Run-time Support to Comorbidities in GLARE-SSCPM

Alessio Bottrighi, Luca Piovesan and Paolo Terenziani

DISIT, Universita’ del Piemonte Orientale, Viale Teresa Michel 11, Alessandria, Italy

Keywords: Clinical Guidelines, Comorbid Patients, Interaction Analysis and Management.

Abstract: Comorbidities play a relevant role in healthcare, so that, in the last years, several approaches Medical

Informatics and Artificial Intelligence have developed software tools to support physicians in the treatment

of comorbid patients. Computer Interpretable Guidelines (CIGs) are consolidated decision support tools to

help physicians, but they are devoted to provide evidence-based recommendations for one specific disease.

In order to support the treatment of patient affected by multiple diseases, challenging additional problems

have to be addressed, such as (i) the detection of the interactions between CIG actions, (ii) their

management, and, finally, (ii) the “merge” of CIGs. Several CIG approaches have been recently extended in

order to face (at least one of) such challenging problems, and one of them is GLARE (GuideLine

Acquisition Representation and Execution). However, such approaches have mostly focused on the “a-

priori” treatment of such problems, while addressing them “run-time” (i.e., to support physicians during the

execution of the CIGs on a specific patient) involves additional challenges, and requires additional

methodologies. In this paper we take advantage of previous extensions of GLARE (to cope with issues (i),

(ii), (iii)), and propose a new knowledge-based, “focused” and interactive management of comorbid

patients.

1 INTRODUCTION

The term comorbidity indicates the co-occurrence of

more than one disease in a patient. They are quite

frequent (an average of 25% of the population), thus

constituting an important problem from different

viewpoints.

Evidence-based decision making is a quite

consolidated practice in healthcare, since it exploits

the evidence and knowledge provided by clinical

trials, and by previous experiences. One of the main

methodologies to put evidence-based medicine into

practice is the development of Clinical Practice

Guidelines (CPGs). CPGs are defined as

“systematically developed statements to assist

practitioner and patient decisions about appropriate

health care in specific clinical circumstances”

(Institute of Medicine (US), 1990). Generally, CPGs

are elaborated by national or international teams of

specialists, and collect and organize in a textual form

the knowledge available in literature to manage a

specific clinical circumstance. They play a major

role in modern healthcare, and thousands of CPGs

have been devised in the last few years. For instance,

the Guideline International Network, which groups

97 organizations from all the continents, provides a

library of more than 6000 CPGs.

Additionally, in the last 30 years or so, the

research in Artificial Intelligence and in Medical

Informatics has shown that software tools can be

designed to increase the practical impact of CPGs in

healthcare. Specifically several software tools have

been devised in order to acquire, represent, execute

and reason with the so-called Computer-

Interpretable Guidelines (CIGs henceforth; see, for

example, the surveys (Peleg, 2013; Ten Teije et al.,

2008)).

1.1 CIG and Comorbidities

Unfortunately, CPGs provide evidence-based

information of interventions, but only on individual

pathologies. The simple solution of applying

multiple CPGs (one for each disease) to a patient

does not work: the treatments recommended by

different CPGs may interact with each other, and

such interactions may be (very) dangerous for

patients. The approach of considering all the

possible combinations of diseases is not only

difficult, but also impractical. Such considerations

498

Bottrighi, A., Piovesan, L. and Terenziani, P.

Run-time Suppor t to Comorbidities in GLARE-SSCPM.

DOI: 10.5220/0007685004980505

In Proceedings of the 12th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2019), pages 498-505

ISBN: 978-989-758-353-7

highlight the importance of developing

methodologies to merge CPGs for single disease

interventions to provide professionals' assistance to

comorbid patients (Riaño and Collado, 2013). The

development of such methods has been identified as

one of the “grand challenges” for clinical decision

support (Sittig et al., 2008). Since the early 2010's,

the research in Computer Science has been very

active in such a challenging area of research.

1.2 State of the Art

In general, the approaches devised in such an area

have specialized on the treatment of two different

subproblems:

(i) the detection of interactions between CIGs, and

their management (i.e., how to “solve”

interactions), and

(ii) the “merge” of CIGs.

