Available online at www.sciencedirect.

com
ScienceDirect
ScienceDirect
Procedia Computer Science 00 (2019) 000–000
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2019) 000–000 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
ScienceDirect
Procedia Computer Science 164 (2019) 610–617

CENTERIS - International Conference on ENTERprise Information Systems / ProjMAN -

CENTERIS
International - International
Conference Conference
on Project on ENTERprise
MANagement / HCist -Information Systems
International / ProjMAN
Conference -
on Health
International Conference on Project
and Social Care MANagement / HCistand
Information Systems - International
TechnologiesConference on Health
and Social Care Information Systems and Technologies
Analyze in Depth Health Care Business Process and Key
Analyze in Depth Health
Performance Careusing
Indicators Business Process
Process and Key
Mining
Performance Indicators using Process Mining
Emna Ammar Elhadjamora*, Sonia Ayachi Ghannouchib
Emna Ammar Elhadjamora*, Sonia Ayachi Ghannouchib
a
ISITCom Hammam Sousse / RIADI Laboratory-ENSI Manouba, Tunisia
a b
ISITComISG Sousse /Sousse
Hammam RIADI /Laboratory-ENSI Manouba,
RIADI Laboratory-ENSI Tunisia Tunisia
Manouba,
b
ISG Sousse / RIADI Laboratory-ENSI Manouba, Tunisia

Abstract
Abstract
Process mining is applied in literature to many case studies in various domains. It represents an explorative procedure, very
helpful to
Process the domain
mining experts
is applied for analyzing
in literature to manydata
casefrom multiple
studies perspectives.
in various domains.InIt this paper, an
represents weexplorative
concentrateprocedure,
on extracting
very
knowledge
helpful to thefromdomain
event logs related
experts for to a healthdata
analyzing care from
process that is perspectives.
multiple recorded by aInBusiness Process
this paper, ManagementonSystem
we concentrate and
extracting
mining
knowledgeKeyfrom
Performance
event logs Indicators
related tocontent associated
a health to process
care process that isinstances
recordeddata.
by aThus, this work
Business Processcontributes to theSystem
Management evaluation
and
and the enhancement
mining Key Performance of theIndicators
healthcarecontent
processassociated
using Process mininginstances
to process tool. data. Thus, this work contributes to the evaluation
and the enhancement of the healthcare process using Process mining tool.
© 2019 The Authors. Published by Elsevier B.V.
© 2019
© 2019
This is anThe
The Authors.
open Published
accessPublished
Authors. by Elsevier
article under
by Elsevier B.V.
the CC BY-NC-ND
B.V. license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
This is an
Peer-review open access
under article under
responsibility the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
This is an open access article underofthetheCCscientific
BY-NC-ND committee of the CENTERIS - International Conference on ENTERprise
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the CENTERIS -International Conference on ENTERprise
Information Systems
Peer-review / ProjMAN of
under responsibility – International
the scientificConference
committee on of Project MANagement
the CENTERIS / HCist - Conference
- International InternationalonConference
ENTERprise on
Information Systems / ProjMAN - International Conference on Project MANagement / HCist - International Conference on
Health
HealthandandSocial
Information Care
Systems
Social / Information
Care ProjMAN
Information –Systems andand
International
Systems Technologies
Conference
Technologies.on Project MANagement / HCist - International Conference on
Health and Social Care Information Systems and Technologies
Keywords: Healthcare Business Process; Key Performance Indicator; Business Process Management; Process mining
Keywords: Healthcare Business Process; Key Performance Indicator; Business Process Management; Process mining

