Towards a Second Generation of Computer Interpretable Guidelines

Paolo Terenziani, Alessio Bottrighi, Laura Giordano, Giuliana Franceschinis, Stefania Montani,

Luigi Portinale and Daniele Theseider Dupre

DISIT - Computer Science Institute, University of Eastern Piedmont, Alessandria, Italy

Keywords: Computer Interpretable Guidelines, Integration of Different Forms of Medical Knowledge, Logical,

Analogical, Defeasible and Probabilistic Reasoning, Temporal Data.

Abstract: Computer Interpretable Guidelines (CIG) are an emerging area of research, to support medical decision

making through evidence-based recommendations. However, new challenges in the data management field

have to be faced, to integrate CIG management with a proper treatment of patient data, and of other forms of

medical knowledge (e.g., causal and behavioral knowledge). In this position paper, we summarize a

proposal for a research agenda that, in our opinion, can lead to a significant advancement in the field. The

goal of the work is to provide suitable models and reasoning methodologies to cope with the

aforementioned aspects, and to properly integrate them for medical decision support. Achieving such a goal

requires advances in data management, and, in particular, in the treatment of indeterminate valid-time data

in relational databases, of temporal abstraction on time series, of case retrieval on time series, of design-time

and run-time model-based verification of guidelines, of case-based reasoning, of non-monotonic logics, of

formal ontologies, of probabilistic graphical models (Bayesian Networks and Influence Diagrams).

1 INTRODUCTION

This position paper proposes and summarizes a

research agenda we are going to present as a four-

year project proposal for an ERC Advanced Grant

(http://erc.europa.eu/advanced-grants).

Context. A lot of attention has been recently

devoted to the analysis of clinical processes, and

thus to clinical guidelines, which constitute the tools

identified by physicians in order to encode the “best

practice” clinical procedures. Clinical guidelines are

- as defined by US Institute of Medicine -

“systematically developed statements to assist

practitioner and patient decisions about appropriate

health care in specific clinical circumstances”. In the

last years, thousands of clinical guidelines have been

developed by local, national, and international

organizations. Despite such a large effort, it is

widely recognized that clinical guidelines have not

provided all the expected advantages in the clinical

practice. The recent research in medical informatics

has widely demonstrated that Computer Science can

help to drastically improve the impact of clinical

guideline (see, e.g., (Teije et al., 2008)). For such

reasons, research in this sector is becoming more

and more important in the area of Medical

Informatics, and of Computer Science in general. In

particular, in the last twenty years, several

approaches have been developed to deal with

Computer Interpretable Guidelines (henceforth:

CIG; see, e.g., the comparisons in (Peleg et al.,

2004), and the book (Teije et al. 2008)). As a matter

of fact, a wide scientific literature in the last twenty

years have demonstrated that, besides having an

important practical impact, the computer-based

treatment of clinical guidelines arise several

challenging open-problems for the scientific

community, and that much more theoretical and

practical efforts must be spent to achieve all the

potential impact on the clinical practice.

Objectives. The “visionary” idea underlying our

research proposal is that, to achieve a significant

step forward in the state-of-the-art, it is important to

provide a homogeneous approach integrating (at

least) three different aspects:

(1) Computer Interpretable Guidelines

(2) Treatment of patient data

(3) Treatment of “basic” medical knowledge (BMK)

199

Terenziani P., Bottrighi A., Giordano L., Franceschinis G., Montani S., Portinale L. and Theseider Dupre D..

Towards a Second Generation of Computer Interpretable Guidelines.

DOI: 10.5220/0004585501990205

In Proceedings of the 2nd International Conference on Data Technologies and Applications (DATA-2013), pages 199-205

ISBN: 978-989-8565-67-9

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

Such aspects have been often considered in isolation

by the Medical Informatics literature and a step

forward in data management should be achieved in

order to integrate them.

Regarding CIGs, given the dimension of clinical

guidelines, their correctness and their adherence to

specifications cannot be verified by-hand by

physicians. Both design-time and run-time

verification must be considered to cope with the

different phases of guideline life-cycle. Thus, new

formal methodologies must be identified. Different

models, with different features like, logical and

Petri-net-based models, are definitely good

candidates for model-based verification.

