The aim of this study is to provide an extensive systematic literature review about the use of dynamic programming in the management of appointments in health centers, providing a view of the current research environment. Dynamic programming of appointments improves the efficiency through algorithmic decision support tools. This technology has used successfully in other industries such as airlines, car rental agencies and hotels. The application of this technique to the health environment has attracted the interest of many academics in the last 50 years because it is very useful to improve the access and the quality of health systems and also reduce the cost. However, the use of these techniques in health settings is not trivial because every decision is vital for patients. In addition, we must consider other important and complex factors such as emergencies. Therefore, we hope to analyze the current state of this technology in the health environment, identifying keys for future research.
The objective of this work is to provide an extensive review of the literature about the use of dynamic programming in the management of appointments in health centers. This theme has attracted the interest of many academics over the past 50 years, from the pioneering works of Bailey (1952) and Lindley (1952).
Health care providers are under great pressure to reduce costs and improve the quality of service. In recent years, given the greater emphasis on preventive medicine, the ambulatory or the primary level of health are gradually becoming an essential component in health care.
When we add factors to health care budgets such as aging of the population and the increase of the chronic diseases, it is not surprising that there is a growing pressure on health service to improve the efficiency.
Appointment systems can be a source of dissatisfaction for patients and healthcare professionals. Patients often complain about the lack of availability. To mitigate this discontent, we might think of reducing the time of patient care sessions. However, in Spain only six minutes per patient is allocated in average. In fact, health professionals are disappointed with it since they have to make quick decisions about the health of their patients. Thus, the scheduling systems of appointments are in the intersection between efficiency and the correct access to health services.
In this paper we offer a comprehensive research study on the programming of dynamic appointments in health environments. These models have the potential to improve efficiency through algorithmic decision support tools. This technology has used successfully in other industries such as airlines, car rental agencies and hotels (Talluri and Van Ryzin, 2004). We believe that decision support techniques can reduce costs and improve access to health services simultaneously.
Designing a dynamic appointment scheduling system aims to adapt to the demand, with the availability of the resources, and, at the same time, optimize the use of those resources and minimizes the waiting times that patients suffer. In addition, it is imperative to take into account and understand the health environment in which we are, outpatient, hospital, specialized centers, etc. It is necessary to pay special attention to the factors that make appointment scheduling challenging. In conclusion, to offer a roadmap in appointment management design in health centers.
Waiting time and congestion in waiting rooms are two of the few tangible elements of quality. Surveys indicate that excessive waiting time is often the main reason why patients are dissatisfied with the health services offered (Huang, 1994).
Many factors affect the performance of appointments systems. The delay of patients and specialists, as well as possible emergencies are the main factors.
Emergencies are a key factor in the design of a system that allows us to manage efficiently the demand and the resources. A good appointment system should provide convenient access to health services for all patients. However, how do we prioritize emergencies? Patients' needs have different degrees of urgency, and the decisionmaking process we are proposing must be automatic. That is, decisions must be made before having complete information about urgency.
In conclusion, we wish to deepen in this theme and present the general considerations to be taken into account in the modeling of problems. Therefore, we provide a taxonomy of the methodologies used in the existing literature and we can help to understand how to model a dynamic appointment system. 2
This study has been undertaken as a systematic literature review (SLR) based on the original guidelines as proposed by Kitchenham (B.A. Kitchenham et al., 2004) to achieve the goal described in Section 1: provide an extensive review of the literature about the use of dynamic programming in the management of appointments in health centers. A systematic literature review allow identifying, evaluating and interpreting all available research data relevant to a particular research question in a specific investigation area. The guides proposed, which are among the most widely accepted in software engineering, have been followed to carry out this work.
These guidelines establish that a review should comprise three phases: planning, conducting and reporting. The planning activity deals with developing the review protocol as well as deciding how the researchers should work and interact to conduct the review. This protocol prescribes a controlled procedure for executing the review and includes research questions, search and evaluation strategies, inclusion/exclusion criteria, quality assessment, data collection form and methods of analysis. The second phase focuses on executing the protocol as it has been defined. Finally, the last phase describes how the final report has been elaborated. 2.1
To begin the present study, we search similar systematic reviews following the planned searches that its described in the present and the following subsection. The results obtained are described below.
