Improving patient waiting time of centralized front office service in a regional hub hospital using the discrete event simulation model

Abstract

BACKGROUND:

Today, hospital rankings are based not only on basic clinical indicators, but even on quality service indicators such as patient waiting times. Improving these indicators is a very important issue for hospital management, so finding a solution to achieve it in a simple and effective way is one of the greatest goals.

OBJECTIVES:

The aim of this article is to evaluate the use of a discrete event simulation model to improve healthcare processes and reduce waiting time of patients and hospital costs.

METHODS:

The case study proposed in this paper is the reorganization of non-clinical front office operation for the patients (i.e. booking of exams, delivering medical reports, etc.) of the Careggi University Hospital of Florence, to optimize the utilization of the human resources and to improve performances of the process.

RESULTS:

The development and validation of the model was made according to an analysis of real processes and data, pre and post implementation of model outcomes. The new organization shows a decrease of waiting times from an average value of 10 minutes and 37 seconds to 5 minutes and 57 seconds ( $-$ 44%).

CONCLUSIONS:

This paper shows that discrete event simulation could be a precise, cost-limited tool to optimize hospital processes and performance.

Keywords

Discrete event simulation healthcare optimization front office lean thinking

1. Introduction

The organization of front office activity is a problem that every business that provides a public service must deal with with great attention. It is necessary to balance the need of limiting the use of resources, with the aim to provide adequate service to customers. These problems are very important in a critic working environment like public healthcare, in which the need of cost optimization has become increasingly high in the past years. However, the attempt to reduce costs must comply with keeping security and comfort standards for the patients: this is one of the most difficult and important tasks in hospital administration.

The use of simulation models to study and optimize clinic processes is a core part of Lean Thinking approach, whose importance is growing during time [1, 2]. A check-in desk, even inside a hospital, can be studied and analyzed like a classic service to the public [3, 4], where the customers go to desks because they need the provision of a service [5, 6]. Often, the management start or changes this kind of process [7, 8], by delaying at a later time possible actions to contrast organizational problems (lack of resources, high waiting time, etc.) that may be emerged [9, 10]. Instead, it is clearly more efficient to optimize the process before implementation by analyzing in advance the results of a modification [11]. In literature there are many examples of this kind that use simulation models [12, 13]. In a previous article published in 2018, the characteristics and the validation of a simulation model specifically designed to analyze and optimize the front office activity of Careggi Hospital have been introduced [14]. This work, based on the Arena software, developed by Rockwell Automation [15, 16], has been further developed through the research activity described in this paper. The object of this study is to use the simulation model to solve overcrowding and patient waiting time issues related to the front office of Careggi University Hospital of Florence, which provides several administrative services to the users. The quality of service would thus be increased without additional costs for the hospital, despite a significant workload increase due to process re-organization.

In the following paragraph the article outlines the characteristics of front office, based on process analysis, and displays the operation of the model and its validation. Finally, it shows off the outcome of the simulations, the results of the application of the tested scenario and a comparison between the forecast and the reality, especially regarding waiting time of patients.

2. Methods

The goal of this paper is to evaluate the possible advantages of using a simulation model in healthcare. In particular, it aimed to optimize the performance of the front office of Careggi Hospital using a discrete event simulation model. Through an accurate process analysis, the characteristics of service have been identified. This information is essential, because the functioning of the simulation model has to be as close as possible to the actual process.

The front office is open from 7:30 to 18:30 every day from Monday to Friday, and from 7:30 to 13:30 on Saturday, it is closed on Sunday. There are at most 13 desks available. There is no priority or dedicated desk, except for urgencies and acceptation of diabetic patients. The other activities provided are delivery of diagnostic reports, booking of visits and laboratory or diagnostic exams, delivery of laboratory exams, request and delivery of medical records. Through process analysis was also detected a problem of organization. In fact, there wasn’t a planned use of the available human resources. The number of open desks followed the trend of the accesses, so there wasn’t a schedule of work. This situation caused serious organizational problems and worsened the performance of the process.

The records of the database started from 1/2/2016 and ended on 24/11/2016, in this period there were 126,850 users. Thanks to this data was possible to find the time of the arrival of users, the type of service requested and the waiting time. Service time is another core parameter of the process, unfortunately it is impossible to use historical data to estimate it, because the system does not register finish time of service. In this case, it was necessary a direct observation, carried out by the staff, to know the correct numbers to use in the model.

2.1 Simulation model

The simulation model (Fig. 1) manages the entire patient’s journey within the front office, from the moment he accesses the structure until his request will be dealt with and he leaves the pavilion. The construction of the simulation model is strongly linked to the process and data analysis, in fact each part of the model is responsible for the management of a part of the process itself, whose characteristics have been precisely identified on the information and data previously collected.

