PSIP Workshop

1

• DebugIT Building a European distributed clinical
  data mining network to foster the fight against
  microbial diseases

• Christian Lovis, Teodoro Douglas, Emilie
  Pasche,Patrick Ruch, Dirk Colaert, Karl
  Stroetmann

• Presented by Karl Stroetmann

• PSIP Workshop
• Belgirate, Italy, 24-25 September 2009

2
C o n t e n t s

1. The project
2. Conceptual framework technology
3. Challenges
4. Clinical socio-economic impact assessment
5. Outlook

3
The DebugIT project
4
Funding and time schedule

• EU funded IP (integrated project)
• FP7 (Framework Programme 7) - a research
  initiative of the European Union
• The DebugIT project proposal was ranked first
• Start date Jan 1st, 2008
• End date December 31st, 2011
• 11 Partners
• 11 Work packages
• Total EU funding of the project 7m

5
The Partners

• 1 Agfa Agfa HealthCare N.V., Belgium
• 2 HUG Les Hôpitaux universitaires de Genève,
  Switzerland
• 3 UNIGE Université De Genève, CH
• 4 LIU LINKÖPINGS UNIVERSITET, Sweden
• 5 EMP empirica, Bonn, Germany
• 6 UCL University College London, UK
• 7 INSERM Institut National de la Santé et de la
  Recherche Médicale, Paris, France
• 8 UKLFR Universitätsklinikum Freiburg, Germany
• 9 TEILAM TECHNOLOGIKO EKPEDEFTIKO IDRIMA LAMIAS,
  Greece
• 10 IZIP IZIP A.S., Prague, Czech Republic
• 11 GAMA Gama/Sofia Ltd., Sofia, Bulgaria

6
Overview

• DebugIT Detecting and Eliminating Bacteria UsinG
  Information Technology
• Dedicated to infectious diseases
• Aims
• detecting patient safety related patterns and
  trends
• acquiring new knowledge
• using this for better quality healthcare
• Consortium of eleven partners across the EU
• Strong clinical lead assured by
• Clinical Advisory Board (President Prof. Dr.
  Didier Pittet, HUG, Geneva World Alliance for
  Patient Safety, World Health Organisation)
• Scientific Advisory Board

7
Objectives

• Built an advanced tool aiming at infectious
  pathogens across health systems and levels
• Integrate it into clinical information systems of
  participating European hospitals
• Develop generic conceptual base that can be
  easily expanded to other similar medical fields
• Make the tool publicly available

8
Why infectious diseases ?

• Advanced ICT for Risk Assessment and Patient
  Safety project
• -gt main focus on advanced ICT
• Risk assessment and patient safety on a 4 years
  project
• -gt a coherent choice infectious diseases
• usually short life cycles
• measurable results
• data available on the whole range of semantic and
  technical complexity
• lab results, order entry, structured text, free
  text, images
• hot topic for public health and clinical research
• can provide decision support for research,
  clinicians and governance

9
Clinical context

• Antibiotic resistance is a consequence of
  evolution via natural selection
• Antibiotic action is urgently needed to respond
  to environmental pressure
• Patterns of antibiotic usage greatly affect the
  number of resistant organisms which develop
• Overuse of broad-spectrum antibiotics
• Incorrect diagnosis
• Unnecessary prescriptions
• Improper use of antibiotics
• Use of antibiotics as livestock food additives
  for growth promotion
• Counterfeit drugs

10
Clinical context
antibiotic resistance in Salmonella typhimurium
DT104, England and Wales, 1984-1995
WHO Weekly Epidemiological Record, Vol 71, No
18, 1996
11
Main focus for Y2 Closing the Loop As Soon As
Possible

• ? Interoperability platform (WP1)
• ? Data Normalization (WP2)
• Data Analysis (WP3)
• Knowledge extraction (WP3/WP4)
• Knowledge authoring (WP4)
• Inference tools (WP5)
• Clinical decision-support (WP6)

12
Conceptual framework technology
13
Iterative Cycle

• collect routinely stored data from clinical
  systems
• learn by applying advanced data mining techniques
• store the extracted knowledge in repositories
• apply knowledge for decision support and
  monitoring

14
Iterative Cycle
15
Collect clinical data repository

• Routinely stored clinical data is collected and
  aggregated across
• hospitals
• countries
• languages
• information models
• legislations
• via
• commonly agreed data models (minimal data sets)
• standards
• mapping algorithms
• unified and enhanced ontologies

Collect
16
Learn multimodal data mining

• detect relevant patterns
• advanced data mining techniques on multimodal
  multi-source data
• structured data mining
• text mining
• image mining
• create new knowledge using advanced multimodal
  knowledge-driven data mining

Collect
Learn
17
Store medical knowledge repository

• knowledge is
• stored in a distributed repository
• validated by clinicians
• visualised and aggregated together with
  pre-existing medical and biological knowledge
  (guidelines, regulations)
• a consolidated organization in the knowledge
  repository