Of course, CIGs must be executed considering

patients’ data. Supporting a proper treatment of such

data is still a challenging open issue since, for

instance, difficult temporal issues have to be

managed (such as, e.g., temporally indeterminate

patient data). Additionally, in many cases, raw data

cannot be directly used in guidelines. Time series

must be managed and retrieved, usually at different

levels of abstraction. Clinical guidelines capture

medical evidence, thus coping with “typical”

patients, and not with patients’ peculiarities.

However, patient data may be such that the patient is

somehow “not typical”. In such cases, physician

may need to resort to other kinds of BMK, and other

forms of reasoning on it, including medical

ontologies, knowledge from experience and

analogical reasoning (e.g. case-based reasoning),

causal, behavioral and uncertain (probabilistic)

knowledge, and knowledge about patient evolution.

However, although coping separately with each

one of the above aspects is a challenging goal of our

project, our main goals lie in the definition of

appropriate models and methodologies to cope with

their interactions. In particular, the main challenge

concerns the interactions between CIG (which are

executed considering patients data) and BMK.

Indeed, for specific patients, the knowledge in the

guideline and the BMK may suggest different (and

possibly contrasting) actions, and only domain

experts seem able to choose between them, on the

basis of the patient’s status.

Example 1. Clinical Guideline: Patient with acute

myocardial infarction presenting with acute

pulmonary edema; before performing coronary

angiography it is mandatory to treat the acute heart

failure.

BMK: The execution of any clinical guideline may

be suspended, if a problem threatening the patient’s

life suddenly arises.

Example 2. Clinical Guideline: In a patient affected

by unstable angina and advanced predialytic renal

failure, coronary angiography must be executed.

BMK: the contrast media administration may cause

a further final deterioration of the renal functions,

leading the patient to dialysis (see (Bottrighi et al.,

2009) for more examples).

In Example 1 the execution of a CG is suspended,

due to the presence of a problem threatening the

patient’s life. In Example 2 instead the treatment is

performed even if it may involve contrast media

administration, which may dangerous for the patient.

This example shows that not only some guideline’s

prescriptions are “defeasible”, since they may be

overridden by BMK, but the same also holds for part

of BMK. Thus, defeasible (non-monotonic or

probabilistic) reasoning may be applied for the

integration. Indeed, example 2 involves also a form

of cost/benefit analysis, which, in turn, needs to be

supported by physicians’ knowledge about typical

patient evolution.

Providing physicians with decision-support tools

considering these different forms of knowledge and

reasoning about their interactions is the ultimate goal

of our research vision.

Expected Results. The project aims to enhance the

state-of-the-art in several areas of research,

including: the treatment of indeterminate valid-time

data in relational databases, temporal abstraction on

time series, case retrieval on time series, design-time

and run-time model-based verification of guidelines,

case-based reasoning, non-monotonic logics, formal

ontologies, probabilistic graphical models (Bayesian

Networks and Influence Diagrams). Prototypes will

also be developed and tested by physicians, in order

to demonstrate the practical feasibility and the

usefulness of the proposed methodologies.

Methodology. The starting point of our proposal is

GLARE, a prototypical tool for clinical GuideLine

Acquisition, Representation and Execution

(Terenziani et al., 2001). GLARE already interacts

with patient data for CIG execution, and provides

physicians with different advanced facilities, based

on formal Temporal Database and Artificial

Intelligence techniques (Terenziani et al., 2001;

(Anselma et al., 2006); (Terenziani et al., 2007);

(Terenziani et al., 2008); (Leonardi et al., 2012).

The idea is to deeply extend GLARE, by considering

an agenda structured as shown in the following

figure 1. In the rest of the paper, each one of the

tasks in figure 1 will be briefly addressed.

DATA2013-2ndInternationalConferenceonDataManagementTechnologiesandApplications

200

Figure 1: Structure of the research proposal.

2 CIG VERIFICATION

One of the main obstacles for a wide applicability of

clinical guidelines is their complexity and, as a

consequence, the difficulty in defining a

semantically accurate representation, and in

verifying their properties (such as consistency). This

motivates the use of standard verification tools for

the verification of clinical guidelines. In the project,

at least two different types of verification tools will

be explored, and compared: logical-based tools, such

as model checkers and theorem provers, and Petri-

Net based approaches.