*
Corresponding author. Tel.: +0-000-000-0000 ; fax: +0-000-000-0000 .
*
Corresponding [email protected]
E-mail address:author. Tel.: +0-000-000-0000 ; fax: +0-000-000-0000 .
E-mail address: [email protected]
1877-0509© 2019 The Authors. Published by Elsevier B.V.
This is an open
1877-0509© access
2019 Thearticle under
Authors. the CC by
Published BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Elsevier B.V.
Peer-review
This under
is an open responsibility
access of the
article under the scientific
CC BY-NC-NDcommittee of the
license CENTERIS - International Conference on ENTERprise Information Systems /
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
ProjMAN
Peer-review– International Conference
under responsibility of theonscientific
Project MANagement / HCist
committee of the - International
CENTERIS Conference
- International on Health
Conference onand Social Care
ENTERprise Information
Information Systems /
Systems and
ProjMAN TechnologiesConference on Project MANagement / HCist - International Conference on Health and Social Care Information
– International
Systems and Technologies
1877-0509 © 2019 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the CENTERIS -International Conference on ENTERprise Information
Systems / ProjMAN - International Conference on Project MANagement / HCist - International Conference on Health and Social Care
Information Systems and Technologies.
10.1016/j.procs.2019.12.227
 Emna Ammar Elhadjamor et al. / Procedia Computer Science 164 (2019) 610–617 611
2 Author name / Procedia Computer Science 00 (2019) 000–000

1. Introduction

Business Process Management has many advantages such as enhancing processes to suit evolving business needs
and measuring and optimizing processes leading to effective overall company performance and ensuring that the
processes still meet the goals for which they were created. In [1], Van Der Aalst et al define Business Process
Management (BPM) as a discipline that advocates the use of continuous improvement on processes. This discipline
builds upon earlier work on workflow management, operational research, lean manufacturing, six sigma, and
business process re-engineering. It also combines knowledge from information technology, management science,
and industrial engineering.
BPM involves a series of phases which can be depicted in a continuous life-cycle. In literature, there are different
BPM life-cycles proposed by several researchers. The BPM life-cycle [2] defines how to improve processing based
on the knowledge of historical executions. In this paper, we focus on BPM lifecycle described by Weske in [3]. In
the design and analysis phase, the processes are identified, (re)designed and validated using for example simulation
techniques to verify that the process actually exposes the desired behavior. A Business Process Modeling Notation is
used in this phase. In the configuration phase, BP model is implemented by configuring an IT-system. The
enactment phase incorporates the execution of the business process using the system configured. Using the
monitoring component, the administrator can track and visualize the status of each process instances and other
information with the help of the IT-systems. In the evaluation phase, the BP is diagnosed and the information is
collected to evaluate and improve business process model and its implementation. The last phase involves the
continuous monitoring of processes where many techniques can be used such as process mining to analyze the
information stored in event data. Process Mining is a collection of techniques which aim to discover, monitor and
improve business processes [4].
Business Process Management incorporates Business Process Management Systems (BPMSs) as mechanisms to
optimize and automate organizations daily procedures. These BPMSs manage the information related to the
instances (business data) related to the execution of the BP model and recover the information concerning the
process performance (Process Performance Indicators) [5].
Typically, execution logs generated in the enactment of the BPM phase are used to store information about
processes real execution, such as the start or the end of activities. The use of a BPMS facilitates the monitoring of
Key Performance Indicators (KPIs). The definition of Key Performance Indicator has drawn the attention of many
authors in several domains. For instance, in [6] Villalón et al define an indicator as a metric that evaluates the
progress and the performance of the process with respect to some objectives. The authors add later that such metrics
can be directly derived from data or can use a formula to combine values. In [7], Kritikos et al state that KPIs
usually include a metric and a threshold imposed on it, thus defining the respective minimum performance level to
be sustained. In [8], Looy et al note that performance indicator should also have a concretization that makes clear
how it is measured, and its value should be compared to a target value.
BPMSs generate and store information about their performance, such as the number of instances of a business
process execution, the duration time of each activity, who has executed each activity, the resources involved, the
frequency of each activity, and the number of instances [5]. In [9], the authors evaluate each BPMS consistently by
calculating the total support level of every BPMS and sorting evaluated systems according to their level of support
for all evaluation criteria. The best rate is provided by jBPM followed by Camunda, YAWL, Bonita, Activity,
ProcessMaker, and uEngine.
The process considered in this work is the process of healthcare. The proposed business process model is
deployed with the use of jBPM software, where 100 instances have been created through the execution of the
process. There are many measurements that could be used in the evaluation of healthcare process, but after
discussion with experts, we arranged a list of relevant indicators. Our Data comes essentially from BPM database
system and estimated KPI files. The ProM process mining software was chosen in this research. This tool has an
intuitive interface and allows automatically constructing process visualization.
The remainder of this paper is organized as follows: in section 2, we use process mining to analyze emergency
healthcare process. In section 3, we analyze KPI data file. This file contains quantitative and qualitative indicators
target values related to each process instance. The last section gives a brief conclusion.
612 Emna Ammar Elhadjamor et al. / Procedia Computer Science 164 (2019) 610–617
Author name / Procedia Computer Science 00 (2019) 000–000 3

