Argumentation logic for the flexible enactment of goal-based medical guidelines
Maria Adela Grandoa,⇑, David Glasspoolb, Aziz Boxwalaa
aDivision of Biomedical Informatics, School of Medicine, University of California San Diego, 9500 Gilman Drive #0728, La Jolla, CA 92093-0728, USA
bSchool of Informatics, University of Edinburgh, 10 Crichton Street, Edinburgh EH8 9AB, UK
a r t i c l e i n f o
Received 21 October 2011
Accepted 20 March 2012
Available online 29 March 2012
Computer interpretable guidelines
Decision support aid
a b s t r a c t
Research purpose: We have designed a prototype clinical workflow system that allows the specification
and enactment of medical guidelines in terms of clinical goals to be achieved, maintained or avoided
depending on the patient’s disease and treatment evolution. The prototype includes: (1) an argumenta-
tion-based decision support system which can be used both to represent medical decisions within guide-
lines, and to dynamically choose the most suitable plans to achieve clinical goals, and (2) mechanisms to
specify a health organization’s facilities and health workers skills and roles, which can be taken into
account during the decision process in order to improve quality of care.
Results: The framework has been fully implemented in the COGENT formal modeling system. The proto-
type has been evaluated implementing a hypertension guideline.
Conclusions: The framework has shown flexibility and adaptability in (1) advising and tailoring health
care based on a health organization’s resources and a patient’s particular medical condition, (2) delegat-
ing health care, and (3) replanning when unexpected situations arise.
Published by Elsevier Inc.
A gap between medical theory and practice is one of the most
consistent findings in health services research. Care procedures of-
ten differ significantly between locations, yet it is known that pa-
tient outcomes are improved if standard ‘‘best practice’’ is
followed consistently. Computer-based decision-support systems
have been shown to be effective methods for improving adherence
A number of process-based languages have been developed to
allow clinical guidelines to be expressed in well-defined, com-
puter-interpretable formal languages to provide patient-specific
decision-support. According to Peleg et al.  review (including As-
bru , EON , GLIF , GUIDE , PRODIGY  and PROforma 
languages), the standard paradigm for modeling clinical processes
is based on ‘‘networks of tasks’’ in which tasks are modeled a steps
in a care process. Tasks can have a specific goal (for instance in
PROforma) or intention (as in Asbru). However, the need for clinical
guidelines to be adapted for use at different sites, to be tailored to
rapidly changing and unpredictable environments, and to be car-
ried out by distributed clinical teams, implies a degree of flexibility
beyond that of current task-based guideline languages. Addition-
ally, personalization of clinical processes could potentially improve
the quality of care and reduce medical errors, but this is difficult
using current systems.
To address these problems we proposed in , instead of the
traditional task-based approach, a goal-based framework. We
interpreted goals as in both Fox et al.  seminal work on clinical
goals and Asbru language. Goals can be interpreted as temporal
patterns to be achieved, maintained or avoid. But while in both As-
bru and Fox et al. the patterns can refer to both provider actions
(monitor blood glucose once per week) and patient states (patient
has its glucose monitored once per week), for us clinical processes
are always state-based. Also inspired by Fox et al. work, in our
framework the goals of the guidelines are separated from detailed
plans to achieve those goals. When the guideline is applied to a
patient, first the recommended goals for the intervention are
determined based on the clinical context, and then detailed activ-
ities are suggested to achieve those goals. This potentially allows
the same guideline to be applied at different sites, by different
types of clinical teams, and allows the guideline to adapt to the
needs of different patients and to unexpected circumstances.
In this paper we implement and evaluate a practical decision
support system (based on formal principles presented in ) that
suggests the best plans for achieving goals in this system. Our ini-
tial evaluation of the framework, based on the enactment of rec-
ommendations from three medical guidelines (hypertension ,
NICE breast cancer guideline for breast cancer  and part of a
chronic cough guideline  corresponding to recommendations
for inmunocompetent adult patients), seems to show that the
goal-based framework provides flexible mechanisms for the dis-
tributed enactment of clinical process and to address the problems
1532-0464/$ - see front matter Published by Elsevier Inc.
E-mail address: email@example.com (M.A. Grando).