The logical-based verification requires, on the

one hand, the translation of the guideline in the

specification language of the model checker and, on

the other hand, the encoding of the guidelines

properties as temporal formulas. This approach has

been pursued, for instance, in (Giordano et al.,

2006); (Bottrighi et al., 2010), by providing a

translation of GLARE clinical guidelines into the

language Promela of the LTL model checker SPIN.

In this direction, the proposal aims at exploiting the

logical language Temporal ASP (Giordano et al.,

20120, for the specification and the verification of

clinical guidelines. Temporal ASP allows for the

definition of temporal constraints in a domain

specification, as Dynamic Linear Time Temporal

Logic (DLTL) formulas. Such constraint could be

used, together with ASP rules, for a declarative

specification of CIGs. In particular, to do so, a

mapping of GLARE constructs into Temporal ASP

has to be defined. The verification of runtime and

design time properties of the CIGs formulated in

Temporal ASP can then be performed by DLTL

Bounded Model Checking. The runtime verification

that the treatment of a given patient is compliant

with the guideline can be modeled as a satisfiability

problem. Instead, the verification of temporal

properties of the guideline can be modeled as

validity checks.

Petri Net (PN) models and their extensions have

been proposed as a flexible and general formalism

that can be used as a common semantics for different

application-oriented modeling languages. In the

literature the feasibility of the (semi)automatic

translation of CIG models into PNs has been already

illustrated (Beccuti et al., 2009); (Quaglini et al.,

2001); (Peleg et al., 2005): the resulting model may

be used for verification of both qualitative and

quantitative properties. In this context, our research

work will focus on one hand towards extending the

compositional translation rules to represent the

relevant steps in each CIG that produce a request for

resources at given points in time; in other words the

goal is to extract a workload stochastic model from

CIGs represented with GLARE. Multiple

instantiations of sets of similar CIG models can be

composed in a compact way through High Level

Stochastic PNs (in particular Well-Formed Nets,

leading to efficient analysis methods): such a model

can then be merged with another one representing

the available resources and the scheduling

constraints to perform what-if analysis on alternative

ways of organizing healthcare services. In addition

the most suitable (logic or automata based) language

for expressing the non functional properties will be

selected, and tested on case studies. We plan to

define both system oriented performance properties

(useful for the evaluation of costs and resources

utilization) and user oriented performance properties

(useful to evaluate customer satisfaction indicators).

The experiments will be performed mainly through

the GreatSPN tool (Baarir et al., 2009).

3 MODELING PATIENT

CLINICAL DATA

This part of the project focuses on the treatment of

patient clinical data, with specific emphasis on

temporal issues.

3.1 Valid-time Data

In the medical domain, as well as in many other real-

world domains, often the exact time of occurrence of

facts is not known and temporal indeterminacy (i.e.,

“don’t know when”) occurs. Temporal

indeterminacy is widespread within medical data

(consider, e.g., patient complaint in Ex.1).

1. Clinical guideline verification

2. Modeling patient clinical data

2.1 Valid-time data

2.2 Temporal abstraction and time series data

2.3 Querying data and data retrieval

3. Basic medical knowledge (BMK)

3.1 Medical Ontologies

3.2 Experence-based knowledge and past cases

3.3 Causal and behavioral knowledge

3.4 Patient evolution

4. Integration

5. Prototypes

TowardsaSecondGenerationofComputerInterpretableGuidelines

201

Ex.1 On January 1st 2012 Mary had headache

starting between 8am and 9am and ending between

1pm and 2pm.

However, in the area of (relational) databases, the

treatment of temporal indeterminacy has been quite

neglected (see the survey in (Dyreson, 2009) and

(Anselma et al., 2012)). We aim at overcoming such

a limitation by proposing (i) a new data model, to

model temporally indeterminate relational data, and

(ii) a new relational algebra, to query it. To

guarantee implementability on top of current

technologies and interoperability, we will prove that

our data model is a consistent extension of TSQL2’s

one (which, in turn, is a consistent extension of

SQL) (Snodgrass et al., 1995), and our algebra is

reducible to the standard relational one. Also, we

plan to study the interactions between temporal

indeterminacy and telicity, which is important to

characterize patient data (Terenziani et al., 2007).

