Data Mining Predictive Models for Pervasive Intelligent Decision

Support in Intensive Care Medicine

Filipe Portela, Filipe Pinto and Manuel Filipe Santos

Centro Algoritmi, Universidade do Minho, Braga, Portugal

Keywords: Data Mining, KDD, Real-time, Pervasive, Intelligent Decision Support System, Intensive Care.

Abstract: The introduction of an Intelligent Decision Support System (IDSS) in a critical area like the Intensive

Medicine is a complex and difficult process. In this area, their professionals don’t have much time to

document the cases, because the patient direct care is always first. With the objective to reduce significantly

the manual records and, enabling, at the same time, the possibility of developing an IDSS which can help in

the decision making process, all data acquisition process and knowledge discovery in database phases were

automated. From the data acquisition to the knowledge discovering, the entire process is autonomous and

executed in real-time. On-line induced data mining models were used to predict organ failure and outcome.

Preliminary results obtained with a limited population of patients showed that this approach can be applied

successfully.

1 INTRODUCTION

The most important feature for the success of an

Intelligent Decision Support System (IDSS) is its

capability to operate autonomously and in real-time,

providing the results in the right moment of the

decision maker need. IDSS operating in critical

environments must overcome the difficulty in

obtaining the data, processing and transforming

them automatically and in real-time.

INTCare is an IDSS for intensive medicine and

uses Data Mining (DM) models to predict organ

failure and outcome (Gago et al., 2006); (Vilas-Boas

et al., 2010).

In order to meet its goals, it has been necessary

to change the environment and the information

system, developing a real-time data acquisition and

processing system (Portela et al., 2012); (Filipe et

al., 2011); (Santos et al., 2009). This system can

automatically receive and process the patient data,

making it available to obtain further knowledge.

Then, it is important to explore the possibility to

obtain, clean, validate and transform the data

automatically, according to the variables in use by

data mining in an adaptive and real way. The

intensive care units are known as a place where the

patient direct care is in the first place delegating

tasks like the health recording to a secondary

importance (Lyerla et al., 2010). In the past, we have

obtained some good results by using offline data,

where it was correctly identified most of the

variables in use in this project. In an attempt to

transpose this idea to a real environment with real

data obtained in a continuous way, it was verified

that none of variables were regularly collected in an

electronic mode.

In a first step, it was implemented a solution to

obtain all the necessary data. Some changes were

performed and an Electronic Nursing Record (ENR)

has been developed to register and validate the

values. Then, it was necessary to develop a data

processing system. To automate the entire process,

in particular the processing and transformation

phases, it was necessary to deploy a set of agents

(Santos et al., 2011); (Wooldridge, 1999) which

perform some predefined tasks. These agents are

responsible to perform all the tasks which in the past

were done manually. For example, they are

responsible by data acquisition of vital signs, the

validation of the collected values, the data mining

preparation and the models induction. After all the

processes have been totally tested and prepared to be

executed automatically without manual efforts, some

data mining models were developed. The main

objective of these DM Models is the use of the

obtained results to understand the viability to

automate the data transformation process, assessing

Portela F., Pinto F. and Santos M..

Data Mining Predictive Models for Pervasive Intelligent Decision Support in Intensive Care Medicine.

DOI: 10.5220/0004141500810088

In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2012), pages 81-88

ISBN: 978-989-8565-31-0

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

the quality of the results. These results were

compared with the earlier obtained results. All the

tasks were carefully implemented and tested using

the real data related to the patients admitted in ICU

of Hospital Santo António, Centro Hospitalar do

Porto in Portugal. This paper will show the

procedures implemented for the data transformation

process, the first results obtained using DM

techniques and finally the analysis of these results.

The work is organized according to the

Knowledge Discovering Phases. After this

introduction, the second chapter gives an outline of

the paper and an overview of the data acquisition

system in use. The next four chapters explain the

overall KDD phase: at first, it is presented a

summary about the first two phases and then all

automation of the transformation process is exposed.

Fifth chapter introduces the data mining models and

the configurations used. The sixth and seventh

chapters are related to the evaluation of the results

obtained by DM and to the conclusions respectively.

