KNOWLEDGE BASED DECISION SUPPORT FOR THE

MANAGEMENT OF CHRONIC PATIENTS

Sara Colantonio, Massimo Martinelli, Ovidio Salvetti

Institute of Information Science and Tecnologies, ISTI-CNR, Via G. Moruzzi, 1-56124, Pisa, Italy

Giuseppe De Pietro, Massimo Esposito, Alberto Machì

Institute for High Performance Computing and Networking, ICAR-CNR, Via P. Castellino, 111-80131, Napoli, Italy

Keywords: Decision support, Knowledge formalization, Ontologies.

Abstract: Due to the current socio-economic impact of chronic diseases, a strong effort is being spent in the

development of ICT applications able to support a new care paradigm specialized for chronic patients. Such

applications are mainly based on patients’ telemonitoring for the collection of a number of relevant

physiological parameters aimed at identifying and preventing acute events, while maximizing patients’

quality of life and reducing clinical costs. The most advanced and challenging features of these ICT

applications are intelligent services devoted to the interpretation of monitored patients’ data for supporting

clinicians in their routine management of chronic patients. In this paper, a Knowledge-based Clinical

Decision Support System (KB-CDSS) is presented, which is aimed at aiding clinical professionals in

managing chronic patients on a daily basis, by assessing their current status, helping face their worsening

conditions, and preventing disease exacerbation events. The CDSS has been developed by encoding the

relevant knowledge elicited from clinicians who have a large experience in patients’ monitoring. A

formalism based on ontologies and rules was selected to build the Knowledge Base according to a scenario-

based approach. The system is currently under validation for the management of real clinical cases.

1 INTRODUCTION

Chronic diseases are one of the leading causes of

disability and death in most of the industrialized

countries, and have a deep impact on today’s

society, with social health-security systems under

constant pressure, both for financial and

organizational aspects, especially in many European

countries. A chronic disease usually causes major

limitations in patient’s daily living and are

characterized by acute or deterioration events, which

can happen more or less frequently and often cannot

be totally relieved, causing a worsening of patient’s

conditions.

Health organizations all over the world are more

and more focusing on the development of specific

programmes for the management of chronic patients

in the long term. These are mainly based on the

regular collection of information about patients’

status and actions, their compliance to the therapy,

the situation around them, and their interactions with

the environment, in a long-stay setting. Tele- or

home-monitoring programmes are, hence, being

studied and applied for care delivery and

management in cases of chronic diseases (Parè et al.,

2007). ICT solutions are being developed for this

task, and range from (i) “light” applications, based

on video or phone consultations of patients for

assessing their current conditions (Cleland et al.,

2005; Vitacca et al., 2009), to (ii) the continuous or

frequent acquisition of patients’ vital signs through

dedicated sensors, often wearable (e.g., Pentland,

2004), till (iii) complex platforms that merge such a

sensor infrastructure with intelligent services for

data interpretation (e.g., Chiarugi et al., 2010).

The intelligent components are the most

advanced and challenging feature of these

telemonitoring platforms. The intelligence can be

implemented at different level of complexity: (i) as

simple alarming services, which identify variations

in patient’s vital signs, (ii) as interpretation methods,

which recognize exacerbation events, or (iii) as

220

Colantonio S., Martinelli M., Salvetti O., De Pietro G., Esposito M. and Machì A..

KNOWLEDGE BASED DECISION SUPPORT FOR THE MANAGEMENT OF CHRONIC PATIENTS.

DOI: 10.5220/0003657902200225

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2011), pages 220-225

ISBN: 978-989-8425-80-5

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

complex Clinical Decision Support Systems (CDSS),

which are aimed at supplying to clinicians full

assistance in their management of chronic patients.

In this paper, a knowledge-based CDSS is

proposed for managing chronic patients by

interpreting data acquired through a sensor

infrastructure deployed in patients’ normal life

environment. Such a system combines acquired data

with patient’s clinical information, issues possible

alarms and supplies motivated suggestions. Though

the modelling strategy is valid for any chronic

disease, two pathologies were considered for initial

application, namely Chronic Obstructive Pulmonary

Disease (COPD) and Chronic Kidney Disease

(CKD). The system was developed within the EU

IST Project CHRONIOUS which is aimed at

defining a generic platform schema for health status

monitoring, addressed to and specialized for people

suffering from chronic diseases (Rosso et al., 2011).