Issue (i) above has been faced by relatively few

approaches in the CIG literature. In particular, the

approach in (Zamborlini et al., 2014) provides a

knowledge-based solution. It proposes a CIG-

independent conceptual model for medical actions

and reasoning forms operating on it, as well as

domain-independent rules to identify different types

of interactions on the basis of such a knowledge. A

similar approach has been pursued in the GLARE

approach (see Section 2 and (Piovesan et al., 2014),

(Piovesan and Terenziani, 2015)).

On the other hand, issue (ii) has been faced by

several CIG approaches. It is possible to distinguish

between the approaches aiming at achieving

“conservative” CIGs, and those that do not. The

approach in (Sánchez-Garzón et al., 2013), for

instance, belongs to the latter category. It builds ad-

hoc CIGs from scratch, using an agent-based

approach. Agents with hierarchical planning

capabilities represent experts in the treatment of

specific diseases. The CIG coping with the

comorbidity is obtained through the coordination of

all the agents. However, the mainstream is

constituted by conservative techniques, attempting

to merge existing CIGs with limited changes since,

in the real medical practice, physician need to follow

as much as possible evidence-based

recommendations, such as the ones proposed in the

CPGs (and, thus, CIGs) in the literature. The

approaches in such a mainstream mostly assume that

the possible interactions and their managements

have been defined a priori by physicians, and

focuses on CIG merge only. However, quite

different techniques have been proposed. For

instance, in (Wilk et al., 2013), constraint logic

programming (CLP) is adopted. A CLP is derived

from the CIGs, the interactions and their

managements, and a mitigation algorithm is

proposed in order to achieve the merge. Riaño and

Collado (Riaño and Collado, 2013) propose a

model-based approach for the merge. They model

treatments as oriented graphs composed by decisions

and actions. With the help of physicians, they define

a set of operators to merge decisions or actions. The

combination of the original CIGs is obtained through

the application of the operators. On the other hand,

in GLARE, the different management options

applied to independently solve the interactions are

merged through a conciliation module which is

based on CSP (Constraint Satisfaction Problems)

methodologies (Piovesan and Terenziani, 2016).

1.3 “Run-time” Support

Some of the above approaches can be used both (i)

“a-priori”, to analyse interactions between CIGs or

to merge them, without any reference to a specific

patient, and (ii) “run-time”, to support the execution

of CIGs on a specific patient. However, the “run-

time” application of the above methodologies

involves the resolution of new problems: when and

on which parts of the CIGs interaction detection has

to be performed? And the management of the

interactions? And the merge? Such problems are

still open problems in the specialised literature, and

the goal of this paper is to propose a general

methodology to cope with them, thus providing

physicians with an effective, user-friendly and

highly interactive approach supporting physicians in

the run-time treatment of comorbid patients.

Specifically, our approach grounds on GLARE, and

on its extensions (called GLARE-SSCPM) to deal

with comorbidities, which are briefly resumed in

Section 2). However, we emphasize that our

methodology is mostly system-independent, and can

be tuned in order to apply to other approaches to

comorbidities in the literature.

2 BACKGROUND: GLARE AND

GLARE-SSCPM

GLARE Support System for Comorbid Patient

Management (GLARE-SSCPM; (Piovesan et al.,

2018)) is an extension of GLARE (Terenziani et al.,

2001) to support the management of comorbidities,

and which takes into account both the (i) interaction

detection and management, and (ii) the CIG merge.

Run-time Support to Comorbidities in GLARE-SSCPM

499

In the following, we briefly resume such an

approach, which is the basis of the support to the

approach proposed in Section 3.

2.1 GLARE

GLARE (GuideLine Acquisition Representation and

Execution, (Terenziani et al., 2001)) is a well-known

CIG framework, designed in a long term cooperation

between the University of Eastern Piedmont and the

Azienda Ospedaliera San Giovanni Battista in Turin

(one of the largest hospitals in Italy), started in 1997.