Figure 1.

Screenshot of the simulation model.

Since the number of accesses during the day is highly variable, it was not possible to set a standard time interval between an arrival and the next, on the contrary the chosen instrument was the “schedule”. Using this approach, it was possible to realize a variable access program, in this case on an hourly basis, in order to simulate adequately the real trend of the workload, which is concentrated in the central hours of the morning. For each type of service available at the front office a “Create” module was used, with a specially identified schedule. To manage the path that each entity must do within the model, attributes have been assigned in a specific sub-model, in this way it was possible to decide how many users will pay before going to a desk, how many after it and how many will not need it.

The user who for different reasons (has already paid, is exempt, pays after acceptance, etc.) does not have to make any payment, takes the number and waits to be called at the desk. This part of the process is managed with a sub-model that contains seven “Seize” modules, each of which serves to assign a resource (an open desk) to the user based on the type of service. The time that the entity passes in queue waiting for a desk to be freed is the waiting time, which is used for the validation of the model comparing it with the historical data. Proceeding the entity enters a sub-model in which there are seven “Delay” modules each of which simulates the duration of each user’s procedure based on the type of request. The duration of the service has been simulated using an exponential distribution, with average value based on direct observations, because this type of distribution is normally used in cases of simulation of services to the public. Calculate service times required a direct monitoring activity, since no database reference is available. Once the request is completed, the entity goes through a “Release” module that serves to make the resource available again, after which the entity concludes its path in a “Dispose” module. If the user must make the payment, it follows a parallel path, substantially identical to the previous one, but with the creation of a duplicate entity. The number of users of this type is set on the base of direct observation, and their path is managed through a specific attribute assigned to each entity. The purpose of the duplication, which is carried out through a “Separate” module, is to simulate the duration of the payment transaction, while the original entity remains stationary in a “Hold” module, continuing to occupy the desk. When the payment is completed, the duplicate entity is sent via a “Decide” module, controlled by an attribute, to a “Signal” module that sends a signal to the “Hold” module, allowing the original entity to continue the path, terminate the procedure, free the resource in the “Release” module and finally exit the system. The end of the procedure is managed with an additional “Delay” module.

The model was assessed with the help of specialists, to be sure that it was formally correct, and validated comparing simulation’s outputs with historical data. ACT Operation Research, partner of the project and specialized in discrete event simulation application [17], made an evaluation of the model and stated that is working properly, after that a comparison was made with the historical data.

The results show that the estimated values are substantially equal to the real ones, with a difference of less than 5%. The outputs are the mean of 200 replications, each one of them has a length of a week. Weekly workload of the model is equal to reality, this proves that the schedules are correct. Waiting times are like real ones, with a difference that is less than 5% in the worst case, except for urgencies. However, this service has special characteristics, and is so poor in number to be irrelevant for the process. The mean waiting time value of all users calculated by the model is only 2.6% lesser than real one (Table 1). This value is acceptable, especially considering that there are some parameters poorly estimated because of lack of historical data.

Table 1

Confrontation between historical data and model outputs

KPI	Historical data	Model outputs	Difference
Weekly number of accesses	3048	3048	0%
Waiting time for documents delivery	7’41”	7’48”	1.5%
Waiting time for diagnostic exams	7’18”	7’01”	4%
Waiting time for laboratory exams	7’18”	7’00”	4.2%
Waiting time for documents request	8’00”	8’18”	3.7%
Waiting time for urgencies	1’11”	35”	50.6%
Waiting time for booking	7’32”	7’42”	2.3%
Waiting time of all services	7’20”	7’09”	2.6%

2.2 Real case application

In the second semester of 2017 the workload grown from an average value of approximately 600 users to 800 users per day. This change was primarily due to management decision to add the acceptation of medical clinics for diabetic patients. Obviously, this increase caused a deterioration of the performances of the service, especially regarding waiting time of patients. This value was 7’20” in 2016 and became 11’26” at the end of 2017. An increasing like this leads to a worsening of the patients experience and to a discontent among them. So, in order to optimize the performances of the process and to improve patients’ satisfaction, the management decided to use the simulation model previously exposed to identify the best way to use the available resources. Studying historical data of the front office, was detected a trend of accesses, that remains the same every day with a spike in the morning between 10:00 and 12:00. So, the best approach to handle the queues and minimize waiting time is to change the number of operators during the day, with a higher number when there are more accesses and a smaller one when the accesses decrease. After a long and deep analysis, carried forward with the help of the operators, some scenarios were tested with the simulation model to identify the best way to use the available human resources, in order to balance patient waiting time and back office activities, that requires some operators to be done. The characteristics of the selected scenario are shown in Table 2.

