. Jillian and Jonathan are also both members of the Society of Health Systems.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download