Two literature reviews were found. Both are from more than five years ago, but they are very useful to start in the world of dynamic appointment management, as well as recognizing the main properties to take into account. Due to its high interest and although one of them is of the year 2003 and fails one of the quality requirements mentioned in section 2.3, it has been considered.
On the one hand, in the oldest review of the selected
ones, made by
On the other hand, (D Gupta and Denton, 2008) present
a constructive criticism of the studies which are done until
the date. As a consequence, this article proposes other
aspects, besides those mentioned in
From these reviews, we pretend to make a study to see the advances made in the technique to the present date. For that, we focus on all the studies that consider the patient as the main user to take into account. Also, it is important not only to take into account the patient waiting time on the day of their appointment, but also to contemplate the time that passes from an appointment until the date of the appointment, because, surely, it is related to the probabilities of cancellation and delay.
Based on these, some questions are asked to be completed in this new systematic review of the literature.
RQ1. Are there currently applications that use dynamic programming to help manage patients in the healthcare environment?
RQ2. Are the patient's preferences taken into account when making an appointment?
RQ3. What health environments have adopted dynamic programming for managing appointments? Differences to keep in mind?
RQ4. How to mitigate the effects of cancellations and non-presentations of patients?
RQ5. Is the time between the request of the appointment and the day of the appointment (indirect time) taken into account?
RQ6. How to manage emergencies and their priorities?
This section details how the search for related articles has been done. The keywords used and the chosen bibliographic engines are presented.
At first, some keywords were defined which were increasing as related articles were found since we learned from the keywords of those articles found.
After several tests, the keywords are concluded in Table 1.
A1. Health
B1.Dynamic
C1.Appointment In addition, Table 2 presents the search expressions representing the different combinations of keywords done for the search. In order to perform the search, we considered the Google Scholar, Fama (catalog of magazines and books of the University of Seville), IEEE, Scopus and PubMed data sources. In them, the search expressions (cf. Table 2) were introduced for obtaining new revisions and studies that conclude some of the challenges raised in section 2.1.
The articles collected from the different databases were managed through the Mendeley tool.
Thereafter, we define the admission and exclusion criteria. With those criteria, we justify whether to consider or not the articles which are found on the aforementioned search engines. Thus, we can identify the most relevant articles to develop our review of the literature.
The inclusion/exclusion criteria are carried out in five phases presented in Table 3.
Phase
Inclusion/exclusion criteria Article related to dynamic scheduling of appointments in the health field. Inclusion of keywords.
Publications since 2006 Availability of the full text for free.
Not duplicated Number of citations relevant, always keeping in mind its year of publication. If an article has been released in a date close to the present, it is logical to find few citations. However, if an P1 P2 P3 P4 P5 article has been in circulation for several years and its number of citations has not increased in relation, we can consider it as a signal of the quality of that article.
However, it should be mentioned that in the second phase (i.e., publications since 2006), some articles have not been taken into account for the article. Although it does not achieve the requirement, it is very useful to immerse ourselves in the world of dynamic programming in healthcare environments, giving an overview of everything that has to be taken into account. Therefore, it was decided not to exclude them.
In this section, we determine the quality criteria. Once the articles have exceeded the criteria of admission and exclusion, the article’ properties are evaluated to know their quality. Each quality judgment can take the values yes or no and, in some cases, including the value “partially”. All the quality criteria considered and their possible values are explained in Table 4.
QA QA1. Is the text readable? QA2. Limitations have been described? QA3.There are future lines of research? QA4. Have the results been presented? QA5. Has the algorithm been tested? QA6. Has the health application environment been described? QA7. Has the cost function to optimize been defined? Tabla 4. Quality criteria
Values O Yes: it is possible to read and understand without being an expert on the subject.
O Partially: In certain aspects, it is necessary prior experience.
O No: difficult to understand.
O Yes: define the limits of the study done.
O No: does not describe the boundary of the study developed.
O Yes: present a series of ideas to continue the research.
O No: there are not claims on possible innovations.
O Yes: presents the results obtained in the study.
O Partially: reference to other articles.