Table 2
Number of desks open in every time slot

Time slot	Desks without priority	Desks with priority
7:30–8:30	3	2
8:30–9:30	5	2
9:30–10:30	7	2
10:30–11:30	7	3
11:30–12:30	7	3
12:30–13:30	3	2
13:30–14:30	3	2
14:30–15:30	3	2
15:30–16:30	3	2
16:30–17:30	3	1
17:30–18:30	3	1

As you can see the resources are divided in two groups. The first one includes desks used only by activities without priority, while the second group is reserved to diabetic patients, and serves the others only if there are no one of this kind in queue. The simulation model, as exposed in the previous article and before in this paper, uses a schedule of accesses that changes hour by hour for every service available based on historical data. To set the path of the patient, specific attributes based on process analysis information were used. Every service shares the same resources and has the same priority, except for diabetic patients. After creation of the entity, it enters in a sub-model that simulates the duration of every operational procedure according to the type of request. To simulate the service time, was used an exponential distribution, with mean value based on observations. After the completion of request, the entity passes through a module that releases the resource, and then the path is over.

This tool is meant to be used to evaluate the performance of the process changing some parameters. Especially, it is very important to set the correct number of resources in every time slot, to deal with the number of accesses during day, minimizing the costs and improving the KPI of the process. The number of accesses is based on historical data and the model will be used to test how the output of the simulation changes in case of a reorganization of the Service Center. To do this, some scenarios were tested, modifying the number of desks for time slot, to find the best way to face the minimum and maximum number of accesses during day. The selected solution shown in Table 2 is the one with the best performances. Considering the difficulties connected with the absence of a planned schedule of work, it was necessary to do an on-site survey to be sure that the proposed schedule was correctly applicated.

3. Results

The application of the workshift began on May 14 ${}^{\text{th}}$ , 2018 and the results were collected up until July 7 ${}^{\text{th}}$ , 2018. To correctly evaluate the functioning of simulation model, the waiting time of patients was measured, either with or without priority. Later, these data were confronted with the forecast of the model, to assess how reliable the model is to simulate the actual operation of the process, and with the historical data of the end of 2017, to identify the improvement of the performances of the front office.

First, as shown in Table 3, the application of a working schedule that sets a fixed number of desks that must be open in every time slot, has significantly improved the waiting time of every kind of patients. It is also noticeable that this improvement was achieved without adding any new human resource to workforce, but only with a better organization based on correct utilization of the simulation model, developed by deep and precise analysis of process and data.

Table 3
Comparison between mean waiting time of historical data and of experimentation’s outcome

KPI	Oct-Dec 2017	14 May–7 July 2018
Waiting time (total)	10’37”	5’57”
Waiting time (no priority)	11’26”	6’16”
Waiting time (diabetic)	3’54”	2’58”

Waiting time of all patients decreases of 44% and the waiting time of patients without priority scores a reduction of 45.2%. This is very important because this kind of patients are the majority of the total. Also, there was an improvement for diabetic patients too, with a decrease of 24.1%. To emphasize the importance of these results there are more indicators, that show the improvement of the process parameters, especially regarding the waiting time of patients without priority (Table 4).

Table 4

Comparison between historical data and experimentation’s outcome

KPI	Oct–Dec 2017	14 May–7 July 2018
Diabetic patients that wait more than 20 minutes	1.8%	0.2%
Patients without priority that wait more than 20 minutes	19.2%	2.9%
Diabetic patients that wait more than 10 minutes	10.8%	4.1%
Patients without priority that wait more than 10 minutes	47.4%	22.6%
Diabetic patients that wait less than 5 minutes	75%	81.5%
Patients without priority that wait less than 5 minutes	33%	54.5%

Before the implementation of the schedule tested with the simulation model, almost half of patients without priority waited more than 10 minutes to be served. During the experimentation this percentage was more than halved. Moreover, more than half of this kind of patients waited less than 5 minutes in the period of experimentation.

All these data show that it is possible to improve the process using a new organized method to handle the workload. However, it is also important to understand if the forecast made by the simulation model was correct. In this case, has been proved that the tool is functional, and that the hospital management can use it every time there is a need to get better or a necessity to reorganize the front office adding new activities.

As shown in Table 5 the waiting time of patients without priority, predicted by the simulation model, is very close to the real one. The data collected during the experimentation are less than the prediction of simulation model by 9.6%. Considering the complexity to obtain reliable information about service time of every activity and the uncertainty related with the correct application of the schedule, this is a result that can be considered sufficient. For diabetic patients the difference is higher ( $-$ 28.9%), but there are some factors that could explain this result. First of all, the time needed for acceptance was estimated in 2017 at the beginning of this new task. Meanwhile, operators have improved their skills using the software related with this activity, so the real service time for diabetic patient’s acceptance has decreased. Secondly, there are some uncertainties related with the real number of desks with priority, because sometimes are used more resources than expected, to avoid queues.