2. Business Process mining

Before we apply process mining we should prepare and clean the log to answer some questions related to the data
we are interested in. For example, how many process instances/tasks/resources are in the log? What are the most
frequent paths in the process? Are there any loop paths in the process? Thus, in this step, we are based essentially on
Bamtasksummary (figure 1) because it contains information about the dates and times on which tasks were
executed.

Figure 1. Screenshot of BAMTaskSummary Table

To do that we import to ProM the corresponding event log which contains essentially activities that happen, the
process instance identifier, and the start/end date timestamp. Based on the “log visualizer”, we especially focus
on the “Summary” tab and we get the following visualization (figure 2).

Figure 2. Screenshot of process log summary related to the EC process

With ProM, we can also inspect activities occurrences from the event logs. In this view, the user will get
overview information about the number of different classes of registered execution scenarios. We can consider the
role of this log view like the dash builder module provided by jBPM because both of them provide some
performance information about the process. The coloring of activities provided by the software is very important to
see for example how frequently each activity occurs or the main paths of the process flow (figure 3).
 Emna Ammar Elhadjamor et al. / Procedia Computer Science 164 (2019) 610–617 613
4 Author name / Procedia Computer Science 00 (2019) 000–000

Figure 3. Screenshot of possible paths related to the EC process

Based on this view (figure 3), our analysis is based essentially on the completed process instances only. In this
way, the decision maker can see how the process has been actually performed and the most frequent completed
paths and also can analyze the distribution of all cases over the different paths through the process. We can see that
all cases start with the registration activity. Afterward, all patients enter the sorting room for primary exams. In
conclusion, we can say that the main process flows (67% of the log) contain 3 activities which are registration,
sorting, and consultation in box. If we have more complicated BP and more BP traces we can see for example which
process steps happened multiple times? Which activity takes a lot of time before starting? How many actors are
involved in a specific case? In this way, it becomes easy to detect paths or activities that cause a bottleneck. The
more data attributes we have in the logs, the more additional statistics we get. These data are very important for the
process analysis because they hold relevant context information such as: By which actor the activity was handled, in
which department the process was performed, etc.

Figure 4. Screenshot of some cases related to the selected flow

614 Emna Ammar Elhadjamor et al. / Procedia Computer Science 164 (2019) 610–617
Author name / Procedia Computer Science 00 (2019) 000–000 5

We can find out more information when we inspect this log. For example, we can inspect the process instances
and browse the set of cases in the log (figure 4). We can also see the events in a selected case to get a detailed view.
For example, figure 4 shows information related to process instance id number 153.
Second, using process mining tools, we can construct a more critical vision based on Process discovery in order
to suggest a process remodeling or to analyze gaps in the process. The process mining software supports various
actions to mine the control flow perspective of process models in order to see how cases are actually being executed
in the organization. For example, when we use the “α-algorithm” plug-in, it allows to visualize a Petri Net graph
and then we can convert this graph to a BPMN diagram. In this way, we obtain the process model that summarizes
the sequences of activities followed by most/all cases in the log.
Based on all available information, we can set the following points:
- The process takes too much time for patients who have a delayed emergency. Especially, it takes an important
waiting time before the delayed emergency starts.
- The process takes too much time for patients who are oriented to the supervision room. Especially, it takes an
important time to lying waiting activity to end.
- In some cases, the consultation activity takes too much time to end.
When we review these observations with the process stakeholders, we conclude the following explanation:
- From the process manager view, patients who are in the delayed emergency can wait because they are not
urgent.
- For the considerable time occupation in the lying waiting, it is explained by the fact that the health of these
patients is not stable.
- For the important time taken in the consultation in box, it is explained by the fact that some patients required
complementary exams which are frequently made in other departments of the hospital.
Third, in order to improve business processes and KPIs, we consider the mining of KPIs as an explorative and
highly interactive task where the decision makers learn to understand the data in two ways. First by understanding
KPI relationships retrieved from the knowledge database and second by using some data retrieved from KPI
estimated file and process instance log in order to look at the process from multiple views using ProM software.