The kernel of GLARE provides a formalism to

represent CIGs, a tool to acquire them, a mechanism

to execute a CIG on a specific patient. In GLARE,

CIGs are modelled as hierarchical graphs, in which

nodes represent actions or decisions and arcs

represent the control flow relations between nodes.

GLARE distinguishes between atomic actions

(simple steps in a CIG) and composite actions

(plans), which are defined in terms of their

components (thus supporting the definition of CIGs

at different levels of abstraction). Atomic actions

can be work actions (a procedure which must be

executed), pharmacological actions (a drug to be

administered), query actions (retrieval of

information from the clinical record/examinations)

or decision actions (choice among different

alternatives). In particular, GLARE distinguish

among diagnostic and therapeutic decisions (see the

discussion in Section 3.2).

Arcs are used to represent the control flow

relations, and can be annotated with temporal

constraints. In particular, a sequence arc from node

N1 to N2 indicates that the action represented by N1

must terminate before the execution of N2. On the

other hand, constrained arcs represent complex

temporal relations between nodes (e.g., N2 during

N1), and can be used to enforce concurrent

execution of actions.

The kernel of GLARE consists of two main

modules: the acquisition module and the execution

one. The acquisition module proposes a user-

friendly graphical interface for the acquisition of

CIGs, and stores them in an internal format. The

execution module takes in input a CIG and the

clinical record of a specific patient, and supports the

“execution” of the CIG on the patient. The execution

module is based on the “agenda techniques”

(Terenziani et al., 2001): at each time during the

execution of a CIG, GLARE determines (in the

agenda) the set of current actions, each one paired

with a time window, indicating when the action has

to be executed (minimum and maximum time) to

comply the temporal constraints in the CIG.

Notably, GLARE supports concurrent actions.

GLARE’s architecture is open. In the latest

years, several new modules and\or methodologies

have been added to cope with automatic resource-

based contextualization (ADAPT module,

(Terenziani et al., 2004)), temporal reasoning (TR,

(Anselma et al., 2006)), decision making support

(DECIDE_HELP, (Montani et al., 2005)), and

model-based verification (VERIFY, (Bottrighi et al.,

2010)). Recently, GLARE has been extended to

cope with comorbidities (see below).

2.2 GLARE-SSCPM

GLARE-SSCPM (Piovesan et al., 2018) proposes a

set of user-friendly supports to the treatment of

comorbidities. It is a knowledge-based approach,

aimed to support step-by-step physicians in the

treatment of comorbidities.

Operationally speaking, GLARE-SSCPM is

based on a CIG-independent knowledge base of

clinical actions, effects, and interactions, and

supports three main tasks:

(1) The detection of interactions occurring

between CIGs

(2) The management of the interactions

(3) The final merging of the CIGs

Since, in the real practice, interaction occur in

time, all the above tasks can be achieved only if the

temporal dimension is taken into account. Therefore,

also (4) Temporal Reasoning is considered in

GLARE-SSCPM:

Knowledge Base and Reasoning Supports.

GLARE-SSPCM is based on a Knowledge

Manager, i.e., a module coping with additional

(CIG-independent) medical knowledge (Piovesan et

al., 2014). It adopts an OWL ontological model

developed in collaboration with expert physicians,

using Protégé and integrating part of medical

models, such as SNOMED CT and ATC. Each

action in GLARE can be associated with one or

more elements of the ontological model. Such a

knowledge base contains both a general ontology,

describing general notions such as actions, action

intentions\effects, time, interactions, as well as

domain-specific knowledge, such as possible

interactions between specific drug types. Moreover,

the Knowledge Manager module is provided with

standard OWL reasoners providing inferences. Such

inferential mechanisms are used to devise a tool that

navigates the knowledge base and detects which are

the possible interactions (if any) between actions’

effects.

HEALTHINF 2019 - 12th International Conference on Health Informatics

500

Interaction Detection. GLARE-SSCPM Interaction

Detection module (Piovesan et al., 2014) provides a

flexible and interactive focusing tool allowing

physicians to navigate through the different

abstraction levels in the CIGs, to identify the

“relevant” actions. Once the actions of interest are

identified though focusing, interaction detection is

automatic: GLARE-SSCPM exploits the knowledge

provided by the knowledge manager and the OWL

reasoner to retrieve all the interactions between the

intentions, effects and drugs prescribed (in case of

pharmacological actions) of the focused actions.