In this last chapter, it is also presented the future

work.

2 BACKGROUND

2.1 Offline vs Online

The former results obtained using EURICUS

database (Silva et al., 2008) was the main motivation

to develop this work. In this work the variables used

were collected manually and in an offline mode:

“The data was monitored, collected and registered

manually, every hour, all ICU patient biometrics

were recorded in a standardized sheet form by the

nursing staff. The adverse events were also assigned

in a specific sheet at an hourly basis.”(Silva, et al.,

2008). The variables used were: Age, Critical

Events, Admission Variables, Outcome, and SOFA

(Vincent et al., 1998); (Vincent et al., 1996).

The objective is now to obtain all of those

variables automatically and to induce data mining

models using an online approach in order to predict

the organ failure and outcome in real time. The

greater challenge is the development of some

procedures which use all values obtained by the data

acquisition system instead of using hourly values.

This change allows for a continuous data

monitoring.

2.2 INTcare System

INTCare system is composed by four subsystems:

Data acquisition, knowledge management, Inference

and Interface. Each subsystem is autonomous and

uses intelligent agents to perform tasks. Figure 1

presents the current INTCare System and its agents.

In this work some of these Agents were used (Santos

et al., 2011). For data acquisition were used the

following agents: Gateway, Vital Signs Acquisition

agent

ENR agent, LR agent and AIDA.

Furthermore, it was used a pre-processing agent for

the data validation and transformation. Finally, the

induction of data mining models was assured by the

Data Mining agent. During the development of

INTCare, some features were added according to the

pervasive health care concept (Orwat et al., 2008);

(Varshney, 2007), making the system available

anywhere and anytime (Portela et al., 2011).

Figure 1: INTCare system.

2.3 Knowledge Discovery Process

Knowledge Discovery from Databases (KDD)

process is recognized as a process which can obtain

new knowledge using some data. This process is

composed by five stages: Selection, Pre-Processing,

Transformation, Data Mining and Interpretation

(Fayyad et al., 1996).

Figure 2: ICU knowledge discovery in database process.

Figure 2 shows the KDD process for the ICU

data. The database is populated with data from seven

major sources. The data are selected from the data

warehouse to be processed or transformed,

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

depending of the goal to each variable. After this

task, the data are available to be presented by the

ENR and prepared for the creation of Data Mining

Models. Finally, all models are evaluated and the

obtained knowledge is presented in the INTCare

system.

2.4 Data Mining

The use of DM techniques in the medical arena has

been gaining an increasing interest by researchers

but despite the high expectative its application in

real world settings has been limited (Bellazzi and

Zupan, 2008). In this case, and considering the

targets, it should be treated as a classification

problem (Han and Kamber, 2006). Bearing in mind

this point and the idea of having a pervasive and

real-time IDSS, a set of DM solutions were

explored. However, most of them were discarded

due to the fact that they can’t receive data from the

database, process them and store it again into the

database tables in a continuous and quick process.

After achieving a set of evaluation and experiments,

it was decided to develop prevision models using

Oracle Data Mining system (ODM). ODM

(Concepts, 2005); (Tamayo et al., 2005) was chosen

because it gives the possibility to perform all tasks

using the database in real-time and are from the

same supplier of Hospital database. The techniques

of classification used are some of the most used in

DM (Wu et al., 2008) such as Support Vector

Machine, Decision Trees and Naïve Byes. The data

mining results obtained in the past were achieved

using other techniques and data with a semi-

automatic process and offline method.