3. Key Performance Indicator mining

We proceed by using ProM software. In this step, we analyze KPI target values retained from KPI estimated file
(figure 5) and we combine data with the appropriate rows and columns retrieved from process instance log table by
using the same process instance identifier. The KPI file “estimated values” contains all evaluated data after
comparing the real value with the target value or threshold value set by the decision maker. This file contains two
values (“ok” codes the KPI value which is considered by the organization as acceptable and “not ok” corresponds to
the fact that the KPI value is not acceptable).

Figure 5. Screenshot 2 of some KPI estimated values

After preparing the data set, we import the file into ProM. We can get further analysis by analyzing some
relationships between indicators. For example, we analyze the level of satisfaction Quanti_KPI9 and the waiting
time before sorting with Quali_KPI13 installation in the sorting room. The aim of this analysis is to check if the
waiting time before sorting, respects business rules (i.e. evaluated as OK value), also we check if the qualitative
indicators related to this quantitative indicator retained from patient experience are respected (i.e. satisfaction of the
 Emna Ammar Elhadjamor et al. / Procedia Computer Science 164 (2019) 610–617 615
6 Author name / Procedia Computer Science 00 (2019) 000–000

patient). To do that, we focus on log visualization to inspect some statistics about it. Figure 6 shows the
corresponding results.

Figure 6. A screenshot of statistics related to Quanti_KPI9 with Quali_KPI13

In this figure, we can observe that 42% of patients are satisfied to very satisfied about their installation in the
sorting room and the waiting time before sorting also respects the target value (ok value). The waiting time before
sorting doesn’t respect the target value for 23% of cases, however, patients are satisfied to very satisfied with their
installation in the sorting room. We can also note that 16% of cases have a neutral response about this qualitative
aspect and the waiting time before sorting respects the target value. Now if we want to get some statistics in the case
where the quantitative and the qualitative indicators are not respected, we can find 9% of the log file. We can inspect
this statistic more deeply by browsing the set of cases in the log (figure 7). For instance, we can obtain a detailed
view including the corresponding process instances identifiers related to the case where the quantitative and the
qualitative indicators are not respected (green color).

Figure 7. A screenshot of some process instances identifier related to the selected variant

Using ProM and the appropriate file, we can combine as many attributes (indicators) as we want to get some
interesting statistics that help decision maker to evaluate the performance of the process. We take for example the
combination of Quali_KPI3 paramedical staff availability, Quali_KPI4 medical staff availability and Quanti_KPI9
the waiting time before sorting; we obtain the following traces (figure 8).
616 Emna Ammar Elhadjamor et al. / Procedia Computer Science 164 (2019) 610–617
Author name / Procedia Computer Science 00 (2019) 000–000 7

Figure 8. A screenshot of statistics related to Quali_KPI3, Quali_KPI4, and Quanti_KPI9

In figure 8 we focus on four interesting cases. The first one, when all indicators in question respected the desired
value (all of them are ok) which takes the major found percentage (42%). The second one, when patients are
satisfied to very satisfied about the qualitative aspect related to the availability of medical/paramedical staff with the
quantitative aspect related to the waiting time before sorting which is not ok (did not respect the target value). We
can find in this category 32% of the log. The third one, when the patient is not satisfied to very dissatisfied (not ok
value) about the qualitative aspect with the waiting time before sorting which respects the target value, we found
11% of the log in this category. The fourth one, in the case that this combination didn’t respect the desired value
which is a very rare case (takes only 2% of the log). Like the first example, we can browse cases to verify if there
really exists some patient cases that attend less than 5 min before entering the sorting room (Quanti_KPI9 the
waiting time before sorting is ok). However, when they entered to the sorting room, neither paramedical staff nor
medical staff is available (Quali_KPI3 paramedical staff availability and Quali_KPI4 medical staff availability are
not ok). This case is infrequent, only 11% cases were found, but it is important to see concrete examples particularly
for “strange” behavior and to help process stakeholder to find things that are hard to believe until they have drilled
down to an individual example case.
To summarize, the majority of patients were satisfied during their experience in the emergency department and
the majority of quantitative indicators respect the desired target values based on process instances recorded in the
system. However, in rare cases, we observe some deviations. As an example, BP doesn’t take too much time but
patients already had a bad experience. For example, the delayed emergency duration respect the target value.
However, patients are not satisfied about the clarity of information provided by doctors in this activity. Some
patients are not satisfied in the lying waiting because the regularity of doctor visit is not adequate.
We can conclude that process mining software offers different views and different performance information for
all activities of the process and resources included in the data set. Furthermore, data are usually very important for
the process analysis, because they hold relevant context information and orient the decision makers to set additional
questions and goals, such as the following:
-Which type of patient category required more attention?
-In which activity the process was performing (respect the qualitative and the quantitative aspect)
 Emna Ammar Elhadjamor et al. / Procedia Computer Science 164 (2019) 610–617 617
8 Author name / Procedia Computer Science 00 (2019) 000–000