Interaction Management. Once detected and

analysed, interactions must be managed.

Management options are local (and as small as

possible) changes in the original CIGs, which make

the original GIGs executable, avoiding undesirable

interactions and promoting desirable ones. On the

basis of the medical literature, GLARE-SSPCM

propose a wide range of general (i.e., CIG

independent and domain independent) interaction

management options (Piovesan and Terenziani,

2015): Safe Alternative, Replanning, Temporal

Avoidance, Effect Monitoring, Dosage Adjustment,

Interaction Mitigation, Interaction Alignment,

Intention Alignment.

GLARE-SSCPM provides a facility to instantiate

each one of such options, i.e., to apply it to a specific

input interaction, and to modify the CIGs

accordingly. The idea is that, given a specific

interaction, the user-physicians may apply one of the

options, or even trying to apply more than one, in a

“what-if” modality, see what the consequences on

the CIG are, and finally chose an apply the preferred

option in a definitive way.

Merge. Once the interactions have been identified,

and managed in isolation, the union of the original

CIGs with the applied managements is not yet an

executable CIG, since the management options lead

to changes to the original CIGs that are “locally”

consistent, but possibly not consistent with each

other. For these reasons, a final “merging” step is

required. In GLARE-SSCPM such a step is

performed by the CIG Conciliation module

(Piovesan and Terenziani, 2016), which provides as

output a “merged” CIG executable by GLARE.

Temporal Reasoning. GLARE-SSCPM provides

the Temporal module (Piovesan et al., 2015), to cope

with temporal constraints and to perform temporal

reasoning. Such a module operates as a knowledge

server: temporal problems may be demanded to the

Temporal module, which provides them a solution

(or report that there is no solution).

3 RUN-TIME SUPPORT TO

COMORBIDITIES

3.1 Philosophy of the Approach:

Focusing and Interactivity

As discussed in the Introduction, several approaches

in the literature focus on the a-priori “merge” of

CIGs, to avoid dangerous interactions. Such

approaches usually consider whole CIGs, and mostly

operate without interacting with physicians: given

two or more CIGs, to provide to physicians a new

CIG, avoiding dangerous interactions.

GLARE-SSCPM follows a different philosophy:

it provides a highly interactive approach, in which

physicians may (i) focus on specific subparts of the

CIGs, (ii) analyse possible interactions and (iii)

adopt GLARE–SSPCM to check the effects of

applying different management options to deal with

interactions (Piovesan et al., 2018).

In this paper, we propose a methodology to

extend the approaches coping with CIGs and

comorbidities with proper supports for “run-time”

execution, and we follow GLARE-SSCPM

“philosophy”: our methodology supports “focusing”,

and highly interactive with physicians.

Focusing is needed because, when executing

CIGs on a specific patient, physicians are not

interested with the whole CIGs, but only on the

subpart of them that is applicable to the given

patient, given the patient status. Indeed, focusing is

needed along two dimensions:

(i) The dimension of alternative paths in the

CIGs

(ii) The “temporal” dimension

Dimension (i) concern the fact that real CIGs usually

contain many (even hundreds) of different

alternative paths, depending on the different status

that the patient may assume during the CIG

execution. Obviously, only the paths that are

recommended (given the current status of the

patient) are interesting for physicians, and thus have

to be taken into account by decision support tools.

Dimension (ii) regards the fact that physicians do

not usually plan patient treatments far-away in the

future. They consider a limited “window” of the

CIG, usually not exceeding the next decision step in

the CIG. Indeed, taking “future” decisions on the

basis of the current status of the patient is nearly a

“bid”, which is rarely performed by physicians.

Interactivity is needed, in general, because we see

our approach as a support tool, which does not

substitute physicians, but helps them, by providing

Run-time Support to Comorbidities in GLARE-SSCPM

501

additional knowledge and recommendations.

