1
University of Minho School of Engineering Algoritmi Centre Uma Escola a Reinventar o Futuro – Semana da Escola de Engenharia - 24 a 27 de Outubro de 2011 Introduction In the future, the decision-making process and the form how the people make decisions should take into account new requirements like pervasiveness, real- time and online processing. INTCare is an on-going research project involving the Intensive Care Unit of the Porto Hospital Centre whose objective is to implement an Intelligent Decision Support System (IDSS) to predict the dysfunction or failure of six organic systems and the patient outcome in order to help doctors deciding on the better treatments or procedures for the patient. The good results obtained so far motivated the transformation of this system into a pervasive IDSS. With this modification will be possible by the ICU staff access to patient information's and previsions remotely, i.e., any time any place. This poster shows the modifications made, the requirements defined and how can it be integrated in the critical health care arena in order to improve the decision process and the Data Mining results obtained. AIM Change the environment, creating a “smart environment” and make it pervasive; Define a new information system architecture that will be able to work in real-time, online and pervasive mode; Make a process dematerialization, making all process automatic and electronic as possible; Develop prevision Data Mining (DM) models. Method Change the environment and information system architecture, using the Pervasive Health Care features and ICU needs, automatize the data acquisition process, define the Data Mining System and develop / test the models / technique with the best acuity. Results After some interactions with ICU staff and some analyses of the environment, we defined the necessities and requirements for the ICU. We had to change several processes and reformulate the information system (IS) architecture. The figure 1 outlines the actual reality of ICU. Author*: CARLOS FILIPE DA SILVA PORTELA Supervisors: Manuel Filipe Santos, Álvaro Moreira Silva * [email protected] A PERVASIVE APPROACH FOR INTELLIGENT DECISION SUPPORT IN CRITICAL HEALTH CARE The figure 2 illustrates the alterations introduced, the doctors only can connect to the hospital through online applications in this case the interface will be the ENR. Figure 1. ICU Current Environment Figure 2. Hospital Pervasive Access In order to complete the KDD process, the system attends some requirements: Online Learning; Autonomous; Real-Time; Adaptability; Safety; Accurate; Data Mining and Decision Models; Optimization; Intelligent agents; Pervasiveness; Accuracy; Privacy; Secure Access from Exterior; User Policy. Now a new information system architecture is been used by ICU, being the most important the data acquisition sub-system (figure 3). After have all data in real-time and in electronic format the DM system was designed (figure 4). Figure 3. Data Acquisition sub-system Figure 4. INTCare DM System For each target was developed a set of 4 scenarios for some DM techniques: Artificial Neural Network (AAN), Support Vector Machine (SVM) Decision Trees, Regression and Ensembles. Table 1 sums the results and presents, for each organic system, the best scenario and the technique with the best results, in terms of sensitivity. The next figure (5) shows how the entire INTCare system works remotely, valuing the DM Models continuing learning and group and mobile decisions. Figure 5. INTCare Pervasive system Conclusions The features defined and evaluated in this work made possible to transform the system in order to be more secure, robust, easily accessible and intelligent. Is expected the system will improve the patient outcome in the future due to some new facilities like the data availability in online, real-time and in electronic format. With the ICU pervasive access recast, is possible to accede to knowledge portions that can support the decision making process, anytime, anywhere. Future Work Define all processing and transforming task for all data collected, making it an autonomous process by the agents that execute the tasks and rules defined for each variable / source, refining the data quality. Improve the Data Mining Model and continuing the test of prevision models with real data collected in real-time and online mode. Improve the data mining scripts for critical events, ratios and medical scores automatic calculation. Acknowledges The author would like to thank FCT (Foundation of Science and Technology, Portugal) for the financial support through the contract PTDC/EIA/72819/2006. This work was supported by the grant SFRH/BD/70156/2010 from FCT. AdversEvents Drugs Prescriptions Lab Results Nursing Record Vital Sign Data (M onitorized) EHR Procedures Markings Doctor/Nurses Autom atic M anual R egister H ospitalServices A llInform ation O nline ENR O nline Real-Tim e Electronic Doctor/Nurses C onsult Validate Store data D octor2 D octor3 C ollaborative Platform Decision m aking Decision m aking Decision m aking Health Decision C onsult PatientD ata D octor1 C onsult PatientD ata C onsult PatientD ata G et PatientD ata G et PatientD ata Actuation in system orcom municate w ith nurses G et PatientD ata Apply Decision Send Alerts N otify N otify ENR INTC are D atabase D rugs System Vital Signs C ritical Events Adverse Events EHR Lab Results ENR D ata To Transform Process D ata SelectD ata Real D ata Outcom e DM Models O rgan Failure DM Models Prepared Data IN TC are Data W arehouse Processed Data IN TC are System Vital Signs Aquisition Agent LAB Results Agent H L7_M SG Pre Processing Agent AID A EN R Agent LAB Results D atabase Server EN R D rugs System BedSide M onitors EH R U C I_EN R Pre Processing Agent UC I_EH R UC I_M ED Gatew ay (vital Signs) H L7_H D R UCI_LR D octor D rug system N urse C ollaborative Platform Single D ecision YES NO Health Decision D octor D octor Ventilation System IM PRO VIN G PATIENT HEALTH CO NDITIO N Apply Decision Actuation in system orC om m unicate w ith nurses O ther System s IM PRO VIN G PR EVISIO N A N D D EC ISIO N M ODELS Decision M odels Consult Health Decision Consult D atabase G et Prevision Models Store D ata INTCare Server Send D ata G et Prevision Models D ata M ining M odels ConsultD ata Patient Decide DM Engine Decision M aking M odel ENR IN TC are Apply Decision Target Scenario Sensibility Technique Cardiovascula r M1 93.4 ANN Respiratory M2 100.0 SVM Renal M4 100.0 SVM Hepatic M4 100.0 SVM Coagulation M2 99.8 SVM Outcome M1 100.0 SVM

