CONTINUOUS CLINICAL PATHWAYS EVALUATION

BY USING AUTOMATIC LEARNING ALGORITHMS

Carlos Fern

andez-Llatas

, Teresa Meneu

, Jose Miguel Bened

and Vicente Traver

ITACA, Universidad Polit

ecnica de Valencia, Valencia, Spain

DSIC, Universidad Polit

ecnica de Valencia, Valencia, Spain

Keywords:

Process mining, Clinical pathways, Pattern recognition, e-Health, Decision support systems, Process standard-

ization.

Abstract:

The standardization of care processes in medicine, like Clinical pathways, is becoming more and more a

common practice in health care organizations. Nevertheless, their design is not an easy task. Some approaches

in the literature are based on using Workﬂow technology for deﬁning Clinical Pathways. These approaches

allow the creation of unambiguous, complete and automatically executable protocols. In addition to this, the

use of Process Mining technology can help the design using information from real executions of Clinical

Pathways cases. Nevertheless, to ensure a correct continuous evaluation and improvement of care processes,

the creation of a tool that allows to know the current status of the Clinical Pathway execution it’s mandatory. In

this paper, we present a tool able to compare the designed Clinical Pathways with the real implantation cases

in order to detect their differences. This allows Clinical Pathways designers to improve the care protocols

making them more adequate to real cases.

1 INTRODUCTION

Nowadays, current trends on process management are

focused in the standardization of processes in order

to evaluate and improve the strategy to achieve bet-

ter quality and efﬁciency in most common actions. In

Health environments, the care processes standardiza-

tion is becoming more and more present. The use

of standardized protocols, for normalizing the care

process, like Clinical Pathways (Audimoolan et al.,

2005) is more and more adopted by health profession-

als in order to improve the quality of care of patients.

The use of Clinical Pathways for Health Care stan-

dardization is a difﬁcult task. Clinical Pathways are

usually multidisciplinary complex documents that co-

ordinate the actions of patients, doctors and the rest of

involved care process stakeholders. In order to create

efﬁcient care protocols it’s necessary to avoid ambi-

guity, describe completely the whole problem and fa-

cilitate the use of ICT (Information and Communica-

tion Technologies) to make the system usable.

The use of Workﬂow Technology (WfMC, 1999)

can help doctors to design formal, non ambiguous and

automatically executable care protocols to describe

complete Clinical Pathways. In addition, the Process

Mining technology can be used to automatically infer

workﬂows to facilitate the design of the processes in

a formal way (Fern

andez-Llatas et al., 2010).

Nevertheless, the control of the implantation of

Clinical Pathways requires a high level view that sup-

ports the detection of inefﬁcient processes and bottle-

necks. This paper presents a tool created to facilitate

this management and to evaluate the ﬁtness of execut-

ing Clinical pathways by automatically comparing the

real execution care protocols cases with the designed

Clinical Pathway. This tool uses Pattern Recognition

techniques to detect and quantify exceptions occurred

in real cases. These exceptions are divergences from

the ’expected’ path that was designed and the real per-

formance that can be later analyzed by the healthcare

organizations and corrected if needed and possible.

The paper is structured in the following way.

Firstly the solution to Clinical Pathway design us-

ing Workﬂow languages in described. The following

section explains how process Mining techniques can

be used to help the design of Workﬂow based Clin-

ical Pathways. Following this, a tool for the contin-

uous evaluation of ongoing Clinical Pathways is pre-

sented. The last part summarized the conclusions of

this work.

228

Fernández-Llatas C., Meneu T., Benedí J. and Traver V..

CONTINUOUS CLINICAL PATHWAYS EVALUATION BY USING AUTOMATIC LEARNING ALGORITHMS.

DOI: 10.5220/0003153902280234

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2011), pages 228-234

ISBN: 978-989-8425-34-8

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

2 CLINICAL PATHWAYS AS

HEALTH CARE

STANDARDIZATION MODELS

Since the appearance of the Evidence Based

Medicine(EBM) paradigm (Elstein, 2004) that was

born in the early 90’s, the works focused on ap-

plying the scientiﬁc method to medical research are

each year growing signiﬁcantly. The idea of stan-

dardizing health care processes is more and more

adopted to improve daily actions of nurses, doctors

and the rest of stakeholders. In some of the most

commons medical knowledge public repositories, like

COCHRANE (The Cochrane Collaboration, 2010) or

PubMed (PubMed Library, 2010), it’s possible to ﬁnd

detailed care protocols to address the management of

almost any disease. These standardized health care

protocols, also known as Clinical Pathways (Audi-

moolan et al., 2005), deeply describe all the steps that

patients, doctors, nurses and the rest of the stakehold-

ers must follow to perform the care process efﬁciently

and efﬁcaciously.