3 DATA SELECTION &

PRE-PROCESSING

The two initial phases of KDD process (Portela et

al., 2012) use the data acquisition system presented

in the background section (Figure 1) to obtain the

data. The first phase is concerned with the data

selection from database and is in agreement with

what is necessary for the DM Models:

ICU_HL7 {Vital Signs}

ICU_HL7_T {Vital Signs auto validated (real

values)}

ICU_PARAM {ICU Limits (max, min) values}

ICU_LR {All Lab Results}

ICU_DRUGS {All Patient Drugs

administrated}

ICU_ENR {Data validated and provided

from ENR}

ICU_CEVENTS {ICU Critical Patient Events}

EHR_ADMIN {ICU Patient Admission}

EHR_OUT {ICU Patient Outcome}

The second phase is responsible for the automatic

data validation and patient identification, where it is

ensured that all data collected are valid and are

correctly identified, i.e. all values collected are

between the normal ranges of ICU values, and have

a valid patient identification (PID) (Portela et al.,

2011). At pre-processing phase, other procedures are

executed to prepare the Data Mining input table. For

example, only the first five days values collected are

used. When the patient is admitted to ICU, an agent

is responsible to prepare automatically the table

adding 120 rows for that patient. When patient goes

out, if he/she leaves before 120 hours, the rows in

excess are deleted. This table is used as a temporary

table of DM input table. In this table are stored: case

mix values for all 120 lines and the number of

critical events and SOFA (0, 1) values for each hour.

DM agent gets all variables present in temporary

tables and in addition calculates the values in fault.

4 TRANSFORMATION

The third phase of KDD process it is autonomous

and don’t requires a manual action. All tasks were

performed automatically and in real-time, by the

INTCare intelligent agents. The variables in use are:

SOFA

Cardio, Respiratory, Renal, Liver, Coagulation, neurologic

= {0,

1};

Case Mix = {Age (1-4), Admission type (U or P),

Admission from (1-6)};

Critical Events Accumulated (ACE) = {ACE of

Blood Pressure (|N), ACE of Oxygen Saturation

(|N), ACE of Heart Rate (|N), ACE of Urine Output

(|N)}

Ratios1 (R1) = {ACE of BP/elapsed time of stay

(Q+), ACE of SO2/elapsed time of stay (Q+), ACE

of HR/elapsed time of stay (Q+), ACE of Ur/elapsed

time of stay (Q+), Total of ACE / elapsed time of

stay (Q+)}}.

Ratios2 (R2) = {ACE of BP / max number of ACE

of BP (Q+), ACE of SO2/ max number of ACE of

SO2 (Q+), ACE of HR / max number of ACE of HR

(Q+) , ACE of Ur / max number of ACE of Ur (Q+),

Total of ACE (Q+), Total of ACE / Total ACE max

(Q+)}.

DataMiningPredictiveModelsforPervasiveIntelligentDecisionSupportinIntensiveCareMedicine

Ratios (R) = R1 U R2

ACE of HR: Sum of values hour by hour for each

event type. Example if in first hour has 1 event and

in the second hour 2, the ACE of second hour is 3.

Total of ACE is the sum of all ACE of the hour.

Max Number of ACE is the number max of each

variable present in Table 4.

Elapsed Time of Stay: Number of hours elapsed

since patient admission in the moment of ratio is

calculated.

Total ACE Max is always 33 (10+10+6+7) (Table

4).

Outcome = {0, 1}

Table 2 presents the values considered.

Table 1: Variables transforming (example).

ID Variable Min Max Value

Age

- 18 46 1

- 47 65 2

- 66 75 3

- 76 130 4

Admission Type

Urgent - - u

Programed - - p

Admission From

Chirurgic - - 1

Observation - - 2

Emergency - - 3

Other ICU - - 4

Other Hospital - - 5

Other Situation - - 6 or 7

SOFA

Cardio

BP (mean) 0 70 1

Dopamine 0,01 - 1

Dobutamine 0,01 - 1

Epi / Norepi 0,01 - 1

Renal Creatinine 1.2 - 1

Resp Po2/Fio2 0 400 1

Hepatic Bilirubin 1.2 - 1

Coagul Platelets 0 150 1

Neuro Glasgow 3 14 1

The first transformation process is a simple task

for analysing the values collected and for

transforming them according to some rules (if then

else). This process is applied to the variables

presented in Table 2. When there is a case mix, all

variables are inserted in database. When a patient

comes into the ICU, a procedure is executed.

Regarding to the age parameter, the procedure

verifies the patient age. For the admission type and

origin, the admission form is consulted in the EHR.

In all the cases the values are processed and the

value is inserted into DM_INPUT table.