Specifically, in the case of co-morbidities, while we

provide a fully automatic support to the detection of

possible interactions between (the “focused” parts of

the) CIGs, we want to be highly interactive in the

selection of the management options to treat such

interactions. In general, more than one option is

applicable, and we do not want to impose any

specific choice to physicians. On the other hand, we

want to support them in such a choice, by showing

them in an automatic way the consequences of

choosing a given option, or another.

3.2 Scheduled and Candidate Actions

From the practical point of view, a key issue to

realize the notion of “run-time” focusing is the

definition of scheduled (CIG) actions, and of

candidate ones. To propose such definitions, we first

have to point out the different nature of diagnostic vs

therapeutic decisions in CIGs. GLARE (as well as

several other CIG tools), clearly distinguishes

between diagnostic and therapeutic decisions.

Diagnostic decisions discriminate among different

diagnoses on the basis of the patient status,

considering a set of parameters (e.g., blood pressure,

fever, …), which vary from decision to decision. In

GLARE such decisions are represented as scored or

Boolean decisions. The decision criteria are

described within the decision action, and are

automatically evaluated by GLARE, considering the

clinical record describing the status of the patient.

Though in GLARE diagnostic decisions are taken in

a semi-automatic way (since GLARE allows

physicians to over-rule the decision taken

automatically by the system, selecting diagnoses

different from the ones derived from the automatic

evaluation of the decision criteria on the basis of the

status of the patient), such decisions are strictly

related to the state of the patient (so that physicians

cannot freely choose among them, independently of

the patient’s status). On the other hand, in

therapeutic decisions physicians have to choose

among different therapies that are all recommended

(by the CIG) for the given category of patients.

Physicians have usually the “full control” of such

therapeutic decisions, since all CIG alternative

treatments are usually “eligible” for patients. The

choice is done considering a given set of parameters:

effectiveness, cost, side-effects, compliance,

duration. Thus, in GLARE, a therapeutic decision

action is represented by a qualitative evaluation of

each one of the parameters above, for each one of

the alternatives. At run-time, GLARE presents such

evaluations to physicians, who are completely free

to choose among each one of the alternatives.

As a consequence, at any time during the execution

of a CIG on a patient, we distinguish.

(1) the set of current actions

(2) the set of scheduled actions

(3) the set of candidate actions

In Figure 1 in the following, we show a simple

example of GLARE CIGs, to exemplify the

definitions. For the sake of generality, we consider

an abstract example, instead of a concrete one. In the

example, round nodes represent work and

pharmacological actions, red diamond represent

diagnostic decisions, green diamonds represent

therapeutic decisions, and arcs represent the control

flow of actions. For the sake of simplicity, we only

consider sequence arcs (so that, at each time, each

CIG has only a current action).

Definition. Current Actions. The action to be

executed next (multiple next actions are possible, in

case of concurrency)

Example. For instance, in our example, we suppose

that the current action in CIG

is A

Definition. Scheduled Actions. Besides the current

action(s), the set of scheduled actions contains the

set of CIG actions which, if no failure or exception

arise, have necessarily to be executed next, and their

time window.

Specifically, the scheduled actions are all those CIG

actions that can be reached through chains of

Figure 1: Example CIGs. Round blue nodes represent work and pharmacological actions, red diamond represent diagnostic

decisions, green diamonds represent therapeutic decisions, and arcs represent the control flow of actions.

HEALTHINF 2019 - 12th International Conference on Health Informatics

502

sequence and constrained arcs starting from the

current actions, until a decision action is reached.

Example. In CIG

, if the current action is A

, the set

of scheduled actions is {A

, A

, D

}

Defintion. Candidate Actions. The set of candidate

actions contains the set of CIG actions which (if no

failure or exception arise), can possibly be scheduled

for execution, until a new therapeutic decision has to

be taken by physicians.

Candidate actions include all the actions that can be

reached from scheduled actions until a therapeutic

decision has to be taken. The idea is that, while

therapeutic decisions have to be taken by physicians,

diagnostic decisions depends on the status of the

patient. Therefore, the outcome of a future

diagnostic decision cannot be known a-priori, and

physicians may want to consider also the actions that

have to be taken after such decisions.

