caTIES Processing Text to Share Data and Tissue

1
caTIES Processing Text to Share Data and Tissue

• Rebecca Crowley, Kevin Mitchell, and Wendy
  Chapman
• Centers for Oncology and Pathology Informatics
• Center for Biomedical Informatics
• crowleyrs_at_upmc.edu
• June 4, 2006

2
caTIES project goals

• Support research (clinical, translational, basic)
  that uses human tissue or tissue related data by
  providing de-identified, concept-based access to
  surgical pathology report data associated with
  tissue slides and blocks.
• caTIES v2.0 now in place at four institutions
  University of Pittsburgh, University of
  Pennsylvania, Thomas Jefferson University,
  University of Washington St. Louis

3
Potential Uses

• FINAL DIAGNOSIS
• PART 1 LYMPH NODES, LEFT PELVIC,
  LYMPHADENECTOMY
• SIX BENIGN LYMPH NODES WITH NO
  EVIDENCE OF MALIGNANCY.
• PART 2 LYMPH NODES, RIGHT PELVIC,
  LYMPHADENECTOMY
• TWO BENIGN LYMPH NODES WITH NO
  EVIDENCE OF MALIGNANCY.
• PART 3 PROSTATE AND SEMINAL VESICLES,
  RADICAL PROSTATECTOMY
• INVASIVE MODERATELY DIFFERENTIATED PROSTATIC
  ADENOCARCINOMA WITH A COMBINED GLEASON SCORE OF
  336.
• PATHOLOGIC STAGE T2bN0MX
•  

• Case finding for paraffin materials
• Automated annotation for caTISSUE CAE
• Text-mining
• Automation of clinical trials eligibility
  screening
• Cancer Surveillance

4
The tissue bank is.

• Small (hundreds to thousands of specimens)
• Annotated with human abstraction manual data
  entry
• New (less than 10 years old)
• Focused
• Mainly Frozen Tissue
• Storage QA/QC
• RNA studies

5
The paraffin (clinical) archive is

• Large (Millions of Specimens)
• Directly Documented by Pathology
• Old (Decades)
• Ubiquitous General
• DNA and Protein Studies
• TMA
• Rare and Discontinued Tumors
• Tissue Across Generations
• Fairly Intact

6
Problems to solve

• Reduce unnecessary barriers and unnecessary work
• Make entire corpus de-identified and searchable
• Improve retrieval
• Many ways to say the same thing (synonymy)
• Pertinent negatives (negation)
• Discourse inferences
• Information Extraction
• Temporality and aggregation
• Create network of institutions sharing data
  building collaborations
• Interoperable with other systems

7
Long Term Project Goals

• Develop software with value to individual cancer
  centers as well as caBIG matrixed researchers
• Use caTIES as a platform for research - improve
  our ability to extract meaningful information
  from free-text pathology reports

8
Current caTIES functionality

• 1. Secure communication between participating
  organizations coordinated by the Open Grid
  Service Architecture (OGSA) Services for data
  transfer and de-identification
• 2. Concept-coding of surgical pathology reports
  using a GATE-based NLP pipeline
• 3. Databases for private and sharable information
  accessible via OGSA-Data Access and Integration
  services. OGSA-DAI is an extension of OGSA.
• 4. A graphical user interface for concept-based
  querying across the network, role-based
  administration of access, and protocol-based
  ordering of tissue samples
• 5. A set of semantic annotations and services
  based on caCORE and caGRID which foster semantic
  interoperability between caTIES and other caBIG
  applications.

9
Problems to solve

• Reduce unnecessary barriers and unnecessary work
• Make entire corpus de-identified and searchable
• Improve retrieval
• Many ways to say the same thing (synonymy)
• Pertinent negatives (negation)
• Discourse inferences
• Information Extraction
• Temporality and aggregation
• Create network of institutions sharing data
  building collaborations
• Interoperable with other systems

10
Before HIPAA
11
After HIPAA
12
What we hope will happen with caTIES
13
Problems to solve

• Reduce unnecessary barriers and unnecessary work
• Make entire corpus de-identified and searchable
• Improve retrieval
• Many ways to say the same thing (synonymy)
• Pertinent negatives (negation)
• Temporality and aggregation
• Create network of institutions sharing data
  building collaborations
• Interoperable with other systems