O No: does not clarify the results of the investigation.
O Yes: health domain analysis.
O No: the characteristics of the health area are not detailed.
O Yes: health domain analysis.
O No: the characteristics of the health area are not detailed.
O Yes: description of the function to be maximized or minimized.
O No: the optimization function is not detailed.
In order to evaluate and classify the selected articles, a scheme dedicated to organizing and cataloging the information found in the articles is made. Thus, it is easy to have an overview of each article. Based on this scheme, we describe the results obtained responding to the research questions described in section 2.1.
In Table 5 we illustrate the schema definition, and in section 3.2 we instantiate it with the selected articles. Attention to the time passed between the request of the appointment and the day of the appointment Cost function to optimize Results presented Test results Value {} {} {} {} {} {} Prioritization of emergencies {Theoretical, Experimental, No}
With the General Information element we pretend to collect the main characteristics of the selected articles, {Theoretical with reviews to experimental studies, theoretical studies, experimental studies} {External consultations, treatment, surgery, outpatients clinics, nursing, combination of the above} {Theoretical, Experimental, No} {Theoretical, Experimental, No} {Theoretical, Experimental, No} {Theoretical, Experimental, No} {Theoretical, Experimental, No} {Patient waiting time, physician downtime, cost reduction, time from appointment request to appointment day, combination of the above} {Yes (mentioning other studies), Yes, No} {simulation, case study, both, Statistical, No (is a literature review)} including the Author, Year, Title of the article, the Journal and the Source where it was written.
In the Modelling, it is tried to visualize in which articles we can find answers to the research questions RQ2, RQ3, RQ4, RQ5 and RQ6, as well as for QA6 and QA7 quality criteria.
Finally, we expose the form of resolution and the response of the QA4 and QA5 quality criteria in the Result element. 3
This section presents the results obtained after doing the planning described in section 2. First, we explain the results obtained in the various searches performed. After that, we evaluate these articles according to the quality criteria defined in section 2.4. 3.1
The search process was developed with the keywords described in section 2.2 and introducing them into the bibliographic engines specified in section 2.2. Once the results were obtained, it was checked which the inclusion and exclusion criteria. After this, the number of articles was considerably reduced, finding difficulties to select a number of notable articles to perform a good systematic review.
The first search was made in Google Scholar, not finding any new articles in the other databases, besides those already provided by Google Scholar. In addition, FAMA is discarded since it was not possible to collect any article related to the subject.
In Figure 1 and Figure 2 we can graphically display the number of articles included and excluded by inclusion and exclusion criteria respectively. The inclusion/exclusion phase that restricts most the search is the first. With the first criteria, we are sure that we select articles related to the management of dynamic appointments in health environment, and the keywords are chosen are in the article.
Subsequently, phase 2 (i.e., articles published in the last ten years) also excludes a considerable number of studies. This criterion helps us to make a review of the most current literature. 3.2
Finally, we highlight eight articles by instantiating the schema definition designed in section 2.5 (Table 6 – Table 13).
Schema of Appointment scheduling in health care: Challenges and opportunities article
General Author Diwakar Gupta and Brian Denton
Theoretical / Experimental Theoretical with reviews of experimental studies
Area of application Year Title Journal Source Graduate Program in Industrial & Systems Engineering, Department of Mechanical Engineering, University of Minnesota Combination (External consultation, Outpatient clinic and Surgery) Theoretical Theoretical Theoretical Theoretical Theoretical Theoretical Combination of waiting time suffered by the patients and time of inactivity of the doctor) Yes (mentioning other studies) No (it is a literature review) Schema of Outpatient Scheduling in Health Care: a Review of Literature article
General Author Tugba Cayirli and Emre Veral
Year Title Journal Source Number of citations 2003 535 Outpatient Scheduling in Health Care: a Review of Literature Production and Operations Management Society Hofstra University, Department of Management, New York Theoretical / Experimental Theoretical with reviews of experimental studies
Area of application Delays Cancellations Patient’s preferences Emergencies Priority of emergencies Attention to the time passed between the request of the appointment and the day of the appointment Cost function to optimize Results presented Test results External consultation Experimental No No No No Experimental
Schema of Dynamic Scheduling of Outpatient Appointments Under Patient No-Shows and
Nan Liu, Serhan Ziya and Vidyadhar G. Kulkarni Theoretical / Experimental Experimental
Area of application External consultations Dynamic Scheduling of Outpatient Appointments Under Patient No-Shows and Cancellations. Department of Statistics and Operations Research, University of North Carolina Year Title Journal Source Attention to the time passed between the request of the appointment and the day of the appointment 2010 139 simulation Cost function to optimize Cost reduction
Results presented
Test results Schema of Designing appointment scheduling systems for ambulatory care services article
General Author Tugba Cayirli, Emre Veral and Harry Rosen
Designing appointment scheduling systems for ambulatory care services.