Journal of Biomedical Informatics 45 (2012) 938–949
Contents lists available at SciVerse ScienceDirect
Journal of Biomedical Informatics
journal homepage: www.elsevier.com/locate/yjbin
The paper is organized in the following way, we start in Section
2 by explaining in detail our motivation, we continue in Section 3
with an explanation of the methods used. In Section 3.1 we explain
the goal-based framework that we introduced in  for the model-
ing of computer interpretable guidelines (CIGs). In Section 3.2 we
show how we further enhance the flexibility of the framework
by introducing: (1) mechanisms to specify roles, necessary condi-
tions, and health workers skills in order to select the best health
provider for a care service, (2) specifications of the health organiza-
tion’s facilities and the patient’s medical record in order to adapt
and tailor the treatment on the fly, and (3) an argumentation-
based decision support system which can be used to represent
medical decisions within guidelines, and to control the automatic
refinement of plans as a guideline is carried out. We continue in
Section 4.1 explaining a prototype of the framework from Section
3.2 implemented in the COGENT modeling system (http://cogen-
t.psyc.bbk.ac.uk/). In Section 4.2 we explain how we encoded, in
the COGENT-based prototype, a guideline corresponding to the
treatment of high blood pressure . The scenario presented in
Section 4.3 indicates that this framework can be used to enact care
paths that comply with national medical guidelines but are also
tailored to the patient’s treatment evolution maximizing the utility
based on the health organization’s facilities and the skills of the
health workers. Later in Section 5 we share some lessons learned
from evaluating three guidelines in the COGENT prototype pre-
After comparing in Section 6 our approach with the current
state of the art we finish in Section 7 summarizing the main fea-
tures of the prototyped framework and our plans for its further
development into a more scalable language for the specification
and enactment of CIGs.
PROforma  was developed in the 1990s by Cancer Research
UK as an executable process modeling language. PROforma plans
have been successfully built and deployed in a broad range of deci-
sion support systems, guidelines and other clinical applications.
Besides being provided with a formal semantics based on state-
transition systems, PROforma has been used as the basis of aca-
demic and commercial decision support technologies: Tallis and
PROforma, like most of the languages currently available for the
enactment of CIGs, provides a wholly centralized approach for the
enactment of medical applications.
According to :’’Historically the responsibility for the detec-
tion and diagnosis of a patient condition and subsequent treatment
and follow-up were localized, in that the knowledge, actions and
responsibilities required for these duties were centered upon a
specialist team of professionals working in a particular physical
place. However, clinical practice is increasingly complex, distrib-
uted and service oriented’’. Nowadays actions are performed at
numerous specialist sites and responsibilities for care are often dis-
tributed and/or shared between care providers, obliging patients to
move between them to access services.
The challenge is to formally specify specialist services such as
decision support and workflow management, confederating such
systems without compromising the autonomy of each local service,
or the compliance to medical guidelines. In  strong arguments
are given to explain why the traditional wholly centralized ap-
proach for medical applications on which PROforma was based will
not scale well with the growing complexity of the medical con-
texts. On the other hand more recently service-oriented ,
process-like  and multi-agent based systems [16,17] have been
proposed as viable approaches to develop fully distributed
enactment environments. In particular in  a service-oriented
approach to distribute the enactment of PROforma plans was
proposed and evaluated in a cancer treatment scenario.
The purpose of this work is to contribute to the challenge by
proposing a novel approach (more comprehensive and flexible
than ) for distributing the enactment of PROforma-like medical
guidelines between health providers with the aim of providing bet-
ter patient care. Our approach is based on: goal-based modeling of
CIGs as proposed in , shared repositories of specifications of ser-
vices offered by health providers, detailed descriptions of actors
and roles skills, and the use of a decision support system to advise
on the best service provider at run time. By modeling medical
applications at design time in terms of clinical outcomes it is pos-
sible to abstract from implementation details while providing
compliance to medical guidelines. At run time an argumentation-
based decision support system suggests the best way to pull to-
gether available resources to provide the best possible care for
the patient while adapting the guideline’s recommendations to
the realities of actual institutions and hospitals.
Below we explain in detail the formal semantics of the proposed
framework, as an extension of the goal-based formalism intro-
duced in .
3.1. Framework for goal-based workflow enactment
What makes the framework presented in  different from tra-
ditional task-based approaches to model CIGs is the use of goals as
first order objects, the automatic selection of tasks to achieve goals,
and replanning when goals fail or when goals are not achieved
even after enactment of a candidate task.