In the following sections, the strategy followed

is described in more detail, presenting and

motivating the CDSS design, the data it handles and

the knowledge it utilizes. Particular focus is

dedicated to how this knowledge is formalized.

2 THE INTELLIGENT SUPPORT

TO CHRONIC DISEASES

MANAGEMENT

For being effective and profitable, the management

of chronic patients requires monitoring patients in

order to follow up their conditions and detect the

incoming or the occurring of acute events.

For making this possible, a platform of services

was devised to acquire and store data through a

sensing infrastructure consisting of (i) a set of

wearable sensors for the acquisition of patients’

physiological parameters, (ii) a set of environmental

sensors for the acquisition of contextual information

and (iii) a touch-screen device used for the

acquisition of patient’s answers to questionnaires.

Data collected during patient’s clinical visits were,

also, gathered by the platform and stored in the

internal repository (for a more detailed description

of the platform please refer to (Rosso et al., 2011)).

Intelligent services were added into the platform

in order to automatically process the collected sensor

data, identify worrying patient’s conditions and

alerting clinicians. In particular, to reach a high level

of flexibility and reliability, two levels of sensor data

interpretation were conceived: the data acquired by

the sensing infrastructure are firstly processed by a

light intelligence deployed on a smart Personal

Digital Assistance device (PDA) for detecting

possible changes in patient’s vital signs by applying

simple rules related to the collected parameters.

Worrying changes are reported as an alert to clinical

personnel and data are passed to a second level of

intelligence, i.e., the CDSS, which supplies more

accurate suggestions thanks to its ability to process a

larger piece of information. Figure 1 summarizes the

main components of the monitoring platform. Focus

of this paper is on the CDSS described in next

sections.

Figure 1: The main components of the patients’

monitoring platform: the Sensing infrastructure and the

Data Processing infrastructure consisting of two levels of

intelligence.

2.1 The Decision Support System

The CDSS was designed to be really effective,

efficient and to be perceived really useful by the

clinicians it should support. Its main functionalities

can be listed as follows:

 analysing the acquired sensor data for identifying

worrying conditions;

 alerting clinicians when an acute event happens;

 merging the heterogeneous patient’s information

for providing pertinent suggestions.

The data are gathered by a sensorized vest that

collects parameters pertaining patient’s

electrocardiographic activity, respiratory activity,

arterial oxygen saturation, skin temperature. These

data are acquired by the sensors on the vest and,

then, collected by a Data Handler, wired to the vest,

which sends them to the PDA via a Bluetooth

connection. The acquired parameters are the

following:

 heart rate;

 respiration rate, inspiration and expiration time;

 inspiration and expiration volume;

 ambient temperature and humidity;

 motion activity and fall;

KNOWLEDGE BASED DECISION SUPPORT FOR THE MANAGEMENT OF CHRONIC PATIENTS

221

 cough and snoring.

A touch-screen workstation, the Home Patient

Monitor (HPM), is employed to collect a number of

other physiological and contextual parameters. More

in detail, an environmental device, installed in

patient’s living room and cable-connected with the

HPM (USB connection), acquires contextual data

related to ambient light, carbon monoxide, volatile

organic compound and air particle. Body weight,

blood pressure and blood glucose are measured

using commercially available devices, which send

data to the HPM via a Bluetooth connection. Finally,

information pertinent to patient’s lifestyle, food and

drug intake, and psychological conditions is

collected through questionnaires proposed on the

touch-screen of the HPM. All these data are

gathered, on a regular time basis via a wireless

connection, by a PDA assigned to each patient. The

PDA performs a first data processing by applying

simple range checking rules and detects possible

alarming situations, alerting, in this case, the

personnel on duty, and requires an in-depth analysis

of the situation by the CDSS.

Indeed, the CDSS was designed to be invoked

each time new data to be analyzed are available, and

this happens in three scenarios:

 when the PDA detects a worrying condition

and issues an alarm: in this case, the sensor

data collected are sent to the CDSS;

 at the end of each day: when the PDA stores all

the collected data and sends them to the CDSS

for their analysis;

 when a patient undergoes a clinical visit: the

newly collected data are sent to the CDSS for

interpretation.

In all these cases, the CDSS correlates these data

with historical patient’s data according to the

knowledge modeled into its Knowledge Base (KB),

and supplies, as a response, a diagnosis about

current patient’s status, plus suggestions about what

to do. The KB is the main component of the system

and is modeled for inferential reasoning, through a

dedicated inference engine, as described in the next

section.