Hofstra University, Department of Management Theoretical / Experimental Experimental
Area of application Delays Cancellations Patient’s preferences Emergencies Priority of emergencies Attention to the time passed between the request of the appointment and the day of the appointment Cost function to optimize Results presented Test results Outpatients clinics Experimental No No No No Experimental
Combination of waiting time suffered by the patients and time of inactivity of the doctor) Schema of Dynamic multi-appointment patient scheduling for radiation therapy article
General Author Walter J. Gutjahr, Marion S. Raunerb
Attention to the time passed between the request of the appointment and the day of the appointment An ACO algorithm for a dynamic regional nursescheduling problem in Austria Sauder School of Business, University of British Columbia Experimental 2012 41 Elsevier Treatment Experimental No No No Theoretical Experimental Yes simulation Cost function to optimize Waiting time patients
Results presented
Test results Attention to the time passed between the request of the appointment and the day of the appointment 2014 10 Dynamic scheduling with due dates and time windows: an application to chemotherapy patient appointment booking Health Care Management Science Centre for Maintenance Optimization Reliability Engineering, Department of Mechanical Industrial Engineering, University of Toronto and Operations and Logistics Division, Sauder School of Business, University of British Columbia, Experimental Treatment Experimental Experimental No No Experimental Experimental Yes simulation Cost function to optimize Cost reduction
Results presented
Test results Schema of Clinic Overbooking to Improve Patient Access and Increase Provider Productivity article
General Author Linda R. LaGanga and Stephen R. Lawrence
Clinic Overbooking to Improve Patient Access and Increase Provider Productivity Decision Sciences Institute Mental Health Center of Denver and University of Colorado at Boulder Theoretical / Experimental Experimental Attention to the time passed between the request of the appointment and the day of the appointment Cost function to optimize Cost reduction Outpatients clinics Experimental Experimental No No No Experimental Yes simulation
Results presented
Test results Schema of Revenue management for a primary care clinic in the presence of patient choice article Element Characteristic Values General Author Diwakar Gupta and Lei Wang
Revenue management for a primary care clinic in the presence of patient choice.
Operations Research Department of Mechanical Engineering, University of Minnesota, Minneapolis, Minnesota and SmartOps Corporation, Pittsburgh, Pennsylvania Theoretical / Experimental Experimental Attention to the time passed between the request of the appointment and the day of the appointment Cost function to optimize Cost reductions No No No No Experimental Experimental Yes Statistical
Results presented
Test results
We study the distribution of these articles over time thanks to Figure 3.
Although we initially wanted to obtain articles from the last 10 years, we can conclude that this claim was too ambitious. We have got only eight articles and few are close to the current year. On the contrary, it is easier to find articles related to the subject before 2010. However, we wanted to maintain the initial idea, in order to try to write a literature review as current as possible.
With relation to the source of each article, note that all articles have been found in all of the chosen bibliographic sources. That is, no article was discovered exclusively in any of the sources. 3.3
Having commented the selected articles, we proceed to evaluate the quality of these according to the norms defined in section 2.4. This evaluation is shown in Table 14.
QA1 Yes Yes Partially Partially Partially
QA2 Yes
QA3
Yes Yes Yes Yes Yes
Yes Yes Yes
No Partially
Yes
No Partially
Yes
Yes Partially
Yes
No
QA4 Partially Partially Yes Yes Yes Yes Yes Yes
QA5 Partially Partially Yes Yes Yes Yes Yes Yes
QA6 Yes
QA7 Yes Yes Yes Yes
Yes Yes Yes Yes Yes Yes
Yes Yes
Yes No
Yes In Figure 4 we visualize the percentage of quality, differentiating the possible values that can take, indicated in section 2.4.