These protocols are usually described in natural

language. This is due to the high expressivity needed

by Health professionals to describe care protocols.

The situations that a Clinical Pathway can express

are very complex. However, this use of natural lan-

guage supposes a big problem when moving to Clini-

cal Pathways interpretation. Natural language models

have clear ambiguity problems. Different health pro-

fessionals can understand the same phrase in totally

different way. This situation can lead to an erroneous

interpretation of a Clinical Pathway that might affect

the patient’s quality of care. In addition to this, deﬁn-

ing Clinical Pathways using natural language does not

allow to ensure that the process is completely deﬁned.

In that case, it’s possible that the process has unde-

ﬁned states that must be lucubrated by health profes-

sionals.

Other reason to avoid natural language in Clinical

Pathways design it’s the impossibility that this bring

to the automatic execution of guidelines. Currently,

the healthcare organizations are more and more com-

puterized and the data is each time more available

in digital way. The use of dynamic systems that al-

low the computerized guidance of the health profes-

sional in the care process following the Clinical Path-

ways description will suppose a crucial help to the

deployment of those protocols. This is important be-

cause, in some cases like multidisciplinary protocols,

the high bureaucratic load and the required coordina-

tion of speciﬁc teams demands the use of computer

systems in order to allow the standardization of pro-

tocols. Natural language is not understood by com-

puters. For that reason, in order to create systems en-

abling the guidance of health professionals with auto-

matic guidelines it’s necessary to translate the process

deﬁnition to a formal language that could be executed

automatically.

There are some efforts in literature focused in ad-

dressing this problem. Some high level languages

have been deﬁned to formally describe Clinical Path-

ways: GLIF language (Peleg et al., 2001) was de-

ﬁned to describe and evaluate health care protocols;

In Carepaths project (Naranjo et al., 2006) an ontol-

ogy for describing and automatically executing Heart

Failure Clinical Pathways was developed. Other

works are purely based on Workﬂow technology to

describe general purpose and high expressivity lan-

guages for Clinical Pathways deﬁnition (Dominguez

et al., 2008).

Workﬂow Technology is a research ﬁeld focused

on the creation of process speciﬁcation languages and

its dynamic execution. A Workﬂow is deﬁned as the

automation of a business process, in whole or part,

during which documents, information or tasks are

passed from one participant to another for action, ac-

cording to a set of procedural rules (WfMC, 1999).

A Workﬂow is a formal description of a process de-

signed to be automated. A Workﬂow language is a

formalism that allows the deﬁnition of Workﬂows. A

formal Workﬂow can be automatically executed in a

computerized system by using a Workﬂow Engine.

This will allow the automatic execution of standard-

ized protocols and the guidance of processes in com-

puterized environments.

The advantages of the use of Workﬂow technology

over other techniques in Clinical Pathways deﬁnition

are several. In the literature, we can ﬁnd available

some Workﬂow engines for automatic execution of

processes like jBPM from jBoss or Workﬂow Foun-

dation from Microsoft. Workﬂow languages have a

great understandability. The majority of Workﬂow

Languages have a graphical view and it’s possible to

design the execution of processes. A graphical deﬁni-

tion of a Workﬂow can be seen on Figure 2. In addi-

tion, there is the possibility to measure the expressiv-

ity of a Workﬂow language according to the Workﬂow

Patterns (van der Aalst et al., 2003a) to decide the

most adequate language to be used in the formal de-

scription. Workﬂow Patterns are ﬂow structures that

describe the state change rules in a process. Examples

of Workﬂow Patterns are sequences, choices, paral-

lel actions... etc. The more Workﬂow Patterns a lan-

guage is able to deﬁne, the better expressivity it has.