Example. In CIG

, if the current action is A

, the set

of candidate actions is {A

, T

, A

, T

}

3.3 GLARE Extensions

GLARE supports the execution of multiple CIGs.

For each CIG to be executed on a patient, GLARE

provides physicians with an Executor module

supporting the execution. In our approach to run-

time management of comorbidities, GLARE

executor has been enriched with the possibility of

sending and receiving messages to\from a

Comorbidity Master Module, and to activate

GLARE-SSCPM Interaction Management Module

and Conciliation Module.

In the following we consider a single patient (the

extension to multiple patients is trivial).

3.4 Extensions to the Executor

The modifications to GLARE’s original Executor

module are quite limited: it is extended to

communicate with the Comorbidity Master Module.

In particular, the Executor module sends to the

Comorbidity Master Module a message

(i) when it is created (i.e., when the execution of a

new CIG is started on the patient)

(ii) when the execution of a CIG action is

(successfully) terminated. In case the action is

a decision, also the selected path is sent to the

Comorbidity Master Module.

It receives from the Comorbidity Master Module a

message whenever

(iii) one or more interactions have to be managed

When the Executor receives a message that there are

interactions between scheduled actions, the standard

execution is stopped, until all interactions have been

managed. On the other hand, the treatment of

interactions between candidate actions is not

necessary, since such actions will not necessarily

have to be executed on the patient (their execution

depends on the future status of the patient, after the

execution of the scheduled actions). However, it is

important that physicians are notified soon that such

interactions may have to be faced in a near future.

3.5 Treatment of the Interactions

The management of interactions is performed by the

physicians with the support of GLARE-SSCPM

Interaction Management module. Given an

interaction, such a model provides physicians with

the possibility of choosing the most appropriate

management, and helps them in its application to the

original CIGs. Specifically, a result of the

application of the Interaction Management module,

the physician can see how the original CIGs are

modified when applying the chosen interaction

management operation to the given interaction. Such

a process can be iterated, until one of the possible

management is chosen by the physicians.

Notably, in case more than one interaction has to

be managed, the Conciliation Module is invoked, in

order to check the consistency of the different

modifications to the original CIGs. In case they are

consistent, the CIGs in CIG

pat

are updated with the

selected managements, and the Executor Modules

can re-start execution on the updated CIGs. In case

they are not consistent, physicians are requested to

backtrack to the management of the interactions to

consider alternative management options, until a

consistent set of managements is determined.

3.6 Comorbidity Master Module

A new dedicated module has to be introduced, in

order to support the run-time management of

comorbidities. In the following, we informally

describe it (called Comorbidity Master module).

The Comorbidity Master Module takes in input

(1) the clinical record of the patient

(2) the CIGs currently under execution (indicated

by CIG

pat

henceforth)

(3) the current action in each CIG in CIG

pat

and manages for each CIG

in (2), two local

data structures:

(4) the set SA

of scheduled actions

(5) the set CA

of candidate actions

Run-time Support to Comorbidities in GLARE-SSCPM

503

When the Comorbidity Master Module receives in

input (from the Executor of one of the CIGs) a

message that the execution of an action A

in the

CIG CIG

in CIG

pat

has terminated, and the action

is not a decision action, it simply updates the set

by deleting A

from it.

On the other hand, in cases

(i) it receives in input a message that a new CIG

CIG

has been activated on the patient (so that

CIG

is added to CIG

pat

)

(ii) it receives in input a message that the execution

of an action A

in the CIG CIG

has

terminated, and A

is a decision action, and

Path

has been selected

several operations have to be performed, for the

“run-time” identification and resolution of possible

interactions. In such cases, the Comorbidity Master

Module

1. evaluates the new set SA

of scheduled actions

(and their temporal windows)

2. evaluates the new set CA

of scheduled actions

(and their temporal windows).