The QA2 and QA7 quality criteria are perfectly satisfied with 100% of acceptance. On the one hand, we found, thanks to QA1, that reading about 60% of the articles is a bit difficult if one is not an expert in these topics. On the other hand, almost 40% of the articles do not present any review to future investigations, fact remarkable. Related to the results and the test of algorithms (i.e., QA4 and QA5), we were satisfied with the conclusion of the evaluation, since all of the selected articles present results or they are tested.
Finally, in terms of QA6, we find that only an article does not describe the sanitary environment of application. 4
In this section, we answer the research question defined in section 2.1 considering the knowledge assimilated from the set of articles. Note that the main objective is to know the current state of dynamic programming in the management in health centers, focusing on certain challenges launched by (D Gupta and Denton, 2008). Therefore, the analysis will divide this section into parts, each one corresponding to one of the research questions.
Despite intense research on software that supports dynamic appointment management in health environments, no article has been got in which the implantation of this technique in a real sanitary environment is reflected.
A fundamental idea, launched by (D Gupta and Denton 2008). It is a factor that should not be neglected because it directly affects the probability of cancelling an appointment or being delayed. This relationship is completely logical. If the patient's preferences are not taken into account, the patient can lose the interest in the appointment, favouring delays or forgetfulness.
However, few articles discuss this subject. Only three
articles of the eight contemplated, mention the preferences
of the patient, two theoretically and one experimentally.
This may be due, as mentioned
Therefore, we should continue to investigate and research on how to keep patient preferences in the model.
Finally, it should be emphasized that none of the three articles say that contemplating these priorities can produce penalties in the function to be optimized. It's just a design challenge.
Practically all studies have been done in all possible health areas: outpatient clinics, external consultations, nursing, treatment and surgery. We have not found articles that have tried to simulate the operation of a hospital. It is logical because is easier to start reproducing smaller environments.
In Figure 5 we show the distribution of articles by area of application.
It is not surprising that the sum of each block shown in Figure 6 illustrates more than eight articles because several articles dealt with more than one application domain. Studying this graphic we discerned that we find more articles related to the Ambulatory environment, followed closely by the External Consultations since both domains are similar.
On the contrary, Surgery and Treatment are environments more hostile. More characteristics that are vital must be taken into account in the other domains. For example, for treatments, we include the articles (Gocgun and Puterman, 2014; Saur et al., 2012), which are intended to manage appointments for the treatment of chemotherapy and radiotherapy respectively. In this scenario, times are essential to ensure the best saving percentage. Without forgetting the urgencies. Moreover, cancellations and delays have a great impact on time and, significantly, on costs because they are expensive treatments.
About surgery, thanks to one of our basic articles (D Gupta and Denton, 2008), we know the difficulty to model an appointment management system for that environment. The main impediment is the impossibility of generalizing the operating times. Not all bodies are the same, is possible to appear complications in the surgery, and not all surgeons operate at the same speed. Of course, without forgetting the emergencies. The emergencies are vital so we should minimize the waiting time of them, without neglecting or disfavouring the rest of the patients.
Therefore, although all are domains of the health world and the properties to contemplate are similar, when designing them it is essential to obtain close appointments, not very far in time, and it is indispensable to quickly manage urgencies.
About the possible cancellation or non-presentation of the patient, we show in Figure 6 the attribute of the schema definition designed in Section 2.5, with the values obtained for the set of articles formed.
Note that half of the selected articles (four) do not contemplate the possible cancellations, a very common fact and for which the dynamic scheduling of appointments would be a great help. If a patient cancels his appointment, and we have a dynamic appointment manager, that gap will be reused efficiently.
The experimental study
In the one hand, only
On the other hand, (D Gupta and Denton 2008) claims to investigate the reason for such cancellations or absences.
It is an essential aspect in the sanitary domains and it is difficult to model. In Figure 7 we show a comparison of both properties, urgencies and its prioritization, looking to observe how many of the selected articles perform experimental tests contemplating emergencies and also prioritization.