CONTINUOUS CLINICAL PATHWAYS EVALUATION BY USING AUTOMATIC LEARNING ALGORITHMS

229

07/05/2010 10:22:59 => i:1 BeginAction: Admission

07/05/2010 10:34:01 => i:1 EndAction: Admission Res: OK

07/05/2010 10:35:03 => i:1 BeginAction: Triage

07/05/2010 10:46:06 => i:1 EndAction: Triage Res: InHospital

07/05/2010 10:56:07 => i:1 BeginAction: TNS

07/05/2010 10:56:08 => i:1 BeginAction: TMP

07/05/2010 11:23:10 => i:1 EndAction: TNS Res: OK

07/05/2010 11:33:12 => i:1 EndAction: TMP Res: Fever

07/05/2010 11:43:19 => i:1 BeginAction: TMP

07/05/2010 11:50:19 => i:1 EndAction: TMP Res: OK

07/05/2010 11:51:24 => i:1 BeginAction: QualityTest

07/05/2010 12:00:28 => i:1 EndAction: QualityTest Res: OK

07/05/2010 12:15:29 => i:1 BeginAction: Discharge

07/05/2010 12:20:30 => i:1 EndAction: Discharge Res: OK

Figure 1: Example of Clinical Pathway case log.

3 PATTERN RECOGNITION

TECHNIQUES FOR CLINICAL

PATHWAYS AUTOMATIC

LEARNING

The use of Clinical Pathways does not ensure that the

processes was always executed in the same way. In

this line, it’s very common that the care process of a

patient is changed during execution. This change can

be motivated by the daily problems in the care process

(like administrative errors, malfunctioning or unavail-

ability of devices to perform medical tests described

on the care protocol, etc) and the variability of the

patient illnesses. For example, a pluripathologic pa-

tient can suffer changes in his treatment protocol due

to an allergy or due to contraindications of the pro-

tocol with the recommendations for other illness not

documented on that Clinical Pathway.

These exceptions to the care protocol can be

very useful to detect design problems in the Clin-

ical Pathway or underperformance of the care pro-

cesses. Knowing that information would enable the

correction and ﬁne-tuning of the Clinical Pathways in

an iterative process. The correction could be done

in the design or in the real execution problem, de-

pending on the cause for the divergences but the im-

portant touch-point would be to know when, how

and under which conditions those divergences hap-

pen. Using this model on each iteration the errors

detected will be used to correct the Clinical Path-

way that will be changed for the next iteration. This

problem can be approached in an automatic way us-

ing Process Mining Technology. The Process Min-

ing idea (also known as Workﬂow Mining) (van der

Aalst et al., 2003b) is the automatic learning of Work-

ﬂows for business process inference. Process Min-

ing algorithms use the execution samples to infer the

Workﬂow that describe the real process. Using this

technique the logs of care protocols actions applied to

each patient can be used to learn Workﬂows that for-

mally represent those Clinical Pathways. This helps

Clinical Pathways designers to modify previous iter-

ation processes according to the real deployment in-

corporating the new exceptions and correcting design

errors.

In (Fern

andez-Llatas et al., 2010) a Process Min-

ing algorithm (called PALIA) has been proposed to

discover the execution model of the Clinical Pathway

from the data gathered about the actions executed of

the patient past cases. This algorithm takes as en-

try data information about the activities performed by

Clinical Pathways stakeholders in a case assigned to

a patient.

Figure 1 shows an example of the logs of activities

in a case. This log is composed by a set of events that

represents begin action and end action events. The be-

gin action event has the identiﬁcation of the case, the

name of the action and the time when the action was

started. The end action event has the identiﬁcation of

the case, the name of the action, the time when the

action was ﬁnished and the result of this activity.

Analyzing a set of samples like the shown in Fig-

ure 1, using Pattern recognition techniques, PALIA

algorithm is able to infer automatically a Workﬂow

like the shown in Figure 2. This Workﬂow repre-

sents a model that resumes graphically all the possible

ﬂows that the log samples can follow. This formally

deﬁned Workﬂow can be used by Clinical Pathway

designers as the basis for the description of a formally

deﬁned care protocols that explains all the situations

that can occur in reality in a concrete scenario.

Nevertheless, these techniques can be used not

only for helping design of Clinical Pathways but also

to support the continuous evaluation of them. This

allows comparing formal protocols to detect differ-

ences between the designed and the executed one and

facilitate the detection of problems in individual or

HEALTHINF 2011 - International Conference on Health Informatics

230

Figure 2: Original Clinical Pathway deployed.

population cases to improve the efﬁciency. Further-

more, it provides a guideline for managers to under-

stand the reasons for incompliance to protocols and

evaluate if the compliance support strategies put in

place are followed. This is also an important achieve-

ment as compliance of health professionals to care

processes whenever is possible is highly desirable.