Collect
Learn
Store
18
Apply decision support tools

• software tools integrated in clinical and public
  health information systems
• decision support tools
• apply generated knowledge
• help clinicians to provide clinical care
• example choice, dose and administration of
  antibiotics
• predict future outcomes
• monitoring tools for
• research
• epidemiology
• health policy

Collect
Learn
Apply
Store
19
Translational and evidence based medicine

• DebugIT is a nice example of translational
  medicine and evidence based medicine
• clinical care uses knowledge and evidences from
  research(bench to bed)
• research uses real life clinical data (bed to
  bench)
• access to huge amounts of real-world data is a
  welcome addition to expensive traditional
  clinical studies

Collect
Apply
Learn
Store
20
Activities and progress HL7-RIM based common
schema
21
DebugIT CDR architecture
22
Data integration via database federation
SQL endpoint

• First implementation using low performance
  machine
• Many problems with performance
• Constant use of disk temporary tables, indexes
  problems (losing key because disk was full)
• Change to a better server with 8 GB of memory, 4
  processors, SCSI drivers
• Query speed has improved significantly
• Complex queries between 2 centres executed in 1
  min

Ready
Almost ready
Good progress
23
Activities and progress

• Knowledge authoring tool
• Generation assistant

The user writes some parameters
Different methods of generation
List of recommandations
24
Activities and progress

• Knowledge authoring tool
• Validation assistant

The user writes a rule
Different methods of validation
Trend-based validation
Text-based validation
25
SQL endpoint multiple site visualization
Demonstration of CDR query distributed between
LiU and HUG
Yearly resistance of Ecoli to TMP/SMX
26
Challenges
27
Interoperability

• Language independent formal vocabulary as input
  for data analysis data mining
• Formal semantics and textual descriptions to
  precisely describe abstracted meanings
• Extraction of heterogeneous structured and
  unstructured EPR content
• Semantic standard for project-wide information
  Clinical Data Repository Formalism

28
Data mining

• Data aggregation from heterogeneous sources
• Management of data quality and reliability
• Integration and mining of multimodal data,
  including images
• Knowledge-driven data mining
• Advanced data mining, (bio)statistics, signal
  theory, lexical analysis and ontological analysis
• Multi-axial mining, temporal, multimodal, case
  and cohort base

29
Knowledge and inference

• Federated knowledge repository
• heterogeneous sources, variable level of
  certainty
• representation of knowledge and rules
• Reasoning
• statistical logical
• performance
• formalism and decidability
• reliability for case based decision support

30
Example of data mining challenge
31
Impact assessment
32
Impact assessment framework

• Project evaluation
• impact on scientific community
• impact on EC initiatives
• ...

• Outcome assessment
• cost benefit analysis
• clinical impact (DSS)
• technology

What to measure why
How to measure
What to measure why
How to measure
Measurements
Data collection methods
Indicators
Measurements
Indicators
Data collection methos
33
Project evaluation

• Impact on scientific community
• Value of individual project outputs
• Transferability to other research areas
• Type of scientific progress achieved
• Impact on research capacity
• Impact on efficiency of future research
• Impact on scientific technological objectives
  of EC initiatives with regards to
• Macroeconomic development
• Private Sector (Industry SMEs)
• Research initiatives
• Health Sector / eHealth

34
Outcome assessment

• Clinical and socio-economic impact assessment
  based on benefit cost analysis
• Identification of positive (benefits) negative
  (costs) impacts to all relevant stakeholders
• Quantification in terms of monetary units in
  order to derive total net benefit for society
• Development of individual scenarios based on life
  cycle approach, time horizons, and diffusion
  speed
• Capturing uncertainty/risk and prospective nature
  of analysis

35
Risk uncertainty

• Range of probable outcomes

• 99

• 90

• Mean

36
Outlook
37
Summary

• Focus on large existing, heterogeneous clinical
  data repositories
• Building an interoperability platform that is
  usable for the whole infectious domain
• Creation of a federated clinical data repository
  that enables knowledge-driven data mining
• Leverage of patient data with existing knowledge
  and merger into a clinical knowledge repository
• Exploitation of newly generated knowledge with a
  clinical decision support system to loop back to
  clinical practice
• Serious advance in building a large IT
  infrastructure creating knowledge in the fight
  against infectious diseases
• Reusable for other diseases and contexts

38
Acknowledgement and disclaimer

• DebugIT is a project co-funded by the European
  Commissions Seventh FRAMEWORK PROGRAMME.
• The research reported upon in this presentation
  has either directly or indirectly been supported
  by the European Commission, Directorate General
  Information Society and Media, Brussels.
• The results, analyses and conclusions derived
  there from reflect solely the views of its
  authors and of the presenter.
• The European Community is not liable for any use
  that may be made of the information contained
  therein.

39
Thank you for your attention
More info ? http//www.debugit.eu