3.2 Temporal Abstractions and Time

Series Data

In medical applications, most patient data are

naturally collected in the form of time series. Within

the project, we aim at developing a user-friendly,

interactive and flexible support for querying and

retrieving time series data based on Temporal

Abstraction (TA) (Shahar, 1997); (Bellazzi et al.,

1998) for dimensionality reduction. Specifically, we

aim at supporting multi-level abstractions of the

original data (Montani et al., 2013). We foresee to

abstract (and query) time series data at finer or

coarser detail levels, according to two dimensions:

(1) a taxonomy of symbols, and (2) a taxonomy of

time granularities. Dimension (1) refers to the

qualitative detail level we are interested in. As an

example, referring to trend TA, we might want to

identify intervals of increasing trend, or be more

specific, by distinguishing between intervals of

strongly increasing trend, and weakly increasing

trend. As regards dimension (2), we could be very

specific, and abstract intervals expressed at a

“small” time granularity (e.g. minutes – depending

on the application), or be coarser, and just look for

the abstract behavior on a “larger” time scale (e.g.

hours). We plan to convert our raw data into

abstracted sequences of symbols, calculated at the

finest level of detail according to both dimensions.

However, to support flexible retrieval, users will be

allowed to query the database according to all levels,

in both dimensions.

3.3 Querying Data and Data Retrieval

The data structures (i.e. the taxonomies) and the

abstraction and distance functions studied in 3.2 will

be exploited to implement a flexible and efficient

retrieval strategy. Namely, we plan to resort to

proper index structures, able to support early

pruning, defined on the basis of the symbol and time

granularity taxonomies. The goal is to allow queries

to be issued at any level of detail, according to all

defined dimensions. A second goal will be to answer

such queries efficiently both in terms of time and

space. User-flexibility will also be an important

issue, since we aim at: (1) allowing users to express

queries in the form of regular expressions; (2)

allowing users to query for sub-series.

4 BASIC MEDICAL

KNOWLEDGE (BMK)

Different forms of BMK have to be investigated.

4.1 Medical Ontologies

Ontological knowledge is a relevant part of BMK. In

particular, formal ontologies are important to

provide a standard terminology with a clear formal

semantics, to be used in the representation of all the

forms of knowledge (including GIG content). A

specific challenge here is that of defining tractable

non-monotonic extensions of such DLs, to capture

prototypical properties of concepts and inheritance

with exceptions. To capture the degree of

“fuzziness” of medical knowledge, we will make

innovative proposals related to fuzzy and

probabilistic ontologies (Klinov and Parsia, 2008)

and ontologies based on different types of logics,

including, e.g., preferential logics (Giordano et al.,

2009).

4.2 Experience-based Knowledge

and Past Cases

Maintaining patient data as cases allows the decision

maker to access at least in implicit form the

experience-based knowledge contained in such

cases. Case-Based Reasoning (CBR) (Aamodt and

Plaza, 1994) transforms such implicit chunks of

information into more effective and explicit

knowledge for decision making (Schmidt et al.,

2001). A specific goal of our proposal will then be to

integrate our treatment of time series (based on

DATA2013-2ndInternationalConferenceonDataManagementTechnologiesandApplications

202

temporal abstraction and multiple abstraction

levels), into the CBR loop, in such a way that both

the relevant static knowledge about patients, as well

as dynamic temporal information can be integrated

with the objective of a final optimal decision.

Furthermore, CBR has proved to be well suited for

managing exceptional situations which can be

neither foreseen nor preplanned. In the literature

cases have often been resorted to describe

exceptions, in various domains, including medical

domains (Bellazzi et al., 1999). In the management

and the interaction of patient data and guidelines, we

aim at exploiting a CBR approach to support

guideline exception management and adaptation at

the local environment characteristics.

4.3 Causal and Behavioural Knowledge

Modeling BMK definitely implies to have the

capabilities of representing, at some level of details,

basic medical, clinical processes and cause-effect

associations. In the project, we will explore different

kinds of formalisms to properly model BMK.