A workflow is a network of nodes called keystones with unique
starting and ending points, connected by scheduling constraints.
A keystone is an abstraction of tasks and goals. At run-time a
keystone is assigned a state. As is shown in Fig. 1, initially a key-
stone’s state is Dormant and it changes state to InProgress when
it is ready for enactment. The keystone only changes from InPro-
gress to Completed when the successCondition is satisfied. While
a keystone is InProgress it can be Discarded. If a keystone is Dis-
carded it needs to be reinitialized before it can be enacted. A key-
stone can be assigned a precondition, i.e. a predicate that must be
satisfied before the keystone becomes InProgress. A keystone can
be cyclic, in which case the cycleInterval indicates its frequency
of iteration. Time constraints can be attached to a keystone: startAt,
duration and finishAt, specifying the starting time, duration and
completion time for the keystone.
When a keystone is a goal and it is InProgress the goal’s
successCondition, abortCondition and invariantCondition (in case of
goals of type maintain) are monitored. If the successCondition is
Fig. 1. State-based transition system for keystones in our proposed framework.
M.A. Grando et al./Journal of Biomedical Informatics 45 (2012) 938–949
In the case of the hypertension guideline we provided concrete
examples of this sort in Sections 4.3.1 and 4.3.2.
5.2. Delegating health care based on resource availability or quality of
For instance for some treatment options, like SLN biopsy sur-
gery, not all the NHS hospitals have the required clinical staff (sur-
geon specialized in breast surgeries) and therefore the patient care
has to be delegated to other hospital. In the case of the hyperten-
sion guideline we provided a concrete example of this type in Sec-
5.3. Replanning when unexpected situations arise
For instance the patient was being treated for breast cancer, but
after receiving chemotherapy treatment she develops metastasis
and the disease stage is no more an early breast cancer. Therefore
the new clinician’s goal is to treat the patient’ advanced cancer. In
the case of the hypertension guideline we provided a concrete
example of replanning on the fly in Section 4.3.1.
We found out that the three medical guidelines considered for
the prototype evaluation provided different scheme for classifying
the evidence that supported their medical recommendations. But
for the three different evidence scheme considered in guidelines
[10–12], mapping the medical evidence into argument’s support
was straightforward. See for instance in Section 4.2.1 how we
mapped the evidence scheme provided by the hypertension guide-
line  into arguments’ support.
6. Discussion: related approaches
If we have to compare our approach with available languages
for the specification and enactment of medical guidelines, then
we could do it in terms of the same criteria used for measuring
the flexibility and adaptability of our framework:
6.1. Advising and tailoring health care based on patient’s medical
conditions and organization’s resources
Our formalism provides mechanisms to facilitate the specifica-
tion of information related to actor, roles, health organizations and
patient’s medical records. This information could be used to
dynamically select at run-time the best available treatment based
on the patient’s medical conditions, the health organizations’
facilities and the available medical staff. But according to  in
the available languages for the enactment of clinical guidelines is
it possible to specify decision aid based on switch constructs, argu-
mentations schema, decision trees, calls to external functions and
influence diagrams. Therefore we assume that, the same as we
did here, the available languages could be extended with more de-
tailed information related to actor, roles, health organizations and
patient’s medical records. This information could be used, as we
showed in this paper, to advice and tailor the healthcare based
on patient’s medical conditions and organization’s resources. For
instance in the case of EON, PRODIGY and GLIF they are provided
with very complex patient information models to influence the
process decision. In the case of PROforma, Asbru, and Grando
et al.  it is currently possible to indicate the preferred actor or
role who is responsible for the enactment of a task or goal, though
no additional information on the actor’s and roles’ skills and re-
sources is directly supported in those languages.
6.2. Delegating health care based on resource availability or quality of
Originally the enactment of CIGs was considered a centralized
task. But there have been a few recent attempts to distribute the
enactment of CIGs between different health providers, mainly
based on multi-agent systems  , service-oriented architec-
tures  and distributed process enactment [5,15]. In the case
of the GLARE  formalism, it has been extended in  with
an application-independent methodology to support human inter-
action and human resources coordination for the distributed
enactment of clinical guidelines. Their approach is based on asso-
ciating to task the notion of context (where the action can be exe-
cuted), role (who can execute the action) and competences (actor’s
The above mentioned approaches share a similar strategy for
distribution based on design-time role-based process specifica-
tions that are assigned at run-time to concrete actors (health orga-
agents, etc.) for their enactment. The mechanisms used for the
run-time selection of the actors are mainly based on matching
the actor’s skills specification with the role’s requirements. In the
case of multiple actors offering similar services more sophisticated
mechanism of selection based on trust have been used  in the
distributed enactment of PROforma. Coordination and data sharing
between actors can be achieved through the exchange of message
via communication channels.