Introduction

  • Upload
    hien

  • View
    37

  • Download
    0

Embed Size (px)

DESCRIPTION

A PERVASIVE APPROACH FOR INTELLIGENT DECISION SUPPORT IN CRITICAL HEALTH CARE. Author *: CARLOS FILIPE DA SILVA PORTELA Supervisors: Manuel Filipe Santos, Álvaro Moreira Silva * [email protected]. - PowerPoint PPT Presentation

Citation preview

Page 1: Introduction

University of Minho School of Engineering Algoritmi Centre

Uma Escola a Reinventar o Futuro – Semana da Escola de Engenharia - 24 a 27 de Outubro de 2011

IntroductionIn the future, the decision-making process and the form how the people make decisions should take into account new requirements like pervasiveness, real-time and online processing. INTCare is an on-going research project involving the Intensive Care Unit of the Porto Hospital Centre whose objective is to implement an Intelligent Decision Support System (IDSS) to predict the dysfunction or failure of six organic systems and the patient outcome in order to help doctors deciding on the better treatments or procedures for the patient. The good results obtained so far motivated the transformation of this system into a pervasive IDSS. With this modification will be possible by the ICU staff access to patient information's and previsions remotely, i.e., any time any place.

This poster shows the modifications made, the requirements defined and how can it be integrated in the critical health care arena in order to improve the decision process and the Data Mining results obtained.

AIM Change the environment, creating a “smart environment” and make

it pervasive; Define a new information system architecture that will be able to

work in real-time, online and pervasive mode; Make a process dematerialization, making all process automatic

and electronic as possible; Develop prevision Data Mining (DM) models.

Method Change the environment and information system architecture, using the Pervasive Health Care features and ICU needs, automatize the data acquisition process, define the Data Mining System and develop / test the models / technique with the best acuity.

Results After some interactions with ICU staff and some analyses of the environment, we defined the necessities and requirements for the ICU. We had to change several processes and reformulate the information system (IS) architecture.

The figure 1 outlines the actual reality of ICU. Like the figure shows, the environment has changed and now only some tasks are manual. Be noticed that the data validation kept manual because only the humans can see if the data associated to some patient are correct or not. For the best results, ENR, a web-based touchscreen platform, was been developed and, gives the possibility to monitor, store validate and consult, online, all data about the patient.

Author*: CARLOS FILIPE DA SILVA PORTELA Supervisors: Manuel Filipe Santos, Álvaro Moreira Silva

* [email protected]

A PERVASIVE APPROACH FOR INTELLIGENT DECISION SUPPORT IN CRITICAL HEALTH CARE

The figure 2 illustrates the alterations introduced, the doctors only can connect to the hospital through online applications in this case the interface will be the ENR.