14
(No Transcript)
15
Text Acquisition

• Loads the Identified caTIES MySQL datastore
• Minimize manual processes
• Text acquisition service for bulk loads
• Chunker for text reports
• HL7 acquisition service at Penn
• Cerner/CoPath caBIG Data Extractor allows
  scheduled batch input from Cerner CoPath
• Would like to get other LIS Adapters

16
De-Identification Service

• caTIES De-Identification service scrubs pathology
  report, creates de-identified identifiers, loads
  De-Identified caTIES datastore
• caTIES de-identification service wraps the de-ID
  software (Saul, Cooper, etc)
• Safe-Harbor method removes HIPAA mandated
  identifiers
• Creates tokens for names and preserves temporal
  relationships
• De-ID works with adopters as each site comes
  on-line

17
De-Identification Service
18
Preserving identifying relationship

• De-ID service creates linkage through key pair on
  identified side of the system
• Honest Broker view of data behind the firewall
  allows only HB to see identifiers in order to
  fill requests for tissue

19
Problems to solve

• Reduce unnecessary barriers and unnecessary work
• Make entire corpus de-identified and searchable
• Improve retrieval
• Many ways to say the same thing (synonymy)
• Pertinent negatives (negation)
• Temporality and aggregation
• Create network of institutions sharing data
  building collaborations
• Interoperable with other systems

20
Pipeline Service
21
Current Processing Resources

• (1)   Resetter clears document, deletes
  existing annotations,
• (2)   Tokeniser - tokenizes words, numbers,
  punctuation and spaces,
• (3)   Chunker - parses reports into sections,
  parts, sentences, and phrases,
• (4)   Spell-checker - identifies erroneous
  spelling and suggests frequency based correction
• (5)   RegEx - annotates a pre-defined set of
  attribute and value pairs such as tumor grade and
  stage,
• (6)   Vocabulary Concept Tagger - annotates
  fragments of free text to associated concept from
  controlled terminology
• (7)   Semantic-type Filter - removes concepts
  associated with unwanted semantic types,
• (8)   NegEx - implements NegEx negation detection
  algorithm to tag explicitly negated concepts,
• (9) Semantic-type Categorization - extracts
  organs, procedures, diseases infers topology of
  concepts as modifiers of other concepts

22
Problems to solve

• Reduce unnecessary barriers and unnecessary work
• Make entire corpus de-identified and searchable
• Improve retrieval
• Many ways to say the same thing (synonymy)
• Pertinent negatives (negation)
• Temporality and aggregation
• Create network of institutions sharing data
  building collaborations
• Interoperable with other systems

23
Technology

• Decentralized data management and centralized
  order management
• Grid services
• Security

24
Decentralized Data Centralized Ordering

• Although all report-derived data are
  decentralized, we need a centralized way to
  manage orders

25
Grid services for sharing data

• Globus toolkit
• We are in some ways purposely creating two
  systems, one of which is isolated from the caGrid
• Need for Virtual Organizations not everyone is
  going to want to share their data with everyone
  else!

26
Security

• Physical and network
• Application-level
• Grid-level
• Processes and Policies

27
Processes and Policies

• HIPAA
• Institutional Review Boards
• Security Compliance
• Trust agreements
• Honest Brokers
• Materials transfer agreements for tissue
• caTIES is testbed for new Security Policy

28
Problems to solve

• Reduce unnecessary barriers and unnecessary work
• Make entire corpus de-identified and searchable
• Improve retrieval
• Many ways to say the same thing (synonymy)
• Pertinent negatives (negation)
• Temporality and aggregation
• Create network of institutions sharing data
  building collaborations
• Interoperable with other systems

29
Definition

• Semantic interoperability is the ability of
  information systems to exchange information on
  the basis of shared, pre-established and
  negotiated meanings of concepts, such that the
  input received can be machine processed

30
Acknowledgements

• Kevin Mitchell
• Girish Chavan
• Adi Nemlekar,
• Linda Schmandt
• Eugene Tseytlin
• Dean Brown, Rick Nestler, Zeke Holland (coPath) -
  Acquisition
• Melissa Saul, Paul Hanbury, Greg Cooper DeID
• Steve Merahn DeID
• Wendy Chapman NegEx
• Mike Becich, John Gilbertson, Jim Harrison
• University of Pennsylvania (Mike Feldman, Dave
  Fenstermacher, Tara McSherry)
• Washington University and Thomas Jefferson
  University
• GATE, UMLS Lexical Tools, NCI Enterprise
  Vocabulary Services
• Funding from the Shared Pathology Informatics
  Network (NCI) and caBIG (NCI)