2.2 The Knowledge Base

The clinical knowledge modeled for developing the

KB consists of:

 the structure of the domain knowledge, namely

the declarative knowledge;

 the knowledge about the procedures of the

decision making activity, namely the procedural

knowledge.

In particular, the declarative knowledge concerns

the domain compositional elements, such as raw and

abstract concepts, their properties and inter-relations.

On the other hand, procedural knowledge captures

the behavioral logic and provides more explicit

information about which actions/conclusions can be

taken/drawn from declarative knowledge. The

formalism selected for encoding both these types of

knowledge consisted in one ontology and a set of

production rules (i.e. a set of conditional statements

expressed in form of "if antecedents then

consequent") built on the top of it.

The main purpose of the ontology and rules is to

represent domain-specific knowledge necessary to

remotely support clinical operators in the daily

home-monitoring of chronic patients. The approach

is generally aimed at the chronic disease

management, but specific focus was given to the two

pathologies chosen for system demonstration, i.e.,

COPD and CKD.

The way the knowledge is represented for

clinical decision support is one of the most key

facets for having a successful CDSS, starting from

the analysis and design of the CDSS at the very

beginning and ending to the implementation of the

CDSS at the final stage. Ontologies combined with

production rules seemed the most suitable and up-to-

date methodology for solving this task since easily

understandable by a non-specialized audience, e.g.

clinicians. In this way, they could be involved not

only in the knowledge elicitation and representation,

but also in the process of modification/updating of

existing knowledge.

In fact, the eliciting process ran through several

meetings with clinicians for systematizing the

approach to patients’ monitoring. The list of

monitored parameters was used as the starting point

to formalize all the statements about the different

situations and conditions that a patient can go

through and that can be identified by these

parameters. A great help to this process came from

the fact that clinicians were already skilled in

patients’ telemonitoring and were already trained at

interacting with computerized applications for

processing of clinical data.

The result of the elicitation was the formalization

of evidence-based statements which were used to

define the suggestions that should be provided by

the CDSS. The clinicians supplied these statements

in a rule-like form, written in natural language.

These were discussed and extended for creating a set

of consistent and complete rules to be processed by

an automated rule engine. The ontology was defined

to list up all the relevant concepts, selecting a

KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development

222

terminology recognized and agreed by all the

clinicians involved in the elicitation process.

The Web Ontology Language, OWL (2009), was

chosen as ontology language, in order to grant both

formality and expressive power. As regards to the

rule representation, Jena rule language (Carroll et

al., 2004) was identified as the most appropriate

language for writing rules that can be combined with

OWL ontologies. In particular, such a language

enables to build rules starting from the

terminological elements defined in the OWL

ontology. Thanks to its concise but at the same time

very expressive syntax, Jena rule language was not

only easy to use for writing rules but also extremely

simple to read and understand also for non-technical

users and, thus, it was very adequate to provide

reasoning support that met the CDSS requirements.

Moreover, the Jena framework includes a Rule

Engine used for inference on the base of encoded

rules.

2.2.1 The Ontology

The ontology models all the clinical data coming

from different possible information sources, such as

medical history, patient’s general information,

laboratory assays, patient’s monitoring

measurements or environmental measurements

gathered at the patient's home, questionnaires about

mental problems or symptoms opportunely filled up

by the patient at home. Moreover, the ontology also

models the results of the inferences generated by the

CDSS in terms of suggestions to be reported to the

clinicians. Such suggestions are expressed in the

form of alerts, i.e. messages with a different

severity, varying according to the current patient’s

condition, that can require or not the attention of a

clinical operator. Each suggestion can also indicate a

variation of the patient's health status with respect to

the morbidity he/she is affected by. It includes in

natural language the specific clinical guideline

applied by CDSS and delineates the action which

has to be performed in response to the generated

alert.

According to this high-level description, the

ontology concepts and properties are defined. Figure

2 shows the concept taxonomy.

As regards the properties, it is worth noting that

all the properties defined in the ontology are

datatype properties according to the definition of

OWL language. Figure 3 shows the taxonomy of

properties. The use of only datatype properties

simplifies the writing of rules and their final

structure since in the rule antecedents and

consequents it is necessary to indicate only the

appropriate datatype properties, corresponding to

semantically defined terms, opportunely linked to

the concepts they are associated to.

Figure 2: The concept taxonomy.