In the case of the SOFA, the approach is a little

different. The values are collected in real-time and in

a continuous way. The data mining models only use

one value per hour. All collected values are

considered and the final value is assigned. If more

than one result by hour is verified, only the worst

value of the hour is considered. For example, in the

cardiovascular system, there are five different

possibilities (BP, Dopamine, Dobutamine, Epi and

Norepi) and it is sufficient if one of them is 1. The

SOFA values are then transformed in binary

variables, where 0 describes normality and 1

describes dysfunction/failure and comprises the

original SOFA. By default the sofa value variable is

0 and, if some condition is verified (eg. for

coagulation, platelets <=150) the values are update

to 1. This update has effect in the starting date which

the value was measured. The outcome value (live or

died) is updated according to the patient final state.

The value in the table is always 0 (live) except if the

patient dies. In this case all of the values in the input

table are updated to 1.

The second transformation phase uses critical

events. Firstly, a set of procedures are executed in

order to understand if a value is critical and if the

event is adverse.

Table 2: presents the variables in study and the

min and max values for each case.

Table 2: Data ranges.

EvId Descr

Min

Max

Min

Val

Max

Val

Min

Any

Max

Any

1011 BP 90 180 0 300 60

3000 O2 90 100 0 100 80

2009 HR 60 120 0 300 30 180

DIU UR 30 1000 0 1000 10

At first, it is verified if a value is valid, then it is

analysed if that value is critical and how critical is it.

According to the Table 3, a value can be considered

normal (0), critical (1) or too critical (2). In the case

of the value is critical (1), the event will be only

considered critical if the values collected are verified

during some certain time. If the value is spontaneous

and is too bad (2), the event will be always critical

during the time of the event. This process is

presented in the following pseudocode:

IF VALUE >= MIN_VAL AND <= MAX_VAL

THEN

IF VALUE >= MIN_EC AND <= MIN_EC

THEN

SET EVENT_TYPE TO 1

ELSEIF VALUE <= MIN_ANY OR VALUE >=

MAX_ANY

SET EVENT_TYPE TO 2

ELSE SET EVENT_TYPE TO 0

ENDIF

Then each collected value will be inserted in the

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

events table according to the event type and if the

predecessor event is or not the same. To know if this

event type is of same type a flag will be used.

After know the importance of the values, another

procedure starts. This procedure is used to

understand if the critical values collected may or

may not represent a critical event. To this end, Table

3 is used. To consider an event as a critical event it

is necessary to achieve one of the two

characteristics. For example, in the case of SpO2 it

is necessary to have values <= 90 and >=80 for more

than one hour, or <=80 during all the period.

Table 3: The protocol for the out of range physiologic

measurements (adapted by Álvaro (Silva et al., 2008)).

(mmHg)

SpO2

(%)

(bpm)

(ml/h)

Normal range 90 - 180 >= 90 60 - 120

>= 30

Critical event

>= 1h >= 1h >= 1h >= 2h

Critical event

< 60 <80 <30 V> 180 <= 10

a Defined when continuously out of range.

b Defined anytime.

After all values are correctly inserted in patient

events table, a procedure is executed. This procedure

is responsible for reviewing the values according to

the event time and type; if a critical event is verified,

the event will be inserted in critical events table.

Then, and after have the events are stored in

critical events table, an hourly procedure is

executed. This procedure calculates the

Accumulated Critical Events (ACE) – to reflect the

patients’ clinical evolution/severity of illness by

hour. The next step is the obtaining of the ratios.

This process was one of the most difficult to

implement due to the need of a real-time

calculations in the exact moment when the value is

collected. This process requires much memory and

processing time and can delay other procedures. For

the ratios which use the number of maximum critical

events by hour, it is used the maximum number of

occurrences verified in the past (Table 4). The max

values are updated in the future according to the

number max of events verified by a patient for each

variable.

Table 4: Critical Events daily number (adapted from

(Silva, et al., 2008)).