Figure 1. ICU Current Environment Figure 2. Hospital Pervasive Access

In order to complete the KDD process, the system attends some requirements: Online Learning; Autonomous; Real-Time; Adaptability; Safety; Accurate; Data Mining and Decision Models; Optimization; Intelligent agents; Pervasiveness; Accuracy; Privacy; Secure Access from Exterior; User Policy. Now a new information system architecture is been used by ICU, being the most important the data acquisition sub-system (figure 3). After have all data in real-time and in electronic format the DM system was designed (figure 4).

Figure 3. Data Acquisition sub-system Figure 4. INTCare DM System

For each target was developed a set of 4 scenarios for some DM techniques: Artificial Neural Network (AAN), Support Vector Machine (SVM) Decision Trees, Regression and Ensembles. Table 1 sums the results and presents, for each organic system, the best scenario and the technique with the best results, in terms of sensitivity.

Table 1: Best results for each target by scenario and technique

The next figure (5) shows how the entire INTCare system works remotely, valuing the DM Models continuing learning and group and mobile decisions.

Figure 5. INTCare Pervasive system

ConclusionsThe features defined and evaluated in this work made possible to transform the system in order to be more secure, robust, easily accessible and intelligent. Is expected the system will improve the patient outcome in the future due to some new facilities like the data availability in online, real-time and in electronic format. With the ICU pervasive access recast, is possible to accede to knowledge portions that can support the decision making process, anytime, anywhere.

Future WorkDefine all processing and transforming task for all data collected, making it an autonomous process by the agents that execute the tasks and rules defined for each variable / source, refining the data quality. Improve the Data Mining Model and continuing the test of prevision models with real data collected in real-time and online mode. Improve the data mining scripts for critical events, ratios and medical scores automatic calculation.

Acknowledges The author would like to thank FCT (Foundation of Science and Technology, Portugal) for the financial support through the contract PTDC/EIA/72819/2006. This work was supported by the grant SFRH/BD/70156/2010 from FCT.

Advers EventsDrugs PrescriptionsLab Results Nursing RecordVital Sign Data

(Monitorized)EHR ProceduresMarkings

Doctor / Nurses

Automatic Manual

Register

Hospital Services

All Information

Online

ENR

Online Real-Time Electronic

Doctor / Nurses

Consult

Validate

Store data Doctor 2 Doctor 3Collaborative Platform

Decision making

Decision making Decision making

HealthDecision

Consult Patient Data

Doctor 1

Consult Patient Data

Consult Patient Data

Get Patient Data

Get Patient Data

Actuation in system or communicate

with nurses

Get Patient Data

Apply Decision

Send Alerts

Notify

Notify

ENR

INTCare

Database

Drugs System

VitalSigns

Critical Events

Adverse Events

EHR

Lab Results

ENR

Data To Transform

Process DataSelect Data

Real Data

Outcome DM Models

Organ Failure DM Models

Prepared Data

INTCare

DataWarehouse

Processed Data

INTCareSystem

VitalSigns

AquisitionAgent

LAB Results Agent

HL7_MSG

Pre Processing

Agent

AIDA

ENR Agent

LAB Results

Database ServerDatabase Server

ENR

Drugs System

BedSide Monitors

EHR

UCI_ENR

Pre Processing

Agent

UCI_EHR

UCI_MED

Gateway (vital Signs)

HL7_HDR

UCI_LR

Doctor

Drug system

Nurse

Collaborative Platform

Single Decision

YES NO

HealthDecision

Doctor

Doctor

Ventilation System

IMPROVING PATIENT HEALTH CONDITION

ApplyDecision

Actuation in system or Communicate with nurses

OtherSystems

IMPROVING PREVISION AND DECISION MODELS

Decision Models

Consult

Health Decision

Consult

Database

Get Prevision Models

Store Data

INTCare Server

Send Data

Get Prevision

Models

Data Mining Models

Consult Data Patient Decide

DM Engine

Decision Making Model

ENR INTCare

ApplyDecision

Target Scenario Sensibility TechniqueCardiovascular M1 93.4 ANNRespiratory M2 100.0 SVMRenal M4 100.0 SVMHepatic M4 100.0 SVMCoagulation M2 99.8 SVMOutcome M1 100.0 SVM