Figure 3: A piece of the property taxonomy.

2.2.2 The Base of Rules

Based on the introduced ontology, production rules

were devised for the three scenarios introduced in

the previous section, with the final aim of

representing the procedural knowledge.

Rules are simple conditional declarations that

link a logical combination of antecedent conditions

to a consequent, according to the following “if-then”

structure: If (antecedents) then consequent

As a result, the structure of each rule is

composed of one or more antecedents, expressed in

terms of ontology properties concatenated by logical

conjunctive operators, which can be evaluated to be

either true or false. Disjunction is not supported. As

an example, a rule pertaining to the HypoVolemia

for CKD disease is reported in natural language as

follows:

All individuals with values of systolic blood pressure

<110 mmHg, heart rate > 115 beats/min, symptoms

of nausea or vomiting determined by means of the

KNOWLEDGE BASED DECISION SUPPORT FOR THE MANAGEMENT OF CHRONIC PATIENTS

223

CKD Symptom Questionnaire are classified as being

in an abnormal condition due to HypoVolemia. An

alert with red severity is contextually sent to the

medical doctor to report the situation

The implementation of this rule in accordance

with the Jena rule syntax and in terms of the

ontology concepts and datatype properties

corresponds to a group of rules: one of them has the

following form

HypoVolemia_Guideline_1_rule_3:

(?p rdf:type

CKD:PatientMonitoringMeasurement),

(?p CKD:systolicBloodPressure_curr ?a),

lessThan(?a,110),

(?p CKD:heartRate_curr ?b),

greaterThan(?b,115),

(?q rdf:type CKD:CKDSymptomQuestionnaire),

(?q CKD:neauseaOrVomiting ?c),

equal(?c,’true’^^xsd:boolean),

(?s rdf:type CKD:CDSS_Suggestion) ->

(?s CKD:inferredPatientCondition

‘Suspected Hypovolemia’^^xsd:string)

3 SYSTEM DEVELOPMENT &

RESULTS

The CDSS was modeled as a resource of a

monitoring platform as introduced in the previous

section and it is called on demand to analyze new

data about patient’s situation. This assures the

generality and flexibility of this system, which can

be easily plugged into any similar platforms.

The implementation of the system was organized

in decisional services, which are called when

specific events occur. The granularity of these

services was decided in accordance to the data flow

and the requirements of clinicians about the

intervention of the support system. It appeared, then,

profitable to make a correspondence between the set

of services and the set of identified scenarios

introduced in the previous section, which can be

named as follows:

 the Alarm Checking scenario, which corresponds

to an assessment of patient’s status after an alarm

has been issued by the PDA, for alerting about a

possible exacerbation;

 the Home Monitoring scenario, which

corresponds to a periodic assessment of patient’s

status, more precisely once a day, even without

any alerting exacerbation;

 the Clinical Assessment scenario, which

corresponds to the evaluation of patient’s status

after a clinical visits, i.e., when new data comes

from the clinical environment.

Several advantages are assured by this approach,

since separated and well-focused services are: (i)

simpler to integrate; (ii) more flexible; (iii) can be

straightforwardly modified; (iv) their complexity can

be managed more easily. Moreover, this approach

made the implementation strategy of the system

straightforward: the decisional services were

mapped onto a Service Oriented Architecture

approach, and hence implemented as Web Services.

This assures the interoperability of the CDSS, whose

implementation does not depend on the platform:

thanks to the Web Services approach, the system can

be integrated in any other general platform without

any change to its structure.

The first step in the development of the

decisional services regarded the determination of

their mapping to the clinical domains, i.e. COPD and

CKD. Two services, named COPD_Decisional_

Service and CKD_Decisional_Service, were realized

respectively for COPD and CKD diseases. Each

service was delineated from a functional perspective

in terms of its operations, where each operation is

coarse-grained and models how the CDSS works for

a whole distinct usage scenario. Coarse-grained

services avoid the need to maintain state information

between service invocations, reduce the number of

network interactions required to implement a given

usage scenario, improving, this way, performance

and simplifying recovery in the case of failure.

For the development of the KB behind these

services, as introduced in the previous section, the

ontology was developed using OWL. Currently, it

consists of 28 concepts and 860 properties,

organized as outlined. The rules were structured in

scenarios and divided between the two pathologies:

totally the base of rules contains 435 rules for CKD

and 273 rules for COPD.