Consequently, to allow the use of the described tech-

niques in Clinical Pathways Evaluation, it’s necessary

to create a software tool that incorporate those tech-

niques to allow the inference and evaluation of ongo-

ing Clinical Pathways.

4 USING PROCESS MINING

TECHNIQUES IN CLINICAL

PATHWAYS EVALUATION

In this paper, a tool for Process Mining use in Clini-

cal pathways Evaluation is presented. This prototype

was developed in the framework of the Heart Cycle

European project (Heart Cycle Consortium, 2008).

This tool has some mining algorithms to allow

Clinical Pathway designers to detect errors on execu-

tion time and helps the reﬁnement and redesign of the

process in order to improve the efﬁciency and efﬁcacy

of the systems.

Basically, the tool has two kinds of algorithms im-

plemented.

• An Activity-Based Process Mining algorithm was

implemented to discover the whole process that

is occurring in its real execution. The algo-

rithm implemented is based on Parallel Activity-

based Log Inference Algorithm(PALIA) deﬁni-

tion (Fern

andez and Bened

ı, 2008).

• Two algorithms able to compare the original pro-

tocols with the inferred one were implemented.

The ﬁrst algorithm is able to compare two work-

ﬂows to detect differences between them. This

algorithm is based in classical Error Correct-

ing techniques (Rulot and Vidal, 1987) based on

grammar inference theory. The Second algorithm

is able to compare a workﬂow with a individual

case. This algorithm will allow the systems to

show the path followed by the case in the Clini-

cal Pathway ﬂow.

To allow the interaction of the tool with other

Workﬂow management systems the tool admits dif-

CONTINUOUS CLINICAL PATHWAYS EVALUATION BY USING AUTOMATIC LEARNING ALGORITHMS

231

Figure 3: Clinical Pathway Inferred compared with the Designed one.

ferent input and output formats. The tool imple-

mented accepts log process samples codiﬁed in plain

Activity-based Log Format (Fern

andez and Bened

ı,

2008) or in extended MXML format (van Dongen

et al., 2005) for Activity-Based Logs. The graphi-

cal ﬂows inferred can be exported to dot format (Ell-

son et al., 2004), TPA format (Fern

andez and Bened

ı,

2008) and, as a Petri Net, to PNML format (Billing-

ton et al., 2003).

To explain the functionality of the tool, a Clin-

ical Pathway and some evaluation samples will be

used. This example will be widely explained to show

how the application uses Process Mining techniques

to evaluate the execution of Clinical Pathways.

For this experiment a simpliﬁcation of a general

triage process in a Hospital was used. The formal

care protocol used in the experiment is shown in Fig-

ure 2. This protocol was speciﬁcally selected because

resumes the most complex Workﬂow Patterns that are

used in Workﬂow Patterns according the experience

acquired in Heart Cycle Project (Heart Cycle Con-

sortium, 2008). This Protocol combines the use of

action sequences, parallel activities and joins as well

as complex multichoice Workﬂow Patterns (van der

Aalst et al., 2003a).

The Clinical Pathway of the Figure 2 shows a

triage process in a hospital A modiﬁcation of that pro-

tocol has been used to simulate 100 samples using a

Workﬂow simulation engine (Fern

andez et al., 2008).

Each one of these samples represent a real execution

of the designed protocol. In order to be more realis-

tic in the simulation, the Clinical Pathway simulated

is not exactly equal to base. This allows to represent

how the system detects cases in which the patient does

not follow exactly the same path in the Clinical Path-

way. In this way, in our experiment, on one hand the

protocol will represent the Clinical Pathway designed

to be evaluated, and in the other hand the 100 cases

will represent 100 patients assigned to the protocol.

The ﬁrst step to follow for Evaluating a Clinical

Pathway using the Heart Cycle Pathway Evaluation

tool is to discover the Clinical Pathway that repre-

sents the real protocol described by the 100 patients

assigned to the protocol. In order to do that, the 100

samples are used to infer the real Workﬂow to be com-

pared with the designed one. The Workﬂow algorithm

implemented in the tool allows do that.