Fig. 9. Candidate plan for the goal achieve_correct_remediable_cause. The pink rounded shape corresponds to a decision. The precondition of the goal
achieve_adjust_drug_treatment is that the chosen candidate for the decision is the diagnosis of drug-induced hypertension resistance, the precondition of the goal
achieve_improve_volume_overload is that the chosen candidate is the diagnosis of volume overload, and the precondition of the goal achieve_resolve_associated_conditions is
that the chosen diagnosis corresponds to associated conditions (obesity, excess alcohol intake). (For interpretation of the references to colour in this figure legend, the reader
is referred to the web version of this article.)
M.A. Grando et al./Journal of Biomedical Informatics 45 (2012) 938–949
Our approach for distributed enactment is based too on run-
time matching of actor’s skills with roles. But it differs on assuming
goal-based specifications instead of task-based workflow defini-
tions. Based on this difference our framework can provide the fol-
lowing features for enhancing flexibility:
? communication as an implementation issue that does not need
to be specified at design-time but can freely emerge at run-time
(see for instance in Section 4.3.1 the example of service delega-
? optional use of a decision support system for advising on the
selection of the most suitable service provider for achieving
6.3. Replanning when unexpected situations arise
In our framework goals provide an intuitive way to replan based
on: (re)activating a goal if its precondition is satisfied, checking the
goal’s successCondition to determine if the goal was achieved after
the chosen care plan was completed, selecting a new plan if the
previously selected plan was completed and the goal was not
achieved, and aborting a goal if its abortcondition is satisfied. In
the case of task-based languages something similar can be
achieved by providing Boolean functions to control plan execution.
Like filter, setup, suspend, reactivate, complete and abort expres-
sions in Asbru; or precondition, iteration condition, termination
condition, and abort condition in PROforma.
7. Conclusions, future work
In this paper we propose a way to implement flexible goal-
based plan specialization taking advantage of an argumentation-
based decision model to separate decision-relevant knowledge
from plan specifications.
Our approach allows: (a) compliance to medical recommenda-
tions while tailoring the care plan to the health organization’s local
resources and patient’s response to treatment; (b) monitoring and
manipulating significant clinical goals that are normally implicit in
clinical guidelines rather than modeling guidelines as procedural
plans and (c) delegating/assigning treatments to other health orga-
nizations when the local provider’s resources are not the most
appropriate for the patient’s medical condition.
Besides the hypertension guideline explained in Section 4.2 we
have implemented in the COGENT prototype recommendations
from a chronic cough guideline  for immune-competent adult
patients, and also recommendations from a NICE guideline for
early breast cancer .
One restriction of our prototype is that does not allow assigning
priorities to goals. We are considering as future work the possibil-
ity of adding priorities to the goal and replacing the argumenta-
tion-based decision support system proposed here with a multi-
criteria decision support system that will help the patient and
the clinician to make choices involving several goals with different
levels of priority. Possible candidates are value-based methods that
lead the decision maker(s) through a series of judgments that pro-
duce quantitative scores or rankings over relative priorities of the
decision criteria. For instance how important is effectiveness over
cost, or cost over side effects. Methods in this category include
ordinal weighting methods, direct weighting, methods based on
multiatribute utility theory, and the Analytic Hierarchy Process.
For a review of these methodologies we refer the reader to .
We are currently working on replacing our prototype with a
tool specified in a more scalable programming language than
COGENT. For the new tool we will use Protégé to specify the ontol-
ogies and Java to implement the engine. Besides we will use the
Semantic Web Rule Language (SWRL)  and the Jess reasoner
(http://www.jessrules.com/links/) for defining and evaluating the
Boolean conditions such as: conditions that restrict the workflow
enactment (precondition, postcondition, etc.), arguments, argu-
ment schemes, and the rules that enact the goal-based engine
and the argumentation-based decision support. Protégé, Java, and
SWRL are well developed and mature languages which have suc-
cessfully been used in numerous real applications, so we are very
confident of achieving the scalability, reusability and interoperabil-
ity that we are aiming.