3. Invokes the interaction detection module on the

sets of scheduled actions of the CIGs in CIG

pat

(not considering the decision actions). In case

some interaction is detected, the set INT_sched

of such interactions is sent to the Executors of

the CIGs in CIG

pat

, with the indication that

such interactions occur between scheduled

actions.

4. Invokes the interaction detection module on the

sets of candidate actions of the CIGs in CIG

pat

(not considering the decision actions). In case

some interaction is detected, the set INT_cand

of such interactions is sent to the Executors of

the CIGs in CIG

pat

, with the indication that

such interactions may occur between candidate

actions.

Notably, the detection of interaction is based on the

Knowledge base, and is fully automatic.

In the following, we show two examples of Steps 1

and 2 above. Concrete examples of the management

of CIG interactions have been reported in (Piovesan

and Terenziani, 2015; Piovesan et al., 2018).

Example. Suppose that the CIG

is being executed

on patient 1, and that, when A

is under execution (is

current), the treatment of a new disease, through the

CIG CIG

, is started. The start of the execution of

CIG

triggers the Comorbidity Master Manager for

patient 1. CIG

pat1

={CIG

,CIG

}, and the set of

scheduled and candidate actions are valuated as

follows:

={A

}, CA

={A

}

={A

100

101

100

}, CA

={T

100

104

105

106

101

107

108

}

The Interaction Detection module is activated, and

interactions between A

100

101

(if any) must

be managed by physicians (while the interactions

considering also A

, A

104

, A

105

106

, A

107

108

(if

any) are pointed out to the physicians.

Example. Suppose that CIG

and CIG

are being

executed on patient 1, that the current actions in

CIG

and CIG

are A

and D

100

respectively. We

thus have

={A

}, CA

={A

}

={D

100

}, CA

={A

104

105

, D

101

, A

106

107

, A

108

}

Suppose then that the execution of decision D

100

give as result the path starting with A

106

. Then,

={A

106

101

}, CA

={A

107

, A

108

}

4 CONCLUSIONS

The CIG literature has devoted a considerable

attention to the treatment of comorbid patients.

However, the problem of supporting physicians in

the “run-time” detection and management of CIG

interactions has been quite neglected: in short,

(Zamborlini et al., 2014) copes with knowledge-

based interaction detection (but not with CIG

merge), while the other approaches discussed in

Section 1.2 focus on the merge of whole CIGs,

assuming to have a pre-defined set of possible

interactions, and of the way to treat each of them.

In this paper, we propose a comprehensive

approach to run-time comorbidity management,

based on GLARE and GLARE-SSCPM, which (i)

automatically detects the “relevant” parts of the

CIGs (i.e., scheduled and candidate actions), (ii)

automatically detects possible interactions between

them, (iii) supports physicians in the choice of the

most appropriate management of such interactions.

A prototypical implementation of the proposed

approach is under development. Future works

mainly concern a full realization of a tool, and an

extensive experimentation on different concrete

cases of comorbidity.

REFERENCES

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

2006. Towards a comprehensive treatment of

repetitions, periodicity and temporal constraints in

clinical guidelines. Artif. Intell. Med., Temporal

HEALTHINF 2019 - 12th International Conference on Health Informatics

504

Representation and Reasoning in Medicine 38, 171–

195. https://doi.org/10.1016/j.artmed.2006.03.007

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

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

checking techniques for clinical guidelines

verification. Artif. Intell. Med. 48, 1–19.

https://doi.org/10.1016/j.artmed.2009.09.003

Institute of Medicine (US), 1990. Clinical practice

guidelines directions for a new program. National

Academy Press, Washington, D.C.

Montani, S., Terenziani, P., Bottrighi, A., 2005. Exploiting

Decision Theory for Supporting Therapy Selection in

Computerized Clinical Guidelines, in: Miksch, S.,

Hunter, J., Keravnou, E.T. (Eds.), Artificial

Intelligence in Medicine, 10th Conference on

Artificial Intelligence in Medicine, AIME 2005,

Aberdeen, UK, July 23-27, 2005, Proceedings, Lecture

Notes in Computer Science. Springer, pp. 136–140.

https://doi.org/10.1007/11527770_19

Peleg, M., 2013. Computer-interpretable clinical

guidelines: A methodological review. J. Biomed.