We often need more attributes and more cases that are relevant and can be used in the analysis in order to respond
to these questions. After discussion with patients and decision makers we can state below some interesting
observations:
- The privacy of the patient is not taken into account in the sorting room.
- The necessity of staff to orient the patient according to his/her healthcare status to the appropriate service
location
- The necessity to have a system to ensure that data related to each patient (previous analysis, current healthcare
status …) are collected consistently, both within, and across the healthcare departments. Similarly, diagnosis exam
results are missing or overlooked and as a consequence appropriate treatment is delayed or repeated.

4. Conclusion

Based on KPI improvement (better selection of KPIs), different possibilities of BP improvement become
possible. More precisely, various BP perspectives may be concerned. For example, organizational, behavioral or
informational perspectives may be concerned by such improvement. In fact, based on the most important retained
KPIs and their corresponding values, BP improvement may concern the assignment of tasks to actors, the order in
which tasks have to be fulfilled or input data that will be taken into account for specific tasks.
Therefore, performance indicators can be searched according to that occurrence. Furthermore, decision-makers
prefer describing performance measurement indicators that are relevant to their BP. In most cases, decision makers
can include managers who don’t have to know the details of the indicator. In short, there are still pressing needs for
a view that can support performance measurement and BP activities.
Globally, we conclude that the majority of patients were satisfied to very satisfied in the emergency department.
However, improvement suggestions are essential for this process, as they enable patients and healthcare providers to
effectively identify where the performance of healthcare services requires improvement.

References

[1] Van Der Aalst, W. M., La Rosa, M., & Santoro, F. M. (2016). , ‘Business Process Management: Don’t Forget to Improve the Process!’, Bus.
Inf. Syst. Eng., vol. 58, no. 1, pp. 1–6.
[2] Dumas, M., La Rosa, M., Mendling, J., & Reijers, H. A. (2013). Fundamentals of business process management (Vol. 1, p. 2). Heidelberg:
Springer.
[3] Weske, M. (2007). Business Process Management–Concepts, Languages, Architectures, Verlag. ISBN 978-3-540-73521-2. (Pages xv, 4, 14,
15, 16, 17, 18, 20).
[4] Van Der Aalst, W. (2011). Process mining: discovery, conformance and enhancement of business processes (Vol. 2). Heidelberg: Springer.
[5] Perez-Alvarez, J. M., Gomez-Lopez, M. T., Parody, L., & Gasca, R. M. (2016, September). Process instance query language to include
process performance indicators in DMN. In 2016 IEEE 20th International Enterprise Distributed Object Computing Workshop (EDOCW) (pp.
1-8). IEEE.
[6] Poveda Villalón, M., Radulovic, F., & García Castro, R. (2014). Developing ontologies for representing data about key performance
indicators. In: "2nd eeBuilding Key Performance Indicators Workshop (KPIs)", Niza, Francia. pp. 1-16.
[7] Kritikos, K., Plexousakis, D., & Woitsch, R. (2017). Towards Semantic KPI Measurement. In CLOSER(pp. 63-74).
[8] Van Looy, A., & Shafagatova, A. (2016). Business process performance measurement: a structured literature review of indicators, measures
and metrics. SpringerPlus, 5(1), 1797.
[9] Meidan, A., García-García, J. A., Escalona, M. J., & Ramos, I. (2017). A survey on business processes management suites. Computer
Standards & Interfaces, 51, 71-86.