Figure 3 shows the results offered by the tool. At

the right of the screen it’s possible to see the compar-

ison between the original workﬂow and the inferred

one. In that image, the transitions and states are la-

belled with numbers. This represents how many times

this state or transition has been accessed. This allows

Clinical Pathways designers to discover bottle necks

and unused actions to make more efﬁcient and sim-

plify the Clinical Pathway’s Design. In this case, it’s

easily detected that only the 23% of samples complete

the care protocol until the end. Other important data

that can be deduced of this example is that the major-

ity of patients are derived to HHU (63 of 87).

HEALTHINF 2011 - International Conference on Health Informatics

232

Figure 4: Clinical Pathway Instance path view.

In this example, it is shown how Clinical Pathways

exceptions affect the Protocol execution. In the exam-

ple, 32 executions of the 82 cases that are assigned for

HHU Care are returned to External Triage. This step,

not covered by the original Workﬂow, is shown in the

example as a green arrow. With this information, the

Clinical Pathways designer can deduce that there is

an error on External Triage or HHU Admission pro-

cess that makes inefﬁcient the HHU Unit producing

an high percentage of exceptions in the HHU Care

process. Then will be the role of the manager of the

process to ﬁnd the best solution for this divergence,

but the proposed tools has highlighted where it hap-

pens and can reproduce and analyze the conditions in

which it happened: typology of patients, time or date

frames, speciﬁc teams or professionals, etc.

In the left part of the screen some statistical in-

formation can offer more data about the efﬁciency

level of the Clinical Pathway execution. The upper

part shows information about the 100 execution cases.

Each one of the actions of the Clinical Pathway has

assigned a relative cost. Using that information the

system can calculate the total cost of each case and

shown they ordered. In addition, the average of cost

by action made in a case is also show. This allow to

detect the kind actions that has a bigger relative per-

centage cost. To make easier the detection of groups

of deﬁcitary cases, the application shows in red the

cases that has an average cost below the mean, and

marks in green those that has an average cost over the

mean.

The lower part shows statistical information about

the actions executed in all the cases. In that case, it

is shown the relative cost of an action executed in all

the cases, and the average cost by case. In the same

way, the application shows in red the actions that has

an average cost below the mean, and marks in green

those that has an average cost over the mean. This

information allows detecting which actions are more

used and its usual cost in their global execution.

In order to facilitate the detection of inefﬁcient

cases, the application allows to Clinical Pathways de-

signers the single view of a patient case. In this case,

selecting a case from the left instances list with the

mouse, the correspondent ﬂow is shown on the right

Figure 4 shows an example of how a Clinical Path-

way execution is displayed in the application. The

Workﬂow shows in green the states and arrows that

are followed by the selected case. This allows health

professionals to detect punctual errors in individual

cases and how those errors affect patients. In this

example, the case does not follow exactly the Clini-

cal Pathway originally designed and provokes an ex-

ception in HHU care process returning to External

triage process and, after that, returning other time to

HHU Admission process. In the example, a simu-

lated administrative error provokes the exception that

throws out the patient from HHU that ﬁnally was de-

tected and corrected by External Triage process. In

that case, the health professionals can revise the HHU

Care process rejection protocol to detect possible er-

rors.

CONTINUOUS CLINICAL PATHWAYS EVALUATION BY USING AUTOMATIC LEARNING ALGORITHMS

233

5 CONCLUSIONS

In this paper, a tool for the continuous evaluation

of Clinical Pathway deployment is presented. This

tool is able to compare real execution cases with the

theoretical designed Clinical Pathways. This allow

to easily discover the differences between the care

protocols predeﬁned and how they are actually ap-

plied. This allows Health professionals detect inef-

ﬁcient processes, exceptions, inconsistencies and bot-

tlenecks in the designed Clinical Pathways or to iden-

tify possible problems in the organization that prevent

the efﬁcient work of the care teams. In addition to

this, the application offers the possibility to associate

cost measurements to the actions and have a view of

the process relative costs in order to quantify the inef-

ﬁciency of the system.

The tool has been tested in laboratory conditions

using a simulated Clinical Pathway and a set of cases

to achieve a ﬁrst evaluation of the system capabilities.

This tool is currently in evaluation and it is

planned to be validated in real systems in Hearth

Cycle (Heart Cycle Consortium, 2008) European

project.