Inform. 46, 744–763.

https://doi.org/10.1016/j.jbi.2013.06.009

Piovesan, L., Anselma, L., Terenziani, P., 2015. Temporal

detection of guideline interactions, in: HEALTHINF

2015. Presented at the International Conference on

Health Informatics 2015, Scitepress, pp. 40–50.

https://doi.org/10.5220/0005186300400050

Piovesan, L., Molino, G., Terenziani, P., 2014. Supporting

Physicians in the Detection of the Interactions between

Treatments of Co-Morbid Patients, in: Healthcare

Informatics and Analytics: Emerging Issues and

Trends. IGI Global, pp. 165–193.

Piovesan, L., Terenziani, P., 2016. A Constraint-Based

Approach for the Conciliation of Clinical Guidelines,

in: Advances in Artificial Intelligence - IBERAMIA

2016, LNCS. Presented at the Ibero-American

Conference on Artificial Intelligence, Springer

International Publishing, pp. 77–88.

https://doi.org/10.1007/978-3-319-47955-2_7

Piovesan, L., Terenziani, P., 2015. A Mixed-Initiative

approach to the conciliation of Clinical Guidelines for

comorbid patients, in: KR4HC 2015, Lecture Notes in

Artificial Intelligence. Springer International

Publishing, Pavia, pp. 95–108.

Piovesan, L., Terenziani, P., Molino, G., 2018. GLARE-

SSCPM: an Intelligent System to Support the

Treatment of Comorbid Patients. IEEE Intell. Syst.

https://doi.org/10.1109/MIS.2018.111144734

Riaño, D., Collado, A., 2013. Model-Based Combination

of Treatments for the Management of Chronic

Comorbid Patients, in: Artificial Intelligence in

Medicine. Springer Berlin Heidelberg, Berlin,

Heidelberg, pp. 11–16.

Sánchez-Garzón, I., Fernández-Olivares, J., Onaindía, E.,

Milla, G., Jordán, J., Castejón, P., 2013. A Multi-agent

Planning Approach for the Generation of Personalized

Treatment Plans of Comorbid Patients, in: AIME

2013, Lecture Notes in Computer Science. Springer

Berlin Heidelberg, pp. 23–27.

https://doi.org/10.1007/978-3-642-38326-7_4

Sittig, D.F., Wright, A., Osheroff, J.A., Middleton, B.,

Teich, J.M., Ash, J.S., Campbell, E., Bates, D.W.,

2008. Grand challenges in clinical decision support. J.

Biomed. Inform. 41, 387–392.

https://doi.org/10.1016/j.jbi.2007.09.003

Ten Teije, A., Miksch, S., Lucas, P. (Eds.), 2008.

Computer-based medical guidelines and protocols: a

primer and current trends, Studies in health technology

and informatics. IOS Press, Amsterdam.

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

approach for representing and executing clinical

guidelines. Artif. Intell. Med. 23, 249–276.

https://doi.org/10.1016/S0933-3657(01)00087-2

Terenziani, P., Montani, S., Bottrighi, A., Torchio, M.,

Molino, G., Correndo, G., 2004. A context-adaptable

approach to clinical guidelines. Stud. Health Technol.

Inform. 107, 169–173.

Wilk, S., Michalowski, W., Michalowski, M., Farion, K.,

Hing, M.M., Mohapatra, S., 2013. Mitigation of

adverse interactions in pairs of clinical practice

guidelines using constraint logic programming. J.

Biomed. Inform. 46, 341–353.

https://doi.org/10.1016/j.jbi.2013.01.002

Zamborlini, V., da Silveira, M., Pruski, C., ten Teije, A.,

van Harmelen, F., 2014. Towards a Conceptual Model

for Enhancing Reasoning About Clinical Guidelines,

in: Miksch, S., Riaño, D., ten Teije, A. (Eds.),

Knowledge Representation for Health Care. Springer

International Publishing, Cham, pp. 29–44.

Run-time Support to Comorbidities in GLARE-SSCPM

505