Modeling BMK is a challenging goal, since, e.g., it

requires the ability to reason about incomplete and

defeasible knowledge. Furthermore, the properties of

a given patient can be incompletely specified and,

nevertheless, we may want to derive conclusions

concerning that patient. One of the current trends of

modeling defeasible knowledge, that will be

explored in the project, is represented by Answer Set

Programming (ASP) (Gelfond, 2007); (Bonatti,

2010), a logic programming language which allows

for the specification of defeasible rules under the

stable model semantics. ASP allows for a declarative

representation of knowledge and the presence of

default negation allows for the formalization of rules

with exceptions. Preferences among rules can be

modeled in ASP (see, for instance the approaches in

(Brewka, 2004)). Also, several state of the art ASP

solvers are nowadays available for computing

answer sets, among which DLV, Smodels, and

Clingo.

Among the other non-monotonic approaches, the

project will also explore the adoption of the Event

Calculus. In particular, REC (Chesani et al., 2009), a

reactive axiomatization of Event Calculus, seems to

be promising not only in order to capture BMK, but

also to study their interaction with guidelines

knowledge (Bottrighi et al., 2009).

On the other hand, when uncertainty is a primary

issue, Probabilistic Graphical Models (PGMs)

(Jensen and Nielsen, 2007) are a reference

framework for causal knowledge. There are several

points that can be of great relevance in medical

decision making. For instance, the level of

representation is currently assumed propositional,

while in medical setting knowledge about evolutions

could require more powerful representation

schemes. Also, we aim at investigating both the role

of probabilistic logics with first-order semantics

(like Markov logic (Domingos and Lowd, 2009),

probabilistic Horn abduction (Poole, 1993) or

independent choice logic (Poole, 1997)) and that of

Relational PGMs (Getoor et al., 2007) in the clinical

decision making setting. A further challenge is the

possibility of representing and reasoning with

continuous variables.

4.4 Patient Evolution

In order to build flexible and useful decision support

systems for therapeutic purposes, having knowledge

about the patient evolution (i.e. prognostic

information) is very relevant. To this extent, many

different Probabilistic Graphical Models (PGM)

(Koller and Friedman, 2009) have been already

developed in the literature. PGMs can be suitably

applied to this task. However, there are several

points that deserve more attention:

(1) modeling temporal processes is one of the

greatest challenges of decision networks. We aim at

studying the applicability of non markovian PGMs

to the clinical decision making setting. Moreover,

we aim at investigating the potentiality of anytime

algorithms providing approximate decision-making

strategies incrementally refined in order to deal with

time- or modeling-pressured situations;

(2) the majority of dynamic models rely on a

discrete-time model; A further goal of the present

research will be the investigation and the study of

suitable inference algorithms for continuous-time

PGMs, by putting emphasis also on representational

issues relevant for the medical applications.

5 INTEGRATION

Different forms of reasoning will be investigated for

integrating the different knowledge sources (and, in

particular, CIGs and BMK).

Given the defeasible character of medical

knowledge, non-monotonic logical reasoning seems

to be a good candidate for the integration. However,

the complexity of the problem demands for the

exploration of different approaches, and for the

definition of new logics. For instance, the project

TowardsaSecondGenerationofComputerInterpretableGuidelines

203

will explore whether a language combining ASP

with some form of temporal reasoning (or reasoning

about actions) can be suited for the specification of

both CIGs and the BMK, and for the integrated

reasoning about them. An example of such a

language is given by REC (Chesani et al., 2009), a

reactive axiomatization of Event Calculus. Another

example is given by Temporal ASP (Giordano et al.,

2012), a temporal extension of Answer Set

Programming (ASP) (Gelfond, 2007), which

combines ASP with temporal constraints in LTL.

Besides non-monotonic and defeasible

reasoning, also utility-based reasoning will play an

important role in the integration since, in many

practical situations, decisions concerning patient

diagnosis or therapy must be grounded on

quantitative cost/benefit information. We aim at

resorting to reasoning based on decision theory, by

explicitly taking into account uncertainty via

probabilistic modeling, combined with utilities of

outcomes. Since PGMs will be introduced in the

BMK setting, we rely on the features of (dynamic)

decision networks to allow clinical guidelines to

locally exploit optimal decisions from the

underlying network model, when different

alternative actions are possible and have to be

evaluated by considering both uncertainty in the data

or in the evolution and cost/utilities of the outcomes.

