Use Case Tutorial - Health Care (1/7)

epts event processing technical society

eptsevent processing technical society

Pedro Bizarro (University of Coimbra)Dieter Gawlick (Oracle)

Health Care Use Case


2

“I am sitting on amountain of data

hidden behindprocedural code”

Dr. KimballUniversity of Utah

Medical Center


Medical Objectives

Decrease morbidity and mortality

Identify situations of concern as they happen

Identify situations of concern before they happen

Alert medical personal of time critical situations

3


IT Objectives

Extract information fromraw data in real-time

- pull and push -

Disseminate time critical information

Highly extensible - meta-data driven

Avoid alert fatigue - customizability

4


5

Overview

Patient DataMedical

PersonnelHistory

VitalsIns/Outs

BloodTests

Radio-logy

OtherInput*

* Observations Notes Questionnaires Diagnosis Treatment ….

Import/Export

Domain Knowledge Vocabularies Classifications Rules Medical Administrative (Predictive) Models Suggestions/What if Visualizations

Deduced PatientInformation

and State

Push Pull Alerts

structured,automatic,real-time

structured,manual,

small delay

structured,manual,big delay

unstructured,manual &automatic,big delay

History - Incidents


7

Domain Knowledge

• Vocabularies – The data structures of health records

• Classifications– The expression used by medical personnel to qualify data; e.g.,

critical, rapidly deteriorating blood pressure

• Rules – A condition (state of the patient) that the medical personnel has to

be made aware of

• Models– Objects that captures a (complex) state of concern. Models will be

derived through data mining and will be supervised and improved– Models will be scored when conditions require to do so

• All elements of the domain knowledge can be shared between institutions and can be customized


8

System Overview – Conceptual View

Event Service- preferable as extension of data base technology -

Data

History

Incidents

Meta-Data (Medical Knowhow)Taxonomies/ClassificationsRegistered Queries (Rules)

(Predictive) Models

RegistriesMedical

Personnel/ExternalServices

…

ToolsUserDev.

Admin

HumanInteraction Monitors

MedicalServices

(external)

Infra-structure (SW/HW)

Applications- Minimal or no procedural code -

Message basedC/S based


9

Steps of Processing

• Capture– Capture and keep all raw data

• Analyze– Applies all rules and data mining models on incoming data

• Identify situation of interest– Capture any match, alert doctor if situation is time critical

– Explain / provide background

• Investigate/Suggest– Provide access to any patient information

– Support investigation with guided resolution

• React– Determine/adjust treatment

– Records who/when/if alerts are dealt with


10

Events, Alerts, Notifications

• There are two event types

– New raw data are entered - this will trigger the evaluation of all rules and may trigger the scoring of models

– An interval/timeout has expired – this will handle the non-events

• Incidents

– If an instance of a pattern/a high enough scoring has been found, capture information in an incident object

– Actionable incidents have to be reacted to in time; otherwise a reminder will be send

• Alerts

– Only if an incident requires immediate attention a notification should be sent to a pager

– High selectivity/customization of alerts should be used to prevent alert fatigue


11

Functional Requirements

• Information storage with easy access– Current state/history of raw data, aggregations, classifications, …– Pull and push with highly selective notification

• Rich type system– SQL, XML(HL7), RDF, ..., DICOM, extensibility

• Support of data mining– (Predictive) pattern detection (e.g. cardiac arrest prediction)– Scoring

• Application development based on declarative constructs only– Rapid deployment, low maintenance cost and extensibility – High flexibility

• Customization - hospital, doctor, patient


12

Operational Characteristics

• Always available

– Recovery/restart/fault tolerance

• Scalability

– Large amount of (historical) data

– Large amount of rules/models

• Security – to control access to medical records

– Fine grain

– Contextual

• Auditing

– History of all data and rules

– Record of all accesses to data

There is at least one lawyer behind each doctor

when things go wrong


13

Acknowledgement/More Information

• Diogo Guerra - prototype to be published as Master Thesis (University of Coimbra)

– Reference to be added when available (est. end of July 2009)

– DEBS 2009 demo

– Diogo worked through the integration of OLTP, temporal, OLAP, data mining and (complex) event processing technology

• Ute Gawlick - SICU research project (University of Utah Medical Center)

– Reference to be added when available (est. end of July 2009)

– Ute provided the medical knowledge, formalized aspects of medical terminology and focused the project on leveraging data mining in conjunction with event processing

Please attend Diogo’s presentation

Please look at Diogo’s demo


A DemoA DemoPreviewPreview