Variables Min Max

Daily number of critical blood pressure events 0 10

Daily number of critical heart rate events 0 10

Daily number of critical oxygen events 0 6

Daily number of critical urine events 0 7

The next procedure is a selection of the code to

calculate the ACE and all ratios to DM model.

During all the processes described above, a

procedure is responsible to get all data generated and

store them in a specific table for the DM task.

Finally, and after having all values correctly inserted

in DM input table, another procedure runs to clean

the bad values. This procedure is responsible to

delete all rows which have null / incorrect values.

5 REAL-TIME DATA MINING

After all data have been processed, some models

were developed and induced. Only in this step,

appears the first manual operation, i.e., the data

mining models must be manually configured. 108

models were developed (6 targets x 6 models x 3

techniques). The data mining models were induced

in two steps: the first is responsible to prepare the

final data to be used by the prediction models; in the

second, the data obtained by the first stage are used

by DM techniques to predict the probability of

failure of each organ and patient outcome.

In the first step, the data stored in the DM input

table is loaded. The numbers (ACE and ratios) are

distributed using the Bin Top 7 method. The other

values are maintained as they are, and a final table is

generated. This table are then used to predict the

targets and six models M1 to M6 are induced for

each target (renal, hepatic, coagulation,

cardiovascular, respiratory and outcome) by each

technique (DT, NB, SVM):

M1 CM + ACE

M2 CM + ACE + R

M3 CM + ACE + R1

M4 CM + ACE + SOFA

M5 CM + ACE + SOFA + R

M6 CM + ACE + SOFA + R2

M7 CM + ACE + SOFA + R1

M7  NOT CONSIDERED (RESULTS = M5)

During the modulation of DM, the neurologic

system was not considered due to the high number

of data in fault.

Figure 3 provides an overview of Data Mining

modulation. In this figure it can be observed that the

data preparation module is executed using the data

stored in database. After the transformation phase,

the data are stored in DM_INPUT_DB table. This

table is then used to predict the value of six target

variables. Afterwards, the obtained results are

applied into the prediction table of the patients

admitted in ICU (UCI_PATIENT_5DAYS_AG).

Then, six new columns are added containing the

DataMiningPredictiveModelsforPervasiveIntelligentDecisionSupportinIntensiveCareMedicine

prediction of happen 1 for each target. In the data

mining engine, each target is an individual process.

Figure 3: ODM model.

In order to also automate this process, some

researche has been done to find how to induce DM

models automatically. As result it was possible to

develop a procedure to execute the DM engine in

real time. The DM agent is responsible to run the

engine whenever a request is made. This procedure

gets the data stored in UCI_PATIENT_5DAYS_AG

table and stores them into a new table

(INTCARE_RESULTS_APPLY) combinded with

the probability to be 1 for each target.

6 EVALUATION

In order to evaluate the models, a test phase using

online learning was performed. The Data Mining

techniques were applied in the following dataset:

Data Description:

Collection Time: 89 days

Patients Number: 109

Data Considered: Values of five first days

Exclusion criterion I: Patient with data collecting intermittent, i.e.,

the collection system failed at least more

than one hour in a continuous way;

Exclusion criterion II: Existence of null values;

Data Input Distribution of the Inputs and

Targets:

NULL RECORDS: 0,00%

AGE = 1:

AGE = 2:

AGE = 3:

AGE = 4:

ADMIN FROM = 1:

ADMIN FROM = 2:

ADMIN FROM = 3:

ADMIN FROM = 4:

ADMIN FROM = 5:

ADMIN FROM = 6:

ADMIN FROM = 7:

ADMIN TYPE = U:

ADMIN TYPE = P:

0,00%

39,64%

22,85%

37,51%

50,45%

0,00%

9,51%

3,79%

2,64%

21,01%

40,06%

59,94%

40,06%

OUTCOME = 1: 12,58%

SOFA COAG = 1: 9,22%

SOFA HEPATIC = 1: 9,85%

SOFA RENAL = 1: 5,16%

SOFA RESPIRATORY = 1: 3,45%

SOFA CARDIO = 1: 27,37%

After the DM engine has been run, the best results

obtained for each target are presented in Table 5.

Table 5: Results by organ systems and outcome (%).