 Peleg M, Tu S, Bury J, Ciccarese P, Fox J, Greenes RA, et al. Comparing computer-
interpretable guideline models: a case-study approach. J Am Med Inform Assoc
 Shahar Y, Miksch S, Johnson P. The Asgaard project: a task-specific framework
for the application and critiquing of time-oriented clinical guidelines. Artif
Intell Med 1998;14(1–2):29–51.
 Tu SW, Musen MA. A flexible approach to guideline modeling. In: Proc AMIA
symp; 1999. pp. 420–4.
 Peleg M, Boxwala AA, Ogunyemi O, Zeng Q, Tu S, Lacson R, et al. GLIF3: the
evolution of a guideline representation format. In: Proc AMIA symp; 2000. p.
 Ciccarese P, Caffi E, Boiocchi L, Quaglini S, Stefanelli M. A guideline
management system. Stud Health Technol Inform 2004;107(1):28–32.
 Johnson PD, Tu S, Booth N, Sugden B, Purves IN. Using scenarios in chronic
disease management guidelines for primary care. In: Proc AMIA symp; 2000. p.
 Fox J, Sutton DR. The syntax and semantics of the PROforma guideline
modeling language. J Am Med Inform Assoc 2003;10(5):433–43.
 Grando A, Peleg M, Glasspool D. A goal-oriented framework for specifying
clinical guidelinesand handling
 Fox J, Alabassi A, Patkar V, Rose T, Black E. An ontological approach to
modelling tasks and goals. Comput Biol Med 2006;36(7-8):837–56.
 US Department of Health and Human Services. The seventh report of the joint
national committee on prevention, detection, evaluation, and treatment of the
high blood pressure; August 2004.
 Early and Locally Advanced Breast Cancer – NCBI Bookshelf. <http://
www.ncbi.nlm.nih.gov/books/NBK11643/> [accessed 29.10.11].
 Irwin RS, Boulet LP, Cloutier MM, Fuller R, Gold PM, Hoffstein V, et al.
Managing cough as a defense mechanism and as a symptom. A consensus
panelreport ofthe American
 Fox J, Glasspool D, Patkar V, Austin M, Black E, South M, et al. Delivering
clinical decision support services: there is nothing so practical as a good
theory. J Biomed Inform 2010;43:831–43.
 Besana P, Patkar V, Barker A, Robertson D, Glasspool D. Sharing Choreographies
in OpenKnowledge: a novel approach to interoperability. J Software
 Mans RS, Russell NC, van der Aalst WMP, Bakker PJM, Moleman AJ, Jaspers
MWM. Proclets in healthcare. J Biomed Inform 2010;43(4):632–49.
 Han B, Song S, Lee K, Jang K, Shin D. Multi-agent system based efficient
healthcare service, Adv Commun Technol (ICACT); 2006.
 Tolchisnky P, Atkinson K, McBurney P, Modgil S, Cortes U. Agents deliberating
over action proposals using the ProCLAIM Model. In: Presented at the 5th
international Central and Eastern European conference on multi-agent
systems, vol. 4696; 2007. p. 32–41.
 Fox J, Glasspool D, Grecu D, Modgil S, South M, Patkar V. Argumentation-based
inference and decision making–a medical perspective. IEEE Intell Syst
 Grando MA, Peleg M, Cuggia M, Glasspool D. Patterns for collaborative work in
health care teams. Artif Intell Med 2011;53(3):139–60.
 Terenziani P, Montani S, Bottrighi A, Torchio M, Molino G, Correndo G. The
GLARE approach to clinical guidelines: main features. Stud Health Technol
 Bottrighi A, Torchio M, Montani S, Molino G, Terenziani P. Supporting human
interaction and humanresources
guidelines. Stud Health Technol Inform 2010;160(1):319–23.
 Dolan JG, evidence-based Multi-criteria clinical decision support: a primer on
the use of multi criteria decision making methods to promote. patient-
centered healthcare. Patient 2010;3(4):229–48.
 Horrocks I, Patel-Schneider P, Boley H, Tabet S, Grosof B, Dean M. SWRL: a
semantic web rule language combining OWL and RuleML; May 2004.
Collegeof ChestPhysicians. Chest
coordinationin distributed clinical
M.A. Grando et al./Journal of Biomedical Informatics 45 (2012) 938–949