Moreover, a specific goal of the proposal is the

study of the integration of analogical (Case-Based)

reasoning with the above forms of reasoning.

Usually, CBR tools are able to extract relevant

knowledge, but that leave to the user the

responsibility of providing its interpretation and of

formulating the ﬁnal decision. This is because a

strict interaction with the BMK has to be

established. Another goal of the proposal will be to

study such an interaction, both at the most general

level, as well as at the level of specific medical

applications.

To demonstrate the practical feasibility of our

approach, prototypes will be developed, for each

specific task in the agenda. We propose to identify a

set of case studies, in strict cooperation with the

Hospitals and Health Agencies that will take part to

the work. Such case studies will constitute the glue

to relate the different prototypes, and, thus the

different approaches and methodologies developed

within the research work.

ACKNOWLEDGEMENTS

The work described in paper was partially supported

by Compagnia di San Paolo, in the Ginseng project.

REFERENCES

Aamodt, A., Plaza, E., 1994. Case-based reasoning:

foundational issues, methodological variations and

systems approaches. AI Communications 7, 39–59

Montani, S., Bottrighi, A., 2006. Towards a

Comprehensive Treatment of Repetitions, Periodicity

and Temporal Constraints in Clinical Guidelines. AIIM

38, Elsevier, 171-195,

Anselma, L., Terenziani, P., Snodgrass, R. T., 2012.

Valid-Time Indeterminacy in Temporal Relational

Databases: Semantics and Representations, IEEE

TKDE, vol. 99, PrePrints.

Baarir, S., Beccuti, M., Cerotti, D., De Pierro, M.,

Donatelli, S., Franceschinis, G., 2009. The GreatSPN

tool: recent enhancements. ACM SIGMETRICS

Performance Evaluation Review, Vol.36 Issue 4, 4-9

Beccuti, M., Bottrighi, A., Franceschinis, G., Montani, S. ,

Terenziani, P. 2009. Modeling and verifying clinical

guidelines through Petri Nets, Proc. AIME’09, LNAI

5651.

Bellazzi, R., Larizza, C., Riva, A., 1998. Temporal

abstractions for interpreting diabetic patients

monitoring data. Intelligent Data Analysis 2, 97–122.

Bellazzi, R., Montani, S., Portinale, L., Riva, A. 1999.

Integrating Rule-Based and Case-Based Decision

Making in Diabetic Patient Management, Proc.

ICCBR'99, Munich, LNCS 1650, 386-400.

Bonatti, P., Calimeri, F., Leone, N., Ricca, F. 2010.

Answer Set Programming, chapter 5. LNCS Volume

6000 of Dovier and Pontelli,.

Bottrighi, A., Chesani, F., Mello, P., Molino, G., Montali,

M., Montani, M., Storari, S., Terenziani, P., Torchio,

M., 2009. An Hybrid Approach to Clinical Guideline

Conformance. Proc. AIME'09, LNAI 5651.

Bottrighi, A., Giordano, L., Molino, G., Montani, S..

P.Terenziani, M.Torchio, 2010. Adopting model

checking techniques for clinical guidelines

verification, AIIM 48(1), 1-19.

Brewka, G., 2004.: Complex Preferences for Answer Set

Optimization, KR 2004, 213-223.

Chesani, F., Mello, P., Montali, M., Torroni, P., 2009.

Commitment tracking via the reactive event calculus.

In Proc. IJCAI 2009, 91-96.

Domingos P., Lowd, D. 2009. Markov Logic: an Interface

Layer for Artificial Intelligence. Morgan & Claypool

Publ.

Dyreson, C.. 2009.Temporal Indeterminacy. In L. Liu and

M. Tamer Özsu. Encyclopedia of Database Systems,

Springer.

Gelfond, M., 2007. Handbook of Knowledge

Representation, chapter 7, Answer Sets. Elsevier.

Getoor, L., Friedman, N., Koller, D., Pfeffer, A., Taskar,

B., 2007. Probabilistic Relational Models. In L.