Target Technique Scenario Sensibility Accuracy

Cardio SVM M6 67.0 87.3

Respirat SVM M4 99.8 91.5

Renal DT M5 96.4 99.3

Liver SVM M2 100 100

Coag SVM M6 97.5 92.6

Out SVM M4 100 99.5

Table 6: Variation of the obtained results.

Target

SENS

OLD

ACC

OLD

SENS

VAR

ACC

VAR

Cardio

93.4 92.7 - 26.4 -5.4

Respirat

96.2 98.1 3.6 -6.6

Renal

98.1 96.9 -1.7 2.4

Liver

98.3 98.1 1.7 1.9

Coag

97.5 98.2 0 -5.6

Out

98.3 97.4 1.7 2.1

Comparing the results (Table 6) obtained

previously (Filipe Portela et al., 2010) and those

observed in this study (table 5) it is possible to

conclude that some targets (respiratory, liver,

coagulation and outcome) present better results. In

the opposite direction the cardiovascular target

worsened. The models will be used according to the

results obtained in terms of sensibility. In a general

way, the results are widely satisfactory. Due to the

limited dataset (89 patients), we believe that it is

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

possible to improve the results in the future. Duo to

the frequency of admissions in the ICU, a more

comprehensive study may require at least one year

of continuous data acquisition.

7 CONCLUSIONS

The process of development of an IDSS which can

be autonomous and able to operate in real-time is a

big challenge. In this work, the most difficult task

has been the transformation of the data process, and

the process to obtain critical events and their ratios.

The results proved that it is possible to

implement an IDSS in critical health environments

minimizing the human intervention.

The greatest contribution of this work is the

automation of the entire KDD process, in particular,

the data transformation and the data mining

processes. INTCare become an autonomous system

and is able to, in a automatically and in real-time

way, predict the organ failure and outcome for the

next 24 hours for the patients admitted in ICU. The

DM engine operates autonomously. This engine runs

the models and makes the probability results

available through the INTCare system.

Experimental work has been conducted in order

to compare the sensibility and accuracy values

attained using online DM models with those

obtained in offline.

In the work has been considered data from a

limited number of patients. The obtained results

were very good and, in general, similar to the values

obtained in the past, using a semi-autonomous

process. The doctors can now access to patient data

anywhere and anytime, through the electronic

nursing record, and they can consult the probability

of organ failure and the outcome in an intuitive,

quick and easy way. A questionnaire has been

prepared in order to understand the level of

acceptance of the system. The answers provided by

the doctors and by the nurses were very satisfactory

and motivating.

Further work will include more data to optimize

the DM models. Complementarily, it will be

considered the development of an ensemble.

ACKNOWLEDGEMENTS

The authors would like to thank FCT (Foundation of

Science and Technology, Portugal) for the financial

support through the contract PTDC/EIA/72819/

2006. The work of Filipe Portela was supported by

the grant SFRH/BD/70156/2010 from FCT.

REFERENCES

Bellazzi, R., & Zupan, B., (2008). Predictive data mining

in clinical medicine: current issues and guidelines.

International journal of medical informatics, 77(2),

81-97.

Concepts, O. D. M., (2005). 11g Release 1 (11.1). Oracle

Corp, 2007.

Fayyad, U. M., Piatetsky-Shapiro, G., & Smyth, P. (1996).

From data mining to knowledge discovery: an

overview.

Gago, P., Santos, M. F., Silva, Á., Cortez, P., Neves, J., &

Gomes, L., (2006). INTCare: a knowledge discovery

based intelligent decision support system for intensive

care medicine. Journal of Decision Systems.

Han, J., & Kamber, M., (2006). Data mining: concepts

and techniques: Morgan Kaufmann.

Lyerla, F., LeRouge, C., Cooke, D. A., Turpin, D., &

Wilson, L., (2010). A Nursing Clinical Decision

Support System and potential predictors of Head-of-

Bed position for patients receiving Mechanical

Ventilation. American Journal of Critical Care, 19(1),

39-47.