Table 5

Comparison between simulation model forecast and experimentation’s outcome

KPI	Forecast	Experimentation
Waiting time (no priority)	6’55”	6’16”
Waiting time (diabetic)	4’10”	2’58”

4. Conclusions

The simulation model presented in this paper is an appropriate tool to support decision making in healthcare processes. The results obtained by applying discrete event simulation model to the front office process have highlighted an improvement of over 40% of waiting time for users without priority. This result is even more important considering that has been achieved only with a more organized and standardized process, without increasing the available resources of front office and the costs for the hospital. Moreover, the forecast of model has been substantially in line with the real results, and the difference observed has been satisfactorily explained. Therefore, the tool will likely be used in coincidence with a possible reorganization of the service. The results achieved during the above described case study encourage to extend the application of such approach to other hospital processes as well, especially to those services in direct contact to citizens, both clinical and non-clinical.

Footnotes

Acknowledgments

This project was co-funded by Fondazione Cassa di Risparmio di Firenze. The authors thank ACT Operation Research who provided insight and expertise that greatly assisted the research.

Conflict of interest

The authors declare that they have no conflicts of interest.

References

Kagioglou

Sapountzis

. Application of lean thinking a briefing document. Health and Care Infrastructure Research and Innovation Centre (HaCIRIC): University of Salford; 2007.

Kalbasi

Krishnamurthy

Rolia

Singhal

. Simulation by example for complex systems. In: Proceedings of the 2014 Winter Simulation Conference; 2014; pp. 974-985.

Guizzi

Murino

Romano

. A Discrete Event Simulation to model Passenger Flow in the Airport Terminal. Mathematical methods and applied computing. 2008; 427-434.

Tan

Chai

Hawari

. Modeling and Simulation of Queuing System for Customer Service Improvement: A Case Study. In: The 4th International Conference on Quantitative Sciences and Its Applications; 2016.

Ghaleba

Suryahatmajab

Alharkan

. Modeling and Simulation of Queuing Systems Using Arena Software: A Case Study. In: Proceedings of the 2015 International Conference on Industrial Engineering and Operations Management; 2015.

Destin

, Machfud Arkeman

. Queue system improvement of certificate checking for increasing public service in Bogor district land registry office. International Journal of Scientific and Research Publications; 5(10); 2015; 1-7.

Converso

. Dualità delle logiche lean-agile nel dimensionamento delle risorse per l’erogazione dei servizi ospedalieri ad alto contenuto tecnologico. University of Naples Federico II: PhD thesis; 2015.

Torri

. Utilizzo della simulazione ad eventi discreti come strumento per l’analisi ed il re-engineering dei processi sanitari. University of Naples Federico II: PhD thesis; 2015.

Chen

Aurelius

Robielos

Kate

Palañ

Lorenzo

Valencia

Chen

. Scheduling patients’ appointments: Allocation of healthcare service using simulation optimization. Journal of Healthcare Engineering; 6(2); 2015; 259-280.

10.

Mocarzel

Shelton

Uyan

Pérez

Jimenez

De Pagter

. Modeling and simulation of patient admission services in a multi-specialty outpatient clinic. In: Proceedings of the 2013 Winter Simulation Conference; 2013; pp. 2309-2319.

11.

Šteins

. Discrete-Event Simulation for Hospital Resource Planning – Possibilities and Requirements. Linköping University: thesis; 2010.

12.

Duguay

Chetouane

. Modeling and improving emergency department systems using discrete event simulation. Sage publications: Simulation; 83(4); 2007; 311-320.

13.

Vijayan

. Effect of appointment schedules on the operational performance of a university medical clinic. Louisiana State University: Master’s thesis; 2015.

14.

Cocchi

Frosini

Ciagli

Tortoli

Carpini

Cirone

Belardinelli

. Discrete event simulation model for the analysis of centralized front office service in a regional hub hospital. World Congress on Medical Physics and Biomedical Engineering 2018: IFMBE Proceedings; 68(1); 2018; 559-562.

15.

Harpring

Evans

Barber

Deck

. Improving efficiency in social services with discrete event simulation. Computers & Industrial Engineering; 70; 2014; 159-167.

16.

Rinaldi

Montanari

Bottani

. Improving the efficiency of public administrations through business process reengineering and simulation. Business Process Management Journal; 21(2); 2015; 419-462.

17.

Act Operation Research [homepage on internet]. Available from: www.act-operationsresearch.com/.