Although five of the articles selected, out of a total of eight, perform experimental tests attending urgencies, only two of them consider their prioritization.
However, such studies warn of the complexity of
modelling a dynamic appointment management system that
accepts urgency.
After performing the analysis, we highlight the key points to take into account to model an appointment management system for a health environment.
The main point is to know the characteristics of the sanitary environment that we wish to model. In this systematic review of the literature, we have found modeling about outpatient centers, outpatient consultations, treatments and surgery. Among them, they present great differences that have to be considered for a correct modeling of their appointment’s manager. In short, the challenges of the health center, as well as its objective, must be clear.
For example, the goal of an outpatient may be to plan the maximum possible number appointments in the day (always on a minimum quality of care). A cancellation per day in this environment may not have much influence on costs.
However, if we move to the management of chemotherapy appointments, the indirect waiting time of the patient (time it takes from the date that a patient requests for an appointment until the date of the appointment) begins to become vitally important, since the percentage of salvation depends on it. In addition, cancellations in such expensive treatments are costly.
As already noted (D Gupta and Denton 2008), the indirect
waiting time and the probability of cancellation or
nonpresentation of the patient are closely related. However,
after this study, we have not found any methodology to
follow to solve this problem. For example, if we try to
decrease the probability of cancellations, paradoxically the
indirect waiting time increases according to
As future work, it could be interesting to carry out an analysis of the cancellations and non-presentations main causes, in order to better address the problem.
About patient preferences as a method to reduce the probability of cancellation or non-presentation of the patient, its modeling and its execution are complex. However, we believe that it should continue to be a line of future research. One line of research could be the restructuring of the management of appointments depending on the social actor who requests it. For example, preventing that retired people collapse the earliest appointments in the morning, when they may be the most accessible for workers. A study of such strategies would be desirable to see if they would improve the probability of cancellations and forgetting.
In short, we must continue to investigate new ways to avoid cancellations and non-presentations by patients, always keeping in mind their close relationship with the indirect waiting time.
Other key pillars in modeling any appointment manager for a health care environment are urgencies and their prioritization. In this regard, we must continue to investigate since the solution proposed by Caylani, Veral, and Rosen 2006, Gurgun and Puterman 2014, Diwakar Gupta and Law 2008, Laganga and Lawrence 2007, Saur et al. Al., 2012) not used the full potential of dynamic programming. After all, they are recovering time gaps for possible emergencies, and if they are not used, we will lose that time.
Despite the problems still present, decision-making techniques are fully applicable in health domains. Although the results indicate that the current state of technology is on the right way, there is still a long way to obtain reliable software.
In fact, nowadays many clinics continue to manage their appointments under the supervision of a person, without any support at all. So, it is necessary to emphasize that this technology is not being developed to supplant people, only to help them. As stated, this is a domain in which the mistakes are paid expensive, so it would be helpful to have a software that calculates the most optimal appointment for the patient who requests it, without losing the person in charge of full control of the schedule. That is, health professionals should not view this technique as a threat because is only a support for them. So, we must continue working on improving this technique, as well as giving visibility to its advantages in the health world.
Harry Rosen.,
Systems for Management [Gocgun, Yasin, and Martin L. Puterman, 2014] “Dynamic Scheduling with Due Dates and Time Windows: An Application to Chemotherapy Patient Appointment Booking.” Health Care Management Science 17(1): 60–76. [Gupta, D, and B Denton, 2008] “Appointment Scheduling in Health Care: Challenges and Opportunities.” IIE Transactions 40(9): 800–819. [Gupta, Diwakar, and Wang Lei, 2008] “Revenue Management for a Primary-Care Clinic in the Presence of Patient Choice.” Operations Research 56(3): 576–92. http://search.ebscohost.com/login.aspx?direct=true&db=bth &AN=33204706&site=ehost-live. [Huang, X, 1994] “Patient Attitude Towards Waiting in an Outpatient Clinic and Its Applications,” Health Services Management Research,7,2–8. [Laganga, Linda R., and Stephen R. Lawrence., 2007.]“Clinic Overbooking to Improve Patient Access and