Orwat, C., Graefe, A., & Faulwasser, T., (2008). Towards

pervasive computing in health care - A literature

review. [10.1186/1472-6947-8-26]. BMC Medical

Informatics and Decision Making, 8(1), 26.

Portela, F., Santos, M., Vilas-Boas, M., Rua, F., Silva, Á.,

& Neves, J., (2010). Real-time Intelligent decision

support in intensive medicine. Paper presented at the

KMIS 2010- International Conference on Knowledge

Management and Information Sharing.

Portela, F., Santos, M. F., Gago, P., Silva, Á., Rua, F.,

Abelha, A., et al., (2011). Enabling real-time

intelligent decision support in intensive care. Paper

presented at the 25th European Simulation and

Modelling Conference- ESM'2011, Guimarães,

Portugal.

Portela, F., Santos, M. F., Silva, Á., Machado, J., &

Abelha, A., (2011). Enabling a Pervasive approach

for Intelligent Decision Support in Critical Health

Care. Paper presented at the HCist 2011 –

International Workshop on Health and Social Care

Information Systems and Technologies.

Portela, F., Santos, M. F., & Vilas-Boas, M., (2012). A

Pervasive Approach to a Real-Time Intelligent

Decision Support System in Intensive Medicine. In

Springer (Ed.), Communications in Computer and

Information Science (Vol. 0272, pp. 14).

Santos, M. F., Portela, F., Vilas-Boas, M., Machado, J.,

Abelha, A., & Neves, J., (2011). INTCARE - Multi-

agent approach for real-time Intelligent Decision

Support in Intensive Medicine. Paper presented at the

3rd International Conference on Agents and Artificial

Intelligence (ICAART), Rome, Italy.

DataMiningPredictiveModelsforPervasiveIntelligentDecisionSupportinIntensiveCareMedicine

Santos, M. F., Portela, F., Vilas-Boas, M., Machado, J.,

Abelha, A., Neves, J., et al., (2009). Information

Modeling for Real-Time Decision Support in Intensive

Medicine. In S. Y. Chen & Q. Li (Eds.), Proceedings

of the 8th Wseas International Conference on Applied

Computer and Applied Computational Science -

Applied Computer and Applied Computational Science

(pp. 360-365). Athens: World Scientific and

Engineering Acad and Soc.

Silva, Á., Cortez, P., Santos, M. F., Gomes, L., & Neves,

J., (2008). Rating organ failure via adverse events

using data mining in the intensive care unit. Artificial

Intelligence in Medicine, 43(3), 179-193.

Tamayo, P., Berger, C., Campos, M., Yarmus, J.,

Milenova, B., Mozes, A., et al., (2005). Oracle Data

Mining. Data Mining and Knowledge Discovery

Handbook, 1315-1329.

Varshney, U., (2007). Pervasive healthcare and wireless

health monitoring. Mobile Networks and Applications,

12(2-3), 113-127.

Vilas-Boas, M., Santos, M. F., Portela, F., Silva, Á., &

Rua, F., (2010). Hourly prediction of organ failure

and outcome in intensive care based on data mining

techniques. Paper presented at the 12th International

Conference on Enterprise Information Systems.

Vincent, J., Mendonça, A., Cantraine, F., Moreno, R.,

Takala, J., Suter, P., et al., (1998). Use of the SOFA

score to assess the incidence of organ

dysfunction/failure in intensive care units : Results of

a multicenter, prospective study. Critical care

medicine, 26, 1793-1800.

Vincent, J. L., Moreno, R., Takala, J., Willatts, S., De

Mendonca, A., Bruining, H., et al., (1996). The SOFA

(Sepsis-related Organ Failure Assessment) score to

describe organ dysfunction/failure. Intensive care

medicine, 22(7), 707-710.

Wooldridge, M., (1999). Intelligent agents Multiagent

systems: a modern approach to distributed artificial

intelligence (pp. 27-77): MIT Press.

Wu, X., Kumar, V., Ross Quinlan, J., Ghosh, J., Yang, Q.,

Motoda, H., et al., (2008). Top 10 algorithms in data

mining. Knowledge and Information Systems, 14(1), 1-

37.

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing