DeIdentification Methodologies and DeIDTM Software

1
De-Identification Methodologies and De-IDTM
Software

• De-ID Data Corp, LLC
• Dan Wasserstrom
• President
• Steven Merahn, MD
• Chief Medical Officer

2
Overview

• Why de-identify?
• What is de-identification?
• Considerations in choosing a de-identification
  methodology
• Automated vs manual methods
• DE-IDTM software overview

3
Why De-Identifiy?

• De-identification is required in order to protect
  patient privacy and advance the progress of
  clinical, quality and population research
• A reliable and valid de-identification
  methodology can increase opportunities expand
  data access and to leverage and manage data
  assets
• De-identification is required by the Privacy Act
  of HIPAA

4
What is De-identification?

• A well-defined, but limited, step in a broader
  research workflow or protocol
• Patient privacy must be a consideration for the
  complete workflow
• The defined nature of the step includes the
  obfuscation of individually identifiable
  information in records and reports
• Obfuscation schema includes redaction,
  elimination, categorical replacement (e.g.,
  place), and replacement with proxies (Dr X), and
  offsets (day 1)

5
Considerations

• When choosing a de-identification methodology,
  four things need consideration
• What is the reliability and validity of the
  methodology?
• Can the method maintain its specificity and
  sensitivity in local use?
• What are the limitations of the methodology?
• Can files be re-identified?

6
Reliability and Validity

• Despite some claims, manual de-identification is
  not the gold standard
• Problems with inter-rater reliability, manpower
  resource and time constraints, over-marking and
  under-marking
• Automated de-identification eliminates the issues
  of time, costs and manpower and offers
  consistency of quality
• The issue then becomes the quality of the quality
  -- over-marketing (specificity) and
  under-marketing (sensitivity)
• What are acceptable levels of sensitivity and
  specificity?
• 100 for sensitivity for names
• What is the benchmark?
• What is the value of consistency?

7
Local Use

• While some methods may have good numbers, will
  they hold up in local use?
• Every community has its own acronyms, place names
  and other local vocabulary
• Can your methodology be customized to meet local
  needs?
• What is the protocol to manage local quality?
• Regular checks against manual review
• Formal evaluation research

8
Limitations

• Based on the use case, different methods may have
  limitations
• Are files in PDF format?
• Are photographs involved?
• What about other images, captions?
• Know the limits of your methodology
• Manual de-identification personnel may need to be
  trained in the handling of radiology images
• Automated tools may only de-identify text reports
• Match the features of your method against your
  use case

9
Re-identification

• There are two types of re-identification
  deliberate and inadvertent
• Deliberate re-identification involves the use of
  trusted third parties to protect
  re-identification codes
• Review of de-identified files may reveal
  eligibility criteria for clinical trial
• Inadvertent re-identification may occur with
  under-marking, or when special case information
  outside of safe harbor guidelines inadvertently
  allow patient ID
• Patient with Wilsons Disease
• Patient privacy issues must be a consideration
  throughout the overall research workflow

10
Automated De-Identification

• An application that provides a level of
  consistency for quality, saves time and lowers
  costs
• A gatekeeping function in a broader research
  infrastructure
• Increase opportunities to leverage the value of
  data assets

11
Features of De-IDTM Software

• Accurate, reliable automated de-identification
• Leaves original record intact
• Very simple to install and use
• Network-level operability batch processing
• Works with any text from reports or databases
• Meets all HIPAA safe harbor guidelines option
  of limited data sets and custom fields
• Continuous quality improvement program
• New version from published paper
• Sensitivity for names 100 overall sensitivity
  99.5 specificity 89

12
DE-ID Mechanics

• Heuristics and rule sets identify the presence of
  any of the HIPAA 18 identifiers within text.
• Supplemental dictionaries ensure of locations,
  names can be idenfied
• UMLS utilized to ensure that words or phrases
  that may be medical terms with proper names are
  preserved.
• De-ID replaces the identifiable text with
  specific tags in the form of offsets and proxies.
  
• De-ID can be used as a standalone application a
  Java interface can automatically call other
  applications for automated batch processing

13
OUTPUT OPTIONS
Word processing
Transcribed records or reports
Manual Feed
XML
DE-IDTM
HP, imaging or lab reports
Tab. Comma, delimited
Java Interface
Database storage
HIPAA Boundary
Database outputs, EMR storage
Outputs de-identified records and reports as
individual files or batch reports
Are automatically de-identified based on HIPAA
safe harbor guidelines or limited data sets
Record and reports containing protected health
information
14
Summary

• De-identification is an critical success factor
  to the success of clinical and population
  research and expanded data access
• In choosing a de-identification methodology
  issues to consider include manpower requirements,
  consistency, quality, time and costs
• Automated de-identification methodologies reduce
  issues of manpower, time and cost
• Quality and consistency benchmarks can be met by
  software