REFERENCES

Audimoolan, S., Nair, M., Gaikward, R., and Qing, C.

(2005). The role of clinical pathways. Improving Pa-

tient Outcomes, February:–.

Billington, J., Christensen, S., Van Hee, K., Kindler, E.,

Kummer, O., Petrucci, L., Post, R., Stehno, C., and

Weber, M. (2003). The petri net markup language:

concepts, technology, and tools. In ICATPN’03: Pro-

ceedings of the 24th international conference on Ap-

plications and theory of Petri nets, pages 483–505,

Berlin, Heidelberg. Springer-Verlag.

Dominguez, D., Fern

andez, C., Meneu, T., Mocholi, J. B.,

and Serafﬁn, R. (2008). Medical guidelines for the

patient: Introducing the life assistance protocols. In

Computer-based Medical Guidelines and Protocols:

A Primer and Current Trends, volume 139, pages

282–293. IOS Press.

Ellson, J., Gansner, E., Koutsoﬁos, E., North, S., and Wood-

hull, G. (2004). Graphviz and dynagraph – static

and dynamic graph drawing tools. In Junger, M. and

Mutzel, P., editors, Graph Drawing Software, Math-

ematics and Visualization, pages 127–148. Springer-

Verlag, Berlin/Heidelberg.

Elstein, A. (2004). On the origins and development of

evidence-based medicine and medical decision mak-

ing. Springer, ago 2004:184–189.

Fern

andez, C. and Bened

ı, J. M. (2008). Timed parallel

automaton learning in workﬂow mining problems. In

Ciencia y Tecnolog

ıa en la Frontera, number special

2008, pages 181–187.

Fern

andez, C., Sanchez, C., Traver, V., and Bened

ı, J. M.

(2008). Tpaengine: Un motor de workﬂows basado

en tpas. In Ciencia y Tecnolog

ıa en la Frontera

ISSN:1665-9775.

Fern

andez-Llatas, C., Meneu, T., Bened

ı, J. M., and Traver,

V. (2010). Activity-based process mining for clinical

pathways computer aided design. In Proceeding of

the 32nd Annual International Conference of the IEEE

Engineering in Medicine and Biology Society.

Heart Cycle Consortium (2008). Vii framework program

ist project 216695:compliance and effectiveness in hf

and chd closed-loop management 2008-2011.

Naranjo, J. C., Fernandez, C., Pomes, S., and Valdivieso, B.

(2006). Care-paths: Searching the way to implement

pathways. Computers in Cardiology, 33:285–288.

Peleg, M., Boxwala, A. A., Bernstam, E., Tu, S. W.,

Greenes, R. A., and Shortliffe, E. H. (2001). Sharable

representation of clinical guidelines in glif: Relation-

ship to the arden syntax. Journal of Biomedical Infor-

matics, 34(3):170–181.

PubMed Library (2010). National library of medicine

and the national institutes of health pubmed library:

http://www.pubmed.gov.

Rulot, H. and Vidal, E. (1987). Modelling (sub)string-

length based constraints through a grammatical infer-

ence method. In Proc. of the NATO Advanced Study

Institute on Pattern recognition theory and applica-

tions, pages 451–459, London, UK. Springer-Verlag.

The Cochrane Collaboration (2010). Cochrane library:

http://www.cochrane.org/index.htm.

Van der Aalst, W. M. P., Barros, A. P., ter Hofstede, A.

H. M., and Kiepuszewski, B. (2003a). Workﬂow pat-

terns. Distributed and Parallel Databases, page 70.

Van der Aalst, W. M. P., van Dongen, B. F., Herbst, J.,

Maruster, L., Schimm, G., and Weijters, A. J. M. M.

(2003b). Workﬂow mining: A survey of issues and

aproaches. Data and Knowledge Engineering 47

2003, pages 237–267.

Van Dongen, B. F., de Medeiros, A. K. A., Verbeek, H.

M. W., Weijters, A. J. M. M., and van der Aalst, W.

M. P. (2005). The prom framework: A new era in pro-

cess mining tool support. In ICATPN, pages 444–454.

WfMC (1999). Workﬂow Management Coalition Terminol-

ogy Glossary. WFMC-TC-1011, Document Status Is-

sue 3.0.

HEALTHINF 2011 - International Conference on Health Informatics

234