Getoor and B. Taskar, editors, Introduction to

Statistical Relational Learning

DATA2013-2ndInternationalConferenceonDataManagementTechnologiesandApplications

204

Giordano, L., Terenziani, P., Bottrighi, A., Montani, S.,

Donzella, L., 2006. Model Checking for Clinical

Guidelines: an Agent-based Approach. in Proc.

AMIA’06, Washington.

Giordano, L., Gliozzi, V., Olivetti, N., Pozzato, G. L.,

2009. ALC + T: a Preferential Extension of

Description Logics. Fundam. Inform. 96(3): 341-372.

Giordano, L., Martelli, A., Theseider Dupré, D., 2012.

Reasoning about actions with Temporal Answer Sets.

In Theory and Practice of Logic Programming. 2012,

to appear. Available at http://arxiv.org/abs/1110.3672

Jensen, F.V., Nielsen, T.D., 2007. Bayesian Networks and

Decision Graphs. Springer.

Koller, D., Friedman, N., 2009. Probabilistic Graphical

Models, MIT Press.

Klinov, P., Parsia, B., 2008. Optimization and evaluation

of reasoning in probabilistic description logic:

Towards a systematic approach, in International

Semantic Web Conference, LNCS 5318, 213-228.

Springer.

Leonardi, G. , Bottrighi, A., Galliani, G., Terenziani, P.

Messina, A., Della Corte, F., 2012. Exceptions

Handling within GLARE Clinical Guideline

Framework, Proc. AMIA 2012, Distinguished Paper

Award.

Montani, S., Leonardi, G., Bottrighi, A., Portinale, L.,

Terenziani, P., 2013. Supporting Flexible, Efficient

and User-Interpretable Retrieval of Similar Time

Series. IEEE TKDE, ISSN: 1041-4347, doi:

10.1109/TKDE.2011.264

Peleg, M., Tu, S., Bury, J., Ciccarese, P., Fox, J., Greenes,

R.A., Hall, R., Johnson, P.D., Jones, N., Kumar, A.,

Miksch, S., Quaglini, S., Seyfang, A., Shortliffe, E.H.,

Stefanelli, M., 2004. Comparing computer

interpretable GL models: a case-study approach,

JAMIA 10, 52–68.

Peleg, M., Rubin, D., Altman, R.B., 2005. Using petri nets

tools to study properties and dynamics of biological

systems. JAMIA 12:181–199.

Poole, D., 1993. Probabilistic Horn Abduction and

Bayesian Networks. Artificial Intelligence, 64(1): 81-

129.

Poole, D., 1997. The Independent Choice Logic for

modelling multiple agents under uncertainty, Artificial

Intelligence, 94(1-2): 7-56.

Quaglini, S., Stefanelli, M., Lanzola, G., Caporusso, V.,

Panzarasa, S., 2001. Flexible guideline-based patient

careflow systems. AIIM 22:65-80.

Schmidt, R., Montani, S., Bellazzi, R., Portinale, L., Gierl,

L. 2001. Case-Based Reasoning for Medical

Knowledge-based Systems, International Journal of

Medical Informatics 64(2), 355-367.

Shahar, Y., 1997. A framework for knowledge-based

temporal abstractions. Artiﬁcial Intelligence 90, 79–

133.

Snodgrass, R. T. (Editor), 1995: The TSQL2 Temporal

Query Language, Kluwer Academic Publishers.

Ten Teije, A., Miksch, S., Lucas, P., 2008Computer-based

Medical Guidelines and Protocols: A Primer and

Current Trends, Volume 139 Studies in Health

Technology and Informatics, ISBN: 978-1-58603-873-

Terenziani, P., Molino, G., Torchio, M, 2001. A Modular

Approach for Representing and Executing Clinical

Guidelines. AIIM 23, 249-276.

Terenziani, P. , Snodgrass, R.T., Bottrighi, A., Torchio,

M., Molino, G., 2007. Extending temporal database to

deal with telic/atelic medical data, AIIM 39(2) 113-

126.

Terenziani, P., Montani, S., Bottrighi, A., Molino, G.,

Torchio, M., 2008. Applying Artificial Intelligence to

Clinical Guidelines: the GLARE Approach, in (Tejie

et al., 2008).

TowardsaSecondGenerationofComputerInterpretableGuidelines

205