Results provided by the CDSS consist in an

advice about the status of the patient and a

suggestion about the action to be undertaken for

managing the situation. In agreement with

clinicians’ requirements, the results are shown as a

kind of alert/alarm through a Clinician GUI, listing

up a number of information. More in detail, the

CDSS results report:

 an advice about patient’s status diagnosis;

 a suggestion about what to do;

 some additional information, for better

presenting such results;

 a colour that indicates the severity of the advice

or alarm;

 an explanation of the advice and suggestion

produced.

As an example, Figure 4 shows the results inferred

KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development

224

by CKD_Decisional_Service in terms of suggestions

generated in response to abnormal values firstly

identified by the PDA and then passed to the CDSS.

The system has been released to the clinicians

and the validation phase has started for testing the

system in action.

Figure 4: An example of results inferred by

CKD_Decisional_Service. The Explanation field states

that the suggestion is issued according to the guidelines

defined within the project, after an alert sent by PDA (i.e.,

the scenario) and the natural language form of the

guideline.

4 CONCLUSIONS

Due to the socio-economic impact of chronic

diseases, a large effort is being spent to develop

monitoring platforms able to follow up chronic

patients and support clinicians in their management.

In this paper, a knowledge based Clinical Decision

Support System has been presented which encodes

the relevant knowledge elicited from clinicians who

have a large experience in patients’ monitoring.

A formalism based on ontologies and rules was

selected for its expressive power and, at the same

time, ease of use and understanding also by non

technical users that can be involved in an eventual

upgrade process. A scenario-based approach was

adopted to implement the system by mapping each

scenario on a decisional service. Web Services were,

then, used to implement such services, assuring, in

this way, interoperability of the system and the

possibility to plug it into any monitoring platform of

the same kind.

The system was developed within the EU project

CHRONIOUS which is, currently, starting the

validation phase, during which the system will be

deployed to the clinical sites and its functioning

precisely tested.

ACKNOWLEDGEMENTS

This work has been partly funded by the EC IST

Project FP7-ICT-2007–1–216461 CHRONIOUS:

http://www.chronious.eu

. The authors wish to

acknowledge their gratitude and appreciation to all

the project partners for the development of the ideas

and concepts presented in this paper.

In particular, the authors would like to thank

Prof. M. Vitacca from Fondazione Salvatore

Maugeri, and Prof. Cusi and his team from the

University of Milan. The authors want to thanks all

the partners for their support. This paper was written

as a joint contribution; authors are listed in

alphabetical order and grouped by their affiliations,

besides they have equally contributed to this paper.

REFERENCES

Carroll J. J., et al. Jena: Implementing the semantic web

recommendations. In Proceedings of the 13

Int.l

World Wide Web conference on Alternate track

(WWW Alt.'04). ACM, New York, USA, 74-83

Chiarugi F., Colantonio S., Emmanouilidou D., Martinelli

M., Moroni D., Salvetti O. Decision support in heart

failure through processing of electro- and

echocardiograms. In: Artificial Intelligence in

Medicine, vol. 50 pp. 95 - 104. Elsevier, 2010

Cleland J. G. F., et al. 2005. Noninvasive Home

Telemonitoring for Patients with Heart Failure at High

Risk of Recurrent Admission and Death: The Trans-

European Network-Home-Care Management System

(TEN-HMS) study. Journal of the American College

of Cardiology, Volume 45, Issue 10, 17 May 2005,

Pages 1654-1664

OWL 2009. http://www.w3.org/TR/owl-features/

Paré G., Jaana M., Sicotte C., 2007. Systematic Review of

Home Telemonitoring for Chronic Diseases: The

Evidence Base. J. of the American Medical

Informatics Association, Vol. 14, Issue 3, May-June

2007, Pages 269-277

Pentland, S., 2004. Healthwear: Medical Technology

Becomes Wearable. in IEEE Computer, 37(5). May

2004. 34-41

Rosso R., Munaro G., Salvetti O., Colantonio S., Ciancitto

F., 2010 CHRONIOUS. In: EMBC 2010 pp. 6850 -

6853. IEEE.

Vitacca M., Bianchi L., Guerra A., Fracchia C.,

Spanevello A., Balbi B., Scalvini S., 2009. Tele-

assistance in chronic respiratory failure patients: a

randomised clinical trial. Eur Respir J; 33: 411–418

KNOWLEDGE BASED DECISION SUPPORT FOR THE MANAGEMENT OF CHRONIC PATIENTS

225